From c2ce9e53ed31306d95e6ae08c44a588684a5de06 Mon Sep 17 00:00:00 2001
From: Yifan Zhao <yifanz16@illinois.edu>
Date: Wed, 22 Jan 2020 14:02:25 -0600
Subject: [PATCH] Revert "Merge branch 'hpvm-devisc' into 'hpvm-reorg-9'"
This reverts merge request !9
---
.gitignore | 2 +-
hpvm/include/BuildDFG/BuildDFG.h | 10 +-
.../{GenHPVM/GenHPVM.h => GenVISC/GenVISC.h} | 22 +-
hpvm/include/SupportHPVM/HPVMTimer.h | 151 -
.../{SupportHPVM => SupportVISC}/DFG2LLVM.h | 118 +-
.../DFGTreeTraversal.h | 0
.../{SupportHPVM => SupportVISC}/DFGraph.h | 94 +-
.../HPVMHint.h => SupportVISC/VISCHint.h} | 10 +-
hpvm/include/SupportVISC/VISCTimer.h | 151 +
.../HPVMUtils.h => SupportVISC/VISCUtils.h} | 148 +-
hpvm/lib/Transforms/BuildDFG/BuildDFG.cpp | 68 +-
hpvm/lib/Transforms/CMakeLists.txt | 2 +-
hpvm/lib/Transforms/ClearDFG/ClearDFG.cpp | 10 +-
.../DFG2LLVM_NVPTX/DFG2LLVM_NVPTX.cpp | 3139 ++++++++---------
.../Transforms/DFG2LLVM_X86/DFG2LLVM_X86.cpp | 378 +-
hpvm/lib/Transforms/GenHPVM/GenHPVM.cpp | 894 -----
.../{GenHPVM => GenVISC}/CMakeLists.txt | 4 +-
hpvm/lib/Transforms/GenVISC/GenVISC.cpp | 866 +++++
.../GenVISC.exports} | 0
.../{GenHPVM => GenVISC}/LLVMBuild.txt | 4 +-
hpvm/lib/Transforms/LocalMem/LocalMem.cpp | 22 +-
hpvm/llvm_installer/llvm_installer.sh | 8 +-
hpvm/llvm_patches/apply_patch.sh | 2 +-
hpvm/llvm_patches/include/IR/Attributes.td | 2 +-
hpvm/llvm_patches/include/IR/Intrinsics.td | 2 +-
.../llvm_patches/include/IR/IntrinsicsHPVM.td | 208 --
.../llvm_patches/include/IR/IntrinsicsVISC.td | 208 ++
hpvm/llvm_patches/lib/AsmParser/LLLexer.cpp | 2 +-
hpvm/llvm_patches/lib/AsmParser/LLParser.cpp | 6 +-
hpvm/llvm_patches/lib/AsmParser/LLToken.h | 2 +-
.../lib/Bitcode/Reader/BitcodeReader.cpp | 2 +-
.../lib/Bitcode/Writer/BitcodeWriter.cpp | 2 +-
hpvm/llvm_patches/lib/IR/Attributes.cpp | 2 +-
hpvm/projects/hpvm-rt/CMakeLists.txt | 22 -
hpvm/projects/visc-rt/CMakeLists.txt | 22 +
.../deviceStatusSwitchIntervals.txt | 0
.../{hpvm-rt => visc-rt}/device_abstraction.h | 2 +-
hpvm/projects/{hpvm-rt => visc-rt}/makefile | 2 +-
hpvm/projects/{hpvm-rt => visc-rt}/policy.h | 0
.../hpvm-rt.cpp => visc-rt/visc-rt.cpp} | 402 +--
.../{hpvm-rt/hpvm-rt.h => visc-rt/visc-rt.h} | 100 +-
hpvm/test/CTestSuite/Makefile | 8 +-
hpvm/test/CTestSuite/RUN.script | 2 +-
hpvm/test/CTestSuite/gemm.c | 12 +-
hpvm/test/CTestSuite/gemm_2.c | 12 +-
hpvm/test/hpvm-cava/.gitignore | 2 +-
hpvm/test/hpvm-cava/Makefile | 38 +-
hpvm/test/hpvm-cava/Makefile.config.example | 18 +-
hpvm/test/hpvm-cava/README.md | 2 +-
hpvm/test/hpvm-cava/src/main.c | 1452 ++++----
hpvm/test/hpvm-cava/src/pipe_stages.c | 340 +-
hpvm/test/include/hpvm.h | 73 -
hpvm/test/include/visc.h | 73 +
hpvm/test/parboil/RUN.parboil.script | 2 +-
hpvm/test/parboil/benchmarks/bfs/Makefile | 4 +-
.../src/opencl_cpu_baseline/kernel-spir64.ll | 2 +-
.../bfs/src/opencl_cpu_baseline/main.cpp | 4 +-
.../benchmarks/bfs/src/opencl_nvidia/main.cpp | 10 +-
.../bfs/src/{hpvm => visc}/Makefile | 4 +-
.../bfs/src/{hpvm => visc}/config.h | 0
.../bfs/src/{hpvm => visc}/main.cpp | 226 +-
hpvm/test/parboil/benchmarks/cutcp/Makefile | 4 +-
.../src/opencl_cpu_baseline/cutoff6overlap.c | 2 +-
.../{kernel_hpvm.cl => kernel_visc.cl} | 0
...{kernel_hpvm_x64.ll => kernel_visc_x64.ll} | 2 +-
...nel_hpvm_x64.spir => kernel_visc_x64.spir} | Bin
.../src/opencl_cpu_baseline/kernel_x64.ll | 2 +-
.../cutcp/src/opencl_nvidia/cutoff6overlap.c | 2 +-
.../cutcp/src/{hpvm => visc}/Makefile | 4 +-
.../cutcp/src/{hpvm => visc}/atom.h | 0
.../cutcp/src/{hpvm => visc}/cutcpu.c | 0
.../cutcp/src/{hpvm => visc}/cutoff.h | 0
.../cutcp/src/{hpvm => visc}/cutoff6overlap.c | 0
.../cutcp/src/{hpvm => visc}/excl.c | 0
.../cutcp/src/{hpvm => visc}/kernel.cl | 0
.../cutcp/src/{hpvm => visc}/macros.h | 0
.../cutcp/src/{hpvm => visc}/main.cpp | 210 +-
.../benchmarks/cutcp/src/{hpvm => visc}/ocl.c | 0
.../benchmarks/cutcp/src/{hpvm => visc}/ocl.h | 0
.../cutcp/src/{hpvm => visc}/output.c | 0
.../cutcp/src/{hpvm => visc}/output.h | 0
.../cutcp/src/{hpvm => visc}/readatom.c | 0
hpvm/test/parboil/benchmarks/lbm/Makefile | 4 +-
.../lbm/src/opencl_cpu_baseline/main.c | 4 +-
.../benchmarks/lbm/src/opencl_cpu_long/main.c | 10 +-
.../lbm/src/opencl_cpu_short/main.c | 10 +-
.../benchmarks/lbm/src/opencl_nvidia/main.c | 2 +-
.../lbm/src/opencl_nvidia_long/main.c | 10 +-
.../lbm/src/opencl_nvidia_short/main.c | 10 +-
.../lbm/src/{hpvm => visc}/Makefile | 4 +-
.../lbm/src/{hpvm => visc}/layout_config.h | 0
.../benchmarks/lbm/src/{hpvm => visc}/lbm.cpp | 0
.../benchmarks/lbm/src/{hpvm => visc}/lbm.h | 0
.../lbm/src/{hpvm => visc}/lbm_macros.h | 0
.../lbm/src/{hpvm => visc}/main.cpp | 88 +-
.../benchmarks/lbm/src/{hpvm => visc}/main.h | 0
hpvm/test/parboil/benchmarks/sgemm/Makefile | 4 +-
.../sgemm/src/hpvm_vec_opt/Makefile | 8 -
.../benchmarks/sgemm/src/opencl_base/main.cc | 4 +-
.../sgemm/src/opencl_base_opt/main.cc | 4 +-
.../sgemm/src/opencl_base_vec/main.cc | 4 +-
.../benchmarks/sgemm/src/opencl_cpu/main.cc | 4 +-
.../sgemm/src/opencl_cpu_4K/main.cc | 6 +-
.../sgemm/src/opencl_cpu_baseline/main.cc | 6 +-
.../sgemm/src/opencl_cpu_medium/main.cc | 6 +-
.../sgemm/src/opencl_cpu_sm/kernel-spir64.ll | 2 +-
.../sgemm/src/opencl_cpu_sm/main.cc | 2 +-
.../sgemm/src/opencl_cpu_sm/test-spir64.ll | 2 +-
.../sgemm/src/opencl_nvidia/main.cc | 2 +-
.../benchmarks/sgemm/src/opencl_opt_8/main.cc | 4 +-
.../sgemm/src/opencl_opt_8_4K/main.cc | 6 +-
.../sgemm/src/opencl_opt_8_medium/main.cc | 6 +-
.../sgemm/src/opencl_opt_8_vec/main.cc | 4 +-
.../sgemm/src/{hpvm => visc}/Makefile | 4 +-
.../benchmarks/sgemm/src/{hpvm => visc}/io.cc | 0
.../sgemm/src/{hpvm => visc}/kernel.cl | 0
.../sgemm/src/{hpvm => visc}/main.cc | 150 +-
.../sgemm/src/{hpvm_tc => visc_opt}/Makefile | 4 +-
.../sgemm/src/{hpvm_opt => visc_opt}/io.cc | 0
.../src/{hpvm_opt => visc_opt}/kernel.cl | 0
.../sgemm/src/{hpvm_opt => visc_opt}/main.cc | 30 +-
.../sgemm/src/{hpvm_sh => visc_sh}/Makefile | 4 +-
.../sgemm/src/{hpvm_sh => visc_sh}/io.cc | 0
.../sgemm/src/{hpvm_sh => visc_sh}/main.cc | 170 +-
.../sgemm/src/{hpvm_vec => visc_tc}/Makefile | 4 +-
.../sgemm/src/{hpvm_tc => visc_tc}/io.cc | 0
.../sgemm/src/{hpvm_tc => visc_tc}/main.cc | 28 +-
.../src/{hpvm_tc_vec => visc_tc_vec}/Makefile | 4 +-
.../src/{hpvm_tc_vec => visc_tc_vec}/io.cc | 0
.../src/{hpvm_tc_vec => visc_tc_vec}/main.cc | 28 +-
.../sgemm/src/visc_tc_vec/main.visc.ll | 894 +++++
.../sgemm/src/{hpvm_opt => visc_vec}/Makefile | 4 +-
.../sgemm/src/{hpvm_vec => visc_vec}/io.cc | 0
.../src/{hpvm_vec => visc_vec}/kernel.cl | 0
.../sgemm/src/{hpvm_vec => visc_vec}/main.cc | 30 +-
.../sgemm/src/visc_vec/main.visc.ll | 869 +++++
.../sgemm/src/visc_vec_opt/Makefile | 8 +
.../src/{hpvm_vec_opt => visc_vec_opt}/io.cc | 0
.../{hpvm_vec_opt => visc_vec_opt}/kernel.cl | 0
.../{hpvm_vec_opt => visc_vec_opt}/main.cc | 32 +-
.../sgemm/src/visc_vec_opt/main.visc.ll | 889 +++++
hpvm/test/parboil/benchmarks/spmv/Makefile | 4 +-
.../benchmarks/spmv/src/opencl_cpu/main.c | 10 +-
.../spmv/src/opencl_cpu_baseline/main.c | 2 +-
.../spmv/src/opencl_cpu_huge/main.c | 10 +-
.../spmv/src/opencl_cpu_large/main.c | 10 +-
.../benchmarks/spmv/src/opencl_nvidia/main.c | 2 +-
.../spmv/src/opencl_nvidia_huge/main.c | 10 +-
.../spmv/src/opencl_nvidia_large/main.c | 10 +-
.../spmv/src/{hpvm => visc}/Makefile | 4 +-
.../spmv/src/{hpvm => visc}/file.cpp | 0
.../benchmarks/spmv/src/{hpvm => visc}/file.h | 0
.../spmv/src/{hpvm => visc}/gpu_info.cpp | 0
.../spmv/src/{hpvm => visc}/gpu_info.h | 0
.../spmv/src/{hpvm => visc}/kernel.cl | 0
.../spmv/src/{hpvm => visc}/main.cpp | 172 +-
.../spmv/src/visc/main.visc.ll.kernels.bc | Bin 0 -> 2464 bytes
.../spmv/src/visc/main.visc.ll.kernels.ll | 138 +
hpvm/test/parboil/benchmarks/stencil/Makefile | 4 +-
.../benchmarks/stencil/src/opencl_base/main.c | 2 +-
.../stencil/src/opencl_base_default/main.c | 10 +-
.../stencil/src/opencl_base_large/main.c | 10 +-
.../stencil/src/opencl_base_strided/main.c | 10 +-
.../stencil/src/opencl_base_vec/main.c | 10 +-
.../benchmarks/stencil/src/opencl_cpu/main.c | 10 +-
.../stencil/src/opencl_cpu_baseline/main.c | 6 +-
.../src/opencl_cpu_default/kernel-spir64.ll | 2 +-
.../stencil/src/opencl_cpu_default/main.c | 12 +-
.../src/opencl_cpu_large/kernel-spir64.ll | 2 +-
.../stencil/src/opencl_cpu_large/main.c | 12 +-
.../stencil/src/{hpvm => visc}/Makefile | 4 +-
.../stencil/src/{hpvm => visc}/common.h | 0
.../stencil/src/{hpvm => visc}/file.cc | 0
.../stencil/src/{hpvm => visc}/file.h | 0
.../stencil/src/{hpvm => visc}/kernel.cl | 0
.../stencil/src/{hpvm => visc}/stencil.cpp | 140 +-
.../src/{hpvm_vec => visc_vec}/common.h | 0
.../src/{hpvm_vec => visc_vec}/stencil.c | 26 +-
.../stencil/src/visc_vec/stencil.visc.ll | 673 ++++
hpvm/test/parboil/benchmarks/tpacf/Makefile | 4 +-
.../benchmarks/tpacf/src/opencl_base/main.cc | 2 +-
.../tpacf/src/opencl_cpu_base/main.cc | 2 +-
.../tpacf/src/{hpvm => visc}/Makefile | 4 +-
.../tpacf/src/{hpvm => visc}/args.cc | 0
.../tpacf/src/{hpvm => visc}/args.h | 0
.../tpacf/src/{hpvm => visc}/main.cc | 128 +-
.../tpacf/src/{hpvm => visc}/model.cc | 0
.../tpacf/src/{hpvm => visc}/model.h | 0
hpvm/test/parboil/common/include/parboil.h | 32 +-
.../parboil/common/mk/{hpvm.mk => visc.mk} | 58 +-
.../{hpvm.default.mk => visc.default.mk} | 18 +-
hpvm/test/parboil/driver/options.py | 40 +-
hpvm/test/parboil/parboilParser.py | 16 +-
hpvm/test/pipeline/Makefile | 34 +-
hpvm/test/pipeline/copyToVersions.sh | 6 +-
...adient.visc.merged.experiments.notimer.ll} | 200 +-
...lacian.visc.merged.experiments.notimer.ll} | 204 +-
hpvm/test/pipeline/run.sh | 2 +-
hpvm/test/pipeline/runscript.sh | 18 +-
hpvm/test/pipeline/src/Makefile | 4 +-
hpvm/test/pipeline/src/main.cc | 452 +--
hpvm/test/unitTests/CreateNodeAndEdge.c | 38 +-
hpvm/test/unitTests/Makefile | 4 +-
hpvm/test/unitTests/MallocIntrinsic.c | 18 +-
hpvm/test/unitTests/PipelineIntrinsics.c | 16 +-
.../unitTests/PipelineIntrinsics.malloc.c | 16 +-
hpvm/test/unitTests/temp/3level.ll | 50 +-
hpvm/test/unitTests/temp/Makefile | 4 +-
hpvm/test/unitTests/temp/query2D.ll | 52 +-
hpvm/test/unitTests/temp/query3D.ll | 56 +-
hpvm/test/unitTests/temp/queryNodeInst.ll | 44 +-
hpvm/test/unitTests/temp/queryNumDim.ll | 46 +-
hpvm/test/unitTests/temp/queryNumNodeInst.ll | 50 +-
hpvm/test/unitTests/temp/singleNode.ll | 30 +-
hpvm/test/unitTests/temp/singleNodeStream.ll | 56 +-
hpvm/test/unitTests/temp/twoLaunch.ll | 38 +-
hpvm/test/unitTests/temp/twoNode.ll | 34 +-
hpvm/test/unitTests/temp/twoNodeConnect.ll | 36 +-
hpvm/test/unitTests/temp/twoNodeQuery.ll | 46 +-
hpvm/test/unitTests/temp/twoNodeStream.ll | 50 +-
220 files changed, 9761 insertions(+), 6419 deletions(-)
rename hpvm/include/{GenHPVM/GenHPVM.h => GenVISC/GenVISC.h} (67%)
delete mode 100644 hpvm/include/SupportHPVM/HPVMTimer.h
rename hpvm/include/{SupportHPVM => SupportVISC}/DFG2LLVM.h (82%)
rename hpvm/include/{SupportHPVM => SupportVISC}/DFGTreeTraversal.h (100%)
rename hpvm/include/{SupportHPVM => SupportVISC}/DFGraph.h (94%)
rename hpvm/include/{SupportHPVM/HPVMHint.h => SupportVISC/VISCHint.h} (78%)
create mode 100644 hpvm/include/SupportVISC/VISCTimer.h
rename hpvm/include/{SupportHPVM/HPVMUtils.h => SupportVISC/VISCUtils.h} (84%)
delete mode 100644 hpvm/lib/Transforms/GenHPVM/GenHPVM.cpp
rename hpvm/lib/Transforms/{GenHPVM => GenVISC}/CMakeLists.txt (74%)
create mode 100644 hpvm/lib/Transforms/GenVISC/GenVISC.cpp
rename hpvm/lib/Transforms/{GenHPVM/GenHPVM.exports => GenVISC/GenVISC.exports} (100%)
rename hpvm/lib/Transforms/{GenHPVM => GenVISC}/LLVMBuild.txt (88%)
delete mode 100644 hpvm/llvm_patches/include/IR/IntrinsicsHPVM.td
create mode 100644 hpvm/llvm_patches/include/IR/IntrinsicsVISC.td
delete mode 100644 hpvm/projects/hpvm-rt/CMakeLists.txt
create mode 100644 hpvm/projects/visc-rt/CMakeLists.txt
rename hpvm/projects/{hpvm-rt => visc-rt}/deviceStatusSwitchIntervals.txt (100%)
rename hpvm/projects/{hpvm-rt => visc-rt}/device_abstraction.h (96%)
rename hpvm/projects/{hpvm-rt => visc-rt}/makefile (97%)
rename hpvm/projects/{hpvm-rt => visc-rt}/policy.h (100%)
rename hpvm/projects/{hpvm-rt/hpvm-rt.cpp => visc-rt/visc-rt.cpp} (82%)
rename hpvm/projects/{hpvm-rt/hpvm-rt.h => visc-rt/visc-rt.h} (72%)
delete mode 100644 hpvm/test/include/hpvm.h
create mode 100644 hpvm/test/include/visc.h
rename hpvm/test/parboil/benchmarks/bfs/src/{hpvm => visc}/Makefile (81%)
rename hpvm/test/parboil/benchmarks/bfs/src/{hpvm => visc}/config.h (100%)
rename hpvm/test/parboil/benchmarks/bfs/src/{hpvm => visc}/main.cpp (70%)
rename hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/{kernel_hpvm.cl => kernel_visc.cl} (100%)
rename hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/{kernel_hpvm_x64.ll => kernel_visc_x64.ll} (99%)
rename hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/{kernel_hpvm_x64.spir => kernel_visc_x64.spir} (100%)
rename hpvm/test/parboil/benchmarks/cutcp/src/{hpvm => visc}/Makefile (85%)
rename hpvm/test/parboil/benchmarks/cutcp/src/{hpvm => visc}/atom.h (100%)
rename hpvm/test/parboil/benchmarks/cutcp/src/{hpvm => visc}/cutcpu.c (100%)
rename hpvm/test/parboil/benchmarks/cutcp/src/{hpvm => visc}/cutoff.h (100%)
rename hpvm/test/parboil/benchmarks/cutcp/src/{hpvm => visc}/cutoff6overlap.c (100%)
rename hpvm/test/parboil/benchmarks/cutcp/src/{hpvm => visc}/excl.c (100%)
rename hpvm/test/parboil/benchmarks/cutcp/src/{hpvm => visc}/kernel.cl (100%)
rename hpvm/test/parboil/benchmarks/cutcp/src/{hpvm => visc}/macros.h (100%)
rename hpvm/test/parboil/benchmarks/cutcp/src/{hpvm => visc}/main.cpp (82%)
rename hpvm/test/parboil/benchmarks/cutcp/src/{hpvm => visc}/ocl.c (100%)
rename hpvm/test/parboil/benchmarks/cutcp/src/{hpvm => visc}/ocl.h (100%)
rename hpvm/test/parboil/benchmarks/cutcp/src/{hpvm => visc}/output.c (100%)
rename hpvm/test/parboil/benchmarks/cutcp/src/{hpvm => visc}/output.h (100%)
rename hpvm/test/parboil/benchmarks/cutcp/src/{hpvm => visc}/readatom.c (100%)
rename hpvm/test/parboil/benchmarks/lbm/src/{hpvm => visc}/Makefile (85%)
rename hpvm/test/parboil/benchmarks/lbm/src/{hpvm => visc}/layout_config.h (100%)
rename hpvm/test/parboil/benchmarks/lbm/src/{hpvm => visc}/lbm.cpp (100%)
rename hpvm/test/parboil/benchmarks/lbm/src/{hpvm => visc}/lbm.h (100%)
rename hpvm/test/parboil/benchmarks/lbm/src/{hpvm => visc}/lbm_macros.h (100%)
rename hpvm/test/parboil/benchmarks/lbm/src/{hpvm => visc}/main.cpp (86%)
rename hpvm/test/parboil/benchmarks/lbm/src/{hpvm => visc}/main.h (100%)
delete mode 100644 hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec_opt/Makefile
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm => visc}/Makefile (83%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm => visc}/io.cc (100%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm => visc}/kernel.cl (100%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm => visc}/main.cc (69%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_tc => visc_opt}/Makefile (83%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_opt => visc_opt}/io.cc (100%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_opt => visc_opt}/kernel.cl (100%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_opt => visc_opt}/main.cc (90%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_sh => visc_sh}/Makefile (86%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_sh => visc_sh}/io.cc (100%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_sh => visc_sh}/main.cc (65%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_vec => visc_tc}/Makefile (83%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_tc => visc_tc}/io.cc (100%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_tc => visc_tc}/main.cc (90%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_tc_vec => visc_tc_vec}/Makefile (83%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_tc_vec => visc_tc_vec}/io.cc (100%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_tc_vec => visc_tc_vec}/main.cc (90%)
create mode 100644 hpvm/test/parboil/benchmarks/sgemm/src/visc_tc_vec/main.visc.ll
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_opt => visc_vec}/Makefile (83%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_vec => visc_vec}/io.cc (100%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_vec => visc_vec}/kernel.cl (100%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_vec => visc_vec}/main.cc (90%)
create mode 100644 hpvm/test/parboil/benchmarks/sgemm/src/visc_vec/main.visc.ll
create mode 100644 hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/Makefile
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_vec_opt => visc_vec_opt}/io.cc (100%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_vec_opt => visc_vec_opt}/kernel.cl (100%)
rename hpvm/test/parboil/benchmarks/sgemm/src/{hpvm_vec_opt => visc_vec_opt}/main.cc (91%)
create mode 100644 hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/main.visc.ll
rename hpvm/test/parboil/benchmarks/spmv/src/{hpvm => visc}/Makefile (88%)
rename hpvm/test/parboil/benchmarks/spmv/src/{hpvm => visc}/file.cpp (100%)
rename hpvm/test/parboil/benchmarks/spmv/src/{hpvm => visc}/file.h (100%)
rename hpvm/test/parboil/benchmarks/spmv/src/{hpvm => visc}/gpu_info.cpp (100%)
rename hpvm/test/parboil/benchmarks/spmv/src/{hpvm => visc}/gpu_info.h (100%)
rename hpvm/test/parboil/benchmarks/spmv/src/{hpvm => visc}/kernel.cl (100%)
rename hpvm/test/parboil/benchmarks/spmv/src/{hpvm => visc}/main.cpp (68%)
create mode 100644 hpvm/test/parboil/benchmarks/spmv/src/visc/main.visc.ll.kernels.bc
create mode 100644 hpvm/test/parboil/benchmarks/spmv/src/visc/main.visc.ll.kernels.ll
rename hpvm/test/parboil/benchmarks/stencil/src/{hpvm => visc}/Makefile (80%)
rename hpvm/test/parboil/benchmarks/stencil/src/{hpvm => visc}/common.h (100%)
rename hpvm/test/parboil/benchmarks/stencil/src/{hpvm => visc}/file.cc (100%)
rename hpvm/test/parboil/benchmarks/stencil/src/{hpvm => visc}/file.h (100%)
rename hpvm/test/parboil/benchmarks/stencil/src/{hpvm => visc}/kernel.cl (100%)
rename hpvm/test/parboil/benchmarks/stencil/src/{hpvm => visc}/stencil.cpp (66%)
rename hpvm/test/parboil/benchmarks/stencil/src/{hpvm_vec => visc_vec}/common.h (100%)
rename hpvm/test/parboil/benchmarks/stencil/src/{hpvm_vec => visc_vec}/stencil.c (90%)
create mode 100644 hpvm/test/parboil/benchmarks/stencil/src/visc_vec/stencil.visc.ll
rename hpvm/test/parboil/benchmarks/tpacf/src/{hpvm => visc}/Makefile (82%)
rename hpvm/test/parboil/benchmarks/tpacf/src/{hpvm => visc}/args.cc (100%)
rename hpvm/test/parboil/benchmarks/tpacf/src/{hpvm => visc}/args.h (100%)
rename hpvm/test/parboil/benchmarks/tpacf/src/{hpvm => visc}/main.cc (76%)
rename hpvm/test/parboil/benchmarks/tpacf/src/{hpvm => visc}/model.cc (100%)
rename hpvm/test/parboil/benchmarks/tpacf/src/{hpvm => visc}/model.h (100%)
rename hpvm/test/parboil/common/mk/{hpvm.mk => visc.mk} (81%)
rename hpvm/test/parboil/common/platform/{hpvm.default.mk => visc.default.mk} (61%)
rename hpvm/test/pipeline/{gradient.hpvm.merged.experiments.notimer.ll => gradient.visc.merged.experiments.notimer.ll} (95%)
rename hpvm/test/pipeline/{laplacian.hpvm.merged.experiments.notimer.ll => laplacian.visc.merged.experiments.notimer.ll} (95%)
diff --git a/.gitignore b/.gitignore
index 0da6a36714..a17e2716a5 100644
--- a/.gitignore
+++ b/.gitignore
@@ -34,5 +34,5 @@ hpvm/install/
hpvm/llvm/
hpvm/llvm-*.src.tar.xz
hpvm/llvm-*.src/
-hpvm/projects/hpvm-rt/hpvm-rt.ll
+hpvm/projects/visc-rt/visc-rt.ll
hpvm/test/**/build/
diff --git a/hpvm/include/BuildDFG/BuildDFG.h b/hpvm/include/BuildDFG/BuildDFG.h
index ca4c616da5..28230e135b 100644
--- a/hpvm/include/BuildDFG/BuildDFG.h
+++ b/hpvm/include/BuildDFG/BuildDFG.h
@@ -10,7 +10,7 @@
//
//===----------------------------------------------------------------------===//
-#include "SupportHPVM/DFGraph.h"
+#include "SupportVISC/DFGraph.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
@@ -58,10 +58,10 @@ public:
// Functions
virtual bool runOnModule(Module &M);
- static bool isHPVMLaunchIntrinsic(Instruction *I);
- static bool isHPVMGraphIntrinsic(Instruction *I);
- static bool isHPVMQueryIntrinsic(Instruction *I);
- static bool isHPVMIntrinsic(Instruction *I);
+ static bool isViscLaunchIntrinsic(Instruction *I);
+ static bool isViscGraphIntrinsic(Instruction *I);
+ static bool isViscQueryIntrinsic(Instruction *I);
+ static bool isViscIntrinsic(Instruction *I);
static bool isTypeCongruent(Type *L, Type *R);
// TODO: Maybe make these fields const
diff --git a/hpvm/include/GenHPVM/GenHPVM.h b/hpvm/include/GenVISC/GenVISC.h
similarity index 67%
rename from hpvm/include/GenHPVM/GenHPVM.h
rename to hpvm/include/GenVISC/GenVISC.h
index 24798bc274..1db9929be7 100644
--- a/hpvm/include/GenHPVM/GenHPVM.h
+++ b/hpvm/include/GenVISC/GenVISC.h
@@ -1,4 +1,4 @@
-//== GenHPVM.h - Header file for "LLVM IR to HPVM IR Pass" =//
+//== GenVISC.h - Header file for "LLVM IR to VISC IR Pass" =//
//
// The LLVM Compiler Infrastructure
//
@@ -7,7 +7,7 @@
//
//===----------------------------------------------------------------------===//
-#include "SupportHPVM/HPVMTimer.h"
+#include "SupportVISC/VISCTimer.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/Instructions.h"
@@ -18,24 +18,24 @@
using namespace llvm;
-namespace genhpvm {
-// GenHPVM - The first implementation.
-struct GenHPVM : public ModulePass {
+namespace genvisc {
+// GenVISC - The first implementation.
+struct GenVISC : public ModulePass {
static char ID; // Pass identification, replacement for typeid
- GenHPVM() : ModulePass(ID) {}
+ GenVISC() : ModulePass(ID) {}
private:
// Member variables
Module *M;
- FunctionCallee llvm_hpvm_initializeTimerSet;
- FunctionCallee llvm_hpvm_switchToTimer;
- FunctionCallee llvm_hpvm_printTimerSet;
+ FunctionCallee llvm_visc_initializeTimerSet;
+ FunctionCallee llvm_visc_switchToTimer;
+ FunctionCallee llvm_visc_printTimerSet;
GlobalVariable *TimerSet;
// Functions
void initializeTimerSet(Instruction *);
- void switchToTimer(enum hpvm_TimerID, Instruction *);
+ void switchToTimer(enum visc_TimerID, Instruction *);
void printTimerSet(Instruction *);
Value *getStringPointer(const Twine &S, Instruction *InsertBefore,
const Twine &Name = "");
@@ -45,4 +45,4 @@ public:
virtual bool runOnModule(Module &M);
};
-} // namespace genhpvm
+} // namespace genvisc
diff --git a/hpvm/include/SupportHPVM/HPVMTimer.h b/hpvm/include/SupportHPVM/HPVMTimer.h
deleted file mode 100644
index 05b24d41d6..0000000000
--- a/hpvm/include/SupportHPVM/HPVMTimer.h
+++ /dev/null
@@ -1,151 +0,0 @@
-//===------------ HPVMTimer.h - Header file for "HPVM Timer API" ----------===//
-//
-// The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef HPVM_TIMER_HEADER
-#define HPVM_TIMER_HEADER
-
-/************************** Timer Routines ***************************/
-extern "C" {
-
-/* A time or duration. */
-//#if _POSIX_VERSION >= 200112L
-typedef unsigned long long hpvm_Timestamp; /* time in microseconds */
-//#else
-//# error "Timestamps not implemented"
-//#endif
-
-enum hpvm_TimerState {
- hpvm_Timer_STOPPED,
- hpvm_Timer_RUNNING,
-};
-
-struct hpvm_Timer {
- enum hpvm_TimerState state;
- hpvm_Timestamp elapsed; /* Amount of time elapsed so far */
- hpvm_Timestamp init; /* Beginning of the current time interval,
- * if state is RUNNING. End of the last
- * recorded time interfal otherwise. */
-};
-
-/* Reset a timer.
- * Use this to initialize a timer or to clear
- * its elapsed time. The reset timer is stopped.
- */
-void hpvm_ResetTimer(struct hpvm_Timer *timer);
-
-/* Start a timer. The timer is set to RUNNING mode and
- * time elapsed while the timer is running is added to
- * the timer.
- * The timer should not already be running.
- */
-void hpvm_StartTimer(struct hpvm_Timer *timer);
-
-/* Stop a timer.
- * This stops adding elapsed time to the timer.
- * The timer should not already be stopped.
- */
-void hpvm_StopTimer(struct hpvm_Timer *timer);
-
-/* Get the elapsed time in seconds. */
-double hpvm_GetElapsedTime(struct hpvm_Timer *timer);
-
-/* Execution time is assigned to one of these categories. */
-enum hpvm_TimerID {
- hpvm_TimerID_NONE = 0,
- hpvm_TimerID_IO, /* Time spent in input/output */
- hpvm_TimerID_KERNEL, /* Time spent computing on the device,
- * recorded asynchronously */
- hpvm_TimerID_COPY, /* Time spent synchronously moving data
- * to/from device and allocating/freeing
- * memory on the device */
- hpvm_TimerID_DRIVER, /* Time spent in the host interacting with the
- * driver, primarily for recording the time
- * spent queueing asynchronous operations */
- hpvm_TimerID_COPY_ASYNC, /* Time spent in asynchronous transfers */
- hpvm_TimerID_COMPUTE, /* Time for all program execution other
- * than parsing command line arguments,
- * I/O, kernel, and copy */
- hpvm_TimerID_OVERLAP, /* Time double-counted in asynchronous and
- * host activity: automatically filled in,
- * not intended for direct usage */
- // GPU FUNCTION
- hpvm_TimerID_INIT_CTX,
- hpvm_TimerID_CLEAR_CTX,
- hpvm_TimerID_COPY_SCALAR,
- hpvm_TimerID_COPY_PTR,
- hpvm_TimerID_MEM_FREE,
- hpvm_TimerID_READ_OUTPUT,
- hpvm_TimerID_SETUP,
- hpvm_TimerID_MEM_TRACK,
- hpvm_TimerID_MEM_UNTRACK,
- hpvm_TimerID_MISC,
- // LAUNCH FUNCTION
- hpvm_TimerID_PTHREAD_CREATE,
- hpvm_TimerID_ARG_PACK,
- hpvm_TimerID_ARG_UNPACK,
- hpvm_TimerID_COMPUTATION,
- hpvm_TimerID_OUTPUT_PACK,
- hpvm_TimerID_OUTPUT_UNPACK,
-
- hpvm_TimerID_LAST /* Number of timer IDs */
-};
-
-/* Dynamic list of asynchronously tracked times between events */
-struct hpvm_async_time_marker_list {
- char *label; // actually just a pointer to a string
- enum hpvm_TimerID timerID; /* The ID to which the interval beginning
- * with this marker should be attributed */
- void *marker;
- // cudaEvent_t marker; /* The driver event for this marker */
- struct hpvm_async_time_marker_list *next;
-};
-
-struct hpvm_SubTimer {
- char *label;
- struct hpvm_Timer timer;
- struct hpvm_SubTimer *next;
-};
-
-struct hpvm_SubTimerList {
- struct hpvm_SubTimer *current;
- struct hpvm_SubTimer *subtimer_list;
-};
-
-/* A set of timers for recording execution times. */
-struct hpvm_TimerSet {
- enum hpvm_TimerID current;
- struct hpvm_async_time_marker_list *async_markers;
- hpvm_Timestamp async_begin;
- hpvm_Timestamp wall_begin;
- struct hpvm_Timer timers[hpvm_TimerID_LAST];
- struct hpvm_SubTimerList *sub_timer_list[hpvm_TimerID_LAST];
-};
-
-/* Reset all timers in the set. */
-void hpvm_InitializeTimerSet(struct hpvm_TimerSet *timers);
-
-void hpvm_AddSubTimer(struct hpvm_TimerSet *timers, char *label,
- enum hpvm_TimerID hpvm_Category);
-
-/* Select which timer the next interval of time should be accounted
- * to. The selected timer is started and other timers are stopped.
- * Using hpvm_TimerID_NONE stops all timers. */
-inline void hpvm_SwitchToTimer(struct hpvm_TimerSet *timers,
- enum hpvm_TimerID timer);
-
-void hpvm_SwitchToSubTimer(struct hpvm_TimerSet *timers, char *label,
- enum hpvm_TimerID category);
-
-/* Print timer values to standard output. */
-void hpvm_PrintTimerSet(struct hpvm_TimerSet *timers);
-
-/* Release timer resources */
-void hpvm_DestroyTimerSet(struct hpvm_TimerSet *timers);
-}
-#endif // HPVM_RT_HEADER
diff --git a/hpvm/include/SupportHPVM/DFG2LLVM.h b/hpvm/include/SupportVISC/DFG2LLVM.h
similarity index 82%
rename from hpvm/include/SupportHPVM/DFG2LLVM.h
rename to hpvm/include/SupportVISC/DFG2LLVM.h
index 07147c6d90..b9e4cc4158 100644
--- a/hpvm/include/SupportHPVM/DFG2LLVM.h
+++ b/hpvm/include/SupportVISC/DFG2LLVM.h
@@ -1,7 +1,7 @@
#ifndef __DFG2LLVM_H__
#define __DFG2LLVM_H__
-//===---- DFG2LLVM.h - Header file for "HPVM Dataflow Graph to Target" ----===//
+//===---- DFG2LLVM.h - Header file for "VISC Dataflow Graph to Target" ----===//
//
// The LLVM Compiler Infrastructure
//
@@ -11,9 +11,9 @@
//===----------------------------------------------------------------------===//
#include "BuildDFG/BuildDFG.h"
-#include "SupportHPVM/HPVMHint.h"
-#include "SupportHPVM/HPVMTimer.h"
-#include "SupportHPVM/HPVMUtils.h"
+#include "SupportVISC/VISCHint.h"
+#include "SupportVISC/VISCTimer.h"
+#include "SupportVISC/VISCUtils.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
@@ -26,7 +26,7 @@ using namespace builddfg;
#define TIMER(X) \
do { \
- if (HPVMTimer) { \
+ if (VISCTimer) { \
X; \
} \
} while (0)
@@ -37,8 +37,8 @@ using namespace builddfg;
namespace dfg2llvm {
// Helper Functions
-static inline ConstantInt *getTimerID(Module &, enum hpvm_TimerID);
-static inline ConstantInt *getTimerID(Module &, enum hpvm::Target);
+static inline ConstantInt *getTimerID(Module &, enum visc_TimerID);
+static inline ConstantInt *getTimerID(Module &, enum visc::Target);
bool hasAttribute(Function *, unsigned, Attribute::AttrKind);
@@ -69,7 +69,7 @@ protected:
// Member variables
Module &M;
BuildDFG &DFG;
- bool HPVMTimer = false;
+ bool VISCTimer = false;
std::string TargetName = "None";
// Map from Old function associated with DFNode to new cloned function with
@@ -78,12 +78,12 @@ protected:
// "Have we visited this function before?")
DenseMap<DFNode *, Value *> OutputMap;
- // HPVM Runtime API
+ // VISC Runtime API
std::unique_ptr<Module> runtimeModule;
- FunctionCallee llvm_hpvm_initializeTimerSet;
- FunctionCallee llvm_hpvm_switchToTimer;
- FunctionCallee llvm_hpvm_printTimerSet;
+ FunctionCallee llvm_visc_initializeTimerSet;
+ FunctionCallee llvm_visc_switchToTimer;
+ FunctionCallee llvm_visc_printTimerSet;
GlobalVariable *TimerSet;
GlobalVariable *GraphIDAddr;
Instruction *InitCall;
@@ -109,7 +109,7 @@ protected:
// Virtual Functions
virtual void initializeTimerSet(Instruction *);
- virtual void switchToTimer(enum hpvm_TimerID, Instruction *);
+ virtual void switchToTimer(enum visc_TimerID, Instruction *);
virtual void printTimerSet(Instruction *);
virtual ~CodeGenTraversal() {}
@@ -118,9 +118,9 @@ public:
// Constructor
CodeGenTraversal(Module &_M, BuildDFG &_DFG) : M(_M), DFG(_DFG) {}
- static bool checkPreferredTarget(DFNode *N, hpvm::Target T);
- static bool preferredTargetIncludes(DFNode *N, hpvm::Target T);
- hpvm::Target getPreferredTarget(DFNode *N);
+ static bool checkPreferredTarget(DFNode *N, visc::Target T);
+ static bool preferredTargetIncludes(DFNode *N, visc::Target T);
+ visc::Target getPreferredTarget(DFNode *N);
virtual void visit(DFInternalNode *N) {
// If code has already been generated for this internal node, skip the
@@ -157,25 +157,25 @@ public:
// -------------- CodeGenTraversal Implementation -----------------
-bool CodeGenTraversal::checkPreferredTarget(DFNode *N, hpvm::Target T) {
+bool CodeGenTraversal::checkPreferredTarget(DFNode *N, visc::Target T) {
Function *F = N->getFuncPointer();
Module *M = F->getParent();
NamedMDNode *HintNode;
switch (T) {
- case hpvm::GPU_TARGET:
- HintNode = M->getOrInsertNamedMetadata("hpvm_hint_gpu");
+ case visc::GPU_TARGET:
+ HintNode = M->getOrInsertNamedMetadata("visc_hint_gpu");
break;
- case hpvm::SPIR_TARGET:
- HintNode = M->getOrInsertNamedMetadata("hpvm_hint_spir");
+ case visc::SPIR_TARGET:
+ HintNode = M->getOrInsertNamedMetadata("visc_hint_spir");
break;
- case hpvm::CUDNN_TARGET:
- HintNode = M->getOrInsertNamedMetadata("hpvm_hint_cudnn");
+ case visc::CUDNN_TARGET:
+ HintNode = M->getOrInsertNamedMetadata("visc_hint_cudnn");
break;
- case hpvm::PROMISE_TARGET:
- HintNode = M->getOrInsertNamedMetadata("hpvm_hint_promise");
+ case visc::PROMISE_TARGET:
+ HintNode = M->getOrInsertNamedMetadata("visc_hint_promise");
break;
- case hpvm::CPU_TARGET:
- HintNode = M->getOrInsertNamedMetadata("hpvm_hint_cpu");
+ case visc::CPU_TARGET:
+ HintNode = M->getOrInsertNamedMetadata("visc_hint_cpu");
break;
default:
llvm_unreachable("Target Not supported yet!");
@@ -190,37 +190,37 @@ bool CodeGenTraversal::checkPreferredTarget(DFNode *N, hpvm::Target T) {
return false;
}
-hpvm::Target CodeGenTraversal::getPreferredTarget(DFNode *N) {
- return hpvmUtils::getPreferredTarget(N->getFuncPointer());
+visc::Target CodeGenTraversal::getPreferredTarget(DFNode *N) {
+ return viscUtils::getPreferredTarget(N->getFuncPointer());
}
-bool CodeGenTraversal::preferredTargetIncludes(DFNode *N, hpvm::Target T) {
+bool CodeGenTraversal::preferredTargetIncludes(DFNode *N, visc::Target T) {
Function *F = N->getFuncPointer();
Module *M = F->getParent();
std::vector<NamedMDNode *> HintNode;
switch (T) {
- case hpvm::GPU_TARGET:
- HintNode.push_back(M->getOrInsertNamedMetadata("hpvm_hint_gpu"));
- HintNode.push_back(M->getOrInsertNamedMetadata("hpvm_hint_cpu_gpu"));
+ case visc::GPU_TARGET:
+ HintNode.push_back(M->getOrInsertNamedMetadata("visc_hint_gpu"));
+ HintNode.push_back(M->getOrInsertNamedMetadata("visc_hint_cpu_gpu"));
break;
- case hpvm::SPIR_TARGET:
- HintNode.push_back(M->getOrInsertNamedMetadata("hpvm_hint_spir"));
- HintNode.push_back(M->getOrInsertNamedMetadata("hpvm_hint_cpu_spir"));
+ case visc::SPIR_TARGET:
+ HintNode.push_back(M->getOrInsertNamedMetadata("visc_hint_spir"));
+ HintNode.push_back(M->getOrInsertNamedMetadata("visc_hint_cpu_spir"));
break;
- case hpvm::CPU_TARGET:
- HintNode.push_back(M->getOrInsertNamedMetadata("hpvm_hint_cpu"));
- HintNode.push_back(M->getOrInsertNamedMetadata("hpvm_hint_cpu_gpu"));
- HintNode.push_back(M->getOrInsertNamedMetadata("hpvm_hint_cpu_spir"));
+ case visc::CPU_TARGET:
+ HintNode.push_back(M->getOrInsertNamedMetadata("visc_hint_cpu"));
+ HintNode.push_back(M->getOrInsertNamedMetadata("visc_hint_cpu_gpu"));
+ HintNode.push_back(M->getOrInsertNamedMetadata("visc_hint_cpu_spir"));
break;
- case hpvm::CUDNN_TARGET:
- HintNode.push_back(M->getOrInsertNamedMetadata("hpvm_hint_cudnn"));
+ case visc::CUDNN_TARGET:
+ HintNode.push_back(M->getOrInsertNamedMetadata("visc_hint_cudnn"));
break;
- case hpvm::PROMISE_TARGET:
- HintNode.push_back(M->getOrInsertNamedMetadata("hpvm_hint_promise"));
+ case visc::PROMISE_TARGET:
+ HintNode.push_back(M->getOrInsertNamedMetadata("visc_hint_promise"));
break;
- case hpvm::CPU_OR_GPU_TARGET:
- case hpvm::CPU_OR_SPIR_TARGET:
+ case visc::CPU_OR_GPU_TARGET:
+ case visc::CPU_OR_SPIR_TARGET:
assert(false && "Target should be one of CPU/GPU/SPIR\n");
break;
default:
@@ -308,11 +308,11 @@ Function *CodeGenTraversal::addArgument(Function *F, Type *Ty,
Function *newF = Function::Create(FTy, F->getLinkage(),
F->getName() + "_cloned", F->getParent());
renameNewArgument(newF, name);
- newF = hpvmUtils::cloneFunction(F, newF, false);
+ newF = viscUtils::cloneFunction(F, newF, false);
// Check if the function is used by a metadata node
if (F->isUsedByMetadata()) {
- hpvmUtils::fixHintMetadata(*F->getParent(), F, newF);
+ viscUtils::fixHintMetadata(*F->getParent(), F, newF);
}
return newF;
@@ -396,32 +396,32 @@ Argument *CodeGenTraversal::getArgumentAt(Function *F, unsigned offset) {
}
void CodeGenTraversal::initTimerAPI() {
- DECLARE(llvm_hpvm_initializeTimerSet);
- DECLARE(llvm_hpvm_switchToTimer);
- DECLARE(llvm_hpvm_printTimerSet);
+ DECLARE(llvm_visc_initializeTimerSet);
+ DECLARE(llvm_visc_switchToTimer);
+ DECLARE(llvm_visc_printTimerSet);
}
// Timer Routines
// Initialize the timer set
void CodeGenTraversal::initializeTimerSet(Instruction *InsertBefore) {
- // DEBUG(errs() << "Inserting call to: " << *llvm_hpvm_initializeTimerSet <<
+ // DEBUG(errs() << "Inserting call to: " << *llvm_visc_initializeTimerSet <<
// "\n");
TIMER(TimerSet = new GlobalVariable(
M, Type::getInt8PtrTy(M.getContext()), false,
GlobalValue::CommonLinkage,
Constant::getNullValue(Type::getInt8PtrTy(M.getContext())),
- Twine("hpvmTimerSet_") + TargetName);
+ Twine("viscTimerSet_") + TargetName);
DEBUG(errs() << "New global variable: " << *TimerSet << "\n");
- Value *TimerSetAddr = CallInst::Create(llvm_hpvm_initializeTimerSet,
+ Value *TimerSetAddr = CallInst::Create(llvm_visc_initializeTimerSet,
None, "", InsertBefore);
new StoreInst(TimerSetAddr, TimerSet, InsertBefore););
}
-void CodeGenTraversal::switchToTimer(enum hpvm_TimerID timer,
+void CodeGenTraversal::switchToTimer(enum visc_TimerID timer,
Instruction *InsertBefore) {
Value *switchArgs[] = {TimerSet, getTimerID(M, timer)};
- TIMER(CallInst::Create(llvm_hpvm_switchToTimer,
+ TIMER(CallInst::Create(llvm_visc_switchToTimer,
ArrayRef<Value *>(switchArgs, 2), "", InsertBefore));
}
@@ -430,16 +430,16 @@ void CodeGenTraversal::printTimerSet(Instruction *InsertBefore) {
TIMER(TimerName =
getStringPointer(TargetName + Twine("_Timer"), InsertBefore));
Value *printArgs[] = {TimerSet, TimerName};
- TIMER(CallInst::Create(llvm_hpvm_printTimerSet,
+ TIMER(CallInst::Create(llvm_visc_printTimerSet,
ArrayRef<Value *>(printArgs, 2), "", InsertBefore));
}
// Implementation of Helper Functions
-static inline ConstantInt *getTimerID(Module &M, enum hpvm_TimerID timer) {
+static inline ConstantInt *getTimerID(Module &M, enum visc_TimerID timer) {
return ConstantInt::get(Type::getInt32Ty(M.getContext()), timer);
}
-static inline ConstantInt *getTargetID(Module &M, enum hpvm::Target T) {
+static inline ConstantInt *getTargetID(Module &M, enum visc::Target T) {
return ConstantInt::get(Type::getInt32Ty(M.getContext()), T);
}
diff --git a/hpvm/include/SupportHPVM/DFGTreeTraversal.h b/hpvm/include/SupportVISC/DFGTreeTraversal.h
similarity index 100%
rename from hpvm/include/SupportHPVM/DFGTreeTraversal.h
rename to hpvm/include/SupportVISC/DFGTreeTraversal.h
diff --git a/hpvm/include/SupportHPVM/DFGraph.h b/hpvm/include/SupportVISC/DFGraph.h
similarity index 94%
rename from hpvm/include/SupportHPVM/DFGraph.h
rename to hpvm/include/SupportVISC/DFGraph.h
index d904e2401d..0c224a344c 100644
--- a/hpvm/include/SupportHPVM/DFGraph.h
+++ b/hpvm/include/SupportVISC/DFGraph.h
@@ -20,8 +20,8 @@
#ifndef LLVM_IR_DFGRAPH_H
#define LLVM_IR_DFGRAPH_H
-#include "SupportHPVM/HPVMHint.h"
-#include "SupportHPVM/HPVMUtils.h"
+#include "SupportVISC/VISCHint.h"
+#include "SupportVISC/VISCUtils.h"
#include "llvm/ADT/GraphTraits.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/IntrinsicInst.h"
@@ -158,7 +158,7 @@ public:
}
};
-// DFNode represents a single HPVM Dataflow Node in LLVM.
+// DFNode represents a single VISC Dataflow Node in LLVM.
//
// A Dataflow Node basically consists of
// 1. Pointer to a function describing this dataflow node
@@ -210,8 +210,8 @@ private:
///< hierarchy
unsigned Rank; ///< Ordering based on toplogical sort
const DFNodeKind Kind; ///< Kind of Node Internal/Leaf
- hpvm::Target Tag; ///< Code Generated for which backend
- hpvm::Target Hint; ///< To store preferred backend
+ visc::Target Tag; ///< Code Generated for which backend
+ visc::Target Hint; ///< To store preferred backend
public:
virtual ~DFNode() {
@@ -287,13 +287,13 @@ public:
DFNodeKind getKind() const { return Kind; }
- DFNode(IntrinsicInst *_II, Function *_FuncPointer, hpvm::Target _Hint,
+ DFNode(IntrinsicInst *_II, Function *_FuncPointer, visc::Target _Hint,
DFInternalNode *_Parent, unsigned _NumOfDim,
std::vector<Value *> _DimLimits, DFNodeKind _K);
bool isRoot() const {
// It is a root node is it was created from a launch intrinsic
- if (II->getCalledFunction()->getName().equals("llvm.hpvm.launch")) {
+ if (II->getCalledFunction()->getName().equals("llvm.visc.launch")) {
assert(Level == 0 && "Root node's level is zero.");
return true;
}
@@ -326,9 +326,9 @@ public:
unsigned getRank() const { return Rank; }
- void setTag(hpvm::Target T) { Tag = T; }
+ void setTag(visc::Target T) { Tag = T; }
- hpvm::Target getTag() const { return Tag; }
+ visc::Target getTag() const { return Tag; }
void *getProperty(PropertyKind PType) {
assert(PropertyList.count(PType) == 1 &&
@@ -342,24 +342,24 @@ public:
PropertyList[PType] = PValue;
}
- void setGenFunc(Function *F, hpvm::Target T) {
+ void setGenFunc(Function *F, visc::Target T) {
GenFunc = F;
Tag = T;
}
Function *getGenFunc() const { return GenFunc; }
- void setHasX86FuncForTarget(hpvm::Target T, bool isX86Func) {
+ void setHasX86FuncForTarget(visc::Target T, bool isX86Func) {
switch (T) {
- case hpvm::None:
+ case visc::None:
return; // Do nothing.
- case hpvm::CPU_TARGET:
+ case visc::CPU_TARGET:
GenFuncInfo.cpu_hasX86Func = isX86Func;
break;
- case hpvm::GPU_TARGET:
+ case visc::GPU_TARGET:
GenFuncInfo.gpu_hasX86Func = isX86Func;
break;
- case hpvm::CPU_OR_GPU_TARGET:
+ case visc::CPU_OR_GPU_TARGET:
break;
default:
assert(false && "Unknown target\n");
@@ -368,15 +368,15 @@ public:
return;
}
- bool hasX86GenFuncForTarget(hpvm::Target T) const {
+ bool hasX86GenFuncForTarget(visc::Target T) const {
switch (T) {
- case hpvm::None:
+ case visc::None:
return false;
- case hpvm::CPU_TARGET:
+ case visc::CPU_TARGET:
return GenFuncInfo.cpu_hasX86Func;
- case hpvm::GPU_TARGET:
+ case visc::GPU_TARGET:
return GenFuncInfo.gpu_hasX86Func;
- case hpvm::CPU_OR_GPU_TARGET:
+ case visc::CPU_OR_GPU_TARGET:
assert(false && "Single target expected (CPU/GPU/SPIR/CUDNN/PROMISE)\n");
default:
assert(false && "Unknown target\n");
@@ -384,10 +384,10 @@ public:
return false;
}
- void addGenFunc(Function *F, hpvm::Target T, bool isX86Func) {
+ void addGenFunc(Function *F, visc::Target T, bool isX86Func) {
switch (T) {
- case hpvm::CPU_TARGET:
+ case visc::CPU_TARGET:
if (GenFuncs.CPUGenFunc != NULL) {
DEBUG(errs() << "Warning: Second generated CPU function for node "
<< FuncPointer->getName() << "\n");
@@ -395,7 +395,7 @@ public:
GenFuncs.CPUGenFunc = F;
GenFuncInfo.cpu_hasX86Func = isX86Func;
break;
- case hpvm::GPU_TARGET:
+ case visc::GPU_TARGET:
if (GenFuncs.GPUGenFunc != NULL) {
DEBUG(errs() << "Warning: Second generated GPU function for node "
<< FuncPointer->getName() << "\n");
@@ -403,25 +403,25 @@ public:
GenFuncs.GPUGenFunc = F;
GenFuncInfo.gpu_hasX86Func = isX86Func;
break;
- case hpvm::CPU_OR_GPU_TARGET:
+ case visc::CPU_OR_GPU_TARGET:
assert(false && "A node function should be set with a tag specifying its \
type, not the node hint itself\n");
default:
assert(false && "Unknown target for generated function\n");
}
- Tag = hpvmUtils::getUpdatedTag(Tag, T);
+ Tag = viscUtils::getUpdatedTag(Tag, T);
}
- Function *getGenFuncForTarget(hpvm::Target T) const {
+ Function *getGenFuncForTarget(visc::Target T) const {
switch (T) {
- case hpvm::None:
+ case visc::None:
return NULL;
- case hpvm::CPU_TARGET:
+ case visc::CPU_TARGET:
return GenFuncs.CPUGenFunc;
- case hpvm::GPU_TARGET:
+ case visc::GPU_TARGET:
return GenFuncs.GPUGenFunc;
- case hpvm::CPU_OR_GPU_TARGET:
+ case visc::CPU_OR_GPU_TARGET:
assert(false &&
"Requesting genarated node function with dual tag instead of \
CPU/GPU/SPIR/CUDNN/PROMISE\n");
@@ -431,19 +431,19 @@ public:
return NULL;
}
- void removeGenFuncForTarget(hpvm::Target T) {
+ void removeGenFuncForTarget(visc::Target T) {
switch (T) {
- case hpvm::None:
+ case visc::None:
return;
- case hpvm::CPU_TARGET:
+ case visc::CPU_TARGET:
GenFuncs.CPUGenFunc = NULL;
GenFuncInfo.cpu_hasX86Func = false;
break;
- case hpvm::GPU_TARGET:
+ case visc::GPU_TARGET:
GenFuncs.GPUGenFunc = NULL;
GenFuncInfo.gpu_hasX86Func = false;
break;
- case hpvm::CPU_OR_GPU_TARGET:
+ case visc::CPU_OR_GPU_TARGET:
assert(false &&
"Removing genarated node function with dual tag instead of \
CPU/GPU/SPIR/CUDNN/PROMISE\n");
@@ -453,9 +453,9 @@ public:
return;
}
- void setTargetHint(hpvm::Target T) { Hint = T; }
+ void setTargetHint(visc::Target T) { Hint = T; }
- hpvm::Target getTargetHint() const { return Hint; }
+ visc::Target getTargetHint() const { return Hint; }
bool isDummyNode() const { return isEntryNode() || isExitNode(); }
@@ -496,7 +496,7 @@ private:
DFGraph *childGraph; ///< Pointer to dataflow graph
// Constructor
- DFInternalNode(IntrinsicInst *II, Function *FuncPointer, hpvm::Target Hint,
+ DFInternalNode(IntrinsicInst *II, Function *FuncPointer, visc::Target Hint,
DFInternalNode *Parent, int NumOfDim,
std::vector<Value *> DimLimits)
: DFNode(II, FuncPointer, Hint, Parent, NumOfDim, DimLimits,
@@ -508,7 +508,7 @@ private:
public:
static DFInternalNode *
Create(IntrinsicInst *II, Function *FuncPointer,
- hpvm::Target Hint = hpvm::CPU_TARGET, DFInternalNode *Parent = NULL,
+ visc::Target Hint = visc::CPU_TARGET, DFInternalNode *Parent = NULL,
int NumOfDim = 0,
std::vector<Value *> DimLimits = std::vector<Value *>()) {
@@ -539,14 +539,14 @@ class DFLeafNode : public DFNode {
private:
// Constructor
- DFLeafNode(IntrinsicInst *II, Function *FuncPointer, hpvm::Target Hint,
+ DFLeafNode(IntrinsicInst *II, Function *FuncPointer, visc::Target Hint,
DFInternalNode *Parent, int NumOfDim = 0,
std::vector<Value *> DimLimits = std::vector<Value *>())
: DFNode(II, FuncPointer, Hint, Parent, NumOfDim, DimLimits, LeafNode) {}
public:
static DFLeafNode *
- Create(IntrinsicInst *II, Function *FuncPointer, hpvm::Target Hint,
+ Create(IntrinsicInst *II, Function *FuncPointer, visc::Target Hint,
DFInternalNode *Parent, int NumOfDim = 0,
std::vector<Value *> DimLimits = std::vector<Value *>()) {
return new DFLeafNode(II, FuncPointer, Hint, Parent, NumOfDim, DimLimits);
@@ -558,7 +558,7 @@ public:
// void applyDFEdgeVisitor(DFEdgeVisitor &V); /*virtual*/
};
-// DFEdge represents a single HPVM Dataflow Edge in LLVM.
+// DFEdge represents a single VISC Dataflow Edge in LLVM.
//
// A Dataflow Edge basically consists of
// 1. Pointer to the dataflow node that is the source of this edge
@@ -634,8 +634,8 @@ DFGraph::DFGraph(DFInternalNode *P) {
Parent = P;
// Create dummy entry and exit nodes and add them to the graph
Entry =
- DFLeafNode::Create(NULL, Parent->getFuncPointer(), hpvm::None, Parent);
- Exit = DFLeafNode::Create(NULL, Parent->getFuncPointer(), hpvm::None, Parent);
+ DFLeafNode::Create(NULL, Parent->getFuncPointer(), visc::None, Parent);
+ Exit = DFLeafNode::Create(NULL, Parent->getFuncPointer(), visc::None, Parent);
addChildDFNode(Entry);
addChildDFNode(Exit);
}
@@ -655,7 +655,7 @@ bool DFGraph::isStreaming() {
}
//===--------------------- DFNode Outlined Functions --------------===//
-DFNode::DFNode(IntrinsicInst *_II, Function *_FuncPointer, hpvm::Target _Hint,
+DFNode::DFNode(IntrinsicInst *_II, Function *_FuncPointer, visc::Target _Hint,
DFInternalNode *_Parent, unsigned _NumOfDim,
std::vector<Value *> _DimLimits, DFNodeKind _K)
: II(_II), FuncPointer(_FuncPointer), Parent(_Parent), NumOfDim(_NumOfDim),
@@ -663,7 +663,7 @@ DFNode::DFNode(IntrinsicInst *_II, Function *_FuncPointer, hpvm::Target _Hint,
Type *Ty = FuncPointer->getFunctionType()->getReturnType();
- // Allow the return type to be void too, in the hHPVM IR. If return type is
+ // Allow the return type to be void too, in the hVISC IR. If return type is
// void, create an empty struct type and keep that as the return type of the
// node.
if (Ty->isVoidTy())
@@ -683,7 +683,7 @@ DFNode::DFNode(IntrinsicInst *_II, Function *_FuncPointer, hpvm::Target _Hint,
Level = (_Parent) ? _Parent->getLevel() + 1 : 0;
Rank = 0;
- Tag = hpvm::None;
+ Tag = visc::None;
GenFuncs.CPUGenFunc = NULL;
GenFuncs.GPUGenFunc = NULL;
GenFuncs.SPIRGenFunc = NULL;
diff --git a/hpvm/include/SupportHPVM/HPVMHint.h b/hpvm/include/SupportVISC/VISCHint.h
similarity index 78%
rename from hpvm/include/SupportHPVM/HPVMHint.h
rename to hpvm/include/SupportVISC/VISCHint.h
index 1ef4c6eb3b..99266b0718 100644
--- a/hpvm/include/SupportHPVM/HPVMHint.h
+++ b/hpvm/include/SupportVISC/VISCHint.h
@@ -1,4 +1,4 @@
-//===------------ HPVMTimer.h - Header file for "HPVM Timer API" ----------===//
+//===------------ VISCTimer.h - Header file for "VISC Timer API" ----------===//
//
// The LLVM Compiler Infrastructure
//
@@ -7,12 +7,12 @@
//
//===----------------------------------------------------------------------===//
-#ifndef HPVM_HINT_HEADER
-#define HPVM_HINT_HEADER
+#ifndef VISC_HINT_HEADER
+#define VISC_HINT_HEADER
/************************** Hint Routines ***************************/
#ifdef __cplusplus
-namespace hpvm {
+namespace visc {
#endif
enum Target {
@@ -32,4 +32,4 @@ enum Target {
}
#endif
-#endif // HPVM_HINT_HEADER
+#endif // VISC_HINT_HEADER
diff --git a/hpvm/include/SupportVISC/VISCTimer.h b/hpvm/include/SupportVISC/VISCTimer.h
new file mode 100644
index 0000000000..ce3dc8a5e0
--- /dev/null
+++ b/hpvm/include/SupportVISC/VISCTimer.h
@@ -0,0 +1,151 @@
+//===------------ VISCTimer.h - Header file for "VISC Timer API" ----------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef VISC_TIMER_HEADER
+#define VISC_TIMER_HEADER
+
+/************************** Timer Routines ***************************/
+extern "C" {
+
+/* A time or duration. */
+//#if _POSIX_VERSION >= 200112L
+typedef unsigned long long visc_Timestamp; /* time in microseconds */
+//#else
+//# error "Timestamps not implemented"
+//#endif
+
+enum visc_TimerState {
+ visc_Timer_STOPPED,
+ visc_Timer_RUNNING,
+};
+
+struct visc_Timer {
+ enum visc_TimerState state;
+ visc_Timestamp elapsed; /* Amount of time elapsed so far */
+ visc_Timestamp init; /* Beginning of the current time interval,
+ * if state is RUNNING. End of the last
+ * recorded time interfal otherwise. */
+};
+
+/* Reset a timer.
+ * Use this to initialize a timer or to clear
+ * its elapsed time. The reset timer is stopped.
+ */
+void visc_ResetTimer(struct visc_Timer *timer);
+
+/* Start a timer. The timer is set to RUNNING mode and
+ * time elapsed while the timer is running is added to
+ * the timer.
+ * The timer should not already be running.
+ */
+void visc_StartTimer(struct visc_Timer *timer);
+
+/* Stop a timer.
+ * This stops adding elapsed time to the timer.
+ * The timer should not already be stopped.
+ */
+void visc_StopTimer(struct visc_Timer *timer);
+
+/* Get the elapsed time in seconds. */
+double visc_GetElapsedTime(struct visc_Timer *timer);
+
+/* Execution time is assigned to one of these categories. */
+enum visc_TimerID {
+ visc_TimerID_NONE = 0,
+ visc_TimerID_IO, /* Time spent in input/output */
+ visc_TimerID_KERNEL, /* Time spent computing on the device,
+ * recorded asynchronously */
+ visc_TimerID_COPY, /* Time spent synchronously moving data
+ * to/from device and allocating/freeing
+ * memory on the device */
+ visc_TimerID_DRIVER, /* Time spent in the host interacting with the
+ * driver, primarily for recording the time
+ * spent queueing asynchronous operations */
+ visc_TimerID_COPY_ASYNC, /* Time spent in asynchronous transfers */
+ visc_TimerID_COMPUTE, /* Time for all program execution other
+ * than parsing command line arguments,
+ * I/O, kernel, and copy */
+ visc_TimerID_OVERLAP, /* Time double-counted in asynchronous and
+ * host activity: automatically filled in,
+ * not intended for direct usage */
+ // GPU FUNCTION
+ visc_TimerID_INIT_CTX,
+ visc_TimerID_CLEAR_CTX,
+ visc_TimerID_COPY_SCALAR,
+ visc_TimerID_COPY_PTR,
+ visc_TimerID_MEM_FREE,
+ visc_TimerID_READ_OUTPUT,
+ visc_TimerID_SETUP,
+ visc_TimerID_MEM_TRACK,
+ visc_TimerID_MEM_UNTRACK,
+ visc_TimerID_MISC,
+ // LAUNCH FUNCTION
+ visc_TimerID_PTHREAD_CREATE,
+ visc_TimerID_ARG_PACK,
+ visc_TimerID_ARG_UNPACK,
+ visc_TimerID_COMPUTATION,
+ visc_TimerID_OUTPUT_PACK,
+ visc_TimerID_OUTPUT_UNPACK,
+
+ visc_TimerID_LAST /* Number of timer IDs */
+};
+
+/* Dynamic list of asynchronously tracked times between events */
+struct visc_async_time_marker_list {
+ char *label; // actually just a pointer to a string
+ enum visc_TimerID timerID; /* The ID to which the interval beginning
+ * with this marker should be attributed */
+ void *marker;
+ // cudaEvent_t marker; /* The driver event for this marker */
+ struct visc_async_time_marker_list *next;
+};
+
+struct visc_SubTimer {
+ char *label;
+ struct visc_Timer timer;
+ struct visc_SubTimer *next;
+};
+
+struct visc_SubTimerList {
+ struct visc_SubTimer *current;
+ struct visc_SubTimer *subtimer_list;
+};
+
+/* A set of timers for recording execution times. */
+struct visc_TimerSet {
+ enum visc_TimerID current;
+ struct visc_async_time_marker_list *async_markers;
+ visc_Timestamp async_begin;
+ visc_Timestamp wall_begin;
+ struct visc_Timer timers[visc_TimerID_LAST];
+ struct visc_SubTimerList *sub_timer_list[visc_TimerID_LAST];
+};
+
+/* Reset all timers in the set. */
+void visc_InitializeTimerSet(struct visc_TimerSet *timers);
+
+void visc_AddSubTimer(struct visc_TimerSet *timers, char *label,
+ enum visc_TimerID visc_Category);
+
+/* Select which timer the next interval of time should be accounted
+ * to. The selected timer is started and other timers are stopped.
+ * Using visc_TimerID_NONE stops all timers. */
+inline void visc_SwitchToTimer(struct visc_TimerSet *timers,
+ enum visc_TimerID timer);
+
+void visc_SwitchToSubTimer(struct visc_TimerSet *timers, char *label,
+ enum visc_TimerID category);
+
+/* Print timer values to standard output. */
+void visc_PrintTimerSet(struct visc_TimerSet *timers);
+
+/* Release timer resources */
+void visc_DestroyTimerSet(struct visc_TimerSet *timers);
+}
+#endif // VISC_RT_HEADER
diff --git a/hpvm/include/SupportHPVM/HPVMUtils.h b/hpvm/include/SupportVISC/VISCUtils.h
similarity index 84%
rename from hpvm/include/SupportHPVM/HPVMUtils.h
rename to hpvm/include/SupportVISC/VISCUtils.h
index 25b9880180..0efd20b5b5 100644
--- a/hpvm/include/SupportHPVM/HPVMUtils.h
+++ b/hpvm/include/SupportVISC/VISCUtils.h
@@ -1,5 +1,5 @@
//
-//===---- DFG2LLVM.h - Header file for "HPVM Dataflow Graph to Target" ----===//
+//===---- DFG2LLVM.h - Header file for "VISC Dataflow Graph to Target" ----===//
//
// The LLVM Compiler Infrastructure
//
@@ -8,12 +8,12 @@
//
//===----------------------------------------------------------------------===//
-#ifndef HPVM_UTILS_HEADER
-#define HPVM_UTILS_HEADER
+#ifndef VISC_UTILS_HEADER
+#define VISC_UTILS_HEADER
#include <assert.h>
-#include "SupportHPVM/HPVMHint.h"
+#include "SupportVISC/VISCHint.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/InstIterator.h"
#include "llvm/IR/Instructions.h"
@@ -29,31 +29,31 @@
using namespace llvm;
-namespace hpvmUtils {
+namespace viscUtils {
// Helper Functions
-static bool isHPVMCreateNodeIntrinsic(Instruction *I) {
+static bool isViscCreateNodeIntrinsic(Instruction *I) {
if (!isa<IntrinsicInst>(I))
return false;
IntrinsicInst *II = cast<IntrinsicInst>(I);
return (II->getCalledFunction()->getName())
- .startswith("llvm.hpvm.createNode");
+ .startswith("llvm.visc.createNode");
}
-static bool isHPVMCreateNodeCall(Instruction *I) {
+static bool isViscCreateNodeCall(Instruction *I) {
if (!isa<CallInst>(I))
return false;
CallInst *CI = cast<CallInst>(I);
return (CI->getCalledValue()->stripPointerCasts()->getName())
- .startswith("__hpvm__createNode");
+ .startswith("__visc__createNode");
}
-static bool isHPVMLaunchCall(Instruction *I) {
+static bool isViscLaunchCall(Instruction *I) {
if (!isa<CallInst>(I))
return false;
CallInst *CI = cast<CallInst>(I);
return (CI->getCalledValue()->stripPointerCasts()->getName())
- .startswith("__hpvm__launch");
+ .startswith("__visc__launch");
}
// Creates a new createNode intrinsic, similar to II but with different
// associated function F instead
@@ -69,22 +69,22 @@ createIdenticalCreateNodeIntrinsicWithDifferentFunction(Function *F,
ArrayRef<Value *> CreateNodeArgs;
switch (II->getIntrinsicID()) {
- case Intrinsic::hpvm_createNode: {
+ case Intrinsic::visc_createNode: {
CreateNodeArgs = ArrayRef<Value *>(Fp);
break;
}
- case Intrinsic::hpvm_createNode1D: {
+ case Intrinsic::visc_createNode1D: {
Value *CreateNode1DArgs[] = {Fp, II->getArgOperand(1)};
CreateNodeArgs = ArrayRef<Value *>(CreateNode1DArgs, 2);
break;
}
- case Intrinsic::hpvm_createNode2D: {
+ case Intrinsic::visc_createNode2D: {
Value *CreateNode2DArgs[] = {Fp, II->getArgOperand(1),
II->getArgOperand(2)};
CreateNodeArgs = ArrayRef<Value *>(CreateNode2DArgs, 3);
break;
}
- case Intrinsic::hpvm_createNode3D: {
+ case Intrinsic::visc_createNode3D: {
Value *CreateNode3DArgs[] = {Fp, II->getArgOperand(1), II->getArgOperand(2),
II->getArgOperand(3)};
CreateNodeArgs = ArrayRef<Value *>(CreateNode3DArgs, 4);
@@ -101,7 +101,7 @@ createIdenticalCreateNodeIntrinsicWithDifferentFunction(Function *F,
return CreateNodeII;
}
-// Fix HPVM hints for this function
+// Fix VISC hints for this function
void fixHintMetadata(Module &M, Function *F, Function *G) {
Metadata *MD_F = ValueAsMetadata::getIfExists(F);
MDTuple *MDT_F =
@@ -119,9 +119,9 @@ void fixHintMetadata(Module &M, Function *F, Function *G) {
}
};
- FixHint("hpvm_hint_gpu");
- FixHint("hpvm_hint_cpu");
- FixHint("hpvm_hint_cpu_gpu");
+ FixHint("visc_hint_gpu");
+ FixHint("visc_hint_cpu");
+ FixHint("visc_hint_cpu_gpu");
}
// Assuming that the changed function is a node function, it is only used as a
@@ -138,7 +138,7 @@ void replaceNodeFunctionInIR(Module &M, Function *F, Function *G) {
++i) {
Instruction *I = &*i; // Grab pointer to Instruction
- if (isHPVMCreateNodeIntrinsic(I)) {
+ if (isViscCreateNodeIntrinsic(I)) {
IntrinsicInst *II = cast<IntrinsicInst>(I);
// The found createNode is not associated with the changed function
if (II->getArgOperand(0) != F)
@@ -150,7 +150,7 @@ void replaceNodeFunctionInIR(Module &M, Function *F, Function *G) {
createIdenticalCreateNodeIntrinsicWithDifferentFunction(G, II);
II->replaceAllUsesWith(CreateNodeII);
toBeErased.push_back(II);
- } else if (isHPVMCreateNodeCall(I)) {
+ } else if (isViscCreateNodeCall(I)) {
CallInst *CI = cast<CallInst>(I);
// The found createNode is not associated with the changed function
if (CI->getArgOperand(1) != F)
@@ -161,7 +161,7 @@ void replaceNodeFunctionInIR(Module &M, Function *F, Function *G) {
// Replace use of F with use of G
CI->setArgOperand(1, G);
DEBUG(errs() << "Fixed use: " << *CI << "\n");
- } else if (isHPVMLaunchCall(I)) {
+ } else if (isViscLaunchCall(I)) {
CallInst *CI = cast<CallInst>(I);
// The found launch call is not associated with the changed function
if (CI->getArgOperand(1)->stripPointerCasts() != F)
@@ -370,21 +370,21 @@ Function *cloneFunction(Function *F, Function *newF, bool isAddingPtrSizeArg,
//------------------- Helper Functions For Handling Hints -------------------//
// Return true if 1st arg (tag) contains 2nd (target)
-bool tagIncludesTarget(hpvm::Target Tag, hpvm::Target T) {
+bool tagIncludesTarget(visc::Target Tag, visc::Target T) {
switch (Tag) {
- case hpvm::None:
+ case visc::None:
return false;
- case hpvm::CPU_TARGET:
- if (T == hpvm::CPU_TARGET)
+ case visc::CPU_TARGET:
+ if (T == visc::CPU_TARGET)
return true;
return false;
- case hpvm::GPU_TARGET:
- if (T == hpvm::GPU_TARGET)
+ case visc::GPU_TARGET:
+ if (T == visc::GPU_TARGET)
return true;
return false;
- case hpvm::CPU_OR_GPU_TARGET:
- if ((T == hpvm::CPU_TARGET) || (T == hpvm::GPU_TARGET) ||
- (T == hpvm::CPU_OR_GPU_TARGET))
+ case visc::CPU_OR_GPU_TARGET:
+ if ((T == visc::CPU_TARGET) || (T == visc::GPU_TARGET) ||
+ (T == visc::CPU_OR_GPU_TARGET))
return true;
return false;
default:
@@ -392,41 +392,41 @@ bool tagIncludesTarget(hpvm::Target Tag, hpvm::Target T) {
}
}
-bool isSingleTargetTag(hpvm::Target T) {
- return ((T == hpvm::CPU_TARGET) || (T == hpvm::GPU_TARGET));
+bool isSingleTargetTag(visc::Target T) {
+ return ((T == visc::CPU_TARGET) || (T == visc::GPU_TARGET));
}
// Add the specified target to the given tag
-hpvm::Target getUpdatedTag(hpvm::Target Tag, hpvm::Target T) {
- assert(((T == hpvm::CPU_TARGET) || (T == hpvm::GPU_TARGET)) &&
+visc::Target getUpdatedTag(visc::Target Tag, visc::Target T) {
+ assert(((T == visc::CPU_TARGET) || (T == visc::GPU_TARGET)) &&
"The target is only allowed to be a single target: CPU, GPU, SPIR, "
"CUDNN, PROMISE\n");
switch (Tag) {
- case hpvm::None:
+ case visc::None:
return T;
- case hpvm::CPU_TARGET:
- if (T == hpvm::CPU_TARGET)
- return hpvm::CPU_TARGET;
- if (T == hpvm::GPU_TARGET)
- return hpvm::CPU_OR_GPU_TARGET;
+ case visc::CPU_TARGET:
+ if (T == visc::CPU_TARGET)
+ return visc::CPU_TARGET;
+ if (T == visc::GPU_TARGET)
+ return visc::CPU_OR_GPU_TARGET;
return T;
- case hpvm::GPU_TARGET:
- if (T == hpvm::CPU_TARGET)
- return hpvm::CPU_OR_GPU_TARGET;
- if (T == hpvm::GPU_TARGET)
- return hpvm::GPU_TARGET;
+ case visc::GPU_TARGET:
+ if (T == visc::CPU_TARGET)
+ return visc::CPU_OR_GPU_TARGET;
+ if (T == visc::GPU_TARGET)
+ return visc::GPU_TARGET;
return T;
- case hpvm::CPU_OR_GPU_TARGET:
- return hpvm::CPU_OR_GPU_TARGET;
+ case visc::CPU_OR_GPU_TARGET:
+ return visc::CPU_OR_GPU_TARGET;
default:
assert(false && "Unknown Target\n");
}
return T;
}
-// This functions add the hint as metadata in hpvm code
-void addHint(Function *F, hpvm::Target T) {
+// This functions add the hint as metadata in visc code
+void addHint(Function *F, visc::Target T) {
// Get Module
Module *M = F->getParent();
DEBUG(errs() << "Set preferred target for " << F->getName() << ": ");
@@ -434,17 +434,17 @@ void addHint(Function *F, hpvm::Target T) {
// Based on the hint, get the hint metadata
NamedMDNode *HintNode;
switch (T) {
- case hpvm::GPU_TARGET:
+ case visc::GPU_TARGET:
DEBUG(errs() << "GPU Target\n");
- HintNode = M->getOrInsertNamedMetadata("hpvm_hint_gpu");
+ HintNode = M->getOrInsertNamedMetadata("visc_hint_gpu");
break;
- case hpvm::CPU_TARGET:
+ case visc::CPU_TARGET:
DEBUG(errs() << "CPU Target\n");
- HintNode = M->getOrInsertNamedMetadata("hpvm_hint_cpu");
+ HintNode = M->getOrInsertNamedMetadata("visc_hint_cpu");
break;
- case hpvm::CPU_OR_GPU_TARGET:
+ case visc::CPU_OR_GPU_TARGET:
DEBUG(errs() << "CPU or GPU Target\n");
- HintNode = M->getOrInsertNamedMetadata("hpvm_hint_cpu_gpu");
+ HintNode = M->getOrInsertNamedMetadata("visc_hint_cpu_gpu");
break;
default:
llvm_unreachable("Unsupported Target Hint!");
@@ -457,8 +457,8 @@ void addHint(Function *F, hpvm::Target T) {
HintNode->addOperand(N);
}
-// This function removes the hint as metadata in hpvm code
-void removeHint(Function *F, hpvm::Target T) {
+// This function removes the hint as metadata in visc code
+void removeHint(Function *F, visc::Target T) {
// Get Module
Module *M = F->getParent();
DEBUG(errs() << "Remove preferred target for " << F->getName() << ": " << T
@@ -467,14 +467,14 @@ void removeHint(Function *F, hpvm::Target T) {
// Based on the hint, get the hint metadata
NamedMDNode *HintNode;
switch (T) {
- case hpvm::GPU_TARGET:
- HintNode = M->getOrInsertNamedMetadata("hpvm_hint_gpu");
+ case visc::GPU_TARGET:
+ HintNode = M->getOrInsertNamedMetadata("visc_hint_gpu");
break;
- case hpvm::CPU_OR_GPU_TARGET:
- HintNode = M->getOrInsertNamedMetadata("hpvm_hint_cpu_gpu");
+ case visc::CPU_OR_GPU_TARGET:
+ HintNode = M->getOrInsertNamedMetadata("visc_hint_cpu_gpu");
break;
- case hpvm::CPU_TARGET:
- HintNode = M->getOrInsertNamedMetadata("hpvm_hint_cpu");
+ case visc::CPU_TARGET:
+ HintNode = M->getOrInsertNamedMetadata("visc_hint_cpu");
break;
default:
llvm_unreachable("Unsupported Target Hint!");
@@ -501,7 +501,7 @@ void removeHint(Function *F, hpvm::Target T) {
}
}
-hpvm::Target getPreferredTarget(Function *F) {
+visc::Target getPreferredTarget(Function *F) {
DEBUG(errs() << "Finding preferred target for " << F->getName() << "\n");
Module *M = F->getParent();
@@ -517,16 +517,16 @@ hpvm::Target getPreferredTarget(Function *F) {
return false;
};
- if (FoundPrefTarget("hpvm_hint_cpu"))
- return hpvm::CPU_TARGET;
- if (FoundPrefTarget("hpvm_hint_gpu"))
- return hpvm::GPU_TARGET;
- if (FoundPrefTarget("hpvm_hint_cpu_gpu"))
- return hpvm::CPU_OR_GPU_TARGET;
+ if (FoundPrefTarget("visc_hint_cpu"))
+ return visc::CPU_TARGET;
+ if (FoundPrefTarget("visc_hint_gpu"))
+ return visc::GPU_TARGET;
+ if (FoundPrefTarget("visc_hint_cpu_gpu"))
+ return visc::CPU_OR_GPU_TARGET;
- return hpvm::None;
+ return visc::None;
}
-} // namespace hpvmUtils
+} // namespace viscUtils
-#endif // HPVM_UTILS_HEADER
+#endif // VISC_UTILS_HEADER
diff --git a/hpvm/lib/Transforms/BuildDFG/BuildDFG.cpp b/hpvm/lib/Transforms/BuildDFG/BuildDFG.cpp
index be3e6cae3d..058419f1dc 100644
--- a/hpvm/lib/Transforms/BuildDFG/BuildDFG.cpp
+++ b/hpvm/lib/Transforms/BuildDFG/BuildDFG.cpp
@@ -10,8 +10,8 @@
#define DEBUG_TYPE "buildDFG"
#include "BuildDFG/BuildDFG.h"
-#include "SupportHPVM/HPVMHint.h"
-#include "SupportHPVM/HPVMUtils.h"
+#include "SupportVISC/VISCHint.h"
+#include "SupportVISC/VISCUtils.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/IR/InstIterator.h"
#include "llvm/IR/ValueSymbolTable.h"
@@ -35,7 +35,7 @@ bool BuildDFG::runOnModule(Module &M) {
for (inst_iterator i = inst_begin(F), e = inst_end(F); i != e; ++i) {
Instruction *I = &*i; // Grab pointer to Instruction
- if (isHPVMLaunchIntrinsic(I)) {
+ if (isViscLaunchIntrinsic(I)) {
DEBUG(errs() << "------------ Found launch site --------------\n");
II = cast<IntrinsicInst>(I);
@@ -43,7 +43,7 @@ bool BuildDFG::runOnModule(Module &M) {
// Intrinsic Instruction has been initialized from this point on.
Function *F = cast<Function>(II->getOperand(0)->stripPointerCasts());
- Root = DFInternalNode::Create(II, F, hpvmUtils::getPreferredTarget(F));
+ Root = DFInternalNode::Create(II, F, viscUtils::getPreferredTarget(F));
Roots.push_back(Root);
BuildGraph(Root, F);
@@ -118,37 +118,37 @@ void BuildDFG::removeElementFromHandleToDFEdgeMap(Value *V) {
HandleToDFEdgeMap.erase(V);
}
-// Returns true if instruction I is a hpvm launch intrinsic, false otherwise
-bool BuildDFG::isHPVMLaunchIntrinsic(Instruction *I) {
+// Returns true if instruction I is a visc launch intrinsic, false otherwise
+bool BuildDFG::isViscLaunchIntrinsic(Instruction *I) {
if (!isa<IntrinsicInst>(I))
return false;
IntrinsicInst *II = cast<IntrinsicInst>(I);
- return (II->getCalledFunction()->getName()).equals("llvm.hpvm.launch");
+ return (II->getCalledFunction()->getName()).equals("llvm.visc.launch");
}
-// Returns true if instruction I is a hpvm graph intrinsic, false otherwise
-bool BuildDFG::isHPVMGraphIntrinsic(Instruction *I) {
+// Returns true if instruction I is a visc graph intrinsic, false otherwise
+bool BuildDFG::isViscGraphIntrinsic(Instruction *I) {
if (!isa<IntrinsicInst>(I))
return false;
IntrinsicInst *II = cast<IntrinsicInst>(I);
- return (II->getCalledFunction()->getName()).startswith("llvm.hpvm.create") ||
- (II->getCalledFunction()->getName()).startswith("llvm.hpvm.bind");
+ return (II->getCalledFunction()->getName()).startswith("llvm.visc.create") ||
+ (II->getCalledFunction()->getName()).startswith("llvm.visc.bind");
}
-// Returns true if instruction I is a hpvm query intrinsic, false otherwise
-bool BuildDFG::isHPVMQueryIntrinsic(Instruction *I) {
+// Returns true if instruction I is a visc query intrinsic, false otherwise
+bool BuildDFG::isViscQueryIntrinsic(Instruction *I) {
if (!isa<IntrinsicInst>(I))
return false;
IntrinsicInst *II = cast<IntrinsicInst>(I);
- return (II->getCalledFunction()->getName()).startswith("llvm.hpvm.get");
+ return (II->getCalledFunction()->getName()).startswith("llvm.visc.get");
}
-// Returns true if instruction I is a hpvm intrinsic, false otherwise
-bool BuildDFG::isHPVMIntrinsic(Instruction *I) {
+// Returns true if instruction I is a visc intrinsic, false otherwise
+bool BuildDFG::isViscIntrinsic(Instruction *I) {
if (!isa<IntrinsicInst>(I))
return false;
IntrinsicInst *II = cast<IntrinsicInst>(I);
- return (II->getCalledFunction()->getName()).startswith("llvm.hpvm");
+ return (II->getCalledFunction()->getName()).startswith("llvm.visc");
}
// Two types are "congruent" if they are identical, or if they are both
@@ -163,7 +163,7 @@ bool BuildDFG::isTypeCongruent(Type *L, Type *R) {
return PL->getAddressSpace() == PR->getAddressSpace();
}
-// Handles all the createNodeXX hpvm intrinsics.
+// Handles all the createNodeXX visc intrinsics.
void BuildDFG::handleCreateNode(DFInternalNode *N, IntrinsicInst *II) {
bool isInternalNode = false;
@@ -173,7 +173,7 @@ void BuildDFG::handleCreateNode(DFInternalNode *N, IntrinsicInst *II) {
// internal node
for (inst_iterator i = inst_begin(F), e = inst_end(F); i != e; ++i) {
Instruction *I = &*i; // Grab pointer to Instruction
- if (isHPVMGraphIntrinsic(I))
+ if (isViscGraphIntrinsic(I))
isInternalNode = true;
}
@@ -196,14 +196,14 @@ void BuildDFG::handleCreateNode(DFInternalNode *N, IntrinsicInst *II) {
// Create Internal DFNode, add it to the map and recursively build its
// dataflow graph
DFInternalNode *childDFNode = DFInternalNode::Create(
- II, F, hpvmUtils::getPreferredTarget(F), N, numOfDim, dimLimits);
+ II, F, viscUtils::getPreferredTarget(F), N, numOfDim, dimLimits);
N->addChildToDFGraph(childDFNode);
HandleToDFNodeMap[II] = childDFNode;
BuildGraph(childDFNode, F);
} else {
// Create Leaf DFnode and add it to the map.
DFLeafNode *childDFNode = DFLeafNode::Create(
- II, F, hpvmUtils::getPreferredTarget(F), N, numOfDim, dimLimits);
+ II, F, viscUtils::getPreferredTarget(F), N, numOfDim, dimLimits);
N->addChildToDFGraph(childDFNode);
HandleToDFNodeMap[II] = childDFNode;
}
@@ -336,11 +336,11 @@ void BuildDFG::handleBindOutput(DFInternalNode *N, IntrinsicInst *II) {
void BuildDFG::BuildGraph(DFInternalNode *N, Function *F) {
DEBUG(errs() << "FUNCTION: " << F->getName() << "\n");
- // TODO: Place checks for valid hpvm functions. For example one of the
- // check can be that any function that contains hpvm dataflow graph
+ // TODO: Place checks for valid visc functions. For example one of the
+ // check can be that any function that contains visc dataflow graph
// construction intrinsics should not have other llvm IR statements.
- // Iterate over all the instructions of a function and look for hpvm
+ // Iterate over all the instructions of a function and look for visc
// intrinsics.
for (inst_iterator i = inst_begin(F), e = inst_end(F); i != e; ++i) {
Instruction *I = &*i; // Grab pointer to Instruction
@@ -349,25 +349,25 @@ void BuildDFG::BuildGraph(DFInternalNode *N, Function *F) {
DEBUG(errs() << "IntrinsicID = " << II->getIntrinsicID() << ": "
<< II->getCalledFunction()->getName() << "\n");
switch (II->getIntrinsicID()) {
- case Intrinsic::hpvm_createNode:
- case Intrinsic::hpvm_createNode1D:
- case Intrinsic::hpvm_createNode2D:
- case Intrinsic::hpvm_createNode3D:
+ case Intrinsic::visc_createNode:
+ case Intrinsic::visc_createNode1D:
+ case Intrinsic::visc_createNode2D:
+ case Intrinsic::visc_createNode3D:
handleCreateNode(N, II);
break;
- case Intrinsic::hpvm_createEdge:
+ case Intrinsic::visc_createEdge:
handleCreateEdge(N, II);
break;
- case Intrinsic::hpvm_bind_input:
+ case Intrinsic::visc_bind_input:
handleBindInput(N, II);
break;
- case Intrinsic::hpvm_bind_output:
+ case Intrinsic::visc_bind_output:
handleBindOutput(N, II);
break;
// TODO: Reconsider launch within a dataflow graph (recursion?)
- case Intrinsic::hpvm_wait:
- case Intrinsic::hpvm_launch:
+ case Intrinsic::visc_wait:
+ case Intrinsic::visc_launch:
DEBUG(errs()
<< "Error: Launch/wait intrinsic used within a dataflow graph\n\t"
<< *II << "\n");
@@ -375,7 +375,7 @@ void BuildDFG::BuildGraph(DFInternalNode *N, Function *F) {
default:
DEBUG(
- errs() << "Error: Invalid HPVM Intrinsic inside Internal node!\n\t"
+ errs() << "Error: Invalid VISC Intrinsic inside Internal node!\n\t"
<< *II << "\n");
break;
}
diff --git a/hpvm/lib/Transforms/CMakeLists.txt b/hpvm/lib/Transforms/CMakeLists.txt
index 5c9b8b9fe0..68724684e5 100644
--- a/hpvm/lib/Transforms/CMakeLists.txt
+++ b/hpvm/lib/Transforms/CMakeLists.txt
@@ -2,5 +2,5 @@ add_subdirectory(BuildDFG)
add_subdirectory(ClearDFG)
add_subdirectory(DFG2LLVM_NVPTX)
add_subdirectory(DFG2LLVM_X86)
-add_subdirectory(GenHPVM)
+add_subdirectory(GenVISC)
add_subdirectory(LocalMem)
diff --git a/hpvm/lib/Transforms/ClearDFG/ClearDFG.cpp b/hpvm/lib/Transforms/ClearDFG/ClearDFG.cpp
index c23043e782..6dae9e6977 100644
--- a/hpvm/lib/Transforms/ClearDFG/ClearDFG.cpp
+++ b/hpvm/lib/Transforms/ClearDFG/ClearDFG.cpp
@@ -18,7 +18,7 @@
using namespace llvm;
using namespace builddfg;
-// STATISTIC(IntrinsicCounter, "Counts number of hpvm intrinsics greeted");
+// STATISTIC(IntrinsicCounter, "Counts number of visc intrinsics greeted");
namespace {
@@ -101,8 +101,8 @@ bool ClearDFG::runOnModule(Module &M) {
// BuildDFG::HandleToDFNode &HandleToDFNodeMap = DFG.getHandleToDFNodeMap();
// BuildDFG::HandleToDFEdge &HandleToDFEdgeMap = DFG.getHandleToDFEdgeMap();
- Function *VI = M.getFunction("llvm.hpvm.init");
- assert(VI->hasOneUse() && "More than one use of llvm.hpvm.init\n");
+ Function *VI = M.getFunction("llvm.visc.init");
+ assert(VI->hasOneUse() && "More than one use of llvm.visc.init\n");
for (Value::user_iterator ui = VI->user_begin(), ue = VI->user_end();
ui != ue; ui++) {
Instruction *I = dyn_cast<Instruction>(*ui);
@@ -111,8 +111,8 @@ bool ClearDFG::runOnModule(Module &M) {
VI->replaceAllUsesWith(UndefValue::get(VI->getType()));
VI->eraseFromParent();
- Function *VC = M.getFunction("llvm.hpvm.cleanup");
- assert(VC->hasOneUse() && "More than one use of llvm.hpvm.cleanup\n");
+ Function *VC = M.getFunction("llvm.visc.cleanup");
+ assert(VC->hasOneUse() && "More than one use of llvm.visc.cleanup\n");
for (Value::user_iterator ui = VC->user_begin(), ue = VC->user_end();
ui != ue; ui++) {
Instruction *I = dyn_cast<Instruction>(*ui);
diff --git a/hpvm/lib/Transforms/DFG2LLVM_NVPTX/DFG2LLVM_NVPTX.cpp b/hpvm/lib/Transforms/DFG2LLVM_NVPTX/DFG2LLVM_NVPTX.cpp
index 584da07e6e..8a36e3b8af 100644
--- a/hpvm/lib/Transforms/DFG2LLVM_NVPTX/DFG2LLVM_NVPTX.cpp
+++ b/hpvm/lib/Transforms/DFG2LLVM_NVPTX/DFG2LLVM_NVPTX.cpp
@@ -15,39 +15,40 @@
#define SHARED_ADDRSPACE 3
#define DEBUG_TYPE "DFG2LLVM_NVPTX"
-#include "SupportHPVM/DFG2LLVM.h"
-#include "SupportHPVM/HPVMTimer.h"
-#include "SupportHPVM/HPVMUtils.h"
-#include "llvm-c/Core.h"
-#include "llvm/IR/Attributes.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/IRBuilder.h"
-#include "llvm/IR/InstIterator.h"
#include "llvm/IR/Module.h"
-#include "llvm/IRReader/IRReader.h"
-#include "llvm/Linker/Linker.h"
#include "llvm/Pass.h"
-#include "llvm/Support/FileSystem.h"
-#include "llvm/Support/SourceMgr.h"
+#include "llvm/IR/InstIterator.h"
+#include "llvm/Transforms/Utils/ValueMapper.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/Cloning.h"
-#include "llvm/Transforms/Utils/ValueMapper.h"
+#include "llvm/IRReader/IRReader.h"
+#include "llvm/Linker/Linker.h"
+#include "llvm/Support/SourceMgr.h"
+#include "llvm/Support/FileSystem.h"
+#include "llvm/IR/Attributes.h"
+#include "llvm-c/Core.h"
+#include "SupportVISC/VISCTimer.h"
+#include "SupportVISC/DFG2LLVM.h"
+#include "SupportVISC/VISCUtils.h"
#include "llvm/IR/IRPrintingPasses.h"
#include "llvm/IR/LegacyPassManager.h"
-#include "llvm/IR/UseListOrder.h"
#include "llvm/Support/ToolOutputFile.h"
+#include "llvm/IR/UseListOrder.h"
+
#include <sstream>
using namespace llvm;
using namespace builddfg;
using namespace dfg2llvm;
-using namespace hpvmUtils;
+using namespace viscUtils;
-// HPVM Command line option to use timer or not
-static cl::opt<bool> HPVMTimer_NVPTX("hpvm-timers-ptx",
- cl::desc("Enable hpvm timers"));
+// VISC Command line option to use timer or not
+static cl::opt<bool>
+VISCTimer_NVPTX("visc-timers-ptx", cl::desc("Enable visc timers"));
namespace {
// Helper class declarations
@@ -56,88 +57,94 @@ namespace {
// in bytes. Would have preferred to use tuple but support not yet available
class OutputPtr {
public:
- OutputPtr(Value *_h_ptr, Value *_d_ptr, Value *_bytes)
- : h_ptr(_h_ptr), d_ptr(_d_ptr), bytes(_bytes) {}
+ OutputPtr(Value* _h_ptr, Value* _d_ptr, Value* _bytes)
+ : h_ptr(_h_ptr), d_ptr(_d_ptr), bytes(_bytes) {}
- Value *h_ptr;
- Value *d_ptr;
- Value *bytes;
+ Value* h_ptr;
+ Value* d_ptr;
+ Value* bytes;
};
// Class to maintain important kernel info required for generating runtime
// calls
class Kernel {
public:
- Kernel(
- Function *_KF, DFLeafNode *_KLeafNode,
- std::map<unsigned, unsigned> _inArgMap = std::map<unsigned, unsigned>(),
- std::map<unsigned, std::pair<Value *, unsigned>> _sharedInArgMap =
- std::map<unsigned, std::pair<Value *, unsigned>>(),
- std::vector<unsigned> _outArgMap = std::vector<unsigned>(),
- unsigned _gridDim = 0,
- std::vector<Value *> _globalWGSize = std::vector<Value *>(),
- unsigned _blockDim = 0,
- std::vector<Value *> _localWGSize = std::vector<Value *>())
- : KernelFunction(_KF), KernelLeafNode(_KLeafNode), inArgMap(_inArgMap),
- sharedInArgMap(_sharedInArgMap), outArgMap(_outArgMap),
- gridDim(_gridDim), globalWGSize(_globalWGSize), blockDim(_blockDim),
- localWGSize(_localWGSize) {
-
- assert(gridDim == globalWGSize.size() &&
- "gridDim should be same as the size of vector globalWGSize");
- assert(blockDim == localWGSize.size() &&
- "blockDim should be same as the size of vector localWGSize");
+ Kernel(Function* _KF, DFLeafNode* _KLeafNode, std::map<unsigned, unsigned> _inArgMap =
+ std::map<unsigned, unsigned>(),
+ std::map<unsigned, std::pair<Value*, unsigned> > _sharedInArgMap =
+ std::map<unsigned, std::pair<Value*, unsigned> >(),
+ std::vector<unsigned> _outArgMap = std::vector<unsigned>(),
+ unsigned _gridDim = 0, std::vector<Value*> _globalWGSize = std::vector<Value*>(),
+ unsigned _blockDim = 0, std::vector<Value*> _localWGSize = std::vector<Value*>())
+ : KernelFunction(_KF), KernelLeafNode(_KLeafNode), inArgMap(_inArgMap),
+ sharedInArgMap(_sharedInArgMap), outArgMap(_outArgMap), gridDim(_gridDim),
+ globalWGSize(_globalWGSize), blockDim(_blockDim), localWGSize(_localWGSize) {
+
+ assert(gridDim == globalWGSize.size()
+ && "gridDim should be same as the size of vector globalWGSize");
+ assert(blockDim == localWGSize.size()
+ && "blockDim should be same as the size of vector localWGSize");
}
- Function *KernelFunction;
- DFLeafNode *KernelLeafNode;
+ Function* KernelFunction;
+ DFLeafNode* KernelLeafNode;
std::map<unsigned, unsigned> inArgMap;
// Map for shared memory arguments
- std::map<unsigned, std::pair<Value *, unsigned>> sharedInArgMap;
+ std::map<unsigned, std::pair<Value*, unsigned> > sharedInArgMap;
// Fields for (potential) allocation node
- DFLeafNode *AllocationNode;
- Function *AllocationFunction;
+ DFLeafNode* AllocationNode;
+ Function* AllocationFunction;
std::map<unsigned, unsigned> allocInArgMap;
std::vector<unsigned> outArgMap;
unsigned gridDim;
- std::vector<Value *> globalWGSize;
+ std::vector<Value*> globalWGSize;
unsigned blockDim;
- std::vector<Value *> localWGSize;
+ std::vector<Value*> localWGSize;
std::vector<int> localDimMap;
- std::map<unsigned, unsigned> &getInArgMap() { return inArgMap; }
- void setInArgMap(std::map<unsigned, unsigned> map) { inArgMap = map; }
+ std::map<unsigned, unsigned> &getInArgMap() {
+ return inArgMap;
+ }
+ void setInArgMap(std::map<unsigned, unsigned> map) {
+ inArgMap = map;
+ }
- std::map<unsigned, std::pair<Value *, unsigned>> &getSharedInArgMap() {
+ std::map<unsigned, std::pair<Value*, unsigned> > &getSharedInArgMap() {
return sharedInArgMap;
}
- void setSharedInArgMap(std::map<unsigned, std::pair<Value *, unsigned>> map) {
+ void setSharedInArgMap(std::map<unsigned, std::pair<Value*, unsigned> > map) {
sharedInArgMap = map;
}
- std::vector<unsigned> &getOutArgMap() { return outArgMap; }
- void setOutArgMap(std::vector<unsigned> map) { outArgMap = map; }
+ std::vector<unsigned> &getOutArgMap() {
+ return outArgMap;
+ }
+ void setOutArgMap(std::vector<unsigned> map) {
+ outArgMap = map;
+ }
- void setLocalWGSize(std::vector<Value *> V) { localWGSize = V; }
+ void setLocalWGSize(std::vector<Value*> V) {
+ localWGSize = V;
+ }
- bool hasLocalWG() const { return blockDim != 0; }
+ bool hasLocalWG() const {
+ return blockDim != 0;
+ }
};
// Helper function declarations
-static bool canBePromoted(Argument *arg, Function *F);
-static void getExecuteNodeParams(Module &M, Value *&, Value *&, Value *&,
- Kernel *, ValueToValueMapTy &, Instruction *);
-static Value *genWorkGroupPtr(Module &M, std::vector<Value *>,
- ValueToValueMapTy &, Instruction *,
- const Twine &WGName = "WGSize");
-static std::string getPTXFilename(const Module &);
-static std::string getFilenameFromModule(const Module &M);
+static bool canBePromoted(Argument* arg, Function* F);
+static void getExecuteNodeParams(Module &M, Value* &, Value* &, Value* &, Kernel*,
+ ValueToValueMapTy&, Instruction*);
+static Value* genWorkGroupPtr(Module &M, std::vector<Value*>, ValueToValueMapTy&,
+ Instruction*, const Twine& WGName = "WGSize");
+static std::string getPTXFilename(const Module&);
+static std::string getFilenameFromModule(const Module& M);
static void changeDataLayout(Module &);
static void changeTargetTriple(Module &);
static void findReturnInst(Function *, std::vector<ReturnInst *> &);
-static void findIntrinsicInst(Function *, Intrinsic::ID,
- std::vector<IntrinsicInst *> &);
+static void findIntrinsicInst(Function *, Intrinsic::ID, std::vector<IntrinsicInst *> &);
static AtomicRMWInst::BinOp getAtomicOp(Intrinsic::ID);
static std::string getAtomicOpName(Intrinsic::ID);
@@ -147,6 +154,7 @@ struct DFG2LLVM_NVPTX : public DFG2LLVM {
DFG2LLVM_NVPTX() : DFG2LLVM(ID) {}
private:
+
public:
bool runOnModule(Module &M);
};
@@ -155,60 +163,57 @@ public:
class CGT_NVPTX : public CodeGenTraversal {
private:
- // Member variables
+ //Member variables
std::unique_ptr<Module> KernelM;
- DFNode *KernelLaunchNode = NULL;
- Kernel *kernel;
-
- // HPVM Runtime API
- FunctionCallee llvm_hpvm_ocl_launch;
- FunctionCallee llvm_hpvm_ocl_wait;
- FunctionCallee llvm_hpvm_ocl_initContext;
- FunctionCallee llvm_hpvm_ocl_clearContext;
- FunctionCallee llvm_hpvm_ocl_argument_shared;
- FunctionCallee llvm_hpvm_ocl_argument_scalar;
- FunctionCallee llvm_hpvm_ocl_argument_ptr;
- FunctionCallee llvm_hpvm_ocl_output_ptr;
- FunctionCallee llvm_hpvm_ocl_free;
- FunctionCallee llvm_hpvm_ocl_getOutput;
- FunctionCallee llvm_hpvm_ocl_executeNode;
-
- // Functions
+ DFNode* KernelLaunchNode = NULL;
+ Kernel* kernel;
+
+ // VISC Runtime API
+ FunctionCallee llvm_visc_ocl_launch;
+ FunctionCallee llvm_visc_ocl_wait;
+ FunctionCallee llvm_visc_ocl_initContext;
+ FunctionCallee llvm_visc_ocl_clearContext;
+ FunctionCallee llvm_visc_ocl_argument_shared;
+ FunctionCallee llvm_visc_ocl_argument_scalar;
+ FunctionCallee llvm_visc_ocl_argument_ptr;
+ FunctionCallee llvm_visc_ocl_output_ptr;
+ FunctionCallee llvm_visc_ocl_free;
+ FunctionCallee llvm_visc_ocl_getOutput;
+ FunctionCallee llvm_visc_ocl_executeNode;
+
+ //Functions
std::string getKernelsModuleName(Module &M);
- void fixValueAddrspace(Value *V, unsigned addrspace);
- std::vector<unsigned> globalToConstantMemoryOpt(std::vector<unsigned> *,
- Function *);
- Function *changeArgAddrspace(Function *F, std::vector<unsigned> &Ags,
- unsigned i);
- void addCLMetadata(Function *F);
- Function *transformFunctionToVoid(Function *F);
- void insertRuntimeCalls(DFInternalNode *N, Kernel *K, const Twine &FileName);
+ void fixValueAddrspace(Value* V, unsigned addrspace);
+ std::vector<unsigned> globalToConstantMemoryOpt(std::vector<unsigned>*, Function*);
+ Function* changeArgAddrspace(Function* F, std::vector<unsigned> &Ags, unsigned i);
+ void addCLMetadata(Function* F);
+ Function* transformFunctionToVoid(Function* F);
+ void insertRuntimeCalls(DFInternalNode* N, Kernel* K, const Twine& FileName);
// Virtual Functions
void init() {
- HPVMTimer = HPVMTimer_NVPTX;
+ VISCTimer = VISCTimer_NVPTX;
TargetName = "NVPTX";
}
void initRuntimeAPI();
- void codeGen(DFInternalNode *N);
- void codeGen(DFLeafNode *N);
+ void codeGen(DFInternalNode* N);
+ void codeGen(DFLeafNode* N);
public:
+
// Constructor
- CGT_NVPTX(Module &_M, BuildDFG &_DFG)
- : CodeGenTraversal(_M, _DFG), KernelM(CloneModule(_M)) {
+ CGT_NVPTX(Module &_M, BuildDFG &_DFG) : CodeGenTraversal(_M, _DFG), KernelM(CloneModule(_M)) {
init();
initRuntimeAPI();
- DEBUG(errs() << "Old module pointer: " << &_M << "\n");
- DEBUG(errs() << "New module pointer: " << KernelM.get() << "\n");
+ errs() << "Old module pointer: " << &_M << "\n";
+ errs() << "New module pointer: " << KernelM.get() << "\n";
- // Copying instead of creating new, in order to preserve required info
- // (metadata) Remove functions, global variables and aliases
- std::vector<GlobalVariable *> GVVect;
+ // Copying instead of creating new, in order to preserve required info (metadata)
+ // Remove functions, global variables and aliases
+ std::vector<GlobalVariable*> GVVect;
for (Module::global_iterator mi = KernelM->global_begin(),
- me = KernelM->global_end();
- (mi != me); ++mi) {
- GlobalVariable *GV = &*mi;
+ me = KernelM->global_end(); (mi != me); ++mi) {
+ GlobalVariable* GV = &*mi;
GVVect.push_back(GV);
}
for (auto *GV : GVVect) {
@@ -216,10 +221,10 @@ public:
GV->eraseFromParent();
}
- std::vector<Function *> FuncVect;
- for (Module::iterator mi = KernelM->begin(), me = KernelM->end();
- (mi != me); ++mi) {
- Function *F = &*mi;
+ std::vector<Function*> FuncVect;
+ for (Module::iterator mi = KernelM->begin(),
+ me = KernelM->end(); (mi != me); ++mi) {
+ Function* F = &*mi;
FuncVect.push_back(F);
}
for (auto *F : FuncVect) {
@@ -227,11 +232,10 @@ public:
F->eraseFromParent();
}
- std::vector<GlobalAlias *> GAVect;
+ std::vector<GlobalAlias*> GAVect;
for (Module::alias_iterator mi = KernelM->alias_begin(),
- me = KernelM->alias_end();
- (mi != me); ++mi) {
- GlobalAlias *GA = &*mi;
+ me = KernelM->alias_end(); (mi != me); ++mi) {
+ GlobalAlias* GA = &*mi;
GAVect.push_back(GA);
}
for (auto *GA : GAVect) {
@@ -242,69 +246,73 @@ public:
changeDataLayout(*KernelM);
changeTargetTriple(*KernelM);
+
DEBUG(errs() << *KernelM);
+
}
void writeKernelsModule();
};
-// Initialize the HPVM runtime API. This makes it easier to insert these calls
+// Initialize the VISC runtime API. This makes it easier to insert these calls
void CGT_NVPTX::initRuntimeAPI() {
// Load Runtime API Module
SMDiagnostic Err;
- char *LLVM_SRC_ROOT = getenv("LLVM_SRC_ROOT");
+ char* LLVM_SRC_ROOT = getenv("LLVM_SRC_ROOT");
assert(LLVM_SRC_ROOT != NULL && "Define LLVM_SRC_ROOT environment variable!");
Twine llvmSrcRoot = LLVM_SRC_ROOT;
- Twine runtimeAPI =
- llvmSrcRoot + "/../build/tools/hpvm/projects/hpvm-rt/hpvm-rt.bc";
+ Twine runtimeAPI = llvmSrcRoot + "/../build/tools/hpvm/projects/visc-rt/visc-rt.bc";
runtimeModule = parseIRFile(runtimeAPI.str(), Err, M.getContext());
- if (runtimeModule == nullptr) {
+ if(runtimeModule == nullptr) {
DEBUG(errs() << Err.getMessage() << " " << runtimeAPI << "\n");
assert(false && "couldn't parse runtime");
- } else
- DEBUG(errs() << "Successfully loaded hpvm-rt API module\n");
+ }
+ else
+ DEBUG(errs() << "Successfully loaded visc-rt API module\n");
// Get or insert the global declarations for launch/wait functions
- DECLARE(llvm_hpvm_ocl_launch);
- DECLARE(llvm_hpvm_ocl_wait);
- DECLARE(llvm_hpvm_ocl_initContext);
- DECLARE(llvm_hpvm_ocl_clearContext);
- DECLARE(llvm_hpvm_ocl_argument_shared);
- DECLARE(llvm_hpvm_ocl_argument_scalar);
- DECLARE(llvm_hpvm_ocl_argument_ptr);
- DECLARE(llvm_hpvm_ocl_output_ptr);
- DECLARE(llvm_hpvm_ocl_free);
- DECLARE(llvm_hpvm_ocl_getOutput);
- DECLARE(llvm_hpvm_ocl_executeNode);
+ DECLARE(llvm_visc_ocl_launch);
+ DECLARE(llvm_visc_ocl_wait);
+ DECLARE(llvm_visc_ocl_initContext);
+ DECLARE(llvm_visc_ocl_clearContext);
+ DECLARE(llvm_visc_ocl_argument_shared);
+ DECLARE(llvm_visc_ocl_argument_scalar);
+ DECLARE(llvm_visc_ocl_argument_ptr);
+ DECLARE(llvm_visc_ocl_output_ptr);
+ DECLARE(llvm_visc_ocl_free);
+ DECLARE(llvm_visc_ocl_getOutput);
+ DECLARE(llvm_visc_ocl_executeNode);
// Get or insert timerAPI functions as well if you plan to use timers
initTimerAPI();
// Insert init context in main
DEBUG(errs() << "Gen Code to initialize NVPTX Timer\n");
- Function *VI = M.getFunction("llvm.hpvm.init");
- assert(VI->getNumUses() == 1 && "__hpvm__init should only be used once");
+ Function* VI = M.getFunction("llvm.visc.init");
+ assert(VI->getNumUses() == 1 && "__visc__init should only be used once");
InitCall = cast<Instruction>(*VI->user_begin());
initializeTimerSet(InitCall);
- switchToTimer(hpvm_TimerID_INIT_CTX, InitCall);
- CallInst::Create(llvm_hpvm_ocl_initContext,
- ArrayRef<Value *>(getTargetID(M, hpvm::GPU_TARGET)), "",
- InitCall);
- switchToTimer(hpvm_TimerID_NONE, InitCall);
+ switchToTimer(visc_TimerID_INIT_CTX, InitCall);
+ CallInst::Create(llvm_visc_ocl_initContext,
+ ArrayRef<Value*>(getTargetID(M, visc::GPU_TARGET)),
+ "", InitCall);
+ switchToTimer(visc_TimerID_NONE, InitCall);
- // Insert print instruction at hpvm exit
+ // Insert print instruction at visc exit
DEBUG(errs() << "Gen Code to print NVPTX Timer\n");
- Function *VC = M.getFunction("llvm.hpvm.cleanup");
+ Function* VC = M.getFunction("llvm.visc.cleanup");
DEBUG(errs() << *VC << "\n");
- assert(VC->getNumUses() == 1 && "__hpvm__clear should only be used once");
+ assert(VC->getNumUses() == 1 && "__visc__clear should only be used once");
CleanupCall = cast<Instruction>(*VC->user_begin());
printTimerSet(CleanupCall);
+
+
}
// Generate Code to call the kernel
@@ -312,37 +320,36 @@ void CGT_NVPTX::initRuntimeAPI() {
// used to generate a function to associate with this leaf node. The function
// is responsible for all the memory allocation/transfer and invoking the
// kernel call on the device
-void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
- const Twine &FileName) {
+void CGT_NVPTX::insertRuntimeCalls(DFInternalNode* N, Kernel* K, const Twine& FileName) {
// Check if clone already exists. If it does, it means we have visited this
// function before.
- // assert(N->getGenFunc() == NULL && "Code already generated for this node");
+// assert(N->getGenFunc() == NULL && "Code already generated for this node");
- assert(N->getGenFuncForTarget(hpvm::GPU_TARGET) == NULL &&
+ assert(N->getGenFuncForTarget(visc::GPU_TARGET) == NULL &&
"Code already generated for this node");
// Useful values
- Value *True = ConstantInt::get(Type::getInt1Ty(M.getContext()), 1);
- Value *False = ConstantInt::get(Type::getInt1Ty(M.getContext()), 0);
+ Value* True = ConstantInt::get(Type::getInt1Ty(M.getContext()), 1);
+ Value* False = ConstantInt::get(Type::getInt1Ty(M.getContext()), 0);
// If kernel struct has not been initialized with kernel function, then fail
assert(K != NULL && "No kernel found!!");
DEBUG(errs() << "Generating kernel call code\n");
- Function *F = N->getFuncPointer();
+ Function* F = N->getFuncPointer();
+
// Create of clone of F with no instructions. Only the type is the same as F
// without the extra arguments.
- Function *F_X86;
+ Function* F_X86;
// Clone the function, if we are seeing this function for the first time. We
// only need a clone in terms of type.
ValueToValueMapTy VMap;
// Create new function with the same type
- F_X86 =
- Function::Create(F->getFunctionType(), F->getLinkage(), F->getName(), &M);
+ F_X86 = Function::Create(F->getFunctionType(), F->getLinkage(), F->getName(), &M);
// Loop over the arguments, copying the names of arguments over.
Function::arg_iterator dest_iterator = F_X86->arg_begin();
@@ -355,25 +362,26 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
// Add a basic block to this empty function
BasicBlock *BB = BasicBlock::Create(M.getContext(), "entry", F_X86);
- ReturnInst *RI = ReturnInst::Create(
- M.getContext(), UndefValue::get(F_X86->getReturnType()), BB);
+ ReturnInst* RI = ReturnInst::Create(M.getContext(),
+ UndefValue::get(F_X86->getReturnType()), BB);
// FIXME: Adding Index and Dim arguments are probably not required except
// for consistency purpose (DFG2LLVM_X86 does assume that all leaf nodes do
// have those arguments)
// Add Index and Dim arguments except for the root node
- if (!N->isRoot() && !N->getParent()->isChildGraphStreaming())
+ if(!N->isRoot() && !N->getParent()->isChildGraphStreaming())
F_X86 = addIdxDimArgs(F_X86);
BB = &*F_X86->begin();
RI = cast<ReturnInst>(BB->getTerminator());
- // Add the generated function info to DFNode
- // N->setGenFunc(F_X86, hpvm::CPU_TARGET);
- N->addGenFunc(F_X86, hpvm::GPU_TARGET, true);
- DEBUG(errs() << "Added GPUGenFunc: " << F_X86->getName() << " for node "
- << N->getFuncPointer()->getName() << "\n");
+ //Add the generated function info to DFNode
+// N->setGenFunc(F_X86, visc::CPU_TARGET);
+ N->addGenFunc(F_X86, visc::GPU_TARGET, true);
+ errs() << "Added GPUGenFunc: " << F_X86->getName() << " for node "
+ << N->getFuncPointer()->getName() << "\n";
+
// Loop over the arguments, to create the VMap
dest_iterator = F_X86->arg_begin();
@@ -406,53 +414,51 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
break;
}
- assert(C->isDummyNode() == false && "Internal Node only contains dummy
- nodes!");
+ assert(C->isDummyNode() == false && "Internal Node only contains dummy nodes!");
Function* CF = C->getFuncPointer();
*/
- Function *KF = K->KernelLeafNode->getFuncPointer();
+ Function* KF = K->KernelLeafNode->getFuncPointer();
// Initialize context
- // DEBUG(errs() << "Initializing context" << "\n");
- // CallInst::Create(llvm_hpvm_ocl_initContext, None, "", RI);
+ //DEBUG(errs() << "Initializing context" << "\n");
+ //CallInst::Create(llvm_visc_ocl_initContext, None, "", RI);
- DEBUG(errs() << "Initializing commandQ"
- << "\n");
+ DEBUG(errs() << "Initializing commandQ" << "\n");
// Initialize command queue
- switchToTimer(hpvm_TimerID_SETUP, InitCall);
- Value *fileStr = getStringPointer(FileName, InitCall, "Filename");
+ switchToTimer(visc_TimerID_SETUP, InitCall);
+ Value* fileStr = getStringPointer(FileName, InitCall, "Filename");
DEBUG(errs() << "Kernel Filename constant: " << *fileStr << "\n");
- DEBUG(errs() << "Generating code for kernel - "
- << K->KernelFunction->getName() << "\n");
- Value *kernelStr =
- getStringPointer(K->KernelFunction->getName(), InitCall, "KernelName");
-
- Value *LaunchInstArgs[] = {fileStr, kernelStr};
-
- DEBUG(errs() << "Inserting launch call"
- << "\n");
- CallInst *NVPTX_Ctx = CallInst::Create(llvm_hpvm_ocl_launch,
- ArrayRef<Value *>(LaunchInstArgs, 2),
- "graph" + KF->getName(), InitCall);
+ DEBUG(errs() << "Generating code for kernel - " << K->KernelFunction->getName()<< "\n");
+ Value* kernelStr = getStringPointer(K->KernelFunction->getName(), InitCall,"KernelName");
+
+ Value* LaunchInstArgs[] = {fileStr, kernelStr};
+
+ DEBUG(errs() << "Inserting launch call" << "\n");
+ CallInst* NVPTX_Ctx = CallInst::Create(llvm_visc_ocl_launch,
+ ArrayRef<Value*>(LaunchInstArgs, 2),
+ "graph"+KF->getName(),
+ InitCall);
DEBUG(errs() << *NVPTX_Ctx << "\n");
- GraphIDAddr = new GlobalVariable(M, NVPTX_Ctx->getType(), false,
+ GraphIDAddr = new GlobalVariable(M,
+ NVPTX_Ctx->getType(),
+ false,
GlobalValue::CommonLinkage,
Constant::getNullValue(NVPTX_Ctx->getType()),
- "graph" + KF->getName() + ".addr");
+ "graph"+KF->getName()+".addr");
DEBUG(errs() << "Store at: " << *GraphIDAddr << "\n");
- StoreInst *SI = new StoreInst(NVPTX_Ctx, GraphIDAddr, InitCall);
+ StoreInst* SI = new StoreInst(NVPTX_Ctx, GraphIDAddr, InitCall);
DEBUG(errs() << *SI << "\n");
- switchToTimer(hpvm_TimerID_NONE, InitCall);
- switchToTimer(hpvm_TimerID_SETUP, RI);
- Value *GraphID = new LoadInst(GraphIDAddr, "graph." + KF->getName(), RI);
+ switchToTimer(visc_TimerID_NONE, InitCall);
+ switchToTimer(visc_TimerID_SETUP, RI);
+ Value* GraphID = new LoadInst(GraphIDAddr, "graph."+KF->getName(), RI);
- // Iterate over the required input edges of the node and use the hpvm-rt API
+ // Iterate over the required input edges of the node and use the visc-rt API
// to set inputs
- DEBUG(errs() << "Iterate over input edges of node and insert hpvm api\n");
+ DEBUG(errs() << "Iterate over input edges of node and insert visc api\n");
std::vector<OutputPtr> OutputPointers;
- // Vector to hold the device memory object that need to be cleared before we
- // release context
- std::vector<Value *> DevicePointers;
+ // Vector to hold the device memory object that need to be cleared before we release
+ // context
+ std::vector<Value*> DevicePointers;
std::map<unsigned, unsigned> &kernelInArgMap = K->getInArgMap();
/*
@@ -464,134 +470,133 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
*/
- for (auto &InArgMapPair : kernelInArgMap) {
+ for(auto &InArgMapPair : kernelInArgMap) {
unsigned i = InArgMapPair.first;
- Value *inputVal = getArgumentAt(F_X86, InArgMapPair.second);
- DEBUG(errs() << "\tArgument " << i << " = " << *inputVal << "\n");
+ Value* inputVal = getArgumentAt(F_X86, InArgMapPair.second);
+ DEBUG(errs() << "\tArgument "<< i<< " = " << *inputVal << "\n");
// input value has been obtained.
// Check if input is a scalar value or a pointer operand
// For scalar values such as int, float, etc. the size is simply the size of
// type on target machine, but for pointers, the size of data would be the
// next integer argument
- if (inputVal->getType()->isPointerTy()) {
+ if(inputVal->getType()->isPointerTy()) {
- switchToTimer(hpvm_TimerID_COPY_PTR, RI);
+ switchToTimer(visc_TimerID_COPY_PTR, RI);
// Pointer Input
// CheckAttribute
- Value *isOutput = (hasAttribute(KF, i, Attribute::Out)) ? True : False;
- Value *isInput = ((hasAttribute(KF, i, Attribute::Out)) &&
- !(hasAttribute(KF, i, Attribute::In)))
- ? False
- : True;
-
- Argument *A = getArgumentAt(KF, i);
- if (isOutput == True) {
+ Value* isOutput = (hasAttribute(KF, i, Attribute::Out))? True : False;
+ Value* isInput = ((hasAttribute(KF, i, Attribute::Out))
+ && !(hasAttribute(KF, i, Attribute::In)))? False : True;
+
+ Argument* A = getArgumentAt(KF, i);
+ if(isOutput == True) {
DEBUG(errs() << *A << " is an OUTPUT argument\n");
}
- if (isInput == True) {
+ if(isInput == True) {
DEBUG(errs() << *A << " is an INPUT argument\n");
}
- Value *inputValI8Ptr = CastInst::CreatePointerCast(
- inputVal, Type::getInt8PtrTy(M.getContext()),
- inputVal->getName() + ".i8ptr", RI);
+
+ Value* inputValI8Ptr = CastInst::CreatePointerCast(inputVal,
+ Type::getInt8PtrTy(M.getContext()),
+ inputVal->getName()+".i8ptr",
+ RI);
// Assert that the pointer argument size (next argument) is in the map
- assert(kernelInArgMap.find(i + 1) != kernelInArgMap.end());
-
- Value *inputSize = getArgumentAt(F_X86, kernelInArgMap[i + 1]);
- assert(
- inputSize->getType() == Type::getInt64Ty(M.getContext()) &&
- "Pointer type input must always be followed by size (integer type)");
- Value *setInputArgs[] = {
- GraphID,
- inputValI8Ptr,
- ConstantInt::get(Type::getInt32Ty(M.getContext()), i),
- inputSize,
- isInput,
- isOutput};
- Value *d_ptr =
- CallInst::Create(llvm_hpvm_ocl_argument_ptr,
- ArrayRef<Value *>(setInputArgs, 6), "", RI);
+ assert(kernelInArgMap.find(i+1) != kernelInArgMap.end());
+
+ Value* inputSize = getArgumentAt(F_X86, kernelInArgMap[i+1]);
+ assert(inputSize->getType() == Type::getInt64Ty(M.getContext())
+ && "Pointer type input must always be followed by size (integer type)");
+ Value* setInputArgs[] = {GraphID,
+ inputValI8Ptr,
+ ConstantInt::get(Type::getInt32Ty(M.getContext()),i),
+ inputSize,
+ isInput,
+ isOutput
+ };
+ Value* d_ptr = CallInst::Create(llvm_visc_ocl_argument_ptr,
+ ArrayRef<Value*>(setInputArgs, 6), "", RI);
DevicePointers.push_back(d_ptr);
// If this has out attribute, store the returned device pointer in
// memory to read device memory later
- if (isOutput == True)
- OutputPointers.push_back(OutputPtr(inputValI8Ptr, d_ptr, inputSize));
- } else {
- switchToTimer(hpvm_TimerID_COPY_SCALAR, RI);
+ if(isOutput == True) OutputPointers.push_back(OutputPtr(inputValI8Ptr, d_ptr, inputSize));
+ }
+ else {
+ switchToTimer(visc_TimerID_COPY_SCALAR, RI);
// Scalar Input
// Store the scalar value on stack and then pass the pointer to its
// location
- AllocaInst *inputValPtr = new AllocaInst(
- inputVal->getType(), 0, inputVal->getName() + ".ptr", RI);
- StoreInst *SI = new StoreInst(inputVal, inputValPtr, RI);
-
- Value *inputValI8Ptr = CastInst::CreatePointerCast(
- inputValPtr, Type::getInt8PtrTy(M.getContext()),
- inputVal->getName() + ".i8ptr", RI);
-
- Value *setInputArgs[] = {
- GraphID, inputValI8Ptr,
- ConstantInt::get(Type::getInt32Ty(M.getContext()), i),
- ConstantExpr::getSizeOf(inputVal->getType())};
- CallInst::Create(llvm_hpvm_ocl_argument_scalar,
- ArrayRef<Value *>(setInputArgs, 4), "", RI);
+ AllocaInst* inputValPtr = new AllocaInst(inputVal->getType(), 0, inputVal->getName()+".ptr", RI);
+ StoreInst* SI = new StoreInst(inputVal, inputValPtr, RI);
+
+ Value* inputValI8Ptr = CastInst::CreatePointerCast(inputValPtr,
+ Type::getInt8PtrTy(M.getContext()),
+ inputVal->getName()+".i8ptr",
+ RI);
+
+ Value* setInputArgs[] = {GraphID,
+ inputValI8Ptr,
+ ConstantInt::get(Type::getInt32Ty(M.getContext()),i),
+ ConstantExpr::getSizeOf(inputVal->getType())
+ };
+ CallInst::Create(llvm_visc_ocl_argument_scalar,
+ ArrayRef<Value*>(setInputArgs, 4), "", RI);
}
}
- DEBUG(
- errs() << "Setup shared memory arguments of node and insert hpvm api\n");
+ DEBUG(errs() << "Setup shared memory arguments of node and insert visc api\n");
// Check to see if all the allocation sizes are constant (determined
// statically)
bool constSizes = true;
- for (auto &e : K->getSharedInArgMap()) {
+ for (auto& e: K->getSharedInArgMap()) {
constSizes &= isa<Constant>(e.second.first);
}
// If the sizes are all constant
if (constSizes) {
- for (auto &e : K->getSharedInArgMap()) {
+ for (auto& e: K->getSharedInArgMap()) {
unsigned argNum = e.first;
- Value *allocSize = e.second.first;
+ Value* allocSize = e.second.first;
- DEBUG(errs() << "\tLocal Memory at " << argNum
- << ", size = " << *allocSize << "\n");
+ DEBUG(errs() << "\tLocal Memory at "<< argNum << ", size = " << *allocSize << "\n");
if (KF->getFunctionType()->getParamType(argNum)->isPointerTy()) {
// Shared memory ptr argument - scalar at size position
- switchToTimer(hpvm_TimerID_COPY_SCALAR, RI);
+ switchToTimer(visc_TimerID_COPY_SCALAR, RI);
- assert(isa<Constant>(allocSize) &&
- "Constant shared memory size is expected");
+ assert(isa<Constant>(allocSize) && "Constant shared memory size is expected");
- Value *setInputArgs[] = {
- GraphID, ConstantInt::get(Type::getInt32Ty(M.getContext()), argNum),
- allocSize};
- CallInst::Create(llvm_hpvm_ocl_argument_shared,
- ArrayRef<Value *>(setInputArgs, 3), "", RI);
- } else {
+ Value* setInputArgs[] = {GraphID,
+ ConstantInt::get(Type::getInt32Ty(M.getContext()),argNum),
+ allocSize
+ };
+ CallInst::Create(llvm_visc_ocl_argument_shared,
+ ArrayRef<Value*>(setInputArgs, 3), "", RI);
+ }
+ else {
// Sharem memory size argument - scalar at address position
- switchToTimer(hpvm_TimerID_COPY_SCALAR, RI);
+ switchToTimer(visc_TimerID_COPY_SCALAR, RI);
// Store the scalar value on stack and then pass the pointer to its
// location
- AllocaInst *allocSizePtr =
- new AllocaInst(allocSize->getType(), 0,
- allocSize->getName() + ".sharedMem.ptr", RI);
- StoreInst *SI = new StoreInst(allocSize, allocSizePtr, RI);
-
- Value *allocSizeI8Ptr = CastInst::CreatePointerCast(
- allocSizePtr, Type::getInt8PtrTy(M.getContext()),
- allocSize->getName() + ".sharedMem.i8ptr", RI);
-
- Value *setInputArgs[] = {
- GraphID, allocSizeI8Ptr,
- ConstantInt::get(Type::getInt32Ty(M.getContext()), argNum),
- ConstantExpr::getSizeOf(allocSize->getType())};
- CallInst::Create(llvm_hpvm_ocl_argument_scalar,
- ArrayRef<Value *>(setInputArgs, 4), "", RI);
+ AllocaInst* allocSizePtr = new AllocaInst(allocSize->getType(), 0,
+ allocSize->getName()+".sharedMem.ptr", RI);
+ StoreInst* SI = new StoreInst(allocSize, allocSizePtr, RI);
+
+ Value* allocSizeI8Ptr = CastInst::CreatePointerCast(allocSizePtr,
+ Type::getInt8PtrTy(M.getContext()),
+ allocSize->getName()+".sharedMem.i8ptr",
+ RI);
+
+ Value* setInputArgs[] = {GraphID,
+ allocSizeI8Ptr,
+ ConstantInt::get(Type::getInt32Ty(M.getContext()),argNum),
+ ConstantExpr::getSizeOf(allocSize->getType())
+ };
+ CallInst::Create(llvm_visc_ocl_argument_scalar,
+ ArrayRef<Value*>(setInputArgs, 4), "", RI);
}
}
} else {
@@ -612,64 +617,68 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
ExtractValueInstVec.push_back(EI);
}
- for (auto &e : K->getSharedInArgMap()) {
+ for (auto& e: K->getSharedInArgMap()) {
unsigned argNum = e.first;
- Value *allocSize = ExtractValueInstVec[e.second.second / 2];
+ Value* allocSize = ExtractValueInstVec[e.second.second/2];
- DEBUG(errs() << "\tLocal Memory at " << argNum
- << ", size = " << *allocSize << "\n");
+ DEBUG(errs() << "\tLocal Memory at "<< argNum << ", size = " << *allocSize << "\n");
if (KF->getFunctionType()->getParamType(argNum)->isPointerTy()) {
// Shared memory ptr argument - scalar at size position
- switchToTimer(hpvm_TimerID_COPY_SCALAR, RI);
-
- Value *setInputArgs[] = {
- GraphID, ConstantInt::get(Type::getInt32Ty(M.getContext()), argNum),
- allocSize};
- CallInst::Create(llvm_hpvm_ocl_argument_shared,
- ArrayRef<Value *>(setInputArgs, 3), "", RI);
- } else {
+ switchToTimer(visc_TimerID_COPY_SCALAR, RI);
+
+ Value* setInputArgs[] = {GraphID,
+ ConstantInt::get(Type::getInt32Ty(M.getContext()),argNum),
+ allocSize
+ };
+ CallInst::Create(llvm_visc_ocl_argument_shared,
+ ArrayRef<Value*>(setInputArgs, 3), "", RI);
+ }
+ else {
// Sharem memory size argument - scalar at address position
- switchToTimer(hpvm_TimerID_COPY_SCALAR, RI);
+ switchToTimer(visc_TimerID_COPY_SCALAR, RI);
// Store the scalar value on stack and then pass the pointer to its
// location
- AllocaInst *allocSizePtr =
- new AllocaInst(allocSize->getType(), 0,
- allocSize->getName() + ".sharedMem.ptr", RI);
- StoreInst *SI = new StoreInst(allocSize, allocSizePtr, RI);
-
- Value *allocSizeI8Ptr = CastInst::CreatePointerCast(
- allocSizePtr, Type::getInt8PtrTy(M.getContext()),
- allocSize->getName() + ".sharedMem.i8ptr", RI);
-
- Value *setInputArgs[] = {
- GraphID, allocSizeI8Ptr,
- ConstantInt::get(Type::getInt32Ty(M.getContext()), argNum),
- ConstantExpr::getSizeOf(allocSize->getType())};
- CallInst::Create(llvm_hpvm_ocl_argument_scalar,
- ArrayRef<Value *>(setInputArgs, 4), "", RI);
+ AllocaInst* allocSizePtr = new AllocaInst(allocSize->getType(), 0,
+ allocSize->getName()+".sharedMem.ptr", RI);
+ StoreInst* SI = new StoreInst(allocSize, allocSizePtr, RI);
+
+ Value* allocSizeI8Ptr = CastInst::CreatePointerCast(allocSizePtr,
+ Type::getInt8PtrTy(M.getContext()),
+ allocSize->getName()+".sharedMem.i8ptr",
+ RI);
+
+ Value* setInputArgs[] = {GraphID,
+ allocSizeI8Ptr,
+ ConstantInt::get(Type::getInt32Ty(M.getContext()),argNum),
+ ConstantExpr::getSizeOf(allocSize->getType())
+ };
+ CallInst::Create(llvm_visc_ocl_argument_scalar,
+ ArrayRef<Value*>(setInputArgs, 4), "", RI);
}
}
}
- DEBUG(errs() << "Setup output edges of node and insert hpvm api\n");
+
+ DEBUG(errs() << "Setup output edges of node and insert visc api\n");
// Set output if struct is not an empty struct
- StructType *OutputTy = K->KernelLeafNode->getOutputType();
- std::vector<Value *> d_Outputs;
- if (!OutputTy->isEmptyTy()) {
- switchToTimer(hpvm_TimerID_COPY_PTR, RI);
+ StructType* OutputTy = K->KernelLeafNode->getOutputType();
+ std::vector<Value*> d_Outputs;
+ if(!OutputTy->isEmptyTy()) {
+ switchToTimer(visc_TimerID_COPY_PTR, RI);
// Not an empty struct
// Iterate over all elements of the struct and put them in
- for (unsigned i = 0; i < OutputTy->getNumElements(); i++) {
- unsigned outputIndex = KF->getFunctionType()->getNumParams() + i;
- Value *setOutputArgs[] = {
- GraphID,
- ConstantInt::get(Type::getInt32Ty(M.getContext()), outputIndex),
- ConstantExpr::getSizeOf(OutputTy->getElementType(i))};
-
- CallInst *d_Output = CallInst::Create(llvm_hpvm_ocl_output_ptr,
- ArrayRef<Value *>(setOutputArgs, 3),
- "d_output." + KF->getName(), RI);
+ for(unsigned i=0; i < OutputTy->getNumElements(); i++) {
+ unsigned outputIndex = KF->getFunctionType()->getNumParams()+i;
+ Value* setOutputArgs[] = {GraphID,
+ ConstantInt::get(Type::getInt32Ty(M.getContext()),outputIndex),
+ ConstantExpr::getSizeOf(OutputTy->getElementType(i))
+ };
+
+ CallInst* d_Output = CallInst::Create(llvm_visc_ocl_output_ptr,
+ ArrayRef<Value*>(setOutputArgs, 3),
+ "d_output."+KF->getName(),
+ RI);
d_Outputs.push_back(d_Output);
}
}
@@ -679,41 +688,50 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
// Allocate size_t[numDims] space on stack. Store the work group sizes and
// pass it as an argument to ExecNode
- switchToTimer(hpvm_TimerID_MISC, RI);
+ switchToTimer(visc_TimerID_MISC, RI);
Value *workDim, *LocalWGPtr, *GlobalWGPtr;
getExecuteNodeParams(M, workDim, LocalWGPtr, GlobalWGPtr, K, VMap, RI);
- switchToTimer(hpvm_TimerID_KERNEL, RI);
- Value *ExecNodeArgs[] = {GraphID, workDim, LocalWGPtr, GlobalWGPtr};
- CallInst *Event = CallInst::Create(llvm_hpvm_ocl_executeNode,
- ArrayRef<Value *>(ExecNodeArgs, 4),
- "event." + KF->getName(), RI);
+ switchToTimer(visc_TimerID_KERNEL, RI);
+ Value* ExecNodeArgs[] = {GraphID,
+ workDim,
+ LocalWGPtr,
+ GlobalWGPtr
+ };
+ CallInst* Event = CallInst::Create(llvm_visc_ocl_executeNode,
+ ArrayRef<Value*>(ExecNodeArgs, 4),
+ "event."+KF->getName(),
+ RI);
DEBUG(errs() << "Execute Node Call: " << *Event << "\n");
// Wait for Kernel to Finish
- CallInst::Create(llvm_hpvm_ocl_wait, ArrayRef<Value *>(GraphID), "", RI);
+ CallInst::Create(llvm_visc_ocl_wait,
+ ArrayRef<Value*>(GraphID),
+ "",
+ RI);
- switchToTimer(hpvm_TimerID_READ_OUTPUT, RI);
+ switchToTimer(visc_TimerID_READ_OUTPUT, RI);
// Read Output Struct if not empty
- if (!OutputTy->isEmptyTy()) {
- std::vector<Value *> h_Outputs;
- Value *KernelOutput = UndefValue::get(OutputTy);
- for (unsigned i = 0; i < OutputTy->getNumElements(); i++) {
- Value *GetOutputArgs[] = {
- GraphID, Constant::getNullValue(Type::getInt8PtrTy(M.getContext())),
- d_Outputs[i], ConstantExpr::getSizeOf(OutputTy->getElementType(i))};
- CallInst *h_Output = CallInst::Create(
- llvm_hpvm_ocl_getOutput, ArrayRef<Value *>(GetOutputArgs, 4),
- "h_output." + KF->getName() + ".addr", RI);
+ if(!OutputTy->isEmptyTy()) {
+ std::vector<Value*>h_Outputs;
+ Value* KernelOutput = UndefValue::get(OutputTy);
+ for(unsigned i=0; i < OutputTy->getNumElements(); i++) {
+ Value* GetOutputArgs[] = {GraphID,
+ Constant::getNullValue(Type::getInt8PtrTy(M.getContext())),
+ d_Outputs[i],
+ ConstantExpr::getSizeOf(OutputTy->getElementType(i))
+ };
+ CallInst* h_Output = CallInst::Create(llvm_visc_ocl_getOutput,
+ ArrayRef<Value*>(GetOutputArgs, 4),
+ "h_output."+KF->getName()+".addr",
+ RI);
// Read each device pointer listed in output struct
// Load the output struct
- CastInst *BI = BitCastInst::CreatePointerCast(
- h_Output, OutputTy->getElementType(i)->getPointerTo(), "output.ptr",
- RI);
-
- Value *OutputElement = new LoadInst(BI, "output." + KF->getName(), RI);
- KernelOutput = InsertValueInst::Create(KernelOutput, OutputElement,
- ArrayRef<unsigned>(i),
- KF->getName() + "output", RI);
+ CastInst* BI = BitCastInst::CreatePointerCast(h_Output,
+ OutputTy->getElementType(i)->getPointerTo(), "output.ptr", RI);
+
+ Value* OutputElement = new LoadInst(BI, "output."+KF->getName(), RI);
+ KernelOutput = InsertValueInst::Create(KernelOutput, OutputElement, ArrayRef<unsigned>(i),
+ KF->getName()+"output", RI);
}
OutputMap[K->KernelLeafNode] = KernelOutput;
}
@@ -728,76 +746,75 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
DEBUG(errs() << "\tTo: " << *output.h_ptr << "\n");
DEBUG(errs() << "\t#bytes: " << *output.bytes << "\n");
- Value* GetOutputArgs[] = {GraphID, output.h_ptr, output.d_ptr,
- output.bytes}; CallInst* CI = CallInst::Create(llvm_hpvm_ocl_getOutput,
+ Value* GetOutputArgs[] = {GraphID, output.h_ptr, output.d_ptr, output.bytes};
+ CallInst* CI = CallInst::Create(llvm_visc_ocl_getOutput,
ArrayRef<Value*>(GetOutputArgs, 4),
"", RI);
}*/
- switchToTimer(hpvm_TimerID_MEM_FREE, RI);
+ switchToTimer(visc_TimerID_MEM_FREE, RI);
// Clear Context and free device memory
- DEBUG(errs() << "Clearing context"
- << "\n");
+ DEBUG(errs() << "Clearing context" << "\n");
// Free Device Memory
- for (auto d_ptr : DevicePointers) {
- CallInst::Create(llvm_hpvm_ocl_free, ArrayRef<Value *>(d_ptr), "", RI);
+ for(auto d_ptr: DevicePointers) {
+ CallInst::Create(llvm_visc_ocl_free, ArrayRef<Value*>(d_ptr), "", RI);
}
- switchToTimer(hpvm_TimerID_CLEAR_CTX, CleanupCall);
+ switchToTimer(visc_TimerID_CLEAR_CTX, CleanupCall);
// Clear Context
- LoadInst *LI = new LoadInst(GraphIDAddr, "", CleanupCall);
- CallInst::Create(llvm_hpvm_ocl_clearContext, ArrayRef<Value *>(LI), "",
- CleanupCall);
- switchToTimer(hpvm_TimerID_NONE, CleanupCall);
+ LoadInst* LI = new LoadInst(GraphIDAddr, "", CleanupCall);
+ CallInst::Create(llvm_visc_ocl_clearContext, ArrayRef<Value*>(LI), "", CleanupCall);
+ switchToTimer(visc_TimerID_NONE, CleanupCall);
- switchToTimer(hpvm_TimerID_MISC, RI);
+ switchToTimer(visc_TimerID_MISC, RI);
DEBUG(errs() << "*** Generating epilogue code for the function****\n");
// Generate code for output bindings
// Get Exit node
- DFNode *C = N->getChildGraph()->getExit();
+ DFNode* C = N->getChildGraph()->getExit();
// Get OutputType of this node
- StructType *OutTy = N->getOutputType();
+ StructType* OutTy = N->getOutputType();
Value *retVal = UndefValue::get(F_X86->getReturnType());
// Find the kernel's output arg map, to use instead of the bindings
std::vector<unsigned> outArgMap = kernel->getOutArgMap();
// Find all the input edges to exit node
- for (unsigned i = 0; i < OutTy->getNumElements(); i++) {
+ for (unsigned i=0; i < OutTy->getNumElements(); i++) {
DEBUG(errs() << "Output Edge " << i << "\n");
// Find the incoming edge at the requested input port
- DFEdge *E = C->getInDFEdgeAt(i);
+ DFEdge* E = C->getInDFEdgeAt(i);
assert(E && "No Binding for output element!");
// Find the Source DFNode associated with the incoming edge
- DFNode *SrcDF = E->getSourceDF();
+ DFNode* SrcDF = E->getSourceDF();
- DEBUG(errs() << "Edge source -- " << SrcDF->getFuncPointer()->getName()
- << "\n");
+ DEBUG(errs() << "Edge source -- " << SrcDF->getFuncPointer()->getName() << "\n");
// If Source DFNode is a dummyNode, edge is from parent. Get the
// argument from argument list of this internal node
- Value *inputVal;
- if (SrcDF->isEntryNode()) {
+ Value* inputVal;
+ if(SrcDF->isEntryNode()) {
inputVal = getArgumentAt(F_X86, i);
- DEBUG(errs() << "Argument " << i << " = " << *inputVal << "\n");
- } else {
+ DEBUG(errs() << "Argument "<< i<< " = " << *inputVal << "\n");
+ }
+ else {
// edge is from a internal node
// Check - code should already be generated for this source dfnode
// FIXME: Since the 2-level kernel code gen has aspecific structure, we
// can assume the SrcDF is same as Kernel Leaf node.
// Use outArgMap to get correct mapping
SrcDF = K->KernelLeafNode;
- assert(OutputMap.count(SrcDF) &&
- "Source node call not found. Dependency violation!");
+ assert(OutputMap.count(SrcDF)
+ && "Source node call not found. Dependency violation!");
// Find Output Value associated with the Source DFNode using OutputMap
- Value *CI = OutputMap[SrcDF];
+ Value* CI = OutputMap[SrcDF];
// Extract element at source position from this call instruction
std::vector<unsigned> IndexList;
// i is the destination of DFEdge E
// Use the mapping instead of the bindings
- // IndexList.push_back(E->getSourcePosition());
+// IndexList.push_back(E->getSourcePosition());
IndexList.push_back(outArgMap[i]);
- DEBUG(errs() << "Going to generate ExtarctVal inst from " << *CI << "\n");
- ExtractValueInst *EI = ExtractValueInst::Create(CI, IndexList, "", RI);
+ DEBUG(errs() << "Going to generate ExtarctVal inst from "<< *CI <<"\n");
+ ExtractValueInst* EI = ExtractValueInst::Create(CI, IndexList,
+ "",RI);
inputVal = EI;
}
std::vector<unsigned> IdxList;
@@ -806,33 +823,31 @@ void CGT_NVPTX::insertRuntimeCalls(DFInternalNode *N, Kernel *K,
}
DEBUG(errs() << "Extracted all\n");
- switchToTimer(hpvm_TimerID_NONE, RI);
+ switchToTimer(visc_TimerID_NONE, RI);
retVal->setName("output");
- ReturnInst *newRI = ReturnInst::Create(F_X86->getContext(), retVal);
+ ReturnInst* newRI = ReturnInst::Create(F_X86->getContext(), retVal);
ReplaceInstWithInst(RI, newRI);
}
+
// Right now, only targeting the one level case. In general, device functions
// can return values so we don't need to change them
-void CGT_NVPTX::codeGen(DFInternalNode *N) {
- DEBUG(errs() << "Inside internal node: " << N->getFuncPointer()->getName()
- << "\n");
- if (KernelLaunchNode == NULL)
- DEBUG(errs() << "No kernel launch node\n");
+void CGT_NVPTX::codeGen(DFInternalNode* N) {
+ errs () << "Inside internal node: " << N->getFuncPointer()->getName() << "\n";
+ if(KernelLaunchNode == NULL)
+ errs () << "No kernel launch node\n";
else {
- DEBUG(errs() << "KernelLaunchNode: "
- << KernelLaunchNode->getFuncPointer()->getName() << "\n");
+ errs() << "KernelLaunchNode: " << KernelLaunchNode->getFuncPointer()->getName() << "\n";
}
if (!KernelLaunchNode) {
- DEBUG(errs()
- << "No code generated (host code for kernel launch complete).\n");
+ DEBUG(errs() << "No code generated (host code for kernel launch complete).\n");
return;
}
if (N == KernelLaunchNode) {
DEBUG(errs() << "Found kernel launch node. Generating host code.\n");
- // TODO
+ //TODO
// Now the remaining nodes to be visited should be ignored
KernelLaunchNode = NULL;
@@ -847,8 +862,7 @@ void CGT_NVPTX::codeGen(DFInternalNode *N) {
// TODO: Structure assumed: one thread node, one allocation node (at most),
// TB node
std::map<unsigned, unsigned> inmapFinal;
- for (std::map<unsigned, unsigned>::iterator ib = inmap2.begin(),
- ie = inmap2.end();
+ for (std::map<unsigned, unsigned>::iterator ib = inmap2.begin(), ie = inmap2.end();
ib != ie; ++ib) {
inmapFinal[ib->first] = inmap1[ib->second];
}
@@ -865,9 +879,8 @@ void CGT_NVPTX::codeGen(DFInternalNode *N) {
// 0 ... outmap2.size()-1
// The limit is the size of outmap2, because this is the number of kernel
// output arguments for which the mapping matters
- // For now, it reasonable to assume that all the kernel arguments are
- // returned, maybe plys some others from other nodes, thus outmap2.size() <=
- // outmap1.size()
+ // For now, it reasonable to assume that all the kernel arguments are returned,
+ // maybe plys some others from other nodes, thus outmap2.size() <= outmap1.size()
for (unsigned i = 0; i < outmap2.size(); i++) {
outmap1[i] = outmap2[outmap1[i]];
}
@@ -875,14 +888,15 @@ void CGT_NVPTX::codeGen(DFInternalNode *N) {
// Track the source of local dimlimits for the kernel
// Dimension limit can either be a constant or an argument of parent
- // function. Since Internal node would no longer exist, we need to insert
- // the localWGSize with values from the parent of N.
- std::vector<Value *> localWGSizeMapped;
+ // function. Since Internal node would no longer exist, we need to insert the
+ // localWGSize with values from the parent of N.
+ std::vector<Value*> localWGSizeMapped;
for (unsigned i = 0; i < kernel->localWGSize.size(); i++) {
if (isa<Constant>(kernel->localWGSize[i])) {
// if constant, use as it is
localWGSizeMapped.push_back(kernel->localWGSize[i]);
- } else if (Argument *Arg = dyn_cast<Argument>(kernel->localWGSize[i])) {
+ }
+ else if (Argument* Arg = dyn_cast<Argument>(kernel->localWGSize[i])) {
// if argument, find the argument location in N. Use InArgMap of N to
// find the source location in Parent of N. Retrieve the argument from
// parent to insert in the vector.
@@ -892,49 +906,46 @@ void CGT_NVPTX::codeGen(DFInternalNode *N) {
assert(N->getInArgMap().find(argNum) != N->getInArgMap().end());
unsigned parentArgNum = N->getInArgMap()[argNum];
- Argument *A =
- getArgumentAt(N->getParent()->getFuncPointer(), parentArgNum);
+ Argument* A = getArgumentAt(N->getParent()->getFuncPointer(), parentArgNum);
localWGSizeMapped.push_back(A);
- } else {
- assert(
- false &&
- "LocalWGsize using value which is neither argument nor constant!");
+ }
+ else {
+ assert(false && "LocalWGsize using value which is neither argument nor constant!");
}
}
// Update localWGSize vector of kernel
kernel->setLocalWGSize(localWGSizeMapped);
}
+
}
-void CGT_NVPTX::codeGen(DFLeafNode *N) {
- DEBUG(errs() << "Inside leaf node: " << N->getFuncPointer()->getName()
- << "\n");
+void CGT_NVPTX::codeGen(DFLeafNode* N) {
+ errs () << "Inside leaf node: " << N->getFuncPointer()->getName() << "\n";
// Skip code generation if it is a dummy node
- if (N->isDummyNode()) {
+ if(N->isDummyNode()) {
DEBUG(errs() << "Skipping dummy node\n");
return;
}
// Skip code generation if it is an allocation node
- if (N->isAllocationNode()) {
+ if(N->isAllocationNode()) {
DEBUG(errs() << "Skipping allocation node\n");
return;
}
// Generate code only if it has the right hint
- // if(!checkPreferredTarget(N, hpvm::GPU_TARGET)) {
- // errs() << "Skipping node: "<< N->getFuncPointer()->getName() << "\n";
- // return;
- // }
- if (!preferredTargetIncludes(N, hpvm::GPU_TARGET)) {
- DEBUG(errs() << "Skipping node: " << N->getFuncPointer()->getName()
- << "\n");
+// if(!checkPreferredTarget(N, visc::GPU_TARGET)) {
+// errs() << "Skipping node: "<< N->getFuncPointer()->getName() << "\n";
+// return;
+// }
+ if(!preferredTargetIncludes(N, visc::GPU_TARGET)) {
+ errs() << "Skipping node: "<< N->getFuncPointer()->getName() << "\n";
return;
}
// Checking which node is the kernel launch
- DFNode *PNode = N->getParent();
+ DFNode* PNode = N->getParent();
int pLevel = PNode->getLevel();
int pReplFactor = PNode->getNumOfDim();
@@ -942,40 +953,42 @@ void CGT_NVPTX::codeGen(DFLeafNode *N) {
// (1) Parent is the top level node i.e., Root of DFG
// OR
// (2) Parent does not have multiple instances
- DEBUG(errs() << "pLevel = " << pLevel << "\n");
- DEBUG(errs() << "pReplFactor = " << pReplFactor << "\n");
+ errs() << "pLevel = " << pLevel << "\n";
+ errs() << "pReplFactor = " << pReplFactor << "\n";
assert((pLevel > 0) && "Root not allowed to be chosen as Kernel Node.");
// Only these options are supported
- enum XLevelHierarchy { ONE_LEVEL, TWO_LEVEL } SelectedHierarchy;
- if (pLevel == 1 || !pReplFactor) {
- DEBUG(errs()
- << "*************** Kernel Gen: 1-Level Hierarchy **************\n");
+ enum XLevelHierarchy{ONE_LEVEL, TWO_LEVEL} SelectedHierarchy;
+ if(pLevel == 1 || !pReplFactor) {
+ errs() << "*************** Kernel Gen: 1-Level Hierarchy **************\n";
SelectedHierarchy = ONE_LEVEL;
KernelLaunchNode = PNode;
- kernel = new Kernel(NULL, N, N->getInArgMap(), N->getSharedInArgMap(),
- N->getOutArgMap(), N->getNumOfDim(), N->getDimLimits());
- } else {
+ kernel = new Kernel(NULL,
+ N,
+ N->getInArgMap(),
+ N->getSharedInArgMap(),
+ N->getOutArgMap(),
+ N->getNumOfDim(),
+ N->getDimLimits());
+ }
+ else {
// Converting a 2-level DFG to opencl kernel
- DEBUG(errs()
- << "*************** Kernel Gen: 2-Level Hierarchy **************\n");
- assert((pLevel >= 2) &&
- "Selected node not nested deep enough to be Kernel Node.");
+ errs() << "*************** Kernel Gen: 2-Level Hierarchy **************\n";
+ assert((pLevel >= 2) && "Selected node not nested deep enough to be Kernel Node.");
SelectedHierarchy = TWO_LEVEL;
KernelLaunchNode = PNode->getParent();
- assert((PNode->getNumOfDim() == N->getNumOfDim()) &&
- "Dimension number must match");
+ assert((PNode->getNumOfDim() == N->getNumOfDim()) && "Dimension number must match");
// Contains the instructions generating the kernel configuration parameters
- kernel = new Kernel(NULL, // kernel function
- N, // kernel leaf node
- N->getInArgMap(), // kenel argument mapping
+ kernel = new Kernel(NULL, // kernel function
+ N, // kernel leaf node
+ N->getInArgMap(), // kenel argument mapping
N->getSharedInArgMap(),
- N->getOutArgMap(), // kernel output mapping from the
- // leaf to the interemediate node
- PNode->getNumOfDim(), // gridDim
- PNode->getDimLimits(), // grid size
- N->getNumOfDim(), // blockDim
- N->getDimLimits()); // block size
+ N->getOutArgMap(), // kernel output mapping from the leaf to the interemediate node
+ PNode->getNumOfDim(), // gridDim
+ PNode->getDimLimits(),// grid size
+ N->getNumOfDim(), // blockDim
+ N->getDimLimits()); // block size
+
}
std::vector<Instruction *> IItoRemove;
@@ -987,62 +1000,58 @@ void CGT_NVPTX::codeGen(DFLeafNode *N) {
// Look up if we have visited this function before. If we have, then just
// get the cloned function pointer from DFNode. Otherwise, create the cloned
// function and add it to the DFNode GenFunc.
- // Function *F_nvptx = N->getGenFunc();
- Function *F_nvptx = N->getGenFuncForTarget(hpvm::GPU_TARGET);
+// Function *F_nvptx = N->getGenFunc();
+ Function *F_nvptx = N->getGenFuncForTarget(visc::GPU_TARGET);
- assert(F_nvptx == NULL &&
- "Error: Visiting a node for which code already generated");
+ assert(F_nvptx == NULL && "Error: Visiting a node for which code already generated");
// Clone the function
ValueToValueMapTy VMap;
- // F_nvptx->setName(FName+"_nvptx");
+ //F_nvptx->setName(FName+"_nvptx");
Twine FName = F->getName();
StringRef fStr = FName.getSingleStringRef();
- Twine newFName = Twine(fStr, "_nvptx");
+ Twine newFName = Twine(fStr, "_nvptx");
F_nvptx = CloneFunction(F, VMap);
F_nvptx->setName(newFName);
+
// errs() << "Old Function Name: " << F->getName() << "\n";
// errs() << "New Function Name: " << F_nvptx->getName() << "\n";
F_nvptx->removeFromParent();
+
// Insert the cloned function into the kernels module
KernelM->getFunctionList().push_back(F_nvptx);
- // TODO: Iterate over all the instructions of F_nvptx and identify the
- // callees and clone them into this module.
+
+ //TODO: Iterate over all the instructions of F_nvptx and identify the
+ //callees and clone them into this module.
DEBUG(errs() << *F_nvptx->getType());
DEBUG(errs() << *F_nvptx);
// Transform the function to void and remove all target dependent attributes
// from the function
F_nvptx = transformFunctionToVoid(F_nvptx);
+
+ //Add generated function info to DFNode
+// N->setGenFunc(F_nvptx, visc::GPU_TARGET);
+ N->addGenFunc(F_nvptx, visc::GPU_TARGET, false);
- // Add generated function info to DFNode
- // N->setGenFunc(F_nvptx, hpvm::GPU_TARGET);
- N->addGenFunc(F_nvptx, hpvm::GPU_TARGET, false);
-
- DEBUG(
- errs()
- << "Removing all attributes from Kernel Function and adding nounwind\n");
- F_nvptx->removeAttributes(AttributeList::FunctionIndex,
- F_nvptx->getAttributes().getFnAttributes());
+ DEBUG(errs() << "Removing all attributes from Kernel Function and adding nounwind\n");
+ F_nvptx->removeAttributes(AttributeList::FunctionIndex, F_nvptx->getAttributes().getFnAttributes());
F_nvptx->addAttribute(AttributeList::FunctionIndex, Attribute::NoUnwind);
- // FIXME: For now, assume only one allocation node
+ //FIXME: For now, assume only one allocation node
kernel->AllocationNode = NULL;
- for (DFNode::const_indfedge_iterator ieb = N->indfedge_begin(),
- iee = N->indfedge_end();
+ for (DFNode::const_indfedge_iterator ieb = N->indfedge_begin(), iee = N->indfedge_end();
ieb != iee; ++ieb) {
DFNode *SrcDFNode = (*ieb)->getSourceDF();
- DEBUG(errs() << "Found edge from node: "
- << " " << SrcDFNode->getFuncPointer()->getName() << "\n");
+ DEBUG(errs() << "Found edge from node: " << " " << SrcDFNode->getFuncPointer()->getName() << "\n");
DEBUG(errs() << "Current Node: " << N->getFuncPointer()->getName() << "\n");
- DEBUG(errs() << "isAllocationNode = " << SrcDFNode->isAllocationNode()
- << "\n");
+ DEBUG(errs() << "isAllocationNode = "<< SrcDFNode->isAllocationNode() << "\n");
if (!SrcDFNode->isDummyNode()) {
assert(SrcDFNode->isAllocationNode());
kernel->AllocationNode = dyn_cast<DFLeafNode>(SrcDFNode);
@@ -1056,22 +1065,19 @@ void CGT_NVPTX::codeGen(DFLeafNode *N) {
// If no allocation node was found, SharedMemArgs is empty
if (kernel->AllocationNode) {
-
ValueToValueMapTy VMap;
- Function *F_alloc =
- CloneFunction(kernel->AllocationNode->getFuncPointer(), VMap);
- // F_alloc->removeFromParent();
+ Function *F_alloc = CloneFunction(kernel->AllocationNode->getFuncPointer(), VMap);
+ //F_alloc->removeFromParent();
// Insert the cloned function into the kernels module
- // M.getFunctionList().push_back(F_alloc);
+ //M.getFunctionList().push_back(F_alloc);
- std::vector<IntrinsicInst *> HPVMMallocInstVec;
- findIntrinsicInst(F_alloc, Intrinsic::hpvm_malloc, HPVMMallocInstVec);
+ std::vector<IntrinsicInst *> ViscMallocInstVec;
+ findIntrinsicInst(F_alloc, Intrinsic::visc_malloc, ViscMallocInstVec);
- for (unsigned i = 0; i < HPVMMallocInstVec.size(); i++) {
- IntrinsicInst *II = HPVMMallocInstVec[i];
- assert(II->hasOneUse() && "hpvm_malloc result is used more than once");
- II->replaceAllUsesWith(
- ConstantPointerNull::get(Type::getInt8PtrTy(M.getContext())));
+ for (unsigned i = 0; i < ViscMallocInstVec.size(); i++) {
+ IntrinsicInst *II = ViscMallocInstVec[i];
+ assert(II->hasOneUse() && "visc_malloc result is used more than once");
+ II->replaceAllUsesWith(ConstantPointerNull::get(Type::getInt8PtrTy(M.getContext())));
II->eraseFromParent();
}
kernel->AllocationFunction = F_alloc;
@@ -1086,19 +1092,15 @@ void CGT_NVPTX::codeGen(DFLeafNode *N) {
assert(RetStructTy && "Allocation node does not return a struct type");
unsigned numFields = RetStructTy->getNumElements();
*/
- std::map<unsigned, std::pair<Value *, unsigned>> sharedInMap =
- kernel->getSharedInArgMap();
- AllocationNodeProperty *APN =
- (AllocationNodeProperty *)kernel->AllocationNode->getProperty(
- DFNode::Allocation);
- for (auto &AllocPair : APN->getAllocationList()) {
+ std::map<unsigned, std::pair<Value*, unsigned> > sharedInMap = kernel->getSharedInArgMap();
+ AllocationNodeProperty* APN =
+ (AllocationNodeProperty*) kernel->AllocationNode->getProperty(DFNode::Allocation);
+ for (auto& AllocPair: APN->getAllocationList()) {
unsigned destPos = AllocPair.first->getDestPosition();
unsigned srcPos = AllocPair.first->getSourcePosition();
SharedMemArgs.push_back(destPos);
- sharedInMap[destPos] =
- std::pair<Value *, unsigned>(AllocPair.second, srcPos + 1);
- sharedInMap[destPos + 1] =
- std::pair<Value *, unsigned>(AllocPair.second, srcPos + 1);
+ sharedInMap[destPos] = std::pair<Value *, unsigned>(AllocPair.second, srcPos+1);
+ sharedInMap[destPos+1] = std::pair<Value *, unsigned>(AllocPair.second, srcPos+1);
}
kernel->setSharedInArgMap(sharedInMap);
}
@@ -1108,14 +1110,12 @@ void CGT_NVPTX::codeGen(DFLeafNode *N) {
// global address space
unsigned argIndex = 0;
std::vector<unsigned> GlobalMemArgs;
- for (Function::arg_iterator ai = F_nvptx->arg_begin(),
- ae = F_nvptx->arg_end();
- ai != ae; ++ai) {
- if (ai->getType()->isPointerTy()) {
- // If the arguement is already chosen for shared memory arguemnt list,
- // skip. Else put it in Global memory arguement list
- if (std::count(SharedMemArgs.begin(), SharedMemArgs.end(), argIndex) ==
- 0) {
+ for(Function::arg_iterator ai = F_nvptx->arg_begin(), ae = F_nvptx->arg_end();
+ ai != ae; ++ai) {
+ if (ai->getType()->isPointerTy()) {
+ // If the arguement is already chosen for shared memory arguemnt list, skip.
+ // Else put it in Global memory arguement list
+ if(std::count(SharedMemArgs.begin(), SharedMemArgs.end(), argIndex) == 0) {
GlobalMemArgs.push_back(argIndex);
}
}
@@ -1129,21 +1129,20 @@ void CGT_NVPTX::codeGen(DFLeafNode *N) {
// Optimization: Gloabl memory arguments, which are not modified and whose
// loads are not dependent on node id of current node, should be moved to
// constant memory, subject to size of course
- std::vector<unsigned> ConstantMemArgs =
- globalToConstantMemoryOpt(&GlobalMemArgs, F_nvptx);
+ std::vector<unsigned> ConstantMemArgs = globalToConstantMemoryOpt(&GlobalMemArgs, F_nvptx);
F_nvptx = changeArgAddrspace(F_nvptx, ConstantMemArgs, GLOBAL_ADDRSPACE);
F_nvptx = changeArgAddrspace(F_nvptx, SharedMemArgs, SHARED_ADDRSPACE);
F_nvptx = changeArgAddrspace(F_nvptx, GlobalMemArgs, GLOBAL_ADDRSPACE);
- // Function to replace call instructions to functions in the kernel
+// Function to replace call instructions to functions in the kernel
std::map<Function *, Function *> OrgToClonedFuncMap;
std::vector<Function *> FuncToBeRemoved;
- auto CloneAndReplaceCall = [&](CallInst *CI, Function *OrgFunc) {
- Function *NewFunc;
+ auto CloneAndReplaceCall = [&] (CallInst *CI, Function *OrgFunc) {
+ Function* NewFunc;
// Check if the called function has already been cloned before.
auto It = OrgToClonedFuncMap.find(OrgFunc);
- if (It == OrgToClonedFuncMap.end()) {
+ if(It == OrgToClonedFuncMap.end()) {
ValueToValueMapTy VMap;
NewFunc = CloneFunction(OrgFunc, VMap);
OrgToClonedFuncMap[OrgFunc] = NewFunc;
@@ -1152,48 +1151,43 @@ void CGT_NVPTX::codeGen(DFLeafNode *N) {
NewFunc = (*It).second;
}
// Replace the calls to this function
- std::vector<Value *> args;
- for (unsigned i = 0; i < CI->getNumArgOperands(); i++) {
+ std::vector<Value*> args;
+ for(unsigned i=0; i < CI->getNumArgOperands(); i++) {
args.push_back(CI->getArgOperand(i));
}
- CallInst *Inst = CallInst::Create(
- NewFunc, args,
- OrgFunc->getReturnType()->isVoidTy() ? "" : CI->getName(), CI);
+ CallInst* Inst = CallInst::Create(NewFunc, args,
+ OrgFunc->getReturnType()->isVoidTy()? "" : CI->getName(), CI);
CI->replaceAllUsesWith(Inst);
IItoRemove.push_back(CI);
return NewFunc;
};
+
// Go through all the instructions
- for (inst_iterator i = inst_begin(F_nvptx), e = inst_end(F_nvptx); i != e;
- ++i) {
+ for (inst_iterator i = inst_begin(F_nvptx), e = inst_end(F_nvptx); i != e; ++i) {
Instruction *I = &(*i);
- // Leaf nodes should not contain HPVM graph intrinsics or launch
- assert(!BuildDFG::isHPVMLaunchIntrinsic(I) &&
- "Launch intrinsic within a dataflow graph!");
- assert(!BuildDFG::isHPVMGraphIntrinsic(I) &&
- "HPVM graph intrinsic within a leaf dataflow node!");
+ // Leaf nodes should not contain VISC graph intrinsics or launch
+ assert(!BuildDFG::isViscLaunchIntrinsic(I) && "Launch intrinsic within a dataflow graph!");
+ assert(!BuildDFG::isViscGraphIntrinsic(I) && "VISC graph intrinsic within a leaf dataflow node!");
- if (BuildDFG::isHPVMIntrinsic(I)) {
- IntrinsicInst *II = dyn_cast<IntrinsicInst>(I);
- IntrinsicInst *ArgII;
- DFNode *ArgDFNode;
+ if (BuildDFG::isViscIntrinsic(I)) {
+ IntrinsicInst* II = dyn_cast<IntrinsicInst>(I);
+ IntrinsicInst* ArgII;
+ DFNode* ArgDFNode;
- /************************ Handle HPVM Query intrinsics
- * ************************/
+ /************************ Handle VISC Query intrinsics ************************/
switch (II->getIntrinsicID()) {
- /**************************** llvm.hpvm.getNode()
- * *****************************/
- case Intrinsic::hpvm_getNode: {
+ /**************************** llvm.visc.getNode() *****************************/
+ case Intrinsic::visc_getNode: {
DEBUG(errs() << F_nvptx->getName() << "\t: Handling getNode\n");
// add mapping <intrinsic, this node> to the node-specific map
Leaf_HandleToDFNodeMap[II] = N;
IItoRemove.push_back(II);
- } break;
- /************************* llvm.hpvm.getParentNode()
- * **************************/
- case Intrinsic::hpvm_getParentNode: {
+ }
+ break;
+ /************************* llvm.visc.getParentNode() **************************/
+ case Intrinsic::visc_getParentNode: {
DEBUG(errs() << F_nvptx->getName() << "\t: Handling getParentNode\n");
// get the parent node of the arg node
// get argument node
@@ -1206,10 +1200,10 @@ void CGT_NVPTX::codeGen(DFLeafNode *N) {
Leaf_HandleToDFNodeMap[II] = ArgDFNode->getParent();
IItoRemove.push_back(II);
- } break;
- /*************************** llvm.hpvm.getNumDims()
- * ***************************/
- case Intrinsic::hpvm_getNumDims: {
+ }
+ break;
+ /*************************** llvm.visc.getNumDims() ***************************/
+ case Intrinsic::visc_getNumDims: {
DEBUG(errs() << F_nvptx->getName() << "\t: Handling getNumDims\n");
// get node from map
// get the appropriate field
@@ -1217,48 +1211,47 @@ void CGT_NVPTX::codeGen(DFLeafNode *N) {
ArgDFNode = Leaf_HandleToDFNodeMap[ArgII];
int numOfDim = ArgDFNode->getNumOfDim();
DEBUG(errs() << "\t Got node dimension : " << numOfDim << "\n");
- IntegerType *IntTy = Type::getInt32Ty(KernelM->getContext());
- ConstantInt *numOfDimConstant =
- ConstantInt::getSigned(IntTy, (int64_t)numOfDim);
+ IntegerType* IntTy = Type::getInt32Ty(KernelM->getContext());
+ ConstantInt* numOfDimConstant = ConstantInt::getSigned(IntTy, (int64_t) numOfDim);
// Replace the result of the intrinsic with the computed value
II->replaceAllUsesWith(numOfDimConstant);
IItoRemove.push_back(II);
- } break;
- /*********************** llvm.hpvm.getNodeInstanceID()
- * ************************/
- case Intrinsic::hpvm_getNodeInstanceID_x:
- case Intrinsic::hpvm_getNodeInstanceID_y:
- case Intrinsic::hpvm_getNodeInstanceID_z: {
- DEBUG(errs() << F_nvptx->getName() << "\t: Handling getNodeInstanceID\n"
- << "\t: " << *II << "\n");
+ }
+ break;
+ /*********************** llvm.visc.getNodeInstanceID() ************************/
+ case Intrinsic::visc_getNodeInstanceID_x:
+ case Intrinsic::visc_getNodeInstanceID_y:
+ case Intrinsic::visc_getNodeInstanceID_z: {
+ DEBUG(errs() << F_nvptx->getName() << "\t: Handling getNodeInstanceID\n" << "\t: " << *II << "\n");
ArgII = cast<IntrinsicInst>((II->getOperand(0))->stripPointerCasts());
ArgDFNode = Leaf_HandleToDFNodeMap[ArgII];
assert(ArgDFNode && "Arg node is NULL");
// A leaf node always has a parent
- DFNode *ParentDFNode = ArgDFNode->getParent();
+ DFNode* ParentDFNode = ArgDFNode->getParent();
assert(ParentDFNode && "Parent node of a leaf is NULL");
// Get the number associated with the required dimension
// FIXME: The order is important!
// These three intrinsics need to be consecutive x,y,z
- uint64_t dim =
- II->getIntrinsicID() - Intrinsic::hpvm_getNodeInstanceID_x;
+ uint64_t dim = II->getIntrinsicID() -
+ Intrinsic::visc_getNodeInstanceID_x;
assert((dim < 3) && "Invalid dimension argument");
DEBUG(errs() << "\t dimension = " << dim << "\n");
// Argument of the function to be called
- ConstantInt *DimConstant =
- ConstantInt::get(Type::getInt32Ty(KernelM->getContext()), dim);
- // ArrayRef<Value *> Args(DimConstant);
+ ConstantInt * DimConstant =
+ ConstantInt::get(Type::getInt32Ty(KernelM->getContext()), dim);
+ //ArrayRef<Value *> Args(DimConstant);
// The following is to find which function to call
- Function *OpenCLFunction;
+ Function * OpenCLFunction;
- FunctionType *FT =
- FunctionType::get(Type::getInt64Ty(KernelM->getContext()),
- Type::getInt32Ty(KernelM->getContext()), false);
+ FunctionType* FT =
+ FunctionType::get(Type::getInt64Ty(KernelM->getContext()),
+ Type::getInt32Ty(KernelM->getContext()),
+ false);
if (SelectedHierarchy == ONE_LEVEL && ArgDFNode == N) {
// We only have one level in the hierarchy or the parent node is not
// replicated. This indicates that the parent node is the kernel
@@ -1267,867 +1260,838 @@ void CGT_NVPTX::codeGen(DFLeafNode *N) {
// itself
DEBUG(errs() << "Substitute with get_global_id()\n");
DEBUG(errs() << *II << "\n");
- OpenCLFunction = cast<Function>(
- (KernelM->getOrInsertFunction(StringRef("get_global_id"), FT))
- .getCallee());
+ OpenCLFunction = cast<Function>
+ ((KernelM->getOrInsertFunction(StringRef("get_global_id"), FT)).getCallee());
} else if (Leaf_HandleToDFNodeMap[ArgII] == N) {
- // DEBUG(errs() << "Here inside cond 2\n");
+ //DEBUG(errs() << "Here inside cond 2\n");
// We are asking for this node's id with respect to its parent
// this is a local id call
- OpenCLFunction = cast<Function>(
- (KernelM->getOrInsertFunction(StringRef("get_local_id"), FT))
- .getCallee());
- // DEBUG(errs() << "exiting condition 2\n");
+ OpenCLFunction = cast<Function>
+ ((KernelM->getOrInsertFunction(StringRef("get_local_id"), FT)).getCallee());
+ //DEBUG(errs() << "exiting condition 2\n");
} else if (Leaf_HandleToDFNodeMap[ArgII] == N->getParent()) {
// We are asking for this node's parent's id with respect to its
// parent: this is a group id call
- OpenCLFunction = cast<Function>(
- (KernelM->getOrInsertFunction(StringRef("get_group_id"), FT))
- .getCallee());
+ OpenCLFunction = cast<Function>
+ ((KernelM->getOrInsertFunction(StringRef("get_group_id"), FT)).getCallee());
} else {
- DEBUG(errs() << N->getFuncPointer()->getName() << "\n");
- DEBUG(errs() << N->getParent()->getFuncPointer()->getName() << "\n");
- DEBUG(errs() << *II << "\n");
+ errs() << N->getFuncPointer()->getName() << "\n";
+ errs() << N->getParent()->getFuncPointer()->getName() << "\n";
+ errs() << *II << "\n";
assert(false && "Unable to translate getNodeInstanceID intrinsic");
}
- // DEBUG(errs() << "Create call instruction, insert it before the
- // instrinsic\n"); DEBUG(errs() << "Function: " << *OpenCLFunction <<
- // "\n"); DEBUG(errs() << "Arguments size: " << Args.size() << "\n");
- // DEBUG(errs() << "Argument: " << Args[0] << "\n");
- // DEBUG(errs() << "Arguments: " << *DimConstant << "\n");
+ //DEBUG(errs() << "Create call instruction, insert it before the instrinsic\n");
+ //DEBUG(errs() << "Function: " << *OpenCLFunction << "\n");
+ //DEBUG(errs() << "Arguments size: " << Args.size() << "\n");
+ //DEBUG(errs() << "Argument: " << Args[0] << "\n");
+ //DEBUG(errs() << "Arguments: " << *DimConstant << "\n");
// Create call instruction, insert it before the intrinsic and
// replace the uses of the previous instruction with the new one
- CallInst *CI = CallInst::Create(OpenCLFunction, DimConstant, "", II);
- // DEBUG(errs() << "Replace uses\n");
+ CallInst* CI = CallInst::Create(OpenCLFunction, DimConstant, "", II);
+ //DEBUG(errs() << "Replace uses\n");
II->replaceAllUsesWith(CI);
IItoRemove.push_back(II);
- } break;
- /********************** llvm.hpvm.getNumNodeInstances()
- * ***********************/
- case Intrinsic::hpvm_getNumNodeInstances_x:
- case Intrinsic::hpvm_getNumNodeInstances_y:
- case Intrinsic::hpvm_getNumNodeInstances_z: {
+ }
+ break;
+ /********************** llvm.visc.getNumNodeInstances() ***********************/
+ case Intrinsic::visc_getNumNodeInstances_x:
+ case Intrinsic::visc_getNumNodeInstances_y:
+ case Intrinsic::visc_getNumNodeInstances_z: {
// TODO: think about whether this is the best way to go there are hw
// specific registers. therefore it is good to have the intrinsic but
// then, why do we need to keep that info in the graph? (only for the
// kernel configuration during the call)
- DEBUG(errs() << F_nvptx->getName()
- << "\t: Handling getNumNodeInstances\n");
+ DEBUG(errs() << F_nvptx->getName() << "\t: Handling getNumNodeInstances\n");
ArgII = cast<IntrinsicInst>((II->getOperand(0))->stripPointerCasts());
ArgDFNode = Leaf_HandleToDFNodeMap[ArgII];
// A leaf node always has a parent
- DFNode *ParentDFNode = ArgDFNode->getParent();
+ DFNode* ParentDFNode = ArgDFNode->getParent();
assert(ParentDFNode && "Parent node of a leaf is NULL");
// Get the number associated with the required dimension
// FIXME: The order is important!
// These three intrinsics need to be consecutive x,y,z
- uint64_t dim =
- II->getIntrinsicID() - Intrinsic::hpvm_getNumNodeInstances_x;
+ uint64_t dim = II->getIntrinsicID() -
+ Intrinsic::visc_getNumNodeInstances_x;
assert((dim < 3) && "Invalid dimension argument");
DEBUG(errs() << "\t dimension = " << dim << "\n");
// Argument of the function to be called
- ConstantInt *DimConstant =
- ConstantInt::get(Type::getInt32Ty(KernelM->getContext()), dim);
- // ArrayRef<Value *> Args(DimConstant);
+ ConstantInt * DimConstant =
+ ConstantInt::get(Type::getInt32Ty(KernelM->getContext()), dim);
+ //ArrayRef<Value *> Args(DimConstant);
// The following is to find which function to call
- Function *OpenCLFunction;
- FunctionType *FT =
+ Function * OpenCLFunction;
+ FunctionType* FT =
FunctionType::get(Type::getInt64Ty(KernelM->getContext()),
- Type::getInt32Ty(KernelM->getContext()), false);
+ Type::getInt32Ty(KernelM->getContext()),
+ false);
if (N == ArgDFNode && SelectedHierarchy == ONE_LEVEL) {
// We only have one level in the hierarchy or the parent node is not
// replicated. This indicates that the parent node is the kernel
// launch, so the instances are global_size (gridDim x blockDim)
- OpenCLFunction = cast<Function>(
- (KernelM->getOrInsertFunction(StringRef("get_global_size"), FT))
- .getCallee());
+ OpenCLFunction = cast<Function>
+ ((KernelM->getOrInsertFunction(StringRef("get_global_size"), FT)).getCallee());
} else if (Leaf_HandleToDFNodeMap[ArgII] == N) {
// We are asking for this node's instances
// this is a local size (block dim) call
- OpenCLFunction = cast<Function>(
- (KernelM->getOrInsertFunction(StringRef("get_local_size"), FT))
- .getCallee());
+ OpenCLFunction = cast<Function>
+ ((KernelM->getOrInsertFunction(StringRef("get_local_size"), FT)).getCallee());
} else if (Leaf_HandleToDFNodeMap[ArgII] == N->getParent()) {
// We are asking for this node's parent's instances
// this is a (global_size/local_size) (grid dim) call
- OpenCLFunction = cast<Function>(
- (KernelM->getOrInsertFunction(StringRef("get_num_groups"), FT))
- .getCallee());
+ OpenCLFunction = cast<Function>
+ ((KernelM->getOrInsertFunction(StringRef("get_num_groups"), FT)).getCallee());
} else {
assert(false && "Unable to translate getNumNodeInstances intrinsic");
}
// Create call instruction, insert it before the intrinsic and
// replace the uses of the previous instruction with the new one
- CallInst *CI = CallInst::Create(OpenCLFunction, DimConstant, "", II);
+ CallInst* CI = CallInst::Create(OpenCLFunction, DimConstant, "", II);
II->replaceAllUsesWith(CI);
IItoRemove.push_back(II);
- } break;
- case Intrinsic::hpvm_barrier: {
+ }
+ break;
+ case Intrinsic::visc_barrier:
+ {
DEBUG(errs() << F_nvptx->getName() << "\t: Handling barrier\n");
DEBUG(errs() << "Substitute with barrier()\n");
DEBUG(errs() << *II << "\n");
- FunctionType *FT = FunctionType::get(
- Type::getVoidTy(KernelM->getContext()),
- std::vector<Type *>(1, Type::getInt32Ty(KernelM->getContext())),
- false);
- Function *OpenCLFunction = cast<Function>(
- (KernelM->getOrInsertFunction(StringRef("barrier"), FT))
- .getCallee());
- CallInst *CI =
- CallInst::Create(OpenCLFunction,
- ArrayRef<Value *>(ConstantInt::get(
- Type::getInt32Ty(KernelM->getContext()), 1)),
- "", II);
+ FunctionType* FT = FunctionType::get(Type::getVoidTy(KernelM->getContext()),
+ std::vector<Type*>(1, Type::getInt32Ty(KernelM->getContext())),
+ false);
+ Function* OpenCLFunction = cast<Function>
+ ((KernelM->getOrInsertFunction(StringRef("barrier"), FT)).getCallee());
+ CallInst* CI = CallInst::Create(OpenCLFunction,
+ ArrayRef<Value*>(ConstantInt::get(Type::getInt32Ty(KernelM->getContext()), 1)),
+ "", II);
II->replaceAllUsesWith(CI);
IItoRemove.push_back(II);
- } break;
- case Intrinsic::hpvm_atomic_cmpxchg:
- break;
- case Intrinsic::hpvm_atomic_add:
- case Intrinsic::hpvm_atomic_sub:
- case Intrinsic::hpvm_atomic_xchg:
- case Intrinsic::hpvm_atomic_min:
- case Intrinsic::hpvm_atomic_max:
- case Intrinsic::hpvm_atomic_and:
- case Intrinsic::hpvm_atomic_or:
- case Intrinsic::hpvm_atomic_xor:
- // case Intrinsic::hpvm_atomic_inc:
- // case Intrinsic::hpvm_atomic_dec:
- {
- DEBUG(errs() << *II << "\n");
- // Only have support for i32 atomic intrinsics
- assert(II->getType() == Type::getInt32Ty(II->getContext()) &&
- "Only support i32 atomic intrinsics for now");
- // Substitute with atomicrmw instruction
- assert(II->getNumArgOperands() == 2 &&
- "Expecting 2 operands for these atomics");
- Value *Ptr = II->getArgOperand(0);
- Value *Val = II->getArgOperand(1);
- assert(Ptr->getType()->isPointerTy() &&
- "First argument of supported atomics is expected to be a "
- "pointer");
- PointerType *PtrTy = cast<PointerType>(Ptr->getType());
- PointerType *TargetTy =
- Type::getInt32PtrTy(II->getContext(), PtrTy->getAddressSpace());
- if (PtrTy != TargetTy) {
- Ptr = CastInst::CreatePointerCast(Ptr, TargetTy, "", II);
- PtrTy = TargetTy;
- }
-
- std::string name;
- if (II->getIntrinsicID() == Intrinsic::hpvm_atomic_add)
- name = "atomic_add";
- else if (II->getIntrinsicID() == Intrinsic::hpvm_atomic_sub)
- name = "atomic_sub";
- else if (II->getIntrinsicID() == Intrinsic::hpvm_atomic_xchg)
- name = "atomic_xchg";
- else if (II->getIntrinsicID() == Intrinsic::hpvm_atomic_min)
- name = "atomic_min";
- else if (II->getIntrinsicID() == Intrinsic::hpvm_atomic_max)
- name = "atomic_max";
- else if (II->getIntrinsicID() == Intrinsic::hpvm_atomic_and)
- name = "atomic_and";
- else if (II->getIntrinsicID() == Intrinsic::hpvm_atomic_or)
- name = "atomic_or";
- else if (II->getIntrinsicID() == Intrinsic::hpvm_atomic_xor)
- name = "atomic_xor";
- Type *paramTypes[] = {PtrTy, Val->getType()};
- FunctionType *AtomFuncT = FunctionType::get(
- II->getType(), ArrayRef<Type *>(paramTypes, 2), false);
- FunctionCallee AtomFunc =
- KernelM->getOrInsertFunction(name, AtomFuncT);
-
- Value *Params[] = {Ptr, Val};
- CallInst *AtomCI = CallInst::Create(
- AtomFunc, ArrayRef<Value *>(Params, 2), II->getName(), II);
- DEBUG(errs() << "Substitute with: " << *AtomCI << "\n");
- II->replaceAllUsesWith(AtomCI);
- IItoRemove.push_back(II);
- }
- break;
- default:
- llvm_unreachable("Unknown HPVM Intrinsic!");
- break;
- }
-
- } else if (MemCpyInst *MemCpyI = dyn_cast<MemCpyInst>(I)) {
- IRBuilder<> Builder(I);
- Value *Source = MemCpyI->getSource();
- Value *Destination = MemCpyI->getArgOperand(0)->stripPointerCasts();
- Value *Length = MemCpyI->getOperand(2);
- DEBUG(errs() << "Found memcpy instruction: " << *I << "\n");
- DEBUG(errs() << "Source: " << *Source << "\n");
- DEBUG(errs() << "Destination: " << *Destination << "\n");
- DEBUG(errs() << "Length: " << *Length << "\n");
-
- size_t memcpy_length;
- unsigned int memcpy_count;
- if (ConstantInt *CI = dyn_cast<ConstantInt>(Length)) {
- if (CI->getBitWidth() <= 64) {
- memcpy_length = CI->getSExtValue();
- DEBUG(errs() << "Memcpy lenght = " << memcpy_length << "\n");
- Type *Source_Type = Source->getType()->getPointerElementType();
- DEBUG(errs() << "Source Type : " << *Source_Type << "\n");
- memcpy_count =
- memcpy_length / (Source_Type->getPrimitiveSizeInBits() / 8);
- DEBUG(errs() << "Memcpy count = " << memcpy_count << "\n");
- if (GetElementPtrInst *sourceGEPI =
- dyn_cast<GetElementPtrInst>(Source)) {
- if (GetElementPtrInst *destGEPI =
- dyn_cast<GetElementPtrInst>(Destination)) {
- Value *SourcePtrOperand = sourceGEPI->getPointerOperand();
- Value *DestPtrOperand = destGEPI->getPointerOperand();
- for (int i = 0; i < memcpy_count; ++i) {
- Constant *increment;
- LoadInst *newLoadI;
- StoreInst *newStoreI;
- // First, need to increment the correct index for both source
- // and dest This invluves checking to see how many indeces the
- // GEP has Assume for now only 1 or 2 are the viable options.
-
- std::vector<Value *> GEPlIndex;
- if (sourceGEPI->getNumIndices() == 1) {
- Value *Index = sourceGEPI->getOperand(1);
- increment = ConstantInt::get(Index->getType(), i, false);
- Value *incAdd = Builder.CreateAdd(Index, increment);
- DEBUG(errs() << "Add: " << *incAdd << "\n");
- GEPlIndex.push_back(incAdd);
- Value *newGEPIl = Builder.CreateGEP(
- SourcePtrOperand, ArrayRef<Value *>(GEPlIndex));
- DEBUG(errs() << "Load GEP: " << *newGEPIl << "\n");
- newLoadI = Builder.CreateLoad(newGEPIl);
- DEBUG(errs() << "Load: " << *newLoadI << "\n");
- } else {
- llvm_unreachable("Unhandled case where source GEPI has more "
- "than 1 indices!\n");
- }
-
- std::vector<Value *> GEPsIndex;
- if (destGEPI->getNumIndices() == 1) {
-
- } else if (destGEPI->getNumIndices() == 2) {
- Value *Index0 = destGEPI->getOperand(1);
- GEPsIndex.push_back(Index0);
- Value *Index1 = destGEPI->getOperand(2);
- increment = ConstantInt::get(Index1->getType(), i, false);
- Value *incAdd = Builder.CreateAdd(Index1, increment);
- DEBUG(errs() << "Add: " << *incAdd << "\n");
- GEPsIndex.push_back(incAdd);
- Value *newGEPIs = Builder.CreateGEP(
- DestPtrOperand, ArrayRef<Value *>(GEPsIndex));
- DEBUG(errs() << "Store GEP: " << *newGEPIs << "\n");
- newStoreI = Builder.CreateStore(newLoadI, newGEPIs,
- MemCpyI->isVolatile());
- DEBUG(errs() << "Store: " << *newStoreI << "\n");
- } else {
- llvm_unreachable("Unhandled case where dest GEPI has more "
- "than 2 indices!\n");
- }
- }
- IItoRemove.push_back(sourceGEPI);
- IItoRemove.push_back(destGEPI);
- Instruction *destBitcastI =
- dyn_cast<Instruction>(MemCpyI->getArgOperand(0));
- Instruction *sourceBitcastI =
- dyn_cast<Instruction>(MemCpyI->getArgOperand(1));
- IItoRemove.push_back(destBitcastI);
- IItoRemove.push_back(sourceBitcastI);
- IItoRemove.push_back(MemCpyI);
- }
- }
- }
- } else {
- llvm_unreachable("MEMCPY length is not a constant, not handled!\n");
}
- // llvm_unreachable("HERE!");
- }
-
- else if (CallInst *CI = dyn_cast<CallInst>(I)) {
- DEBUG(errs() << "Found a call: " << *CI << "\n");
- Function *calleeF =
- cast<Function>(CI->getCalledValue()->stripPointerCasts());
- if (calleeF->isDeclaration()) {
- // Add the declaration to kernel module
- if (calleeF->getName() == "sqrtf") {
- calleeF->setName(Twine("sqrt"));
- DEBUG(errs() << "CaleeF: " << *calleeF << "\n");
- DEBUG(errs() << "CI: " << *CI << "\n");
- } else if (calleeF->getName() == "rsqrtf") {
- calleeF->setName(Twine("rsqrt"));
- DEBUG(errs() << "CaleeF: " << *calleeF << "\n");
- DEBUG(errs() << "CI: " << *CI << "\n");
- }
- DEBUG(errs() << "Adding declaration to Kernel module: " << *calleeF
- << "\n");
- KernelM->getOrInsertFunction(calleeF->getName(),
- calleeF->getFunctionType());
- } else {
- // Check if the called function has already been cloned before.
- Function *NewFunc = CloneAndReplaceCall(CI, calleeF);
- // Iterate over the new function to see if it calls any other functions
- // in the module.
- for (inst_iterator i = inst_begin(NewFunc), e = inst_end(NewFunc);
- i != e; ++i) {
- if (auto *Call = dyn_cast<CallInst>(&*i)) {
- Function *CalledFunc =
- cast<Function>(Call->getCalledValue()->stripPointerCasts());
- CloneAndReplaceCall(Call, CalledFunc);
- }
- }
- }
- // TODO: how to handle address space qualifiers in load/store
- }
- }
- // search for pattern where float is being casted to int and loaded/stored and
- // change it.
- DEBUG(errs() << "finding pattern for replacement!\n");
- for (inst_iterator i = inst_begin(F_nvptx), e = inst_end(F_nvptx); i != e;
- ++i) {
- bool cont = false;
- bool keepGEPI = false;
- bool keepGEPI2 = false;
- Instruction *I = &(*i);
- GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(I);
-
- if (!GEPI) {
- // did nod find pattern start, continue
- continue;
- }
- // may have found pattern, check
- DEBUG(errs() << "GEPI " << *GEPI << "\n");
- // print whatever we want for debug
- Value *PtrOp = GEPI->getPointerOperand();
- Type *SrcTy = GEPI->getSourceElementType();
- unsigned GEPIaddrspace = GEPI->getAddressSpace();
-
- if (SrcTy->isArrayTy())
- DEBUG(errs() << *SrcTy << " is an array type! "
- << *(SrcTy->getArrayElementType()) << "\n");
- else
- DEBUG(errs() << *SrcTy << " is not an array type!\n");
- // check that source element type is float
- if (SrcTy->isArrayTy()) {
- if (!(SrcTy->getArrayElementType()->isFloatTy())) {
- DEBUG(errs() << "GEPI type is array but not float!\n");
- continue;
- }
- } else if (!(SrcTy->isFPOrFPVectorTy() /*isFloatTy()*/)) {
- DEBUG(errs() << "GEPI type is " << *SrcTy << "\n");
- // does not fit this pattern - no float GEP instruction
- continue;
- }
- // check that addressspace is 1
- // if (GEPIaddrspace != 1) {
- // // does not fit this pattern - addrspace of pointer
- // argument is not global continue;
- // }
- if (!(GEPI->hasOneUse())) {
- // does not fit this pattern - more than one uses
- // continue;
- // Keep GEPI around if it has other uses
- keepGEPI = true;
- }
- DEBUG(errs() << "Found GEPI " << *GEPI << "\n");
-
- // 1st GEPI it has one use
- // assert(GEPI->hasOneUse() && "GEPI has a single use");
-
- // See if it is a bitcast
- BitCastInst *BitCastI;
- for (User *U : GEPI->users()) {
- if (Instruction *ui = dyn_cast<Instruction>(U)) {
- DEBUG(errs() << "--" << *ui << "\n");
- if (isa<BitCastInst>(ui)) {
- BitCastI = dyn_cast<BitCastInst>(ui);
- DEBUG(errs() << "---Found bitcast as only use of GEP\n");
- break;
- }
- }
- DEBUG(errs() << "GEPI does not have a bitcast user, continue\n");
- cont = true;
- }
- // for (Value::user_iterator ui = GEPI->user_begin(),
- // ue = GEPI->user_end(); ui!=ue; ++ui) {
- // DEBUG(errs() << "--" << *ui << "\n");
- // if (isa<BitCastInst>(*ui)) {
- // BitCastI = dyn_cast<BitCastInst>(*ui);
- // DEBUG(errs() << "Found bitcast as only use of GEP\n");
- // }
- // }
-
- if (cont /*!BitCastI*/) {
- continue; // not in pattern
- }
-
- // DEBUG(errs() << *BitCastI << "\n");
- // Otherwise, check that first operand is GEP and 2nd is i32*. 1st Operand
- // has to be the GEP, since this is a use of the GEP.
- Value *Op2 = BitCastI->getOperand(0);
- DEBUG(errs() << "----" << *Op2 << "\n");
- // assert(cast<Type>(Op2) && "Invalid Operand for Bitcast\n");
- // Type *OpTy = cast<Type>(Op2);
- Type *OpTy = BitCastI->getDestTy();
- DEBUG(errs() << "---- Bitcast destination type: " << *OpTy << "\n");
- // DEBUG(errs() << "---- " << *(Type::getInt32PtrTy(M.getContext(),1)) <<
- // "\n");
- if (!(OpTy == Type::getInt32PtrTy(M.getContext(), GEPIaddrspace))) {
- // maybe right syntax is (Type::getInt32Ty)->getPointerTo()
- continue; // not in pattern
- }
-
- DEBUG(errs() << "----Here!\n");
- // We are in GEP, bitcast.
-
- // user_iterator, to find the load.
-
- if (!(BitCastI->hasOneUse())) {
- // does not fit this pattern - more than one uses
- continue;
- }
- DEBUG(errs() << "----Bitcast has one use!\n");
- // it has one use
- assert(BitCastI->hasOneUse() && "BitCastI has a single use");
- LoadInst *LoadI;
- for (User *U : BitCastI->users()) {
- if (Instruction *ui = dyn_cast<Instruction>(U)) {
- DEBUG(errs() << "-----" << *ui << "\n");
- if (isa<LoadInst>(ui)) {
- LoadI = dyn_cast<LoadInst>(ui);
- DEBUG(errs() << "-----Found load as only use of bitcast\n");
- break;
+ break;
+ case Intrinsic::visc_atomic_add:
+ case Intrinsic::visc_atomic_sub:
+ case Intrinsic::visc_atomic_xchg:
+ case Intrinsic::visc_atomic_min:
+ case Intrinsic::visc_atomic_max:
+ case Intrinsic::visc_atomic_and:
+ case Intrinsic::visc_atomic_or:
+ case Intrinsic::visc_atomic_xor:
+ {
+ DEBUG(errs() << *II << "\n");
+ // Only have support for i32 atomic intrinsics
+ assert(II->getType() == Type::getInt32Ty(II->getContext())
+ && "Only support i32 atomic intrinsics for now");
+ // Substitute with atomicrmw instruction
+ assert(II->getNumArgOperands() == 2 && "Expecting 2 operands for these atomics");
+ Value* Ptr = II->getArgOperand(0);
+ Value* Val = II->getArgOperand(1);
+ assert(Ptr->getType()->isPointerTy()
+ && "First argument of supported atomics is expected to be a pointer");
+ PointerType* PtrTy = cast<PointerType>(Ptr->getType());
+ PointerType* TargetTy = Type::getInt32PtrTy(II->getContext(), PtrTy->getAddressSpace());
+ if (PtrTy != TargetTy) {
+ Ptr = CastInst::CreatePointerCast(Ptr, TargetTy, "", II);
+ PtrTy = TargetTy;
}
- }
- DEBUG(errs() << "Bitcast does not have a load user, continue!\n");
- cont = true;
- }
- // for (Value::user_iterator ui = BitCastI->user_begin(),
- // ue = BitCastI->user_end(); ui!=ue; ++ui) {
- // if (isa<LoadInst>(*ui)) {
- // LoadI = dyn_cast<LoadInst>(*ui);
- // errs() << "Found load as only use of bitcast\n";
- // }
- // }
-
- if (cont) {
- continue; // not in pattern
- }
- DEBUG("HERE!\n");
- // check that we load from pointer we got from bitcast - assert - the unique
- // argument must be the use we found it from
- assert(LoadI->getPointerOperand() == BitCastI &&
- "Unexpected Load Instruction Operand\n");
-
- // Copy user_iterator, to find the store.
-
- if (!(LoadI->hasOneUse())) {
- // does not fit this pattern - more than one uses
- continue;
- // TODO: generalize: one load can have more than one store users
- }
-
- // it has one use
- assert(LoadI->hasOneUse() && "LoadI has a single use");
- Value::user_iterator ui = LoadI->user_begin();
- // skipped loop, because is has a single use
- StoreInst *StoreI = dyn_cast<StoreInst>(*ui);
- if (!StoreI) {
- continue; // not in pattern
- }
-
- // Also check that the store uses the loaded value as the value operand
- if (StoreI->getValueOperand() != LoadI) {
- continue;
- }
-
- DEBUG(errs() << "-------Found store instruction\n");
-
- // Look for its bitcast, which is its pointer operand
- Value *StPtrOp = StoreI->getPointerOperand();
- DEBUG(errs() << "-------" << *StPtrOp << "\n");
- BitCastInst *BitCastI2 = dyn_cast<BitCastInst>(StPtrOp);
- DEBUG(errs() << "-------" << *BitCastI2 << "\n");
- if (!BitCastI2) {
- continue; // not in pattern
- }
-
- DEBUG(errs() << "-------- Found Bit Cast of store!\n");
- // found bitcast. Look for the second GEP, its from operand.
- Value *BCFromOp = BitCastI2->getOperand(0);
- GetElementPtrInst *GEPI2 = dyn_cast<GetElementPtrInst>(BCFromOp);
- DEBUG(errs() << "---------- " << *GEPI2 << "\n");
- if (!GEPI2) {
- continue; // not in pattern
- }
-
- if (!(GEPI2->hasOneUse())) {
- // does not fit this pattern - more than one uses
- // continue;
- // Keep GEPI around if it has other uses
- keepGEPI2 = true;
- }
- DEBUG(errs() << "---------- Found GEPI of Bitcast!\n");
-
- Value *PtrOp2 = GEPI2->getPointerOperand();
-
- // Found GEPI2. TODO: kind of confused as o what checks I need to add here,
- // let's add them together- all the code for int-float type checks is
- // already above.
-
- // Assume we found pattern
- if (!keepGEPI) {
- IItoRemove.push_back(GEPI);
- DEBUG(errs() << "Pushing " << *GEPI << " for removal\n");
- } else {
- DEBUG(errs() << "Keeping " << *GEPI << " since it has multiple uses!\n");
- }
- IItoRemove.push_back(BitCastI);
- DEBUG(errs() << "Pushing " << *BitCastI << " for removal\n");
- IItoRemove.push_back(LoadI);
- DEBUG(errs() << "Pushing " << *LoadI << " for removal\n");
- IItoRemove.push_back(GEPI2);
- DEBUG(errs() << "Pushing " << *GEPI2 << " for removal\n");
- IItoRemove.push_back(BitCastI2);
- DEBUG(errs() << "Pushing " << *BitCastI2 << " for removal\n");
- if (!keepGEPI2) {
- IItoRemove.push_back(StoreI);
- DEBUG(errs() << "Pushing " << *StoreI << " for removal\n");
- } else {
-
- DEBUG(errs() << "Keeping " << *StoreI
- << " since it has multiple uses!\n");
- }
-
- std::vector<Value *> GEPlIndex;
- if (GEPI->hasIndices()) {
- for (auto ii = GEPI->idx_begin(); ii != GEPI->idx_end(); ++ii) {
- Value *Index = dyn_cast<Value>(&*ii);
- DEBUG(errs() << "GEP-1 Index: " << *Index << "\n");
- GEPlIndex.push_back(Index);
- }
- }
- // ArrayRef<Value*> GEPlArrayRef(GEPlIndex);
-
- std::vector<Value *> GEPsIndex;
- if (GEPI2->hasIndices()) {
- for (auto ii = GEPI2->idx_begin(); ii != GEPI2->idx_end(); ++ii) {
- Value *Index = dyn_cast<Value>(&*ii);
- DEBUG(errs() << "GEP-2 Index: " << *Index << "\n");
- GEPsIndex.push_back(Index);
- }
- }
- // ArrayRef<Value*> GEPsArrayRef(GEPlIndex);
-
- // ArrayRef<Value*>(GEPI->idx_begin(), GEPI->idx_end());
- GetElementPtrInst *newlGEP = GetElementPtrInst::Create(
- GEPI->getSourceElementType(), // Type::getFloatTy(M.getContext()),
- PtrOp, // operand from 1st GEP
- ArrayRef<Value *>(GEPlIndex), Twine(), StoreI);
- DEBUG(errs() << "Adding: " << *newlGEP << "\n");
- // insert load before GEPI
- LoadInst *newLoadI =
- new LoadInst(Type::getFloatTy(M.getContext()),
- newlGEP, // new GEP
- Twine(), LoadI->isVolatile(), LoadI->getAlignment(),
- LoadI->getOrdering(), LoadI->getSyncScopeID(), StoreI);
- DEBUG(errs() << "Adding: " << *newLoadI << "\n");
- // same for GEP for store, for store operand
- GetElementPtrInst *newsGEP = GetElementPtrInst::Create(
- GEPI2->getSourceElementType(), // Type::getFloatTy(M.getContext()),
- PtrOp2, // operand from 2nd GEP
- ArrayRef<Value *>(GEPsIndex), Twine(), StoreI);
- DEBUG(errs() << "Adding: " << *newsGEP << "\n");
- // insert store before GEPI
- StoreInst *newStoreI =
- new StoreInst(newLoadI,
- newsGEP, // new GEP
- StoreI->isVolatile(), StoreI->getAlignment(),
- StoreI->getOrdering(), StoreI->getSyncScopeID(), StoreI);
- DEBUG(errs() << "Adding: " << *newStoreI << "\n");
- }
+ std::string name;
+ if(II->getIntrinsicID() == Intrinsic::visc_atomic_add)
+ name = "atomic_add";
+ else if(II->getIntrinsicID() == Intrinsic::visc_atomic_sub)
+ name = "atomic_sub";
+ else if(II->getIntrinsicID() == Intrinsic::visc_atomic_xchg)
+ name = "atomic_xchg";
+ else if(II->getIntrinsicID() == Intrinsic::visc_atomic_min)
+ name = "atomic_min";
+ else if(II->getIntrinsicID() == Intrinsic::visc_atomic_max)
+ name = "atomic_max";
+ else if(II->getIntrinsicID() == Intrinsic::visc_atomic_and)
+ name = "atomic_and";
+ else if(II->getIntrinsicID() == Intrinsic::visc_atomic_or)
+ name = "atomic_or";
+ else if(II->getIntrinsicID() == Intrinsic::visc_atomic_xor)
+ name = "atomic_xor";
+ Type* paramTypes[] = {PtrTy, Val->getType()};
+ FunctionType * AtomFuncT = FunctionType::get(II->getType(), ArrayRef<Type*>(paramTypes,2), false);
+ FunctionCallee AtomFunc = KernelM->getOrInsertFunction(name, AtomFuncT);
+
+ Value* Params[] = {Ptr, Val};
+ CallInst* AtomCI = CallInst::Create(AtomFunc, ArrayRef<Value*>(Params,2), II->getName(), II);
+ DEBUG(errs() << "Substitute with: " << *AtomCI << "\n");
+ II->replaceAllUsesWith(AtomCI);
+ IItoRemove.push_back(II);
+ }
+ break;
+ default:
+ llvm_unreachable("Unknown VISC Intrinsic!");
+ break;
+ }
+
+ }
+ else if(MemCpyInst *MemCpyI = dyn_cast<MemCpyInst>(I)) {
+ IRBuilder<> Builder(I);
+ Value *Source = MemCpyI->getSource();
+ Value *Destination = MemCpyI->getArgOperand(0)->stripPointerCasts();
+ Value *Length = MemCpyI->getOperand(2);
+ DEBUG(errs() << "Found memcpy instruction: " << *I << "\n");
+ DEBUG(errs() << "Source: " << *Source << "\n");
+ DEBUG(errs() << "Destination: " << *Destination << "\n");
+ DEBUG(errs() << "Length: " << *Length << "\n");
+
+ size_t memcpy_length;
+ unsigned int memcpy_count;
+ if (ConstantInt* CI = dyn_cast<ConstantInt>(Length)) {
+ if (CI->getBitWidth() <= 64) {
+ memcpy_length = CI->getSExtValue();
+ DEBUG(errs() << "Memcpy lenght = " << memcpy_length << "\n");
+ Type *Source_Type = Source->getType()->getPointerElementType();
+ DEBUG(errs() << "Source Type : " << *Source_Type << "\n");
+ memcpy_count = memcpy_length / (Source_Type->getPrimitiveSizeInBits() / 8);
+ DEBUG(errs() << "Memcpy count = " << memcpy_count << "\n");
+ if (GetElementPtrInst *sourceGEPI = dyn_cast<GetElementPtrInst>(Source)) {
+ if (GetElementPtrInst *destGEPI = dyn_cast<GetElementPtrInst>(Destination)) {
+ Value *SourcePtrOperand = sourceGEPI->getPointerOperand();
+ Value *DestPtrOperand = destGEPI->getPointerOperand();
+ for(int i = 0; i < memcpy_count; ++i) {
+ Constant *increment;
+ LoadInst *newLoadI;
+ StoreInst *newStoreI;
+ // First, need to increment the correct index for both source and dest
+ // This invluves checking to see how many indeces the GEP has
+ // Assume for now only 1 or 2 are the viable options.
+
+ std::vector<Value*> GEPlIndex;
+ if (sourceGEPI->getNumIndices() == 1) {
+ Value *Index = sourceGEPI->getOperand(1);
+ increment = ConstantInt::get(Index->getType(), i, false);
+ Value *incAdd = Builder.CreateAdd(Index, increment);
+ DEBUG(errs() << "Add: " << *incAdd << "\n");
+ GEPlIndex.push_back(incAdd);
+ Value *newGEPIl = Builder.CreateGEP(SourcePtrOperand, ArrayRef<Value*>(GEPlIndex));
+ DEBUG(errs() << "Load GEP: " << *newGEPIl << "\n");
+ newLoadI = Builder.CreateLoad(newGEPIl);
+ DEBUG(errs() << "Load: " << *newLoadI << "\n");
+ } else {
+ llvm_unreachable("Unhandled case where source GEPI has more than 1 indices!\n");
+ }
+
+
+ std::vector<Value*> GEPsIndex;
+ if (destGEPI->getNumIndices() == 1) {
+
+ } else if (destGEPI->getNumIndices() == 2) {
+ Value *Index0 = destGEPI->getOperand(1);
+ GEPsIndex.push_back(Index0);
+ Value *Index1 = destGEPI->getOperand(2);
+ increment = ConstantInt::get(Index1->getType(), i, false);
+ Value *incAdd = Builder.CreateAdd(Index1, increment);
+ DEBUG(errs() << "Add: " << *incAdd << "\n");
+ GEPsIndex.push_back(incAdd);
+ Value *newGEPIs = Builder.CreateGEP(DestPtrOperand, ArrayRef<Value*>(GEPsIndex));
+ DEBUG(errs() << "Store GEP: " << *newGEPIs << "\n");
+ newStoreI = Builder.CreateStore(newLoadI, newGEPIs, MemCpyI->isVolatile());
+ DEBUG(errs() << "Store: " << *newStoreI << "\n");
+ } else {
+ llvm_unreachable("Unhandled case where dest GEPI has more than 2 indices!\n");
+ }
+ }
+ IItoRemove.push_back(sourceGEPI);
+ IItoRemove.push_back(destGEPI);
+ Instruction *destBitcastI = dyn_cast<Instruction>(MemCpyI->getArgOperand(0));
+ Instruction *sourceBitcastI = dyn_cast<Instruction>(MemCpyI->getArgOperand(1));
+ IItoRemove.push_back(destBitcastI);
+ IItoRemove.push_back(sourceBitcastI);
+ IItoRemove.push_back(MemCpyI);
+ }
+ }
+
+ }
+ } else {
+ llvm_unreachable("MEMCPY length is not a constant, not handled!\n");
+ }
+ // llvm_unreachable("HERE!");
+ }
+
+ else if(CallInst* CI = dyn_cast<CallInst>(I)) {
+ DEBUG(errs() << "Found a call: " << *CI << "\n");
+ Function* calleeF = cast<Function>(CI->getCalledValue()->stripPointerCasts());
+ if(calleeF->isDeclaration()) {
+ // Add the declaration to kernel module
+ if (calleeF->getName() == "sqrtf") {
+ calleeF->setName(Twine("sqrt"));
+ DEBUG(errs() << "CaleeF: " << *calleeF << "\n");
+ DEBUG(errs() << "CI: " << *CI << "\n");
+ } else if (calleeF->getName() == "rsqrtf") {
+ calleeF->setName(Twine("rsqrt"));
+ DEBUG(errs() << "CaleeF: " << *calleeF << "\n");
+ DEBUG(errs() << "CI: " << *CI << "\n");
+ }
+ DEBUG(errs() << "Adding declaration to Kernel module: " << *calleeF << "\n");
+ KernelM->getOrInsertFunction(calleeF->getName(), calleeF->getFunctionType());
+ }
+ else {
+ // Check if the called function has already been cloned before.
+ Function *NewFunc = CloneAndReplaceCall(CI, calleeF);
+ // Iterate over the new function to see if it calls any other functions
+ // in the module.
+ for(inst_iterator i = inst_begin(NewFunc), e = inst_end(NewFunc); i != e; ++i) {
+ if(auto *Call = dyn_cast<CallInst>(&*i)) {
+ Function *CalledFunc = cast<Function>(Call->getCalledValue()->stripPointerCasts());
+ CloneAndReplaceCall(Call, CalledFunc);
+ }
+ }
+ }
+ //TODO: how to handle address space qualifiers in load/store
+ }
+
+ }
+ // search for pattern where float is being casted to int and loaded/stored and change it.
+ DEBUG(errs() << "finding pattern for replacement!\n");
+ for (inst_iterator i = inst_begin(F_nvptx), e = inst_end(F_nvptx); i != e; ++i) {
+ bool cont = false;
+ bool keepGEPI = false;
+ bool keepGEPI2= false;
+ Instruction *I = &(*i);
+ GetElementPtrInst* GEPI = dyn_cast<GetElementPtrInst>(I);
+
+ if (!GEPI) {
+ // did nod find pattern start, continue
+ continue;
+ }
+ // may have found pattern, check
+ DEBUG(errs() << "GEPI " << *GEPI << "\n");
+ // print whatever we want for debug
+ Value* PtrOp = GEPI->getPointerOperand();
+ Type *SrcTy = GEPI->getSourceElementType();
+ unsigned GEPIaddrspace = GEPI->getAddressSpace();
+
+ if (SrcTy->isArrayTy())
+ DEBUG(errs() << *SrcTy << " is an array type! " << *(SrcTy->getArrayElementType()) << "\n");
+ else
+ DEBUG(errs() << *SrcTy << " is not an array type!\n");
+ // check that source element type is float
+ if (SrcTy->isArrayTy()) {
+ if (!(SrcTy->getArrayElementType()->isFloatTy())) {
+ DEBUG(errs() << "GEPI type is array but not float!\n");
+ continue;
+ }
+ }
+ else if (!(SrcTy->isFPOrFPVectorTy()/*isFloatTy()*/)) {
+ DEBUG(errs() << "GEPI type is " << *SrcTy << "\n");
+ // does not fit this pattern - no float GEP instruction
+ continue;
+ }
+ // check that addressspace is 1
+ // if (GEPIaddrspace != 1) {
+ // // does not fit this pattern - addrspace of pointer argument is not global
+ // continue;
+ // }
+ if (!(GEPI->hasOneUse())) {
+ // does not fit this pattern - more than one uses
+ //continue;
+ // Keep GEPI around if it has other uses
+ keepGEPI = true;
+ }
+ DEBUG(errs() << "Found GEPI " << *GEPI << "\n");
+
+ // 1st GEPI it has one use
+ // assert(GEPI->hasOneUse() && "GEPI has a single use");
+
+ // See if it is a bitcast
+ BitCastInst *BitCastI;
+ for (User * U : GEPI->users()) {
+ if(Instruction *ui = dyn_cast<Instruction> (U)) {
+ DEBUG(errs() << "--" << *ui << "\n");
+ if (isa<BitCastInst>(ui)) {
+ BitCastI = dyn_cast<BitCastInst>(ui);
+ DEBUG(errs() << "---Found bitcast as only use of GEP\n");
+ break;
+ }
+ }
+ DEBUG(errs() << "GEPI does not have a bitcast user, continue\n");
+ cont = true;
+ }
+ // for (Value::user_iterator ui = GEPI->user_begin(),
+ // ue = GEPI->user_end(); ui!=ue; ++ui) {
+ // DEBUG(errs() << "--" << *ui << "\n");
+ // if (isa<BitCastInst>(*ui)) {
+ // BitCastI = dyn_cast<BitCastInst>(*ui);
+ // DEBUG(errs() << "Found bitcast as only use of GEP\n");
+ // }
+ // }
+
+ if (cont/*!BitCastI*/) {
+ continue; // not in pattern
+ }
+
+ // DEBUG(errs() << *BitCastI << "\n");
+ // Otherwise, check that first operand is GEP and 2nd is i32*. 1st Operand has to be the GEP, since this is a use of the GEP.
+ Value *Op2 = BitCastI->getOperand(0);
+ DEBUG(errs() << "----" << *Op2 << "\n");
+ // assert(cast<Type>(Op2) && "Invalid Operand for Bitcast\n");
+ // Type *OpTy = cast<Type>(Op2);
+ Type *OpTy = BitCastI->getDestTy();
+ DEBUG(errs() << "---- Bitcast destination type: " << *OpTy << "\n");
+ // DEBUG(errs() << "---- " << *(Type::getInt32PtrTy(M.getContext(),1)) << "\n");
+ if (!(OpTy == Type::getInt32PtrTy(M.getContext(), GEPIaddrspace))) {
+ // maybe right syntax is (Type::getInt32Ty)->getPointerTo()
+ continue; // not in pattern
+ }
+
+ DEBUG(errs() << "----Here!\n");
+ // We are in GEP, bitcast.
+
+ // user_iterator, to find the load.
+
+ if (!(BitCastI->hasOneUse())) {
+ // does not fit this pattern - more than one uses
+ continue;
+ }
+ DEBUG(errs() << "----Bitcast has one use!\n");
+ // it has one use
+ assert(BitCastI->hasOneUse() && "BitCastI has a single use");
+ LoadInst *LoadI;
+ for (User * U : BitCastI->users()) {
+ if (Instruction *ui = dyn_cast<Instruction> (U)) {
+ DEBUG(errs() << "-----" << *ui << "\n");
+ if (isa<LoadInst>(ui)) {
+ LoadI = dyn_cast<LoadInst>(ui);
+ DEBUG(errs() << "-----Found load as only use of bitcast\n");
+ break;
+ }
+ }
+ DEBUG(errs() << "Bitcast does not have a load user, continue!\n");
+ cont = true;
+ }
+ // for (Value::user_iterator ui = BitCastI->user_begin(),
+ // ue = BitCastI->user_end(); ui!=ue; ++ui) {
+ // if (isa<LoadInst>(*ui)) {
+ // LoadI = dyn_cast<LoadInst>(*ui);
+ // errs() << "Found load as only use of bitcast\n";
+ // }
+ // }
+
+ if (cont) {
+ continue; // not in pattern
+ }
+
+ DEBUG("HERE!\n");
+ // check that we load from pointer we got from bitcast - assert - the unique argument must be the use we found it from
+ assert(LoadI->getPointerOperand() == BitCastI && "Unexpected Load Instruction Operand\n");
+
+ // Copy user_iterator, to find the store.
+
+ if (!(LoadI->hasOneUse())) {
+ // does not fit this pattern - more than one uses
+ continue;
+ // TODO: generalize: one load can have more than one store users
+ }
+
+ // it has one use
+ assert(LoadI->hasOneUse() && "LoadI has a single use");
+ Value::user_iterator ui = LoadI->user_begin();
+ // skipped loop, because is has a single use
+ StoreInst *StoreI = dyn_cast<StoreInst>(*ui);
+ if (!StoreI) {
+ continue; // not in pattern
+ }
+
+ // Also check that the store uses the loaded value as the value operand
+ if (StoreI->getValueOperand() != LoadI) {
+ continue;
+ }
+
+ DEBUG(errs() << "-------Found store instruction\n");
+
+ // Look for its bitcast, which is its pointer operand
+ Value *StPtrOp = StoreI->getPointerOperand();
+ DEBUG(errs() << "-------" << *StPtrOp << "\n");
+ BitCastInst *BitCastI2 = dyn_cast<BitCastInst>(StPtrOp);
+ DEBUG(errs() << "-------" << *BitCastI2 << "\n");
+ if (!BitCastI2) {
+ continue; //not in pattern
+ }
+
+ DEBUG(errs() << "-------- Found Bit Cast of store!\n" );
+ // found bitcast. Look for the second GEP, its from operand.
+ Value *BCFromOp = BitCastI2->getOperand(0);
+ GetElementPtrInst *GEPI2 = dyn_cast<GetElementPtrInst>(BCFromOp);
+ DEBUG(errs() << "---------- " << *GEPI2 << "\n");
+ if (!GEPI2) {
+ continue; //not in pattern
+ }
+
+ if (!(GEPI2->hasOneUse())) {
+ // does not fit this pattern - more than one uses
+ //continue;
+ // Keep GEPI around if it has other uses
+ keepGEPI2 = true;
+ }
+ DEBUG(errs() << "---------- Found GEPI of Bitcast!\n");
+
+ Value *PtrOp2 = GEPI2->getPointerOperand();
+
+ // Found GEPI2. TODO: kind of confused as o what checks I need to add here, let's add them together- all the code for int-float type checks is already above.
+
+ // Assume we found pattern
+ if (!keepGEPI) {
+ IItoRemove.push_back(GEPI);
+ DEBUG(errs() << "Pushing " << *GEPI << " for removal\n");
+ } else {
+ DEBUG(errs() << "Keeping " << *GEPI << " since it has multiple uses!\n");
+ }
+ IItoRemove.push_back(BitCastI);
+ DEBUG(errs() << "Pushing " << *BitCastI << " for removal\n");
+ IItoRemove.push_back(LoadI);
+ DEBUG(errs() << "Pushing " << *LoadI << " for removal\n");
+ IItoRemove.push_back(GEPI2);
+ DEBUG(errs() << "Pushing " << *GEPI2 << " for removal\n");
+ IItoRemove.push_back(BitCastI2);
+ DEBUG(errs() << "Pushing " << *BitCastI2 << " for removal\n");
+ if (!keepGEPI2) {
+ IItoRemove.push_back(StoreI);
+ DEBUG(errs() << "Pushing " << *StoreI << " for removal\n");
+ } else {
+
+ DEBUG(errs() << "Keeping " << *StoreI << " since it has multiple uses!\n");
+ }
+
+ std::vector<Value*> GEPlIndex;
+ if (GEPI->hasIndices()) {
+ for(auto ii = GEPI->idx_begin(); ii != GEPI->idx_end(); ++ii) {
+ Value *Index = dyn_cast<Value>(&*ii);
+ DEBUG(errs() << "GEP-1 Index: " << *Index << "\n");
+ GEPlIndex.push_back(Index);
+ }
+ }
+ // ArrayRef<Value*> GEPlArrayRef(GEPlIndex);
+
+ std::vector<Value*> GEPsIndex;
+ if (GEPI2->hasIndices()) {
+ for(auto ii = GEPI2->idx_begin(); ii != GEPI2->idx_end(); ++ii) {
+ Value *Index = dyn_cast<Value>(&*ii);
+ DEBUG(errs() << "GEP-2 Index: " << *Index << "\n");
+ GEPsIndex.push_back(Index);
+ }
+ }
+ // ArrayRef<Value*> GEPsArrayRef(GEPlIndex);
+
+
+
+ // ArrayRef<Value*>(GEPI->idx_begin(), GEPI->idx_end());
+ GetElementPtrInst* newlGEP =
+ GetElementPtrInst::Create(GEPI->getSourceElementType(), //Type::getFloatTy(M.getContext()),
+ PtrOp, // operand from 1st GEP
+ ArrayRef<Value*>(GEPlIndex),
+ Twine(),
+ StoreI);
+ DEBUG(errs() << "Adding: " << *newlGEP << "\n");
+ // insert load before GEPI
+ LoadInst *newLoadI =
+ new LoadInst(Type::getFloatTy(M.getContext()),
+ newlGEP, // new GEP
+ Twine(),
+ LoadI->isVolatile(),
+ LoadI->getAlignment(),
+ LoadI->getOrdering(),
+ LoadI->getSyncScopeID(),
+ StoreI);
+ DEBUG(errs() << "Adding: " << *newLoadI << "\n");
+ // same for GEP for store, for store operand
+ GetElementPtrInst* newsGEP =
+ GetElementPtrInst::Create(GEPI2->getSourceElementType(), // Type::getFloatTy(M.getContext()),
+ PtrOp2, // operand from 2nd GEP
+ ArrayRef<Value*>(GEPsIndex),
+ Twine(),
+ StoreI);
+ DEBUG(errs() << "Adding: " << *newsGEP << "\n");
+ // insert store before GEPI
+ StoreInst *newStoreI =
+ new StoreInst(newLoadI,
+ newsGEP, // new GEP
+ StoreI->isVolatile(),
+ StoreI->getAlignment(),
+ StoreI->getOrdering(),
+ StoreI->getSyncScopeID(),
+ StoreI);
+ DEBUG(errs() << "Adding: " << *newStoreI << "\n");
+
+ }
+
+ // We need to do this explicitly: DCE pass will not remove them because we
+ // have assumed theworst memory behaviour for these function calls
+ // Traverse the vector backwards, otherwise definitions are deleted while
+ // their subsequent uses are still around
+ for (auto *I : reverse(IItoRemove)) {
+ DEBUG(errs() << "Erasing: " << *I << "\n");
+ I->eraseFromParent();
+ }
+
+ // Removed the cloned functions from the parent module into the new module
+ for(auto *F : FuncToBeRemoved) {
+ F->removeFromParent(); //TODO: MARIA check
+ KernelM->getFunctionList().push_back(F);
+ }
+
+ addCLMetadata(F_nvptx);
+ kernel->KernelFunction = F_nvptx;
+ errs() << "Identified kernel - " << kernel->KernelFunction->getName() << "\n";
+ DEBUG(errs() << *KernelM);
+
+ return;
+}
- // We need to do this explicitly: DCE pass will not remove them because we
- // have assumed theworst memory behaviour for these function calls
- // Traverse the vector backwards, otherwise definitions are deleted while
- // their subsequent uses are still around
- for (auto *I : reverse(IItoRemove)) {
- DEBUG(errs() << "Erasing: " << *I << "\n");
- I->eraseFromParent();
- }
+bool DFG2LLVM_NVPTX::runOnModule(Module &M) {
+ errs() << "\nDFG2LLVM_NVPTX PASS\n";
- // Removed the cloned functions from the parent module into the new module
- for (auto *F : FuncToBeRemoved) {
- F->removeFromParent(); // TODO: MARIA check
- KernelM->getFunctionList().push_back(F);
- }
+ // Get the BuildDFG Analysis Results:
+ // - Dataflow graph
+ // - Maps from i8* hansles to DFNode and DFEdge
+ BuildDFG &DFG = getAnalysis<BuildDFG>();
- addCLMetadata(F_nvptx);
- kernel->KernelFunction = F_nvptx;
- DEBUG(errs() << "Identified kernel - " << kernel->KernelFunction->getName()
- << "\n");
- DEBUG(errs() << *KernelM);
+ // DFInternalNode *Root = DFG.getRoot();
+ std::vector<DFInternalNode*> Roots = DFG.getRoots();
+ // BuildDFG::HandleToDFNode &HandleToDFNodeMap = DFG.getHandleToDFNodeMap();
+ // BuildDFG::HandleToDFEdge &HandleToDFEdgeMap = DFG.getHandleToDFEdgeMap();
- return;
-}
+ // Visitor for Code Generation Graph Traversal
+ CGT_NVPTX *CGTVisitor = new CGT_NVPTX(M, DFG);
-bool DFG2LLVM_NVPTX::runOnModule(Module &M) {
- DEBUG(errs() << "\nDFG2LLVM_NVPTX PASS\n");
-
- // Get the BuildDFG Analysis Results:
- // - Dataflow graph
- // - Maps from i8* hansles to DFNode and DFEdge
- BuildDFG &DFG = getAnalysis<BuildDFG>();
-
- // DFInternalNode *Root = DFG.getRoot();
- std::vector<DFInternalNode *> Roots = DFG.getRoots();
- // BuildDFG::HandleToDFNode &HandleToDFNodeMap =
- // DFG.getHandleToDFNodeMap(); BuildDFG::HandleToDFEdge &HandleToDFEdgeMap
- // = DFG.getHandleToDFEdgeMap();
-
- // Visitor for Code Generation Graph Traversal
- CGT_NVPTX *CGTVisitor = new CGT_NVPTX(M, DFG);
-
- // Iterate over all the DFGs and produce code for each one of them
- for (auto rootNode : Roots) {
- // Initiate code generation for root DFNode
- CGTVisitor->visit(rootNode);
- }
+ // Iterate over all the DFGs and produce code for each one of them
+ for (auto rootNode: Roots) {
+ // Initiate code generation for root DFNode
+ CGTVisitor->visit(rootNode);
+ }
- CGTVisitor->writeKernelsModule();
+ CGTVisitor->writeKernelsModule();
- // TODO: Edit module epilogue to remove the HPVM intrinsic declarations
- delete CGTVisitor;
+ //TODO: Edit module epilogue to remove the VISC intrinsic declarations
+ delete CGTVisitor;
- return true;
+ return true;
}
std::string CGT_NVPTX::getKernelsModuleName(Module &M) {
- /*SmallString<128> currentDir;
- llvm::sys::fs::current_path(currentDir);
- std::string fileName = getFilenameFromModule(M);
- Twine output = Twine(currentDir) + "/Output/" + fileName + "";
- return output.str().append(".kernels.ll");*/
- std::string mid = M.getModuleIdentifier();
- return mid.append(".kernels.ll");
+ /*SmallString<128> currentDir;
+ llvm::sys::fs::current_path(currentDir);
+ std::string fileName = getFilenameFromModule(M);
+ Twine output = Twine(currentDir) + "/Output/" + fileName + "";
+ return output.str().append(".kernels.ll");*/
+ std::string mid = M.getModuleIdentifier();
+ return mid.append(".kernels.ll");
}
-void CGT_NVPTX::fixValueAddrspace(Value *V, unsigned addrspace) {
- assert(isa<PointerType>(V->getType()) && "Value should be of Pointer Type!");
- PointerType *OldTy = cast<PointerType>(V->getType());
- PointerType *NewTy = PointerType::get(OldTy->getElementType(), addrspace);
- V->mutateType(NewTy);
- for (Value::user_iterator ui = V->user_begin(), ue = V->user_end(); ui != ue;
- ui++) {
- // Change all uses producing pointer type in same address space to new
- // addressspace.
- if (PointerType *PTy = dyn_cast<PointerType>((*ui)->getType())) {
- if (PTy->getAddressSpace() == OldTy->getAddressSpace()) {
- fixValueAddrspace(*ui, addrspace);
- }
- }
- }
+void CGT_NVPTX::fixValueAddrspace(Value* V, unsigned addrspace) {
+ assert(isa<PointerType>(V->getType())
+ && "Value should be of Pointer Type!");
+ PointerType* OldTy = cast<PointerType>(V->getType());
+ PointerType* NewTy = PointerType::get(OldTy->getElementType(), addrspace);
+ V->mutateType(NewTy);
+ for(Value::user_iterator ui = V->user_begin(), ue = V->user_end(); ui != ue; ui++) {
+ // Change all uses producing pointer type in same address space to new
+ // addressspace.
+ if(PointerType* PTy = dyn_cast<PointerType>((*ui)->getType())) {
+ if(PTy->getAddressSpace() == OldTy->getAddressSpace()) {
+ fixValueAddrspace(*ui, addrspace);
+ }
+ }
+ }
}
-std::vector<unsigned>
-CGT_NVPTX::globalToConstantMemoryOpt(std::vector<unsigned> *GlobalMemArgs,
- Function *F) {
- std::vector<unsigned> ConstantMemArgs;
- for (Function::arg_iterator ai = F->arg_begin(), ae = F->arg_end(); ai != ae;
- ++ai) {
- Argument *arg = &*ai;
- std::vector<unsigned>::iterator pos = std::find(
- GlobalMemArgs->begin(), GlobalMemArgs->end(), arg->getArgNo());
- // It has to be a global memory argument to be promotable
- if (pos == GlobalMemArgs->end())
- continue;
-
- // Check if it can/should be promoted
- if (canBePromoted(arg, F)) {
- DEBUG(errs() << "Promoting << " << arg->getName()
- << " to constant memory."
- << "\n");
- ConstantMemArgs.push_back(arg->getArgNo());
- GlobalMemArgs->erase(pos);
- }
- }
- return ConstantMemArgs;
-}
-Function *CGT_NVPTX::changeArgAddrspace(Function *F,
- std::vector<unsigned> &Args,
- unsigned addrspace) {
- unsigned idx = 0;
- std::vector<Type *> ArgTypes;
- for (Function::arg_iterator ai = F->arg_begin(), ae = F->arg_end(); ai != ae;
- ++ai) {
- Argument *arg = &*ai;
- DEBUG(errs() << *arg << "\n");
- unsigned argno = arg->getArgNo();
- if ((idx < Args.size()) && (argno == Args[idx])) {
- fixValueAddrspace(arg, addrspace);
- idx++;
- }
- ArgTypes.push_back(arg->getType());
- }
- FunctionType *newFT = FunctionType::get(F->getReturnType(), ArgTypes, false);
-
- // F->mutateType(PTy);
- Function *newF = cloneFunction(F, newFT, false);
- replaceNodeFunctionInIR(*F->getParent(), F, newF);
+std::vector<unsigned> CGT_NVPTX::globalToConstantMemoryOpt(std::vector<unsigned>* GlobalMemArgs, Function* F) {
+ std::vector<unsigned> ConstantMemArgs;
+ for(Function::arg_iterator ai = F->arg_begin(), ae = F->arg_end();
+ ai != ae; ++ai) {
+ Argument* arg = &*ai;
+ std::vector<unsigned>::iterator pos = std::find(GlobalMemArgs->begin(),
+ GlobalMemArgs->end(), arg->getArgNo());
+ // It has to be a global memory argument to be promotable
+ if(pos == GlobalMemArgs->end())
+ continue;
+
+ // Check if it can/should be promoted
+ if(canBePromoted(arg, F)) {
+ errs() << "Promoting << " << arg->getName() << " to constant memory."<< "\n";
+ ConstantMemArgs.push_back(arg->getArgNo());
+ GlobalMemArgs->erase(pos);
+ }
+ }
+ return ConstantMemArgs;
+}
- DEBUG(errs() << *newF->getFunctionType() << "\n" << *newF << "\n");
- return newF;
+Function* CGT_NVPTX::changeArgAddrspace(Function* F, std::vector<unsigned> &Args, unsigned addrspace) {
+ unsigned idx = 0;
+ std::vector<Type*> ArgTypes;
+ for(Function::arg_iterator ai = F->arg_begin(), ae = F->arg_end();
+ ai != ae; ++ai) {
+ Argument *arg = &*ai;
+ DEBUG(errs() << *arg << "\n");
+ unsigned argno = arg->getArgNo();
+ if ((idx < Args.size()) && (argno == Args[idx])) {
+ fixValueAddrspace(arg, addrspace);
+ idx++;
+ }
+ ArgTypes.push_back(arg->getType());
+ }
+ FunctionType* newFT = FunctionType::get(F->getReturnType(), ArgTypes, false);
+
+ //F->mutateType(PTy);
+ Function* newF = cloneFunction(F, newFT, false);
+ replaceNodeFunctionInIR(*F->getParent(), F, newF);
+
+ DEBUG(errs() << *newF->getFunctionType() << "\n" <<*newF << "\n");
+ return newF;
}
/* Add metadata to module KernelM, for OpenCL kernels */
void CGT_NVPTX::addCLMetadata(Function *F) {
- IRBuilder<> Builder(&*F->begin());
-
- SmallVector<Metadata *, 8> KernelMD;
- KernelMD.push_back(ValueAsMetadata::get(F));
-
- // TODO: There is additional metadata used by kernel files but we skip them as
- // they are not mandatory. In future they might be useful to enable
- // optimizations
-
- MDTuple *MDKernelNode = MDNode::get(KernelM->getContext(), KernelMD);
- NamedMDNode *MDN_kernels =
- KernelM->getOrInsertNamedMetadata("opencl.kernels");
- MDN_kernels->addOperand(MDKernelNode);
-
- KernelMD.push_back(MDString::get(KernelM->getContext(), "kernel"));
- // TODO: Replace 1 with the number of the kernel.
- // Add when support for multiple launces is added
- KernelMD.push_back(ValueAsMetadata::get(
- ConstantInt::get(Type::getInt32Ty(KernelM->getContext()), 1)));
- MDNode *MDNvvmAnnotationsNode = MDNode::get(KernelM->getContext(), KernelMD);
- NamedMDNode *MDN_annotations =
- KernelM->getOrInsertNamedMetadata("nvvm.annotations");
- MDN_annotations->addOperand(MDNvvmAnnotationsNode);
-}
+ IRBuilder<> Builder(&*F->begin());
-void CGT_NVPTX::writeKernelsModule() {
+ SmallVector<Metadata*,8> KernelMD;
+ KernelMD.push_back(ValueAsMetadata::get(F));
- // In addition to deleting all other functions, we also want to spiff it
- // up a little bit. Do this now.
- legacy::PassManager Passes;
+ // TODO: There is additional metadata used by kernel files but we skip them as
+ // they are not mandatory. In future they might be useful to enable
+ // optimizations
- DEBUG(errs() << "Writing to File --- ");
- DEBUG(errs() << getKernelsModuleName(M).c_str() << "\n");
- std::error_code EC;
- ToolOutputFile Out(getKernelsModuleName(M).c_str(), EC, sys::fs::F_None);
- if (EC) {
- DEBUG(errs() << EC.message() << '\n');
- }
+ MDTuple *MDKernelNode = MDNode::get(KernelM->getContext(), KernelMD);
+ NamedMDNode *MDN_kernels = KernelM->getOrInsertNamedMetadata("opencl.kernels");
+ MDN_kernels->addOperand(MDKernelNode);
- Passes.add(createPrintModulePass(Out.os()));
+ KernelMD.push_back(MDString::get(KernelM->getContext(), "kernel"));
+ // TODO: Replace 1 with the number of the kernel.
+ // Add when support for multiple launces is added
+ KernelMD.push_back(ValueAsMetadata::get(ConstantInt::get(Type::getInt32Ty(KernelM->getContext()),1)));
+ MDNode *MDNvvmAnnotationsNode = MDNode::get(KernelM->getContext(), KernelMD);
+ NamedMDNode *MDN_annotations = KernelM->getOrInsertNamedMetadata("nvvm.annotations");
+ MDN_annotations->addOperand(MDNvvmAnnotationsNode);
- Passes.run(*KernelM);
-
- // Declare success.
- Out.keep();
}
-Function *CGT_NVPTX::transformFunctionToVoid(Function *F) {
-
- DEBUG(errs() << "Transforming function to void: " << F->getName() << "\n");
- // FIXME: Maybe do that using the Node?
- StructType *FRetTy = dyn_cast<StructType>(F->getReturnType());
- assert(FRetTy && "Return Type must always be a struct");
+void CGT_NVPTX::writeKernelsModule() {
- // Keeps return statements, because we will need to replace them
- std::vector<ReturnInst *> RItoRemove;
- findReturnInst(F, RItoRemove);
+ // In addition to deleting all other functions, we also want to spiff it
+ // up a little bit. Do this now.
+ legacy::PassManager Passes;
- std::vector<Type *> RetArgTypes;
- std::vector<Argument *> RetArgs;
- std::vector<Argument *> Args;
- // Check for { } return struct, which means that the function returns void
- if (FRetTy->isEmptyTy()) {
+ errs() << "Writing to File --- ";
+ errs() << getKernelsModuleName(M).c_str() << "\n";
+ std::error_code EC;
+ ToolOutputFile Out(getKernelsModuleName(M).c_str(), EC, sys::fs::F_None);
+ if (EC) {
+ errs() << EC.message() << '\n';
+ }
- DEBUG(errs() << "\tFunction output struct is void\n");
- DEBUG(errs() << "\tNo parameters added\n");
+ Passes.add(
+ createPrintModulePass(Out.os()));
- // Replacing return statements with others returning void
- for (auto *RI : RItoRemove) {
- ReturnInst::Create((F->getContext()), 0, RI);
- RI->eraseFromParent();
- }
- DEBUG(errs() << "\tChanged return statements to return void\n");
- } else {
- // The struct has return values, thus needs to be converted to parameter
-
- // Iterate over all element types of return struct and add arguments to the
- // function
- for (unsigned i = 0; i < FRetTy->getNumElements(); i++) {
- Argument *RetArg =
- new Argument(FRetTy->getElementType(i)->getPointerTo(), "ret_arg", F);
- RetArgs.push_back(RetArg);
- RetArgTypes.push_back(RetArg->getType());
- DEBUG(errs() << "\tCreated parameter: " << *RetArg << "\n");
- }
+ Passes.run(*KernelM);
- DEBUG(errs() << "\tReplacing Return statements\n");
- // Replace return statements with extractValue and store instructions
- for (auto *RI : RItoRemove) {
- Value *RetVal = RI->getReturnValue();
- for (unsigned i = 0; i < RetArgs.size(); i++) {
- ExtractValueInst *EI = ExtractValueInst::Create(
- RetVal, ArrayRef<unsigned>(i), RetArgs[i]->getName() + ".val", RI);
- new StoreInst(EI, RetArgs[i], RI);
- }
- // assert(RetVal && "Return value should not be null at this point");
- // StructType* RetType = cast<StructType>(RetVal->getType());
- // assert(RetType && "Return type is not a struct");
-
- ReturnInst::Create((F->getContext()), 0, RI);
- RI->eraseFromParent();
- }
- }
- DEBUG(errs() << "\tReplaced return statements\n");
-
- // Create the argument type list with the added argument's type
- std::vector<Type *> ArgTypes;
- for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end();
- ai != ae; ++ai) {
- ArgTypes.push_back(ai->getType());
- }
- for (auto *RATy : RetArgTypes) {
- ArgTypes.push_back(RATy);
- }
-
- // Creating Args vector to use in cloning!
- for (Function::arg_iterator ai = F->arg_begin(), ae = F->arg_end(); ai != ae;
- ++ai) {
- Args.push_back(&*ai);
- }
- for (auto *ai : RetArgs) {
- Args.push_back(ai);
- }
+ // Declare success.
+ Out.keep();
+}
- // Adding new arguments to the function argument list, would not change the
- // function type. We need to change the type of this function to reflect the
- // added arguments
- Type *VoidRetType = Type::getVoidTy(F->getContext());
- FunctionType *newFT = FunctionType::get(VoidRetType, ArgTypes, F->isVarArg());
-
- // Change the function type
- // F->mutateType(PTy);
- Function *newF = cloneFunction(F, newFT, false, NULL, &Args);
- replaceNodeFunctionInIR(*F->getParent(), F, newF);
- // F->eraseFromParent();
- return newF;
+Function* CGT_NVPTX::transformFunctionToVoid(Function* F) {
+
+ DEBUG(errs() << "Transforming function to void: " << F->getName() << "\n");
+ // FIXME: Maybe do that using the Node?
+ StructType* FRetTy = dyn_cast<StructType>(F->getReturnType());
+ assert(FRetTy && "Return Type must always be a struct");
+
+ // Keeps return statements, because we will need to replace them
+ std::vector<ReturnInst *> RItoRemove;
+ findReturnInst(F, RItoRemove);
+
+ std::vector<Type *> RetArgTypes;
+ std::vector<Argument*> RetArgs;
+ std::vector<Argument*> Args;
+ // Check for { } return struct, which means that the function returns void
+ if (FRetTy->isEmptyTy()) {
+
+ DEBUG(errs() << "\tFunction output struct is void\n");
+ DEBUG(errs() << "\tNo parameters added\n");
+
+ // Replacing return statements with others returning void
+ for (auto *RI : RItoRemove) {
+ ReturnInst::Create((F->getContext()), 0, RI);
+ RI->eraseFromParent();
+ }
+ DEBUG(errs() << "\tChanged return statements to return void\n");
+ }
+ else {
+ // The struct has return values, thus needs to be converted to parameter
+
+ // Iterate over all element types of return struct and add arguments to the
+ // function
+ for (unsigned i=0; i<FRetTy->getNumElements(); i++) {
+ Argument* RetArg = new Argument(FRetTy->getElementType(i)->getPointerTo(), "ret_arg", F);
+ RetArgs.push_back(RetArg);
+ RetArgTypes.push_back(RetArg->getType());
+ DEBUG(errs() << "\tCreated parameter: " << *RetArg << "\n");
+ }
+
+ DEBUG(errs() << "\tReplacing Return statements\n");
+ // Replace return statements with extractValue and store instructions
+ for (auto *RI : RItoRemove) {
+ Value* RetVal = RI->getReturnValue();
+ for(unsigned i = 0; i < RetArgs.size(); i++) {
+ ExtractValueInst* EI = ExtractValueInst::Create(RetVal, ArrayRef<unsigned>(i),
+ RetArgs[i]->getName()+".val", RI);
+ new StoreInst(EI, RetArgs[i], RI);
+ }
+ // assert(RetVal && "Return value should not be null at this point");
+ // StructType* RetType = cast<StructType>(RetVal->getType());
+ // assert(RetType && "Return type is not a struct");
+
+ ReturnInst::Create((F->getContext()), 0, RI);
+ RI->eraseFromParent();
+
+ }
+ }
+ DEBUG(errs() << "\tReplaced return statements\n");
+
+ // Create the argument type list with the added argument's type
+ std::vector<Type*> ArgTypes;
+ for(Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end();
+ ai != ae; ++ai) {
+ ArgTypes.push_back(ai->getType());
+ }
+ for(auto *RATy: RetArgTypes) {
+ ArgTypes.push_back(RATy);
+ }
+
+ // Creating Args vector to use in cloning!
+ for(Function::arg_iterator ai = F->arg_begin(), ae = F->arg_end();
+ ai != ae; ++ai) {
+ Args.push_back(&*ai);
+ }
+ for(auto *ai : RetArgs) {
+ Args.push_back(ai);
+ }
+
+ // Adding new arguments to the function argument list, would not change the
+ // function type. We need to change the type of this function to reflect the
+ // added arguments
+ Type* VoidRetType = Type::getVoidTy(F->getContext());
+ FunctionType* newFT = FunctionType::get(VoidRetType, ArgTypes, F->isVarArg());
+
+ // Change the function type
+ //F->mutateType(PTy);
+ Function* newF = cloneFunction(F, newFT, false, NULL, &Args);
+ replaceNodeFunctionInIR(*F->getParent(), F, newF);
+ //F->eraseFromParent();
+ return newF;
}
/******************************************************************************
@@ -2138,344 +2102,314 @@ Function *CGT_NVPTX::transformFunctionToVoid(Function *F) {
// 1. No stores
// 2. Loads not dependent on getNodeInstanceID itrinsic
-static bool findLoadStoreUses(Value *V, std::vector<Value *> *UseList,
- std::vector<Value *> *VisitedList) {
- if (std::find(VisitedList->begin(), VisitedList->end(), V) !=
- VisitedList->end()) {
- DEBUG(errs() << "\tAlready visited value: " << *V << "\n");
- return false;
- }
- VisitedList->push_back(V);
- for (Value::user_iterator ui = V->user_begin(), ue = V->user_end(); ui != ue;
- ++ui) {
- Instruction *I = dyn_cast<Instruction>(*ui);
- if (!I) {
- // if use is not an instruction, then skip it
- continue;
- }
- DEBUG(errs() << "\t" << *I << "\n");
- if (isa<LoadInst>(I)) {
- DEBUG(errs() << "\tFound load instruction: " << *I << "\n");
- DEBUG(errs() << "\tAdd to use list: " << *V << "\n");
- UseList->push_back(V);
- } else if (isa<StoreInst>(I) || isa<AtomicRMWInst>(I)) {
- // found a store in use chain
- DEBUG(errs() << "Found store/atomicrmw instruction: " << *I << "\n");
- return true;
- } else if (BuildDFG::isHPVMIntrinsic(I)) {
- // If it is an atomic intrinsic, we found a store
- IntrinsicInst *II = dyn_cast<IntrinsicInst>(I);
- assert(II &&
- II->getCalledValue()->getName().startswith("llvm.hpvm.atomic") &&
- "Only hpvm atomic intrinsics can have an argument as input");
- return true;
- } else {
- DEBUG(errs() << "\tTraverse use chain of: " << *I << "\n");
- if (findLoadStoreUses(I, UseList, VisitedList))
- return true;
- }
- }
- return false;
+static bool findLoadStoreUses(Value* V, std::vector<Value*>*UseList, std::vector<Value*>*VisitedList) {
+ if(std::find(VisitedList->begin(), VisitedList->end(), V) != VisitedList->end()) {
+ DEBUG(errs() << "\tAlready visited value: " << *V << "\n");
+ return false;
+ }
+ VisitedList->push_back(V);
+ for(Value::user_iterator ui = V->user_begin(), ue = V->user_end();
+ ui != ue; ++ui) {
+ Instruction* I = dyn_cast<Instruction>(*ui);
+ if(!I) {
+ // if use is not an instruction, then skip it
+ continue;
+ }
+ DEBUG(errs() << "\t" << *I << "\n");
+ if(isa<LoadInst>(I)) {
+ DEBUG(errs() << "\tFound load instruction: " << *I << "\n");
+ DEBUG(errs() << "\tAdd to use list: " << *V << "\n");
+ UseList->push_back(V);
+ }
+ else if(isa<StoreInst>(I) || isa<AtomicRMWInst>(I)) {
+ // found a store in use chain
+ DEBUG(errs() << "Found store/atomicrmw instruction: " << *I << "\n");
+ return true;
+ }
+ else if(BuildDFG::isViscIntrinsic(I)) {
+ // If it is an atomic intrinsic, we found a store
+ IntrinsicInst* II = dyn_cast<IntrinsicInst>(I);
+ assert(II && II->getCalledValue()->getName().startswith("llvm.visc.atomic")
+ && "Only visc atomic intrinsics can have an argument as input");
+ return true;
+ }
+ else {
+ DEBUG(errs() << "\tTraverse use chain of: " << *I << "\n");
+ if(findLoadStoreUses(I, UseList, VisitedList))
+ return true;
+ }
+ }
+ return false;
}
-static bool isDependentOnNodeInstanceID(Value *V,
- std::vector<Value *> *DependenceList) {
- if (std::find(DependenceList->begin(), DependenceList->end(), V) !=
- DependenceList->end()) {
- DEBUG(errs() << "\tAlready visited value: " << *V << "\n");
- return false;
- }
- DependenceList->push_back(V);
- // If not an instruction, then not dependent on node instance id
- if (!isa<Instruction>(V) || isa<Constant>(V)) {
- DEBUG(errs() << "\tStop\n");
- return false;
- }
-
- Instruction *I = cast<Instruction>(V);
- for (unsigned i = 0; i < I->getNumOperands(); i++) {
- Value *operand = I->getOperand(i);
- if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(operand)) {
- if ((II->getIntrinsicID() == Intrinsic::hpvm_getNodeInstanceID_x ||
- II->getIntrinsicID() == Intrinsic::hpvm_getNodeInstanceID_y ||
- II->getIntrinsicID() == Intrinsic::hpvm_getNodeInstanceID_z)) {
- Value *Node = II->getArgOperand(0);
- IntrinsicInst *GN = dyn_cast<IntrinsicInst>(Node);
- assert(
- GN &&
- "NodeInstanceID operande should be node/parent node intrinsic\n");
- if (GN->getIntrinsicID() == Intrinsic::hpvm_getNode) {
- DEBUG(errs() << "\tDependency found on Node instance ID: " << *II
- << "\n");
- return true;
- }
- }
- }
- if (CmpInst *CI = dyn_cast<CmpInst>(operand)) {
- DEBUG(errs() << "Found compare instruction: " << *CI
- << "\nNot following its dependency list\n");
- continue;
- }
- DEBUG(errs() << "\tTraverse the operand chain of: " << *operand << "\n");
- if (isDependentOnNodeInstanceID(operand, DependenceList)) {
- return true;
- }
- }
- return false;
+static bool isDependentOnNodeInstanceID(Value* V, std::vector<Value*>*DependenceList) {
+ if(std::find(DependenceList->begin(), DependenceList->end(), V) != DependenceList->end()) {
+ DEBUG(errs() << "\tAlready visited value: " << *V << "\n");
+ return false;
+ }
+ DependenceList->push_back(V);
+ // If not an instruction, then not dependent on node instance id
+ if(!isa<Instruction>(V) || isa<Constant>(V)) {
+ DEBUG(errs() << "\tStop\n");
+ return false;
+ }
+
+ Instruction* I = cast<Instruction>(V);
+ for(unsigned i = 0; i < I->getNumOperands(); i++) {
+ Value* operand = I->getOperand(i);
+ if(IntrinsicInst* II = dyn_cast<IntrinsicInst>(operand)) {
+ if((II->getIntrinsicID() == Intrinsic::visc_getNodeInstanceID_x
+ || II->getIntrinsicID() == Intrinsic::visc_getNodeInstanceID_y
+ || II->getIntrinsicID() == Intrinsic::visc_getNodeInstanceID_z)) {
+ Value* Node = II->getArgOperand(0);
+ IntrinsicInst* GN = dyn_cast<IntrinsicInst>(Node);
+ assert(GN && "NodeInstanceID operande should be node/parent node intrinsic\n");
+ if(GN->getIntrinsicID() == Intrinsic::visc_getNode) {
+ DEBUG(errs() << "\tDependency found on Node instance ID: " << *II << "\n");
+ return true;
+ }
+ }
+ }
+ if(CmpInst* CI = dyn_cast<CmpInst>(operand)) {
+ DEBUG(errs() << "Found compare instruction: "<< *CI<<"\nNot following its dependency list\n");
+ continue;
+ }
+ DEBUG( errs() << "\tTraverse the operand chain of: " << *operand << "\n");
+ if(isDependentOnNodeInstanceID(operand, DependenceList)) {
+ return true;
+ }
+ }
+ return false;
}
// Function to check if argument arg can be changed to a constant memory pointer
-static bool canBePromoted(Argument *arg, Function *F) {
- DEBUG(errs() << "OPT: Check if Argument " << *arg
- << " can be changed to constant memory\n");
- std::vector<Value *> UseList;
- std::vector<Value *> VisitedList;
- // recursively traverse use chain
- // if find a store instruction return false, everything fails, cannot be
- // promoted
- // if find a load instruction as use, add the GEP instruction to list
- bool foundStore = findLoadStoreUses(arg, &UseList, &VisitedList);
- if (foundStore == true)
- return false;
- // See that the GEP instructions are not dependent on getNodeInstanceID
- // intrinsic
- DEBUG(errs() << foundStore
- << "\tNo Store Instruction found. Check dependence on node "
- "instance ID\n");
- std::vector<Value *> DependenceList;
- for (auto U : UseList) {
- if (isDependentOnNodeInstanceID(U, &DependenceList))
- return false;
- }
- DEBUG(errs() << "\tYes, Promotable to Constant Memory\n");
- return true;
+static bool canBePromoted(Argument* arg, Function* F) {
+ DEBUG(errs() << "OPT: Check if Argument " << *arg << " can be changed to constant memory\n");
+ std::vector<Value*> UseList;
+ std::vector<Value*> VisitedList;
+ // recursively traverse use chain
+ // if find a store instruction return false, everything fails, cannot be
+ // promoted
+ // if find a load instruction as use, add the GEP instruction to list
+ bool foundStore = findLoadStoreUses(arg, &UseList, &VisitedList);
+ if(foundStore == true)
+ return false;
+ // See that the GEP instructions are not dependent on getNodeInstanceID
+ // intrinsic
+ DEBUG(errs() << foundStore << "\tNo Store Instruction found. Check dependence on node instance ID\n");
+ std::vector<Value*>DependenceList;
+ for(auto U: UseList) {
+ if(isDependentOnNodeInstanceID(U, &DependenceList))
+ return false;
+ }
+ DEBUG(errs() << "\tYes, Promotable to Constant Memory\n");
+ return true;
}
+
// Calculate execute node parameters which include, number of diemnsions for
// dynamic instances of the kernel, local and global work group sizes.
-static void getExecuteNodeParams(Module &M, Value *&workDim, Value *&LocalWGPtr,
- Value *&GlobalWGPtr, Kernel *kernel,
- ValueToValueMapTy &VMap, Instruction *IB) {
-
- // Assign number of dimenstions a constant value
- workDim = ConstantInt::get(Type::getInt32Ty(M.getContext()), kernel->gridDim);
-
- // If local work group size if null
- if (!kernel->hasLocalWG()) {
- LocalWGPtr = Constant::getNullValue(Type::getInt64PtrTy(M.getContext()));
- } else {
- for (unsigned i = 0; i < kernel->localWGSize.size(); i++) {
- if (isa<Argument>(kernel->localWGSize[i]))
- kernel->localWGSize[i] = VMap[kernel->localWGSize[i]];
- }
- LocalWGPtr =
- genWorkGroupPtr(M, kernel->localWGSize, VMap, IB, "LocalWGSize");
- }
-
- for (unsigned i = 0; i < kernel->globalWGSize.size(); i++) {
- if (isa<Argument>(kernel->globalWGSize[i]))
- kernel->globalWGSize[i] = VMap[kernel->globalWGSize[i]];
- }
-
- // For OpenCL, global work group size is the total bumber of instances in each
- // dimension. So, multiply local and global dim limits.
- std::vector<Value *> globalWGSizeInsts;
- if (kernel->hasLocalWG()) {
- for (unsigned i = 0; i < kernel->gridDim; i++) {
- BinaryOperator *MulInst =
- BinaryOperator::Create(Instruction::Mul, kernel->globalWGSize[i],
- kernel->localWGSize[i], "", IB);
- globalWGSizeInsts.push_back(MulInst);
- }
- } else {
- globalWGSizeInsts = kernel->globalWGSize;
- }
- GlobalWGPtr = genWorkGroupPtr(M, globalWGSizeInsts, VMap, IB, "GlobalWGSize");
- DEBUG(errs() << "Pointer to global work group: " << *GlobalWGPtr << "\n");
+static void getExecuteNodeParams(Module &M, Value* &workDim, Value* &LocalWGPtr, Value*
+ &GlobalWGPtr, Kernel* kernel, ValueToValueMapTy& VMap, Instruction* IB) {
+
+ // Assign number of dimenstions a constant value
+ workDim = ConstantInt::get(Type::getInt32Ty(M.getContext()), kernel->gridDim);
+
+ // If local work group size if null
+ if(!kernel->hasLocalWG()) {
+ LocalWGPtr = Constant::getNullValue(Type::getInt64PtrTy(M.getContext()));
+ }
+ else {
+ for(unsigned i = 0; i < kernel->localWGSize.size(); i++) {
+ if(isa<Argument>(kernel->localWGSize[i]))
+ kernel->localWGSize[i] = VMap[kernel->localWGSize[i]];
+ }
+ LocalWGPtr = genWorkGroupPtr(M, kernel->localWGSize, VMap, IB, "LocalWGSize");
+ }
+
+ for(unsigned i = 0; i < kernel->globalWGSize.size(); i++) {
+ if(isa<Argument>(kernel->globalWGSize[i]))
+ kernel->globalWGSize[i] = VMap[kernel->globalWGSize[i]];
+ }
+
+ // For OpenCL, global work group size is the total bumber of instances in each
+ // dimension. So, multiply local and global dim limits.
+ std::vector<Value*> globalWGSizeInsts;
+ if(kernel->hasLocalWG()) {
+ for (unsigned i = 0; i < kernel->gridDim; i++) {
+ BinaryOperator* MulInst = BinaryOperator::Create(Instruction::Mul, kernel->globalWGSize[i], kernel->localWGSize[i], "", IB);
+ globalWGSizeInsts.push_back(MulInst);
+ }
+ }
+ else {
+ globalWGSizeInsts = kernel->globalWGSize;
+ }
+ GlobalWGPtr = genWorkGroupPtr(M, globalWGSizeInsts, VMap, IB, "GlobalWGSize");
+ DEBUG(errs() << "Pointer to global work group: " << *GlobalWGPtr << "\n");
}
// CodeGen for allocating space for Work Group on stack and returning a pointer
// to its address
-static Value *genWorkGroupPtr(Module &M, std::vector<Value *> WGSize,
- ValueToValueMapTy &VMap, Instruction *IB,
- const Twine &WGName) {
- Value *WGPtr;
- // Get int64_t and or ease of use
- Type *Int64Ty = Type::getInt64Ty(M.getContext());
-
- // Work Group type is [#dim x i64]
- Type *WGTy = ArrayType::get(Int64Ty, WGSize.size());
- // Allocate space of Global work group data on stack and get pointer to
- // first element.
- AllocaInst *WG = new AllocaInst(WGTy, 0, WGName, IB);
- WGPtr = BitCastInst::CreatePointerCast(WG, Int64Ty->getPointerTo(),
- WG->getName() + ".0", IB);
- Value *nextDim = WGPtr;
- DEBUG(errs() << *WGPtr << "\n");
-
- // Iterate over the number of dimensions and store the global work group
- // size in that dimension
- for (unsigned i = 0; i < WGSize.size(); i++) {
- DEBUG(errs() << *WGSize[i] << "\n");
- assert(WGSize[i]->getType()->isIntegerTy() &&
- "Dimension not an integer type!");
-
- if (WGSize[i]->getType() != Int64Ty) {
- // If number of dimensions are mentioned in any other integer format,
- // generate code to extend it to i64. We need to use the mapped value in
- // the new generated function, hence the use of VMap
- // FIXME: Why are we changing the kernel WGSize vector here?
- DEBUG(errs() << "Not i64. Zero extend required.\n");
- DEBUG(errs() << *WGSize[i] << "\n");
- CastInst *CI =
- BitCastInst::CreateIntegerCast(WGSize[i], Int64Ty, true, "", IB);
- DEBUG(errs() << "Bitcast done.\n");
- StoreInst *SI = new StoreInst(CI, nextDim, IB);
- DEBUG(errs() << "Zero extend done.\n");
- DEBUG(errs() << "\tZero extended work group size: " << *SI << "\n");
- } else {
- // Store the value representing work group size in ith dimension on
- // stack
- StoreInst *SI = new StoreInst(WGSize[i], nextDim, IB);
+static Value* genWorkGroupPtr(Module &M, std::vector<Value*> WGSize, ValueToValueMapTy& VMap, Instruction* IB, const Twine& WGName) {
+ Value* WGPtr;
+ // Get int64_t and or ease of use
+ Type* Int64Ty = Type::getInt64Ty(M.getContext());
+
+ // Work Group type is [#dim x i64]
+ Type* WGTy = ArrayType::get(Int64Ty, WGSize.size());
+ // Allocate space of Global work group data on stack and get pointer to
+ // first element.
+ AllocaInst* WG = new AllocaInst(WGTy, 0, WGName, IB);
+ WGPtr = BitCastInst::CreatePointerCast(WG, Int64Ty->getPointerTo(), WG->getName()+".0", IB);
+ Value* nextDim = WGPtr;
+ DEBUG(errs() << *WGPtr << "\n");
+
+ // Iterate over the number of dimensions and store the global work group
+ // size in that dimension
+ for(unsigned i=0; i < WGSize.size(); i++) {
+ DEBUG(errs() << *WGSize[i] << "\n");
+ assert(WGSize[i]->getType()->isIntegerTy() && "Dimension not an integer type!");
+
+ if(WGSize[i]->getType() != Int64Ty) {
+ // If number of dimensions are mentioned in any other integer format,
+ // generate code to extend it to i64. We need to use the mapped value in
+ // the new generated function, hence the use of VMap
+ // FIXME: Why are we changing the kernel WGSize vector here?
+ DEBUG(errs() << "Not i64. Zero extend required.\n");
+ DEBUG(errs() << *WGSize[i] << "\n");
+ CastInst* CI = BitCastInst::CreateIntegerCast(WGSize[i], Int64Ty, true, "", IB);
+ DEBUG(errs() << "Bitcast done.\n");
+ StoreInst* SI = new StoreInst(CI, nextDim, IB);
+ DEBUG(errs() << "Zero extend done.\n");
+ DEBUG(errs() << "\tZero extended work group size: " << *SI << "\n");
+ } else {
+ // Store the value representing work group size in ith dimension on
+ // stack
+ StoreInst* SI = new StoreInst(WGSize[i], nextDim, IB);
+
+ DEBUG(errs() << "\t Work group size: " << *SI << "\n");
+ }
+ if(i+1 < WGSize.size()) {
+ // Move to next dimension
+ GetElementPtrInst* GEP = GetElementPtrInst::Create(nullptr, nextDim,
+ ArrayRef<Value*>(ConstantInt::get(Int64Ty, 1)),
+ WG->getName()+"."+Twine(i+1),
+ IB);
+ DEBUG(errs() << "\tPointer to next dimension on stack: " << *GEP << "\n");
+ nextDim = GEP;
+ }
+ }
+ return WGPtr;
- DEBUG(errs() << "\t Work group size: " << *SI << "\n");
- }
- if (i + 1 < WGSize.size()) {
- // Move to next dimension
- GetElementPtrInst *GEP = GetElementPtrInst::Create(
- nullptr, nextDim, ArrayRef<Value *>(ConstantInt::get(Int64Ty, 1)),
- WG->getName() + "." + Twine(i + 1), IB);
- DEBUG(errs() << "\tPointer to next dimension on stack: " << *GEP << "\n");
- nextDim = GEP;
- }
- }
- return WGPtr;
}
// Get generated PTX binary name
-static std::string getPTXFilename(const Module &M) {
- std::string moduleID = M.getModuleIdentifier();
- moduleID.append(".kernels.cl");
- return moduleID;
+static std::string getPTXFilename(const Module& M) {
+ std::string moduleID = M.getModuleIdentifier();
+ moduleID.append(".kernels.cl");
+ return moduleID;
}
// Get the name of the input file from module ID
-static std::string getFilenameFromModule(const Module &M) {
- std::string moduleID = M.getModuleIdentifier();
- return moduleID.substr(moduleID.find_last_of("/") + 1);
+static std::string getFilenameFromModule(const Module& M) {
+ std::string moduleID = M.getModuleIdentifier();
+ return moduleID.substr(moduleID.find_last_of("/")+1);
}
// Changes the data layout of the Module to be compiled with NVPTX backend
// TODO: Figure out when to call it, probably after duplicating the modules
static void changeDataLayout(Module &M) {
- std::string nvptx32_layoutStr = "e-p:32:32-i64:64-v16:16-v32:32-n16:32:64";
- std::string nvptx64_layoutStr = "e-i64:64-v16:16-v32:32-n16:32:64";
+ std::string nvptx32_layoutStr = "e-p:32:32-i64:64-v16:16-v32:32-n16:32:64";
+ std::string nvptx64_layoutStr = "e-i64:64-v16:16-v32:32-n16:32:64";
- if (TARGET_PTX == 32)
- M.setDataLayout(StringRef(nvptx32_layoutStr));
- else if (TARGET_PTX == 64)
- M.setDataLayout(StringRef(nvptx64_layoutStr));
- else
- assert(false && "Invalid PTX target");
+ if (TARGET_PTX == 32)
+ M.setDataLayout(StringRef(nvptx32_layoutStr));
+ else if (TARGET_PTX == 64)
+ M.setDataLayout(StringRef(nvptx64_layoutStr));
+ else assert(false && "Invalid PTX target");
- return;
+ return;
}
static void changeTargetTriple(Module &M) {
- std::string nvptx32_TargetTriple = "nvptx--nvidiacl";
- std::string nvptx64_TargetTriple = "nvptx64--nvidiacl";
+ std::string nvptx32_TargetTriple = "nvptx--nvidiacl";
+ std::string nvptx64_TargetTriple = "nvptx64--nvidiacl";
- if (TARGET_PTX == 32)
- M.setTargetTriple(StringRef(nvptx32_TargetTriple));
- else if (TARGET_PTX == 64)
- M.setTargetTriple(StringRef(nvptx64_TargetTriple));
- else
- assert(false && "Invalid PTX target");
+ if (TARGET_PTX == 32)
+ M.setTargetTriple(StringRef(nvptx32_TargetTriple));
+ else if (TARGET_PTX == 64)
+ M.setTargetTriple(StringRef(nvptx64_TargetTriple));
+ else assert(false && "Invalid PTX target");
- return;
+ return;
}
// Helper function, populate a vector with all return statements in a function
-static void findReturnInst(Function *F,
- std::vector<ReturnInst *> &ReturnInstVec) {
- for (auto &BB : *F) {
- if (auto *RI = dyn_cast<ReturnInst>(BB.getTerminator()))
- ReturnInstVec.push_back(RI);
- }
+static void findReturnInst(Function* F, std::vector<ReturnInst *> & ReturnInstVec) {
+ for (auto &BB : *F) {
+ if(auto *RI = dyn_cast<ReturnInst>(BB.getTerminator()))
+ ReturnInstVec.push_back(RI);
+ }
}
-// Helper function, populate a vector with all IntrinsicID intrinsics in a
-// function
-static void findIntrinsicInst(Function *F, Intrinsic::ID IntrinsicID,
- std::vector<IntrinsicInst *> &IntrinsicInstVec) {
- for (inst_iterator i = inst_begin(F), e = inst_end(F); i != e; ++i) {
- Instruction *I = &(*i);
- IntrinsicInst *II = dyn_cast<IntrinsicInst>(I);
- if (II && II->getIntrinsicID() == IntrinsicID) {
- IntrinsicInstVec.push_back(II);
- }
- }
+// Helper function, populate a vector with all IntrinsicID intrinsics in a function
+static void findIntrinsicInst(Function* F, Intrinsic::ID IntrinsicID, std::vector<IntrinsicInst *> & IntrinsicInstVec) {
+ for (inst_iterator i = inst_begin(F), e = inst_end(F); i != e; ++i) {
+ Instruction *I = &(*i);
+ IntrinsicInst* II = dyn_cast<IntrinsicInst>(I);
+ if (II && II->getIntrinsicID() == IntrinsicID) {
+ IntrinsicInstVec.push_back(II);
+ }
+ }
}
-// Helper funtion, returns the atomicrmw op, corresponding to intrinsic atomic
-// op
+// Helper funtion, returns the atomicrmw op, corresponding to intrinsic atomic op
static AtomicRMWInst::BinOp getAtomicOp(Intrinsic::ID ID) {
- switch (ID) {
- case Intrinsic::hpvm_atomic_add:
- return AtomicRMWInst::Add;
- case Intrinsic::hpvm_atomic_sub:
- return AtomicRMWInst::Sub;
- case Intrinsic::hpvm_atomic_min:
- return AtomicRMWInst::Min;
- case Intrinsic::hpvm_atomic_umin:
- return AtomicRMWInst::UMin;
- case Intrinsic::hpvm_atomic_max:
- return AtomicRMWInst::Max;
- case Intrinsic::hpvm_atomic_umax:
- return AtomicRMWInst::UMax;
- // case Intrinsic::hpvm_atomic_inc: return AtomicRMWInst::Inc;
- // case Intrinsic::hpvm_atomic_dec: return AtomicRMWInst::Dec;
- case Intrinsic::hpvm_atomic_xchg:
- return AtomicRMWInst::Xchg;
- case Intrinsic::hpvm_atomic_and:
- return AtomicRMWInst::And;
- case Intrinsic::hpvm_atomic_or:
- return AtomicRMWInst::Or;
- case Intrinsic::hpvm_atomic_xor:
- return AtomicRMWInst::Xor;
- default:
- llvm_unreachable("Unsupported atomic intrinsic!");
- };
+ switch(ID) {
+ case Intrinsic::visc_atomic_add:
+ return AtomicRMWInst::Add;
+ case Intrinsic::visc_atomic_sub:
+ return AtomicRMWInst::Sub;
+ case Intrinsic::visc_atomic_min:
+ return AtomicRMWInst::Min;
+ case Intrinsic::visc_atomic_max:
+ return AtomicRMWInst::Max;
+ case Intrinsic::visc_atomic_xchg:
+ return AtomicRMWInst::Xchg;
+ case Intrinsic::visc_atomic_and:
+ return AtomicRMWInst::And;
+ case Intrinsic::visc_atomic_or:
+ return AtomicRMWInst::Or;
+ case Intrinsic::visc_atomic_xor:
+ return AtomicRMWInst::Xor;
+ default:
+ llvm_unreachable("Unsupported atomic intrinsic!");
+ };
}
+
// Helper funtion, returns the OpenCL function name, corresponding to atomic op
static std::string getAtomicOpName(Intrinsic::ID ID) {
- switch (ID) {
- case Intrinsic::hpvm_atomic_cmpxchg:
- return "atom_cmpxchg";
- case Intrinsic::hpvm_atomic_add:
- return "atom_add";
- case Intrinsic::hpvm_atomic_sub:
- return "atom_sub";
- case Intrinsic::hpvm_atomic_min:
- return "atom_min";
- case Intrinsic::hpvm_atomic_max:
- return "atom_max";
- case Intrinsic::hpvm_atomic_inc:
- return "atom_inc";
- case Intrinsic::hpvm_atomic_dec:
- return "atom_dec";
- case Intrinsic::hpvm_atomic_xchg:
- return "atom_xchg";
- case Intrinsic::hpvm_atomic_and:
- return "atom_and";
- case Intrinsic::hpvm_atomic_or:
- return "atom_or";
- case Intrinsic::hpvm_atomic_xor:
- return "atom_xor";
- default:
- llvm_unreachable("Unsupported atomic intrinsic!");
- };
+ switch(ID) {
+ case Intrinsic::visc_atomic_add:
+ return "atom_add";
+ case Intrinsic::visc_atomic_sub:
+ return "atom_sub";
+ case Intrinsic::visc_atomic_min:
+ return "atom_min";
+ case Intrinsic::visc_atomic_max:
+ return "atom_max";
+ case Intrinsic::visc_atomic_xchg:
+ return "atom_xchg";
+ case Intrinsic::visc_atomic_and:
+ return "atom_and";
+ case Intrinsic::visc_atomic_or:
+ return "atom_or";
+ case Intrinsic::visc_atomic_xor:
+ return "atom_xor";
+ default:
+ llvm_unreachable("Unsupported atomic intrinsic!");
+ };
}
} // End of namespace
@@ -2486,3 +2420,4 @@ static RegisterPass<DFG2LLVM_NVPTX> X("dfg2llvm-nvptx",
false /* does not modify the CFG */,
true /* transformation, *
* not just analysis */);
+
diff --git a/hpvm/lib/Transforms/DFG2LLVM_X86/DFG2LLVM_X86.cpp b/hpvm/lib/Transforms/DFG2LLVM_X86/DFG2LLVM_X86.cpp
index 8152817d9a..a0fa9fcde4 100644
--- a/hpvm/lib/Transforms/DFG2LLVM_X86/DFG2LLVM_X86.cpp
+++ b/hpvm/lib/Transforms/DFG2LLVM_X86/DFG2LLVM_X86.cpp
@@ -8,7 +8,7 @@
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "DFG2LLVM_X86"
-#include "SupportHPVM/DFG2LLVM.h"
+#include "SupportVISC/DFG2LLVM.h"
#include "llvm/IR/Constant.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/InstIterator.h"
@@ -25,29 +25,29 @@ using namespace llvm;
using namespace builddfg;
using namespace dfg2llvm;
-// HPVM Command line option to use timer or not
-static cl::opt<bool> HPVMTimer_X86("hpvm-timers-x86",
- cl::desc("Enable hpvm timers"));
+// VISC Command line option to use timer or not
+static cl::opt<bool> VISCTimer_X86("visc-timers-x86",
+ cl::desc("Enable visc timers"));
// Command line option to enable device abstraction or not
static cl::opt<bool>
- DeviceAbstraction("hpvm-eda", cl::init(false), cl::Hidden,
- cl::desc("Enable hpvm device abstraction"));
+ DeviceAbstraction("visc-eda", cl::init(false), cl::Hidden,
+ cl::desc("Enable visc device abstraction"));
namespace {
// Helper Functions
-static bool isHPVMCall_llvm_hpvm_policy_getVersion(Instruction *I) {
+static bool isVISCCall_llvm_visc_policy_getVersion(Instruction *I) {
if (!isa<CallInst>(I))
return false;
CallInst *CI = cast<CallInst>(I);
return (CI->getCalledValue()->stripPointerCasts()->getName())
- .equals("llvm_hpvm_policy_getVersion");
+ .equals("llvm_visc_policy_getVersion");
}
-CallInst *get_llvm_hpvm_policy_getVersion_call(Function *F) {
+CallInst *get_llvm_visc_policy_getVersion_call(Function *F) {
for (inst_iterator ib = inst_begin(F), ie = inst_end(F); ib != ie; ++ib) {
Instruction *I = &*ib;
- if (isHPVMCall_llvm_hpvm_policy_getVersion(I))
+ if (isVISCCall_llvm_visc_policy_getVersion(I))
return cast<CallInst>(I);
}
return NULL;
@@ -74,27 +74,27 @@ private:
// Member variables
FunctionCallee malloc;
- // HPVM Runtime API
- FunctionCallee llvm_hpvm_x86_launch;
- FunctionCallee llvm_hpvm_x86_wait;
- FunctionCallee llvm_hpvm_x86_argument_ptr;
-
- FunctionCallee llvm_hpvm_streamLaunch;
- FunctionCallee llvm_hpvm_streamPush;
- FunctionCallee llvm_hpvm_streamPop;
- FunctionCallee llvm_hpvm_streamWait;
- FunctionCallee llvm_hpvm_createBindInBuffer;
- FunctionCallee llvm_hpvm_createBindOutBuffer;
- FunctionCallee llvm_hpvm_createEdgeBuffer;
- FunctionCallee llvm_hpvm_createLastInputBuffer;
- FunctionCallee llvm_hpvm_createThread;
- // Constant* llvm_hpvm_freeThreads;
- FunctionCallee llvm_hpvm_bufferPush;
- FunctionCallee llvm_hpvm_bufferPop;
- FunctionCallee llvm_hpvm_x86_dstack_push;
- FunctionCallee llvm_hpvm_x86_dstack_pop;
- FunctionCallee llvm_hpvm_x86_getDimLimit;
- FunctionCallee llvm_hpvm_x86_getDimInstance;
+ // VISC Runtime API
+ FunctionCallee llvm_visc_x86_launch;
+ FunctionCallee llvm_visc_x86_wait;
+ FunctionCallee llvm_visc_x86_argument_ptr;
+
+ FunctionCallee llvm_visc_streamLaunch;
+ FunctionCallee llvm_visc_streamPush;
+ FunctionCallee llvm_visc_streamPop;
+ FunctionCallee llvm_visc_streamWait;
+ FunctionCallee llvm_visc_createBindInBuffer;
+ FunctionCallee llvm_visc_createBindOutBuffer;
+ FunctionCallee llvm_visc_createEdgeBuffer;
+ FunctionCallee llvm_visc_createLastInputBuffer;
+ FunctionCallee llvm_visc_createThread;
+ // Constant* llvm_visc_freeThreads;
+ FunctionCallee llvm_visc_bufferPush;
+ FunctionCallee llvm_visc_bufferPop;
+ FunctionCallee llvm_visc_x86_dstack_push;
+ FunctionCallee llvm_visc_x86_dstack_pop;
+ FunctionCallee llvm_visc_x86_getDimLimit;
+ FunctionCallee llvm_visc_x86_getDimInstance;
// Functions
std::vector<IntrinsicInst *> *getUseList(Value *LI);
@@ -120,7 +120,7 @@ private:
// Virtual Functions
void init() {
- HPVMTimer = HPVMTimer_X86;
+ VISCTimer = VISCTimer_X86;
TargetName = "X86";
}
void initRuntimeAPI();
@@ -177,7 +177,7 @@ bool DFG2LLVM_X86::runOnModule(Module &M) {
return true;
}
-// Initialize the HPVM runtime API. This makes it easier to insert these calls
+// Initialize the VISC runtime API. This makes it easier to insert these calls
void CGT_X86::initRuntimeAPI() {
// Load Runtime API Module
@@ -187,51 +187,51 @@ void CGT_X86::initRuntimeAPI() {
assert(LLVM_SRC_ROOT != NULL && "Define LLVM_SRC_ROOT environment variable!");
Twine llvmSrcRoot = LLVM_SRC_ROOT;
- Twine runtimeAPI = llvmSrcRoot + "/tools/hpvm/projects/hpvm-rt/hpvm-rt.ll";
+ Twine runtimeAPI = llvmSrcRoot + "/tools/hpvm/projects/visc-rt/visc-rt.ll";
runtimeModule = parseIRFile(runtimeAPI.str(), Err, M.getContext());
if (runtimeModule == NULL)
DEBUG(errs() << Err.getMessage());
else
- DEBUG(errs() << "Successfully loaded hpvm-rt API module\n");
+ DEBUG(errs() << "Successfully loaded visc-rt API module\n");
// Get or insert the global declarations for launch/wait functions
- DECLARE(llvm_hpvm_x86_launch);
+ DECLARE(llvm_visc_x86_launch);
DECLARE(malloc);
- DECLARE(llvm_hpvm_x86_wait);
- DECLARE(llvm_hpvm_x86_argument_ptr);
- DECLARE(llvm_hpvm_streamLaunch);
- DECLARE(llvm_hpvm_streamPush);
- DECLARE(llvm_hpvm_streamPop);
- DECLARE(llvm_hpvm_streamWait);
- DECLARE(llvm_hpvm_createBindInBuffer);
- DECLARE(llvm_hpvm_createBindOutBuffer);
- DECLARE(llvm_hpvm_createEdgeBuffer);
- DECLARE(llvm_hpvm_createLastInputBuffer);
- DECLARE(llvm_hpvm_createThread);
- // DECLARE(llvm_hpvm_freeThreads);
- DECLARE(llvm_hpvm_bufferPush);
- DECLARE(llvm_hpvm_bufferPop);
- DECLARE(llvm_hpvm_x86_dstack_push);
- DECLARE(llvm_hpvm_x86_dstack_pop);
- DECLARE(llvm_hpvm_x86_getDimLimit);
- DECLARE(llvm_hpvm_x86_getDimInstance);
+ DECLARE(llvm_visc_x86_wait);
+ DECLARE(llvm_visc_x86_argument_ptr);
+ DECLARE(llvm_visc_streamLaunch);
+ DECLARE(llvm_visc_streamPush);
+ DECLARE(llvm_visc_streamPop);
+ DECLARE(llvm_visc_streamWait);
+ DECLARE(llvm_visc_createBindInBuffer);
+ DECLARE(llvm_visc_createBindOutBuffer);
+ DECLARE(llvm_visc_createEdgeBuffer);
+ DECLARE(llvm_visc_createLastInputBuffer);
+ DECLARE(llvm_visc_createThread);
+ // DECLARE(llvm_visc_freeThreads);
+ DECLARE(llvm_visc_bufferPush);
+ DECLARE(llvm_visc_bufferPop);
+ DECLARE(llvm_visc_x86_dstack_push);
+ DECLARE(llvm_visc_x86_dstack_pop);
+ DECLARE(llvm_visc_x86_getDimLimit);
+ DECLARE(llvm_visc_x86_getDimInstance);
// Get or insert timerAPI functions as well if you plan to use timers
initTimerAPI();
// Insert init context in main
- Function *VI = M.getFunction("llvm.hpvm.init");
- assert(VI->getNumUses() == 1 && "__hpvm__init should only be used once");
+ Function *VI = M.getFunction("llvm.visc.init");
+ assert(VI->getNumUses() == 1 && "__visc__init should only be used once");
DEBUG(errs() << "Inserting x86 timer initialization\n");
Instruction *I = cast<Instruction>(*VI->user_begin());
initializeTimerSet(I);
- switchToTimer(hpvm_TimerID_NONE, I);
+ switchToTimer(visc_TimerID_NONE, I);
// Insert code for initializing the sceduling policy
FunctionCallee IP = M.getOrInsertFunction(
- "llvm_hpvm_policy_init",
- runtimeModule->getFunction("llvm_hpvm_policy_init")->getFunctionType());
+ "llvm_visc_policy_init",
+ runtimeModule->getFunction("llvm_visc_policy_init")->getFunctionType());
CallInst *IPCallInst = CallInst::Create(IP, ArrayRef<Value *>(), "", I);
DEBUG(errs() << *IPCallInst << "\n");
@@ -239,22 +239,22 @@ void CGT_X86::initRuntimeAPI() {
// device status simulation
if (DeviceAbstraction) {
FunctionCallee ID = M.getOrInsertFunction(
- "llvm_hpvm_deviceAbstraction_start",
- runtimeModule->getFunction("llvm_hpvm_deviceAbstraction_start")
+ "llvm_visc_deviceAbstraction_start",
+ runtimeModule->getFunction("llvm_visc_deviceAbstraction_start")
->getFunctionType());
CallInst *IDCallInst = CallInst::Create(ID, ArrayRef<Value *>(), "", I);
DEBUG(errs() << *IDCallInst << "\n");
}
- // Insert print instruction at hpvm exit
- Function *VC = M.getFunction("llvm.hpvm.cleanup");
- assert(VC->getNumUses() == 1 && "__hpvm__cleanup should only be used once");
+ // Insert print instruction at visc exit
+ Function *VC = M.getFunction("llvm.visc.cleanup");
+ assert(VC->getNumUses() == 1 && "__visc__cleanup should only be used once");
// Insert code for clearing the sceduling policy
I = cast<Instruction>(*VC->user_begin());
IP = M.getOrInsertFunction(
- "llvm_hpvm_policy_clear",
- runtimeModule->getFunction("llvm_hpvm_policy_clear")->getFunctionType());
+ "llvm_visc_policy_clear",
+ runtimeModule->getFunction("llvm_visc_policy_clear")->getFunctionType());
IPCallInst = CallInst::Create(IP, ArrayRef<Value *>(), "", I);
DEBUG(errs() << *IPCallInst << "\n");
@@ -265,8 +265,8 @@ void CGT_X86::initRuntimeAPI() {
// device status simulation
if (DeviceAbstraction) {
FunctionCallee ID = M.getOrInsertFunction(
- "llvm_hpvm_deviceAbstraction_end",
- runtimeModule->getFunction("llvm_hpvm_deviceAbstraction_end")
+ "llvm_visc_deviceAbstraction_end",
+ runtimeModule->getFunction("llvm_visc_deviceAbstraction_end")
->getFunctionType());
CallInst *IDCallInst = CallInst::Create(ID, ArrayRef<Value *>(), "", I);
DEBUG(errs() << *IDCallInst << "\n");
@@ -542,7 +542,7 @@ void CGT_X86::startNodeThread(DFNode *C, std::vector<Value *> Args,
// Call runtime to create the thread with these arguments
DEBUG(errs() << "Start Thread for child node: "
<< C->getFuncPointer()->getName() << "\n");
- // DEBUG(errs() << *llvm_hpvm_createThread << "\n");
+ // DEBUG(errs() << *llvm_visc_createThread << "\n");
DEBUG(errs() << *graphID->getType() << "\n");
DEBUG(errs() << *C_Pipeline->getType() << "\n");
DEBUG(errs() << *Struct->getType() << "\n");
@@ -551,7 +551,7 @@ void CGT_X86::startNodeThread(DFNode *C, std::vector<Value *> Args,
Struct->getName(), IB);
Value *CreateThreadArgs[] = {graphID, C_Pipeline, BI};
CallInst *CreateThread = CallInst::Create(
- llvm_hpvm_createThread, ArrayRef<Value *>(CreateThreadArgs, 3), "", IB);
+ llvm_visc_createThread, ArrayRef<Value *>(CreateThreadArgs, 3), "", IB);
}
Function *CGT_X86::createLaunchFunction(DFInternalNode *N) {
@@ -639,17 +639,17 @@ Function *CGT_X86::createLaunchFunction(DFInternalNode *N) {
Type::getInt32Ty(RI->getContext()), Edge->getSourcePosition());
Value *BindInCallArgs[] = {graphID, size, Int_ArgNo};
CI = CallInst::Create(
- llvm_hpvm_createBindInBuffer, ArrayRef<Value *>(BindInCallArgs, 3),
+ llvm_visc_createBindInBuffer, ArrayRef<Value *>(BindInCallArgs, 3),
"BindIn." + Edge->getDestDF()->getFuncPointer()->getName(), RI);
} else if (Edge->getDestDF()->isExitNode()) {
// Bind Output Edge
CI = CallInst::Create(
- llvm_hpvm_createBindOutBuffer, ArrayRef<Value *>(CallArgs, 2),
+ llvm_visc_createBindOutBuffer, ArrayRef<Value *>(CallArgs, 2),
"BindOut." + Edge->getSourceDF()->getFuncPointer()->getName(), RI);
} else {
// Streaming Edge
CI = CallInst::Create(
- llvm_hpvm_createEdgeBuffer, ArrayRef<Value *>(CallArgs, 2),
+ llvm_visc_createEdgeBuffer, ArrayRef<Value *>(CallArgs, 2),
Edge->getSourceDF()->getFuncPointer()->getName() + "." +
Edge->getDestDF()->getFuncPointer()->getName(),
RI);
@@ -668,7 +668,7 @@ Function *CGT_X86::createLaunchFunction(DFInternalNode *N) {
Value *size = ConstantExpr::getSizeOf(Type::getInt64Ty(NF->getContext()));
Value *CallArgs[] = {graphID, size};
CallInst *CI = CallInst::Create(
- llvm_hpvm_createLastInputBuffer, ArrayRef<Value *>(CallArgs, 2),
+ llvm_visc_createLastInputBuffer, ArrayRef<Value *>(CallArgs, 2),
"BindIn.isLastInput." + child->getFuncPointer()->getName(), RI);
NodeLastInputMap[child] = CI;
}
@@ -729,7 +729,7 @@ void CGT_X86::codeGenLaunchStreaming(DFInternalNode *Root) {
DEBUG(errs() << "Substitute launch intrinsic\n");
Value *LaunchInstArgs[] = {RootLaunch, LI->getArgOperand(1)};
CallInst *LaunchInst = CallInst::Create(
- llvm_hpvm_streamLaunch, ArrayRef<Value *>(LaunchInstArgs, 2),
+ llvm_visc_streamLaunch, ArrayRef<Value *>(LaunchInstArgs, 2),
"graph" + Root->getFuncPointer()->getName(), LI);
// ReplaceInstWithInst(LI, LaunchInst);
@@ -742,16 +742,16 @@ void CGT_X86::codeGenLaunchStreaming(DFInternalNode *Root) {
CallInst *CI;
Value *PushArgs[] = {LaunchInst, II->getOperand(1)};
switch (II->getIntrinsicID()) {
- case Intrinsic::hpvm_wait:
- CI = CallInst::Create(llvm_hpvm_streamWait, ArrayRef<Value *>(LaunchInst),
+ case Intrinsic::visc_wait:
+ CI = CallInst::Create(llvm_visc_streamWait, ArrayRef<Value *>(LaunchInst),
"");
break;
- case Intrinsic::hpvm_push:
- CI = CallInst::Create(llvm_hpvm_streamPush,
+ case Intrinsic::visc_push:
+ CI = CallInst::Create(llvm_visc_streamPush,
ArrayRef<Value *>(PushArgs, 2), "");
break;
- case Intrinsic::hpvm_pop:
- CI = CallInst::Create(llvm_hpvm_streamPop, ArrayRef<Value *>(LaunchInst),
+ case Intrinsic::visc_pop:
+ CI = CallInst::Create(llvm_visc_streamPop, ArrayRef<Value *>(LaunchInst),
"");
break;
default:
@@ -771,7 +771,7 @@ void CGT_X86::codeGenLaunch(DFInternalNode *Root) {
DEBUG(errs() << "Generating Launch Function\n");
// Get Launch Instruction
IntrinsicInst *LI = Root->getInstruction();
- switchToTimer(hpvm_TimerID_PTHREAD_CREATE, LI);
+ switchToTimer(visc_TimerID_PTHREAD_CREATE, LI);
DEBUG(errs() << "Generating Launch Function\n");
/* Now we have all the necessary global declarations necessary to generate the
@@ -802,14 +802,14 @@ void CGT_X86::codeGenLaunch(DFInternalNode *Root) {
ReturnInst *RI =
ReturnInst::Create(AppFunc->getContext(),
Constant::getNullValue(AppFunc->getReturnType()), BB);
- switchToTimer(hpvm_TimerID_ARG_UNPACK, RI);
+ switchToTimer(visc_TimerID_ARG_UNPACK, RI);
DEBUG(errs() << "Created Empty Launch Function\n");
// Find the X86 function generated for Root and
// Function* RootF_X86 = Root->getGenFunc();
- Function *RootF_X86 = Root->getGenFuncForTarget(hpvm::CPU_TARGET);
+ Function *RootF_X86 = Root->getGenFuncForTarget(visc::CPU_TARGET);
assert(RootF_X86 && "Error: No generated CPU function for Root node\n");
- assert(Root->hasX86GenFuncForTarget(hpvm::CPU_TARGET) &&
+ assert(Root->hasX86GenFuncForTarget(visc::CPU_TARGET) &&
"Error: Generated Function for Root node with no x86 wrapper\n");
// Generate a call to RootF_X86 with null parameters for now
@@ -837,8 +837,8 @@ void CGT_X86::codeGenLaunch(DFInternalNode *Root) {
CI->setArgOperand(i, elements[i]);
// Add timers around Call to RootF_X86 function
- switchToTimer(hpvm_TimerID_COMPUTATION, CI);
- switchToTimer(hpvm_TimerID_OUTPUT_PACK, RI);
+ switchToTimer(visc_TimerID_COMPUTATION, CI);
+ switchToTimer(visc_TimerID_OUTPUT_PACK, RI);
StructType *RootRetTy =
cast<StructType>(RootF_X86->getFunctionType()->getReturnType());
@@ -888,7 +888,7 @@ void CGT_X86::codeGenLaunch(DFInternalNode *Root) {
new StoreInst(CI, OutputAddrCast, RI);
}
- switchToTimer(hpvm_TimerID_NONE, RI);
+ switchToTimer(visc_TimerID_NONE, RI);
DEBUG(errs() << "Application specific function:\n");
DEBUG(errs() << *AppFunc << "\n");
@@ -896,7 +896,7 @@ void CGT_X86::codeGenLaunch(DFInternalNode *Root) {
// Substitute launch intrinsic main
Value *LaunchInstArgs[] = {AppFunc, LI->getArgOperand(1)};
CallInst *LaunchInst = CallInst::Create(
- llvm_hpvm_x86_launch, ArrayRef<Value *>(LaunchInstArgs, 2),
+ llvm_visc_x86_launch, ArrayRef<Value *>(LaunchInstArgs, 2),
"graph" + Root->getFuncPointer()->getName(), LI);
// ReplaceInstWithInst(LI, LaunchInst);
@@ -907,16 +907,16 @@ void CGT_X86::codeGenLaunch(DFInternalNode *Root) {
IntrinsicInst *II = UseList->at(i);
CallInst *CI;
switch (II->getIntrinsicID()) {
- case Intrinsic::hpvm_wait:
- CI = CallInst::Create(llvm_hpvm_x86_wait, ArrayRef<Value *>(LaunchInst),
+ case Intrinsic::visc_wait:
+ CI = CallInst::Create(llvm_visc_x86_wait, ArrayRef<Value *>(LaunchInst),
"");
break;
- case Intrinsic::hpvm_push:
- CI = CallInst::Create(llvm_hpvm_bufferPush, ArrayRef<Value *>(LaunchInst),
+ case Intrinsic::visc_push:
+ CI = CallInst::Create(llvm_visc_bufferPush, ArrayRef<Value *>(LaunchInst),
"");
break;
- case Intrinsic::hpvm_pop:
- CI = CallInst::Create(llvm_hpvm_bufferPop, ArrayRef<Value *>(LaunchInst),
+ case Intrinsic::visc_pop:
+ CI = CallInst::Create(llvm_visc_bufferPop, ArrayRef<Value *>(LaunchInst),
"");
break;
default:
@@ -970,10 +970,10 @@ void CGT_X86::invokeChild_X86(DFNode *C, Function *F_X86,
Function *CF = C->getFuncPointer();
// Function* CF_X86 = C->getGenFunc();
- Function *CF_X86 = C->getGenFuncForTarget(hpvm::CPU_TARGET);
+ Function *CF_X86 = C->getGenFuncForTarget(visc::CPU_TARGET);
assert(CF_X86 != NULL &&
"Found leaf node for which code generation has not happened yet!\n");
- assert(C->hasX86GenFuncForTarget(hpvm::CPU_TARGET) &&
+ assert(C->hasX86GenFuncForTarget(visc::CPU_TARGET) &&
"The generated function to be called from x86 backend is not an x86 "
"function\n");
DEBUG(errs() << "Invoking child node" << CF_X86->getName() << "\n");
@@ -1040,7 +1040,7 @@ void CGT_X86::invokeChild_X86(DFNode *C, Function *F_X86,
CI->getArgOperand(numArgs - 6 + 2) // iZ
};
- CallInst *Push = CallInst::Create(llvm_hpvm_x86_dstack_push,
+ CallInst *Push = CallInst::Create(llvm_visc_x86_dstack_push,
ArrayRef<Value *>(args, 7), "", CI);
DEBUG(errs() << "Push on stack: " << *Push << "\n");
// Insert call to runtime to pop the dim limits and instanceID from the depth
@@ -1053,7 +1053,7 @@ void CGT_X86::invokeChild_X86(DFNode *C, Function *F_X86,
assert(NextI->getParent() == CI->getParent() &&
"Next Instruction should also belong to the same basic block!");
- CallInst *Pop = CallInst::Create(llvm_hpvm_x86_dstack_pop, None, "", NextI);
+ CallInst *Pop = CallInst::Create(llvm_visc_x86_dstack_pop, None, "", NextI);
DEBUG(errs() << "Pop from stack: " << *Pop << "\n");
DEBUG(errs() << *CI->getParent()->getParent());
}
@@ -1156,7 +1156,7 @@ Function *CGT_X86::createFunctionFilter(DFNode *C) {
"streaming input edges\n");
// First read the termination condition variable islastInput
CallInst *isLastInputPop = CallInst::Create(
- llvm_hpvm_bufferPop, ArrayRef<Value *>(isLastInput), "", RI);
+ llvm_visc_bufferPop, ArrayRef<Value *>(isLastInput), "", RI);
CastInst *BI = BitCastInst::CreateIntegerCast(
isLastInputPop, Type::getInt64Ty(CF_Pipeline->getContext()), false,
@@ -1173,7 +1173,7 @@ Function *CGT_X86::createFunctionFilter(DFNode *C) {
++i) {
if (C->getInDFEdgeAt(i->getArgNo())->isStreamingEdge()) {
CallInst *bufferIn =
- CallInst::Create(llvm_hpvm_bufferPop,
+ CallInst::Create(llvm_visc_bufferPop,
ArrayRef<Value *>(InputArgs[i->getArgNo()]), "", RI);
CastInst *BI;
if (i->getType()->isPointerTy()) {
@@ -1196,7 +1196,7 @@ Function *CGT_X86::createFunctionFilter(DFNode *C) {
// DEBUG(errs() << "Type: " << *C->getGenFunc()->getType() << "\n");
// CallInst* CI = CallInst::Create(C->getGenFunc(), InputArgs,
// C->getGenFunc()->getName()+".output", RI);
- Function *CGenF = C->getGenFuncForTarget(hpvm::CPU_TARGET);
+ Function *CGenF = C->getGenFuncForTarget(visc::CPU_TARGET);
DEBUG(errs() << "Type: " << *CGenF->getType() << "\n");
CallInst *CI =
CallInst::Create(CGenF, InputArgs, CGenF->getName() + ".output", RI);
@@ -1222,7 +1222,7 @@ Function *CGT_X86::createFunctionFilter(DFNode *C) {
// Push to Output buffer
Value *bufferOutArgs[] = {OutputArgs[i], BI};
CallInst *bufferOut = CallInst::Create(
- llvm_hpvm_bufferPush, ArrayRef<Value *>(bufferOutArgs, 2), "", RI);
+ llvm_visc_bufferPush, ArrayRef<Value *>(bufferOutArgs, 2), "", RI);
}
// Add loop around the basic block, which exits the loop if isLastInput is
@@ -1236,9 +1236,9 @@ Function *CGT_X86::createFunctionFilter(DFNode *C) {
CondBB = CondStartI->getParent();
BodyBB = CI->getParent();
Instruction *CntI = NULL;
- CallInst *GetPolicyCI = get_llvm_hpvm_policy_getVersion_call(CGenF);
+ CallInst *GetPolicyCI = get_llvm_visc_policy_getVersion_call(CGenF);
- // If the node function calls the hpvm runtime call to get policy, we update
+ // If the node function calls the visc runtime call to get policy, we update
// it with the counter information. This means we need to pass an additional
// argument to the generated function, that is the iteration number, and then
// use it as an argument to the policy_getVersion call
@@ -1255,14 +1255,14 @@ Function *CGT_X86::createFunctionFilter(DFNode *C) {
}
NewArgTypes.push_back(Type::getInt64Ty(M.getContext()));
FunctionType *NewFT = FunctionType::get(NewRetTy, NewArgTypes, false);
- Function *NewCGenF = hpvmUtils::cloneFunction(CGenF, NewFT, false);
+ Function *NewCGenF = viscUtils::cloneFunction(CGenF, NewFT, false);
// At least one (the last) argument exists (we added it)
Function::arg_iterator ae = NewCGenF->arg_end();
--ae;
Argument *CntArg = &*ae;
CntArg->setName("iteration");
// Replace the old cpu gen func with this one
- C->addGenFunc(NewCGenF, hpvm::CPU_TARGET, true);
+ C->addGenFunc(NewCGenF, visc::CPU_TARGET, true);
// Add counter to the actual parameter list, to create the new call
InputArgs.push_back(CntI);
@@ -1272,7 +1272,7 @@ Function *CGT_X86::createFunctionFilter(DFNode *C) {
// Set second operand of the policy_getVersion call to the last function
// argument
- GetPolicyCI = get_llvm_hpvm_policy_getVersion_call(NewCGenF);
+ GetPolicyCI = get_llvm_visc_policy_getVersion_call(NewCGenF);
GetPolicyCI->setArgOperand(1, CntArg);
}
@@ -1292,13 +1292,13 @@ void CGT_X86::codeGen(DFInternalNode *N) {
// function before and nothing else needs to be done for this leaf node.
// if(N->getGenFunc() != NULL)
// return;
- if (!preferredTargetIncludes(N, hpvm::CPU_TARGET)) {
+ if (!preferredTargetIncludes(N, visc::CPU_TARGET)) {
DEBUG(errs() << "No CPU hint for node " << N->getFuncPointer()->getName()
<< " : skipping it\n");
return;
}
- assert(N->getGenFuncForTarget(hpvm::CPU_TARGET) == NULL &&
+ assert(N->getGenFuncForTarget(visc::CPU_TARGET) == NULL &&
"Error: Visiting a node for which code already generated\n");
// Sort children in topological order before code generation
@@ -1315,7 +1315,7 @@ void CGT_X86::codeGen(DFInternalNode *N) {
if (C->isDummyNode())
continue;
- if (!(C->hasX86GenFuncForTarget(hpvm::CPU_TARGET))) {
+ if (!(C->hasX86GenFuncForTarget(visc::CPU_TARGET))) {
errs() << "No CPU x86 version for child node "
<< C->getFuncPointer()->getName()
<< "\n Skip code gen for parent node "
@@ -1361,8 +1361,8 @@ void CGT_X86::codeGen(DFInternalNode *N) {
RI = cast<ReturnInst>(BB->getTerminator());
// Add generated function info to DFNode
- // N->setGenFunc(F_X86, hpvm::CPU_TARGET);
- N->addGenFunc(F_X86, hpvm::CPU_TARGET, true);
+ // N->setGenFunc(F_X86, visc::CPU_TARGET);
+ N->addGenFunc(F_X86, visc::CPU_TARGET, true);
// Loop over the arguments, to create the VMap.
dest_iterator = F_X86->arg_begin();
@@ -1445,13 +1445,13 @@ void CGT_X86::codeGen(DFInternalNode *N) {
// If not, we see which version exists, check that it is in fact an x86
// function and save it as the CPU_TARGET function
- // TODO: hpvm_id per node, so we can use this for id for policies
+ // TODO: visc_id per node, so we can use this for id for policies
// For now, use node function name and change it later
- Function *CF = N->getGenFuncForTarget(hpvm::CPU_TARGET);
- Function *GF = N->getGenFuncForTarget(hpvm::GPU_TARGET);
+ Function *CF = N->getGenFuncForTarget(visc::CPU_TARGET);
+ Function *GF = N->getGenFuncForTarget(visc::GPU_TARGET);
- bool CFx86 = N->hasX86GenFuncForTarget(hpvm::CPU_TARGET);
- bool GFx86 = N->hasX86GenFuncForTarget(hpvm::GPU_TARGET);
+ bool CFx86 = N->hasX86GenFuncForTarget(visc::CPU_TARGET);
+ bool GFx86 = N->hasX86GenFuncForTarget(visc::GPU_TARGET);
DEBUG(errs() << "Node: " << N->getFuncPointer()->getName() << " with tag "
<< N->getTag() << "\n");
@@ -1460,7 +1460,7 @@ void CGT_X86::codeGen(DFInternalNode *N) {
DEBUG(errs() << "GPU Fun: " << (GF ? GF->getName() : "null") << "\n");
DEBUG(errs() << "hasx86GenFuncForGPU : " << GFx86 << "\n");
- if (N->getTag() == hpvm::None) {
+ if (N->getTag() == visc::None) {
// No code is available for this node. This (usually) means that this
// node is a node that
// - from the accelerator backends has been mapped to an intermediate
@@ -1469,24 +1469,24 @@ void CGT_X86::codeGen(DFInternalNode *N) {
// take place
DEBUG(errs() << "No GenFunc - Skipping CPU code generation for node "
<< N->getFuncPointer()->getName() << "\n");
- } else if (hpvmUtils::isSingleTargetTag(N->getTag())) {
+ } else if (viscUtils::isSingleTargetTag(N->getTag())) {
// There is a single version for this node according to code gen hints.
// Therefore, we do not need to check the policy, we simply use the
// available implementation, whichever target it is for.
// Sanity check - to be removed TODO
switch (N->getTag()) {
- case hpvm::CPU_TARGET:
- assert(N->getGenFuncForTarget(hpvm::CPU_TARGET) && "");
- assert(N->hasX86GenFuncForTarget(hpvm::CPU_TARGET) && "");
- assert(!(N->getGenFuncForTarget(hpvm::GPU_TARGET)) && "");
- assert(!(N->hasX86GenFuncForTarget(hpvm::GPU_TARGET)) && "");
+ case visc::CPU_TARGET:
+ assert(N->getGenFuncForTarget(visc::CPU_TARGET) && "");
+ assert(N->hasX86GenFuncForTarget(visc::CPU_TARGET) && "");
+ assert(!(N->getGenFuncForTarget(visc::GPU_TARGET)) && "");
+ assert(!(N->hasX86GenFuncForTarget(visc::GPU_TARGET)) && "");
break;
- case hpvm::GPU_TARGET:
- assert(!(N->getGenFuncForTarget(hpvm::CPU_TARGET)) && "");
- assert(!(N->hasX86GenFuncForTarget(hpvm::CPU_TARGET)) && "");
- assert(N->getGenFuncForTarget(hpvm::GPU_TARGET) && "");
- assert(N->hasX86GenFuncForTarget(hpvm::GPU_TARGET) && "");
+ case visc::GPU_TARGET:
+ assert(!(N->getGenFuncForTarget(visc::CPU_TARGET)) && "");
+ assert(!(N->hasX86GenFuncForTarget(visc::CPU_TARGET)) && "");
+ assert(N->getGenFuncForTarget(visc::GPU_TARGET) && "");
+ assert(N->hasX86GenFuncForTarget(visc::GPU_TARGET) && "");
break;
default:
assert(false && "Unreachable: we checked that tag was single target!\n");
@@ -1499,8 +1499,8 @@ void CGT_X86::codeGen(DFInternalNode *N) {
if (DeviceAbstraction) {
Function *NodeGenFunc = NULL;
switch (N->getTag()) {
- case hpvm::GPU_TARGET:
- NodeGenFunc = N->getGenFuncForTarget(hpvm::GPU_TARGET);
+ case visc::GPU_TARGET:
+ NodeGenFunc = N->getGenFuncForTarget(visc::GPU_TARGET);
break;
default:
break;
@@ -1512,9 +1512,9 @@ void CGT_X86::codeGen(DFInternalNode *N) {
BasicBlock *BB = &*NodeGenFunc->begin();
std::vector<Value *> Args; // TODO: add the device type as argument?
FunctionCallee RTF = M.getOrInsertFunction(
- "llvm_hpvm_deviceAbstraction_waitOnDeviceStatus",
+ "llvm_visc_deviceAbstraction_waitOnDeviceStatus",
runtimeModule
- ->getFunction("llvm_hpvm_deviceAbstraction_waitOnDeviceStatus")
+ ->getFunction("llvm_visc_deviceAbstraction_waitOnDeviceStatus")
->getFunctionType());
CallInst *RTFInst =
CallInst::Create(RTF, Args, "", BB->getFirstNonPHI());
@@ -1522,17 +1522,17 @@ void CGT_X86::codeGen(DFInternalNode *N) {
}
Function *Ftmp = N->getGenFuncForTarget(N->getTag());
- N->removeGenFuncForTarget(hpvm::GPU_TARGET);
- N->setTag(hpvm::None);
- N->addGenFunc(Ftmp, hpvm::CPU_TARGET, true);
- N->setTag(hpvm::CPU_TARGET);
+ N->removeGenFuncForTarget(visc::GPU_TARGET);
+ N->setTag(visc::None);
+ N->addGenFunc(Ftmp, visc::CPU_TARGET, true);
+ N->setTag(visc::CPU_TARGET);
// Sanity checks - to be removed TODO
- CF = N->getGenFuncForTarget(hpvm::CPU_TARGET);
- GF = N->getGenFuncForTarget(hpvm::GPU_TARGET);
+ CF = N->getGenFuncForTarget(visc::CPU_TARGET);
+ GF = N->getGenFuncForTarget(visc::GPU_TARGET);
- CFx86 = N->hasX86GenFuncForTarget(hpvm::CPU_TARGET);
- GFx86 = N->hasX86GenFuncForTarget(hpvm::GPU_TARGET);
+ CFx86 = N->hasX86GenFuncForTarget(visc::CPU_TARGET);
+ GFx86 = N->hasX86GenFuncForTarget(visc::GPU_TARGET);
DEBUG(errs() << "After editing\n");
DEBUG(errs() << "Node: " << N->getFuncPointer()->getName() << " with tag "
@@ -1545,11 +1545,11 @@ void CGT_X86::codeGen(DFInternalNode *N) {
DEBUG(errs() << "Node Name (for policy) : "
<< N->getFuncPointer()->getName() << "\n");
- Function *CF = N->getGenFuncForTarget(hpvm::CPU_TARGET);
- Function *GF = N->getGenFuncForTarget(hpvm::GPU_TARGET);
+ Function *CF = N->getGenFuncForTarget(visc::CPU_TARGET);
+ Function *GF = N->getGenFuncForTarget(visc::GPU_TARGET);
- bool CFx86 = N->hasX86GenFuncForTarget(hpvm::CPU_TARGET);
- bool GFx86 = N->hasX86GenFuncForTarget(hpvm::GPU_TARGET);
+ bool CFx86 = N->hasX86GenFuncForTarget(visc::CPU_TARGET);
+ bool GFx86 = N->hasX86GenFuncForTarget(visc::GPU_TARGET);
// These assertions express what we can support with the current runtime.
// Code generation works the same way even for other target combinations.
@@ -1610,8 +1610,8 @@ void CGT_X86::codeGen(DFInternalNode *N) {
Args.push_back(
ConstantInt::get(Type::getInt64Ty(M.getContext()), -1, true));
FunctionCallee RTF = M.getOrInsertFunction(
- "llvm_hpvm_policy_getVersion",
- runtimeModule->getFunction("llvm_hpvm_policy_getVersion")
+ "llvm_visc_policy_getVersion",
+ runtimeModule->getFunction("llvm_visc_policy_getVersion")
->getFunctionType());
CallInst *RTFInst = CallInst::Create(RTF, Args, "", BBcurrent);
@@ -1646,9 +1646,9 @@ void CGT_X86::codeGen(DFInternalNode *N) {
// call
std::vector<Value *> Args; // TODO: add the device type as argument?
FunctionCallee RTF = M.getOrInsertFunction(
- "llvm_hpvm_deviceAbstraction_waitOnDeviceStatus",
+ "llvm_visc_deviceAbstraction_waitOnDeviceStatus",
runtimeModule
- ->getFunction("llvm_hpvm_deviceAbstraction_waitOnDeviceStatus")
+ ->getFunction("llvm_visc_deviceAbstraction_waitOnDeviceStatus")
->getFunctionType());
CallInst *RTFInst = CallInst::Create(RTF, Args, "", GenFuncCI);
}
@@ -1673,8 +1673,8 @@ void CGT_X86::codeGen(DFInternalNode *N) {
// Prepare arguments and function for call to wait for device runtime call
// std::vector<Value *> Args; // TODO: add the device type as argument?
// FunctionCallee RTF =
- // M.getOrInsertFunction("llvm_hpvm_deviceAbstraction_waitOnDeviceStatus",
- // runtimeModule->getFunction("llvm_hpvm_deviceAbstraction_waitOnDeviceStatus")->getFunctionType());
+ // M.getOrInsertFunction("llvm_visc_deviceAbstraction_waitOnDeviceStatus",
+ // runtimeModule->getFunction("llvm_visc_deviceAbstraction_waitOnDeviceStatus")->getFunctionType());
// CallInst *RTFInst = CallInst::Create(RTF, Args, "", GenFuncCI);
// }
// }
@@ -1684,9 +1684,9 @@ void CGT_X86::codeGen(DFInternalNode *N) {
// Now, make the node cpu gen func to be this one
// Remove all other versions and update the tag
- N->addGenFunc(F_wrapper, hpvm::CPU_TARGET, true);
- N->removeGenFuncForTarget(hpvm::GPU_TARGET);
- N->setTag(hpvm::CPU_TARGET);
+ N->addGenFunc(F_wrapper, visc::CPU_TARGET, true);
+ N->removeGenFuncForTarget(visc::GPU_TARGET);
+ N->setTag(visc::CPU_TARGET);
// assert(false && "got to the point where we have to combine\n");
}
@@ -1715,7 +1715,7 @@ void CGT_X86::codeGen(DFLeafNode *N) {
// if(N->getGenFunc() != NULL)
// return;
- if (!preferredTargetIncludes(N, hpvm::CPU_TARGET)) {
+ if (!preferredTargetIncludes(N, visc::CPU_TARGET)) {
DEBUG(errs() << "No CPU hint for node " << N->getFuncPointer()->getName()
<< " : skipping it\n");
@@ -1723,10 +1723,10 @@ void CGT_X86::codeGen(DFLeafNode *N) {
<< N->getFuncPointer()->getName() << "\n");
switch (N->getTag()) {
- case hpvm::GPU_TARGET:
+ case visc::GPU_TARGET:
// A leaf node should not have an x86 function for GPU
// by design of DFG2LLVM_NVPTX backend
- assert(!(N->hasX86GenFuncForTarget(hpvm::GPU_TARGET)) && "");
+ assert(!(N->hasX86GenFuncForTarget(visc::GPU_TARGET)) && "");
break;
default:
break;
@@ -1735,7 +1735,7 @@ void CGT_X86::codeGen(DFLeafNode *N) {
return;
}
- assert(N->getGenFuncForTarget(hpvm::CPU_TARGET) == NULL &&
+ assert(N->getGenFuncForTarget(visc::CPU_TARGET) == NULL &&
"Error: Visiting a node for which code already generated\n");
std::vector<IntrinsicInst *> IItoRemove;
@@ -1759,8 +1759,8 @@ void CGT_X86::codeGen(DFLeafNode *N) {
F_X86 = addIdxDimArgs(F_X86);
// Add generated function info to DFNode
- // N->setGenFunc(F_X86, hpvm::CPU_TARGET);
- N->addGenFunc(F_X86, hpvm::CPU_TARGET, true);
+ // N->setGenFunc(F_X86, visc::CPU_TARGET);
+ N->addGenFunc(F_X86, visc::CPU_TARGET, true);
// Go through the arguments, and any pointer arguments with in attribute need
// to have x86_argument_ptr call to get the x86 ptr of the argument
@@ -1768,7 +1768,7 @@ void CGT_X86::codeGen(DFLeafNode *N) {
// Create new BB
BasicBlock *EntryBB = &*F_X86->begin();
BasicBlock *BB =
- BasicBlock::Create(M.getContext(), "getHPVMPtrArgs", F_X86, EntryBB);
+ BasicBlock::Create(M.getContext(), "getVISCPtrArgs", F_X86, EntryBB);
BranchInst *Terminator = BranchInst::Create(EntryBB, BB);
// Insert calls
for (Function::arg_iterator ai = F_X86->arg_begin(), ae = F_X86->arg_end();
@@ -1776,7 +1776,7 @@ void CGT_X86::codeGen(DFLeafNode *N) {
if (F_X86->getAttributes().hasAttribute(ai->getArgNo() + 1,
Attribute::In)) {
assert(ai->getType()->isPointerTy() &&
- "Only pointer arguments can have hpvm in/out attributes ");
+ "Only pointer arguments can have visc in/out attributes ");
Function::arg_iterator aiNext = ai;
++aiNext;
Argument *size = &*aiNext;
@@ -1786,7 +1786,7 @@ void CGT_X86::codeGen(DFLeafNode *N) {
&*ai, Type::getInt8PtrTy(M.getContext()), ai->getName() + ".i8ptr",
Terminator);
Value *ArgPtrCallArgs[] = {BI, size};
- CallInst::Create(llvm_hpvm_x86_argument_ptr,
+ CallInst::Create(llvm_visc_x86_argument_ptr,
ArrayRef<Value *>(ArgPtrCallArgs, 2), "", Terminator);
}
}
@@ -1796,30 +1796,30 @@ void CGT_X86::codeGen(DFLeafNode *N) {
for (inst_iterator i = inst_begin(F_X86), e = inst_end(F_X86); i != e; ++i) {
Instruction *I = &(*i);
DEBUG(errs() << *I << "\n");
- // Leaf nodes should not contain HPVM graph intrinsics or launch
- assert(!BuildDFG::isHPVMLaunchIntrinsic(I) &&
+ // Leaf nodes should not contain VISC graph intrinsics or launch
+ assert(!BuildDFG::isViscLaunchIntrinsic(I) &&
"Launch intrinsic within a dataflow graph!");
- assert(!BuildDFG::isHPVMGraphIntrinsic(I) &&
- "HPVM graph intrinsic within a leaf dataflow node!");
+ assert(!BuildDFG::isViscGraphIntrinsic(I) &&
+ "VISC graph intrinsic within a leaf dataflow node!");
- if (BuildDFG::isHPVMQueryIntrinsic(I)) {
+ if (BuildDFG::isViscQueryIntrinsic(I)) {
IntrinsicInst *II = cast<IntrinsicInst>(I);
IntrinsicInst *ArgII;
DFNode *ArgDFNode;
/***********************************************************************
- * Handle HPVM Query intrinsics *
+ * Handle VISC Query intrinsics *
***********************************************************************/
switch (II->getIntrinsicID()) {
- /**************************** llvm.hpvm.getNode() *******************/
- case Intrinsic::hpvm_getNode: {
+ /**************************** llvm.visc.getNode() *******************/
+ case Intrinsic::visc_getNode: {
// add mapping <intrinsic, this node> to the node-specific map
Leaf_HandleToDFNodeMap[II] = N;
IItoRemove.push_back(II);
break;
}
- /************************* llvm.hpvm.getParentNode() ****************/
- case Intrinsic::hpvm_getParentNode: {
+ /************************* llvm.visc.getParentNode() ****************/
+ case Intrinsic::visc_getParentNode: {
// get the parent node of the arg node
// get argument node
ArgII = cast<IntrinsicInst>((II->getOperand(0))->stripPointerCasts());
@@ -1832,8 +1832,8 @@ void CGT_X86::codeGen(DFLeafNode *N) {
IItoRemove.push_back(II);
break;
}
- /*************************** llvm.hpvm.getNumDims() *****************/
- case Intrinsic::hpvm_getNumDims: {
+ /*************************** llvm.visc.getNumDims() *****************/
+ case Intrinsic::visc_getNumDims: {
// get node from map
// get the appropriate field
ArgII = cast<IntrinsicInst>((II->getOperand(0))->stripPointerCasts());
@@ -1846,10 +1846,10 @@ void CGT_X86::codeGen(DFLeafNode *N) {
IItoRemove.push_back(II);
break;
}
- /*********************** llvm.hpvm.getNodeInstanceID() **************/
- case Intrinsic::hpvm_getNodeInstanceID_x:
- case Intrinsic::hpvm_getNodeInstanceID_y:
- case Intrinsic::hpvm_getNodeInstanceID_z: {
+ /*********************** llvm.visc.getNodeInstanceID() **************/
+ case Intrinsic::visc_getNodeInstanceID_x:
+ case Intrinsic::visc_getNodeInstanceID_y:
+ case Intrinsic::visc_getNodeInstanceID_z: {
ArgII = cast<IntrinsicInst>((II->getOperand(0))->stripPointerCasts());
ArgDFNode = Leaf_HandleToDFNodeMap[ArgII];
@@ -1864,7 +1864,7 @@ void CGT_X86::codeGen(DFLeafNode *N) {
// (dim = 1) => y
// (dim = 2) => z
int dim =
- (int)(II->getIntrinsicID() - Intrinsic::hpvm_getNodeInstanceID_x);
+ (int)(II->getIntrinsicID() - Intrinsic::visc_getNodeInstanceID_x);
assert((dim >= 0) && (dim < 3) &&
"Invalid dimension for getNodeInstanceID_[xyz]. Check Intrinsic "
"ID!");
@@ -1894,7 +1894,7 @@ void CGT_X86::codeGen(DFLeafNode *N) {
Value *args[] = {
ConstantInt::get(Type::getInt32Ty(II->getContext()), parentLevel),
ConstantInt::get(Type::getInt32Ty(II->getContext()), dim)};
- CallInst *CI = CallInst::Create(llvm_hpvm_x86_getDimInstance,
+ CallInst *CI = CallInst::Create(llvm_visc_x86_getDimInstance,
ArrayRef<Value *>(args, 2),
"nodeInstanceID", II);
DEBUG(errs() << *II << " replaced with " << *CI << "\n");
@@ -1903,10 +1903,10 @@ void CGT_X86::codeGen(DFLeafNode *N) {
}
break;
}
- /********************** llvm.hpvm.getNumNodeInstances() *************/
- case Intrinsic::hpvm_getNumNodeInstances_x:
- case Intrinsic::hpvm_getNumNodeInstances_y:
- case Intrinsic::hpvm_getNumNodeInstances_z: {
+ /********************** llvm.visc.getNumNodeInstances() *************/
+ case Intrinsic::visc_getNumNodeInstances_x:
+ case Intrinsic::visc_getNumNodeInstances_y:
+ case Intrinsic::visc_getNumNodeInstances_z: {
ArgII = cast<IntrinsicInst>((II->getOperand(0))->stripPointerCasts());
ArgDFNode = Leaf_HandleToDFNodeMap[ArgII];
@@ -1922,7 +1922,7 @@ void CGT_X86::codeGen(DFLeafNode *N) {
// (dim = 1) => y
// (dim = 2) => z
int dim =
- (int)(II->getIntrinsicID() - Intrinsic::hpvm_getNumNodeInstances_x);
+ (int)(II->getIntrinsicID() - Intrinsic::visc_getNumNodeInstances_x);
assert((dim >= 0) && (dim < 3) &&
"Invalid dimension for getNumNodeInstances_[xyz]. Check "
"Intrinsic ID!");
@@ -1952,7 +1952,7 @@ void CGT_X86::codeGen(DFLeafNode *N) {
Value *args[] = {
ConstantInt::get(Type::getInt32Ty(II->getContext()), parentLevel),
ConstantInt::get(Type::getInt32Ty(II->getContext()), dim)};
- CallInst *CI = CallInst::Create(llvm_hpvm_x86_getDimLimit,
+ CallInst *CI = CallInst::Create(llvm_visc_x86_getDimLimit,
ArrayRef<Value *>(args, 2),
"numNodeInstances", II);
DEBUG(errs() << *II << " replaced with " << *CI << "\n");
@@ -1965,7 +1965,7 @@ void CGT_X86::codeGen(DFLeafNode *N) {
default:
DEBUG(errs() << "Found unknown intrinsic with ID = "
<< II->getIntrinsicID() << "\n");
- assert(false && "Unknown HPVM Intrinsic!");
+ assert(false && "Unknown VISC Intrinsic!");
break;
}
diff --git a/hpvm/lib/Transforms/GenHPVM/GenHPVM.cpp b/hpvm/lib/Transforms/GenHPVM/GenHPVM.cpp
deleted file mode 100644
index 738b39905b..0000000000
--- a/hpvm/lib/Transforms/GenHPVM/GenHPVM.cpp
+++ /dev/null
@@ -1,894 +0,0 @@
-//=== GenHPVM.cpp - Implements "Hierarchical Dataflow Graph Builder Pass" ===//
-//
-// The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#define DEBUG_TYPE "genhpvm"
-#include "GenHPVM/GenHPVM.h"
-
-#include "SupportHPVM/HPVMHint.h"
-#include "SupportHPVM/HPVMUtils.h"
-#include "llvm/ADT/Statistic.h"
-#include "llvm/IR/CallSite.h"
-#include "llvm/IR/DerivedTypes.h"
-#include "llvm/IR/InstIterator.h"
-#include "llvm/IR/Instructions.h"
-#include "llvm/IRReader/IRReader.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/SourceMgr.h"
-#include "llvm/Support/raw_ostream.h"
-#include "llvm/Transforms/Utils/BasicBlockUtils.h"
-#include "llvm/Transforms/Utils/Cloning.h"
-#include "llvm/Transforms/Utils/ValueMapper.h"
-
-#define TIMER(X) \
- do { \
- if (HPVMTimer) { \
- X; \
- } \
- } while (0)
-
-using namespace llvm;
-using namespace hpvmUtils;
-
-// HPVM Command line option to use timer or not
-static cl::opt<bool> HPVMTimer("hpvm-timers-gen",
- cl::desc("Enable GenHPVM timer"));
-
-namespace genhpvm {
-
-// Helper Functions
-
-static inline ConstantInt *getTimerID(Module &, enum hpvm_TimerID);
-static Function *transformReturnTypeToStruct(Function *F);
-static Type *getReturnTypeFromReturnInst(Function *F);
-
-// Check if the dummy function call is a __hpvm__node call
-#define IS_HPVM_CALL(callName) \
- static bool isHPVMCall_##callName(Instruction *I) { \
- if (!isa<CallInst>(I)) \
- return false; \
- CallInst *CI = cast<CallInst>(I); \
- return (CI->getCalledValue()->stripPointerCasts()->getName()) \
- .equals("__hpvm__" #callName); \
- }
-
-static void ReplaceCallWithIntrinsic(Instruction *I, Intrinsic::ID IntrinsicID,
- std::vector<Instruction *> *Erase) {
- // Check if the instruction is Call Instruction
- assert(isa<CallInst>(I) && "Expecting CallInst");
- CallInst *CI = cast<CallInst>(I);
- DEBUG(errs() << "Found call: " << *CI << "\n");
-
- // Find the correct intrinsic call
- Module *M = CI->getParent()->getParent()->getParent();
- Function *F;
- std::vector<Type *> ArgTypes;
- std::vector<Value *> args;
- if (Intrinsic::isOverloaded(IntrinsicID)) {
- // This is an overloaded intrinsic. The types must exactly match. Get the
- // argument types
- for (unsigned i = 0; i < CI->getNumArgOperands(); i++) {
- ArgTypes.push_back(CI->getArgOperand(i)->getType());
- args.push_back(CI->getArgOperand(i));
- }
- F = Intrinsic::getDeclaration(M, IntrinsicID, ArgTypes);
- DEBUG(errs() << *F << "\n");
- } else { // Non-overloaded intrinsic
- F = Intrinsic::getDeclaration(M, IntrinsicID);
- FunctionType *FTy = F->getFunctionType();
- DEBUG(errs() << *F << "\n");
-
- // Create argument list
- assert(CI->getNumArgOperands() == FTy->getNumParams() &&
- "Number of arguments of call do not match with Intrinsic");
- for (unsigned i = 0; i < CI->getNumArgOperands(); i++) {
- Value *V = CI->getArgOperand(i);
- // Either the type should match or both should be of pointer type
- assert((V->getType() == FTy->getParamType(i) ||
- (V->getType()->isPointerTy() &&
- FTy->getParamType(i)->isPointerTy())) &&
- "Dummy function call argument does not match with Intrinsic "
- "argument!");
- // If the types do not match, then both must be pointer type and pointer
- // cast needs to be performed
- if (V->getType() != FTy->getParamType(i)) {
- V = CastInst::CreatePointerCast(V, FTy->getParamType(i), "", CI);
- }
- args.push_back(V);
- }
- }
- // Insert call instruction
- CallInst *Inst = CallInst::Create(
- F, args, F->getReturnType()->isVoidTy() ? "" : CI->getName(), CI);
-
- DEBUG(errs() << "\tSubstitute with: " << *Inst << "\n");
-
- CI->replaceAllUsesWith(Inst);
- // If the previous instruction needs to be erased, insert it in the vector
- // Erased
- if (Erase != NULL)
- Erase->push_back(CI);
-}
-
-IS_HPVM_CALL(launch) /* Exists but not required */
-IS_HPVM_CALL(edge) /* Exists but not required */
-IS_HPVM_CALL(createNodeND)
-// IS_HPVM_CALL(createNode)
-// IS_HPVM_CALL(createNode1D)
-// IS_HPVM_CALL(createNode2D)
-// IS_HPVM_CALL(createNode3D)
-IS_HPVM_CALL(bindIn)
-IS_HPVM_CALL(bindOut)
-IS_HPVM_CALL(push)
-IS_HPVM_CALL(pop)
-IS_HPVM_CALL(getNode)
-IS_HPVM_CALL(getParentNode)
-IS_HPVM_CALL(barrier)
-IS_HPVM_CALL(malloc)
-IS_HPVM_CALL(return )
-IS_HPVM_CALL(getNodeInstanceID_x)
-IS_HPVM_CALL(getNodeInstanceID_y)
-IS_HPVM_CALL(getNodeInstanceID_z)
-IS_HPVM_CALL(getNumNodeInstances_x)
-IS_HPVM_CALL(getNumNodeInstances_y)
-IS_HPVM_CALL(getNumNodeInstances_z)
-// Atomics
-IS_HPVM_CALL(atomic_cmpxchg)
-IS_HPVM_CALL(atomic_add)
-IS_HPVM_CALL(atomic_sub)
-IS_HPVM_CALL(atomic_xchg)
-IS_HPVM_CALL(atomic_inc)
-IS_HPVM_CALL(atomic_dec)
-IS_HPVM_CALL(atomic_min)
-IS_HPVM_CALL(atomic_max)
-IS_HPVM_CALL(atomic_umin)
-IS_HPVM_CALL(atomic_umax)
-IS_HPVM_CALL(atomic_and)
-IS_HPVM_CALL(atomic_or)
-IS_HPVM_CALL(atomic_xor)
-// Misc Fn
-IS_HPVM_CALL(floor)
-IS_HPVM_CALL(rsqrt)
-IS_HPVM_CALL(sqrt)
-IS_HPVM_CALL(sin)
-IS_HPVM_CALL(cos)
-
-IS_HPVM_CALL(init)
-IS_HPVM_CALL(cleanup)
-IS_HPVM_CALL(wait)
-IS_HPVM_CALL(trackMemory)
-IS_HPVM_CALL(untrackMemory)
-IS_HPVM_CALL(requestMemory)
-IS_HPVM_CALL(attributes)
-IS_HPVM_CALL(hint)
-
-// Return the constant integer represented by value V
-static unsigned getNumericValue(Value *V) {
- assert(
- isa<ConstantInt>(V) &&
- "Value indicating the number of arguments should be a constant integer");
- return cast<ConstantInt>(V)->getZExtValue();
-}
-
-// Take the __hpvm__return instruction and generate code for combining the
-// values being returned into a struct and returning it.
-// The first operand is the number of returned values
-static Value *genCodeForReturn(CallInst *CI) {
- LLVMContext &Ctx = CI->getContext();
- assert(isHPVMCall_return(CI) && "__hpvm__return instruction expected!");
-
- // Parse the dummy function call here
- assert(CI->getNumArgOperands() > 0 &&
- "Too few arguments for __hpvm_return call!\n");
- unsigned numRetVals = getNumericValue(CI->getArgOperand(0));
-
- assert(CI->getNumArgOperands() - 1 == numRetVals &&
- "Too few arguments for __hpvm_return call!\n");
- DEBUG(errs() << "\tNum of return values = " << numRetVals << "\n");
-
- std::vector<Type *> ArgTypes;
- for (unsigned i = 1; i < CI->getNumArgOperands(); i++) {
- ArgTypes.push_back(CI->getArgOperand(i)->getType());
- }
- Twine outTyName = "struct.out." + CI->getParent()->getParent()->getName();
- StructType *RetTy = StructType::create(Ctx, ArgTypes, outTyName.str(), true);
-
- InsertValueInst *IV = InsertValueInst::Create(
- UndefValue::get(RetTy), CI->getArgOperand(1), 0, "returnStruct", CI);
- DEBUG(errs() << "Code generation for return:\n");
- DEBUG(errs() << *IV << "\n");
-
- for (unsigned i = 2; i < CI->getNumArgOperands(); i++) {
- IV = InsertValueInst::Create(IV, CI->getArgOperand(i), i - 1, IV->getName(),
- CI);
- DEBUG(errs() << *IV << "\n");
- }
-
- return IV;
-}
-
-// Analyse the attribute call for this function. Add the in and out
-// attributes to pointer parameters.
-static void handleHPVMAttributes(Function *F, CallInst *CI) {
- DEBUG(errs() << "Kernel before adding In/Out HPVM attributes:\n"
- << *F << "\n");
- // Parse the dummy function call here
- unsigned offset = 0;
- // Find number of In pointers
- assert(CI->getNumArgOperands() > offset &&
- "Too few arguments for __hpvm__attributes call!");
- unsigned numInPtrs = getNumericValue(CI->getArgOperand(offset));
- DEBUG(errs() << "\tNum of in pointers = " << numInPtrs << "\n");
-
- for (unsigned i = offset + 1; i < offset + 1 + numInPtrs; i++) {
- Value *V = CI->getArgOperand(i);
- if (Argument *arg = dyn_cast<Argument>(V)) {
- F->addAttribute(1 + arg->getArgNo(), Attribute::In);
- } else {
- DEBUG(errs() << "Invalid argument to __hpvm__attribute: " << *V << "\n");
- llvm_unreachable(
- "Only pointer arguments can be passed to __hpvm__attributes call");
- }
- }
- // Find number of Out Pointers
- offset += 1 + numInPtrs;
- assert(CI->getNumArgOperands() > offset &&
- "Too few arguments for __hpvm__attributes call!");
- unsigned numOutPtrs = getNumericValue(CI->getOperand(offset));
- DEBUG(errs() << "\tNum of out Pointers = " << numOutPtrs << "\n");
- for (unsigned i = offset + 1; i < offset + 1 + numOutPtrs; i++) {
- Value *V = CI->getArgOperand(i);
- if (Argument *arg = dyn_cast<Argument>(V)) {
- F->addAttribute(1 + arg->getArgNo(), Attribute::Out);
- } else {
- DEBUG(errs() << "Invalid argument to __hpvm__attribute: " << *V << "\n");
- llvm_unreachable(
- "Only pointer arguments can be passed to __hpvm__attributes call");
- }
- }
- DEBUG(errs() << "Kernel after adding In/Out HPVM attributes:\n"
- << *F << "\n");
-}
-
-// Public Functions of GenHPVM pass
-bool GenHPVM::runOnModule(Module &M) {
- DEBUG(errs() << "\nGENHPVM PASS\n");
- this->M = &M;
-
- // Load Runtime API Module
- SMDiagnostic Err;
-
- char *LLVM_SRC_ROOT = getenv("LLVM_SRC_ROOT");
- assert(LLVM_SRC_ROOT != NULL && "Define LLVM_SRC_ROOT environment variable!");
-
- Twine llvmSrcRoot = LLVM_SRC_ROOT;
- Twine runtimeAPI =
- llvmSrcRoot + "/../build/tools/hpvm/projects/hpvm-rt/hpvm-rt.bc";
- DEBUG(errs() << llvmSrcRoot << "\n");
-
- std::unique_ptr<Module> runtimeModule =
- parseIRFile(runtimeAPI.str(), Err, M.getContext());
-
- if (runtimeModule == NULL) {
- DEBUG(errs() << Err.getMessage() << " " << runtimeAPI << "\n");
- assert(false && "couldn't parse runtime");
- } else
- DEBUG(errs() << "Successfully loaded hpvm-rt API module\n");
-
- llvm_hpvm_initializeTimerSet = M.getOrInsertFunction(
- "llvm_hpvm_initializeTimerSet",
- runtimeModule->getFunction("llvm_hpvm_initializeTimerSet")
- ->getFunctionType());
- // DEBUG(errs() << *llvm_hpvm_initializeTimerSet);
-
- llvm_hpvm_switchToTimer = M.getOrInsertFunction(
- "llvm_hpvm_switchToTimer",
- runtimeModule->getFunction("llvm_hpvm_switchToTimer")->getFunctionType());
- // DEBUG(errs() << *llvm_hpvm_switchToTimer);
-
- llvm_hpvm_printTimerSet = M.getOrInsertFunction(
- "llvm_hpvm_printTimerSet",
- runtimeModule->getFunction("llvm_hpvm_printTimerSet")->getFunctionType());
- // DEBUG(errs() << *llvm_hpvm_printTimerSet);
-
- // Insert init context in main
- DEBUG(errs() << "Locate __hpvm__init()\n");
- Function *VI = M.getFunction("__hpvm__init");
- assert(VI->getNumUses() == 1 && "__hpvm__init should only be used once");
- Instruction *I = cast<Instruction>(*VI->user_begin());
-
- DEBUG(errs() << "Initialize Timer Set\n");
- initializeTimerSet(I);
- switchToTimer(hpvm_TimerID_NONE, I);
-
- // Insert print instruction at hpvm exit
- DEBUG(errs() << "Locate __hpvm__cleanup()\n");
- Function *VC = M.getFunction("__hpvm__cleanup");
- assert(VC->getNumUses() == 1 && "__hpvm__cleanup should only be used once");
- I = cast<Instruction>(*VC->user_begin());
- printTimerSet(I);
-
- DEBUG(errs() << "-------- Searching for launch sites ----------\n");
-
- std::vector<Instruction *> toBeErased;
- std::vector<Function *> functions;
-
- for (auto &F : M)
- functions.push_back(&F);
-
- // Iterate over all functions in the module
- for (Function *f : functions) {
- DEBUG(errs() << "Function: " << f->getName() << "\n");
-
- // List with the required additions in the function's return type
- std::vector<Type *> FRetTypes;
-
- enum mutateTypeCause {
- mtc_None,
- mtc_BIND,
- mtc_RETURN,
- mtc_NUM_CAUSES
- } bind;
- bind = mutateTypeCause::mtc_None;
-
- // Iterate over all the instructions in this function
- for (inst_iterator i = inst_begin(f), e = inst_end(f); i != e; ++i) {
- Instruction *I = &*i; // Grab pointer to Instruction
- // If not a call instruction, move to next instruction
- if (!isa<CallInst>(I))
- continue;
-
- CallInst *CI = cast<CallInst>(I);
- LLVMContext &Ctx = CI->getContext();
-
- if (isHPVMCall_init(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_init, &toBeErased);
- }
- if (isHPVMCall_cleanup(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_cleanup, &toBeErased);
- }
- if (isHPVMCall_wait(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_wait, &toBeErased);
- }
- if (isHPVMCall_trackMemory(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_trackMemory, &toBeErased);
- }
- if (isHPVMCall_untrackMemory(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_untrackMemory, &toBeErased);
- }
- if (isHPVMCall_requestMemory(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_requestMemory, &toBeErased);
- }
- if (isHPVMCall_hint(I)) {
- assert(isa<ConstantInt>(CI->getArgOperand(0)) &&
- "Argument to hint must be constant integer!");
- ConstantInt *hint = cast<ConstantInt>(CI->getArgOperand(0));
-
- hpvm::Target t = (hpvm::Target)hint->getZExtValue();
- addHint(CI->getParent()->getParent(), t);
- DEBUG(errs() << "Found hpvm hint call: " << *CI << "\n");
- toBeErased.push_back(CI);
- }
- if (isHPVMCall_launch(I)) {
- Function *LaunchF =
- Intrinsic::getDeclaration(&M, Intrinsic::hpvm_launch);
- DEBUG(errs() << *LaunchF << "\n");
- // Get i8* cast to function pointer
- Function *graphFunc = cast<Function>(CI->getArgOperand(1));
- graphFunc = transformReturnTypeToStruct(graphFunc);
- Constant *F =
- ConstantExpr::getPointerCast(graphFunc, Type::getInt8PtrTy(Ctx));
- assert(
- F &&
- "Function invoked by HPVM launch has to be define and constant.");
-
- ConstantInt *Op = cast<ConstantInt>(CI->getArgOperand(0));
- assert(Op && "HPVM launch's streaming argument is a constant value.");
- Value *isStreaming = Op->isZero() ? ConstantInt::getFalse(Ctx)
- : ConstantInt::getTrue(Ctx);
-
- auto *ArgTy = dyn_cast<PointerType>(CI->getArgOperand(2)->getType());
- assert(ArgTy && "HPVM launch argument should be pointer type.");
- Value *Arg = CI->getArgOperand(2);
- if (!ArgTy->getElementType()->isIntegerTy(8))
- Arg = BitCastInst::CreatePointerCast(CI->getArgOperand(2),
- Type::getInt8PtrTy(Ctx), "", CI);
- Value *LaunchArgs[] = {F, Arg, isStreaming};
- CallInst *LaunchInst = CallInst::Create(
- LaunchF, ArrayRef<Value *>(LaunchArgs, 3), "graphID", CI);
- DEBUG(errs() << "Found hpvm launch call: " << *CI << "\n");
- DEBUG(errs() << "\tSubstitute with: " << *LaunchInst << "\n");
- CI->replaceAllUsesWith(LaunchInst);
- toBeErased.push_back(CI);
- }
- if (isHPVMCall_push(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_push, &toBeErased);
- }
- if (isHPVMCall_pop(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_pop, &toBeErased);
- }
- if (isHPVMCall_createNodeND(I)) {
- assert(CI->getNumArgOperands() > 0 &&
- "Too few arguments for __hpvm__createNodeND call");
- unsigned numDims = getNumericValue(CI->getArgOperand(0));
- // We need as meny dimension argments are there are dimensions
- assert(CI->getNumArgOperands() - 2 == numDims &&
- "Too few arguments for __hpvm_createNodeND call!\n");
-
- Function *CreateNodeF;
- switch (numDims) {
- case 0:
- CreateNodeF =
- Intrinsic::getDeclaration(&M, Intrinsic::hpvm_createNode);
- break;
- case 1:
- CreateNodeF =
- Intrinsic::getDeclaration(&M, Intrinsic::hpvm_createNode1D);
- break;
- case 2:
- CreateNodeF =
- Intrinsic::getDeclaration(&M, Intrinsic::hpvm_createNode2D);
- break;
- case 3:
- CreateNodeF =
- Intrinsic::getDeclaration(&M, Intrinsic::hpvm_createNode3D);
- break;
- default:
- llvm_unreachable("Unsupported number of dimensions\n");
- break;
- }
- DEBUG(errs() << *CreateNodeF << "\n");
- DEBUG(errs() << *I << "\n");
- DEBUG(errs() << "in " << I->getParent()->getParent()->getName()
- << "\n");
-
- // Get i8* cast to function pointer
- Function *graphFunc = cast<Function>(CI->getArgOperand(1));
- graphFunc = transformReturnTypeToStruct(graphFunc);
- Constant *F =
- ConstantExpr::getPointerCast(graphFunc, Type::getInt8PtrTy(Ctx));
-
- CallInst *CreateNodeInst;
- switch (numDims) {
- case 0:
- CreateNodeInst = CallInst::Create(CreateNodeF, ArrayRef<Value *>(F),
- graphFunc->getName() + ".node", CI);
- break;
- case 1: {
- assert((CI->getArgOperand(2)->getType() == Type::getInt64Ty(Ctx)) &&
- "CreateNodeND dimension argument, 2, expected to be i64\n");
- Value *CreateNodeArgs[] = {F, CI->getArgOperand(2)};
- CreateNodeInst = CallInst::Create(
- CreateNodeF, ArrayRef<Value *>(CreateNodeArgs, 2),
- graphFunc->getName() + ".node", CI);
- } break;
- case 2: {
- assert((CI->getArgOperand(2)->getType() == Type::getInt64Ty(Ctx)) &&
- "CreateNodeND dimension argument, 2, expected to be i64\n");
- assert((CI->getArgOperand(3)->getType() == Type::getInt64Ty(Ctx)) &&
- "CreateNodeND dimension argument, 3, expected to be i64\n");
- Value *CreateNodeArgs[] = {F, CI->getArgOperand(2),
- CI->getArgOperand(3)};
- CreateNodeInst = CallInst::Create(
- CreateNodeF, ArrayRef<Value *>(CreateNodeArgs, 3),
- graphFunc->getName() + ".node", CI);
- } break;
- case 3: {
- assert((CI->getArgOperand(2)->getType() == Type::getInt64Ty(Ctx)) &&
- "CreateNodeND dimension argument, 2, expected to be i64\n");
- assert((CI->getArgOperand(3)->getType() == Type::getInt64Ty(Ctx)) &&
- "CreateNodeND dimension argument, 3, expected to be i64\n");
- assert((CI->getArgOperand(4)->getType() == Type::getInt64Ty(Ctx)) &&
- "CreateNodeND dimension argument, 4, expected to be i64\n");
- Value *CreateNodeArgs[] = {F, CI->getArgOperand(2),
- CI->getArgOperand(3),
- CI->getArgOperand(4)};
- CreateNodeInst = CallInst::Create(
- CreateNodeF, ArrayRef<Value *>(CreateNodeArgs, 4),
- graphFunc->getName() + ".node", CI);
- } break;
- default:
- llvm_unreachable(
- "Impossible path: number of dimensions is 0, 1, 2, 3\n");
- break;
- }
-
- DEBUG(errs() << "Found hpvm createNode call: " << *CI << "\n");
- DEBUG(errs() << "\tSubstitute with: " << *CreateNodeInst << "\n");
- CI->replaceAllUsesWith(CreateNodeInst);
- toBeErased.push_back(CI);
- }
-
- if (isHPVMCall_edge(I)) {
- Function *EdgeF =
- Intrinsic::getDeclaration(&M, Intrinsic::hpvm_createEdge);
- DEBUG(errs() << *EdgeF << "\n");
- ConstantInt *Op = cast<ConstantInt>(CI->getArgOperand(5));
- ConstantInt *EdgeTypeOp = cast<ConstantInt>(CI->getArgOperand(2));
- assert(Op && EdgeTypeOp &&
- "Arguments of CreateEdge are not constant integers.");
- Value *isStreaming = Op->isZero() ? ConstantInt::getFalse(Ctx)
- : ConstantInt::getTrue(Ctx);
- Value *isAllToAll = EdgeTypeOp->isZero() ? ConstantInt::getFalse(Ctx)
- : ConstantInt::getTrue(Ctx);
- Value *EdgeArgs[] = {CI->getArgOperand(0), CI->getArgOperand(1),
- isAllToAll, CI->getArgOperand(3),
- CI->getArgOperand(4), isStreaming};
- CallInst *EdgeInst = CallInst::Create(
- EdgeF, ArrayRef<Value *>(EdgeArgs, 6), "output", CI);
- DEBUG(errs() << "Found hpvm edge call: " << *CI << "\n");
- DEBUG(errs() << "\tSubstitute with: " << *EdgeInst << "\n");
- CI->replaceAllUsesWith(EdgeInst);
- toBeErased.push_back(CI);
- }
- if (isHPVMCall_bindIn(I)) {
- Function *BindInF =
- Intrinsic::getDeclaration(&M, Intrinsic::hpvm_bind_input);
- DEBUG(errs() << *BindInF << "\n");
- // Check if this is a streaming bind or not
- ConstantInt *Op = cast<ConstantInt>(CI->getArgOperand(3));
- assert(Op && "Streaming argument for bind in intrinsic should be a "
- "constant integer.");
- Value *isStreaming = Op->isZero() ? ConstantInt::getFalse(Ctx)
- : ConstantInt::getTrue(Ctx);
- Value *BindInArgs[] = {CI->getArgOperand(0), CI->getArgOperand(1),
- CI->getArgOperand(2), isStreaming};
- CallInst *BindInInst =
- CallInst::Create(BindInF, ArrayRef<Value *>(BindInArgs, 4), "", CI);
- DEBUG(errs() << "Found hpvm bindIn call: " << *CI << "\n");
- DEBUG(errs() << "\tSubstitute with: " << *BindInInst << "\n");
- CI->replaceAllUsesWith(BindInInst);
- toBeErased.push_back(CI);
- }
- if (isHPVMCall_bindOut(I)) {
- Function *BindOutF =
- Intrinsic::getDeclaration(&M, Intrinsic::hpvm_bind_output);
- DEBUG(errs() << *BindOutF << "\n");
- // Check if this is a streaming bind or not
- ConstantInt *Op = cast<ConstantInt>(CI->getArgOperand(3));
- assert(Op && "Streaming argument for bind out intrinsic should be a "
- "constant integer.");
- Value *isStreaming = Op->isZero() ? ConstantInt::getFalse(Ctx)
- : ConstantInt::getTrue(Ctx);
- Value *BindOutArgs[] = {CI->getArgOperand(0), CI->getArgOperand(1),
- CI->getArgOperand(2), isStreaming};
- CallInst *BindOutInst = CallInst::Create(
- BindOutF, ArrayRef<Value *>(BindOutArgs, 4), "", CI);
- DEBUG(errs() << "Found hpvm bindOut call: " << *CI << "\n");
- DEBUG(errs() << "\tSubstitute with: " << *BindOutInst << "\n");
-
- DEBUG(errs() << "Fixing the return type of the function\n");
- // FIXME: What if the child node function has not been visited already.
- // i.e., it's return type has not been fixed.
- Function *F = I->getParent()->getParent();
- DEBUG(errs() << F->getName() << "\n";);
- IntrinsicInst *NodeIntrinsic =
- cast<IntrinsicInst>(CI->getArgOperand(0));
- assert(NodeIntrinsic &&
- "Instruction value in bind out is not a create node intrinsic.");
- DEBUG(errs() << "Node intrinsic: " << *NodeIntrinsic << "\n");
- assert(
- (NodeIntrinsic->getIntrinsicID() == Intrinsic::hpvm_createNode ||
- NodeIntrinsic->getIntrinsicID() == Intrinsic::hpvm_createNode1D ||
- NodeIntrinsic->getIntrinsicID() == Intrinsic::hpvm_createNode2D ||
- NodeIntrinsic->getIntrinsicID() == Intrinsic::hpvm_createNode3D) &&
- "Instruction value in bind out is not a create node intrinsic.");
- Function *ChildF = cast<Function>(
- NodeIntrinsic->getArgOperand(0)->stripPointerCasts());
- DEBUG(errs() << ChildF->getName() << "\n";);
- int srcpos = cast<ConstantInt>(CI->getArgOperand(1))->getSExtValue();
- int destpos = cast<ConstantInt>(CI->getArgOperand(2))->getSExtValue();
- StructType *ChildReturnTy = cast<StructType>(ChildF->getReturnType());
-
- Type *ReturnType = F->getReturnType();
- DEBUG(errs() << *ReturnType << "\n";);
- assert((ReturnType->isVoidTy() || isa<StructType>(ReturnType)) &&
- "Return type should either be a struct or void type!");
-
- FRetTypes.insert(FRetTypes.begin() + destpos,
- ChildReturnTy->getElementType(srcpos));
- assert(((bind == mutateTypeCause::mtc_BIND) ||
- (bind == mutateTypeCause::mtc_None)) &&
- "Both bind_out and hpvm_return detected");
- bind = mutateTypeCause::mtc_BIND;
-
- CI->replaceAllUsesWith(BindOutInst);
- toBeErased.push_back(CI);
- }
- if (isHPVMCall_attributes(I)) {
- Function *F = CI->getParent()->getParent();
- handleHPVMAttributes(F, CI);
- toBeErased.push_back(CI);
- }
- if (isHPVMCall_getNode(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_getNode, &toBeErased);
- }
- if (isHPVMCall_getParentNode(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_getParentNode, &toBeErased);
- }
- if (isHPVMCall_barrier(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_barrier, &toBeErased);
- }
- if (isHPVMCall_malloc(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_malloc, &toBeErased);
- }
- if (isHPVMCall_return(I)) {
- DEBUG(errs() << "Function before hpvm return processing\n"
- << *I->getParent()->getParent() << "\n");
- // The operands to this call are the values to be returned by the node
- Value *ReturnVal = genCodeForReturn(CI);
- DEBUG(errs() << *ReturnVal << "\n");
- Type *ReturnType = ReturnVal->getType();
- assert(isa<StructType>(ReturnType) &&
- "Return type should be a struct type!");
-
- assert(((bind == mutateTypeCause::mtc_RETURN) ||
- (bind == mutateTypeCause::mtc_None)) &&
- "Both bind_out and hpvm_return detected");
-
- if (bind == mutateTypeCause::mtc_None) {
- // If this is None, this is the first __hpvm__return
- // instruction we have come upon. Place the return type of the
- // function in the return type vector
- bind = mutateTypeCause::mtc_RETURN;
- StructType *ReturnStructTy = cast<StructType>(ReturnType);
- for (unsigned i = 0; i < ReturnStructTy->getNumElements(); i++)
- FRetTypes.push_back(ReturnStructTy->getElementType(i));
- } else { // bind == mutateTypeCause::mtc_RETURN
- // This is not the first __hpvm__return
- // instruction we have come upon.
- // Check that the return types are the same
- assert((ReturnType == FRetTypes[0]) &&
- "Multiple returns with mismatching types");
- }
-
- ReturnInst *RetInst = ReturnInst::Create(Ctx, ReturnVal);
- DEBUG(errs() << "Found hpvm return call: " << *CI << "\n");
- Instruction *oldReturn = CI->getParent()->getTerminator();
- assert(isa<ReturnInst>(oldReturn) &&
- "Expecting a return to be the terminator of this BB!");
- DEBUG(errs() << "Found return statement of BB: " << *oldReturn << "\n");
- DEBUG(errs() << "\tSubstitute return with: " << *RetInst << "\n");
- // CI->replaceAllUsesWith(RetInst);
- toBeErased.push_back(CI);
- ReplaceInstWithInst(oldReturn, RetInst);
- DEBUG(errs() << "Function after hpvm return processing\n"
- << *I->getParent()->getParent() << "\n");
- }
-
- if (isHPVMCall_getNodeInstanceID_x(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_getNodeInstanceID_x,
- &toBeErased);
- }
- if (isHPVMCall_getNodeInstanceID_y(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_getNodeInstanceID_y,
- &toBeErased);
- }
- if (isHPVMCall_getNodeInstanceID_z(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_getNodeInstanceID_z,
- &toBeErased);
- }
- if (isHPVMCall_getNumNodeInstances_x(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_getNumNodeInstances_x,
- &toBeErased);
- }
- if (isHPVMCall_getNumNodeInstances_y(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_getNumNodeInstances_y,
- &toBeErased);
- }
- if (isHPVMCall_getNumNodeInstances_z(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_getNumNodeInstances_z,
- &toBeErased);
- }
- if (isHPVMCall_atomic_add(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_atomic_add, &toBeErased);
- }
- if (isHPVMCall_atomic_sub(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_atomic_sub, &toBeErased);
- }
- if (isHPVMCall_atomic_xchg(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_atomic_xchg, &toBeErased);
- }
- if (isHPVMCall_atomic_min(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_atomic_min, &toBeErased);
- }
- if (isHPVMCall_atomic_max(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_atomic_max, &toBeErased);
- }
- if (isHPVMCall_atomic_and(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_atomic_and, &toBeErased);
- }
- if (isHPVMCall_atomic_or(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_atomic_or, &toBeErased);
- }
- if (isHPVMCall_atomic_xor(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::hpvm_atomic_xor, &toBeErased);
- }
- if (isHPVMCall_sin(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::sin, &toBeErased);
- }
- if (isHPVMCall_cos(I)) {
- ReplaceCallWithIntrinsic(I, Intrinsic::cos, &toBeErased);
- }
- }
-
- // Erase the __hpvm__node calls
- DEBUG(errs() << "Erase " << toBeErased.size() << " Statements:\n");
- for (auto I : toBeErased) {
- DEBUG(errs() << *I << "\n");
- }
- while (!toBeErased.empty()) {
- Instruction *I = toBeErased.back();
- DEBUG(errs() << "\tErasing " << *I << "\n");
- I->eraseFromParent();
- toBeErased.pop_back();
- }
-
- if (bind == mutateTypeCause::mtc_BIND ||
- bind == mutateTypeCause::mtc_RETURN) {
- DEBUG(errs() << "Function before fixing return type\n" << *f << "\n");
- // Argument type list.
- std::vector<Type *> FArgTypes;
- for (Function::const_arg_iterator ai = f->arg_begin(), ae = f->arg_end();
- ai != ae; ++ai) {
- FArgTypes.push_back(ai->getType());
- }
-
- // Find new return type of function
- Type *NewReturnTy;
- if (bind == mutateTypeCause::mtc_BIND) {
-
- std::vector<Type *> TyList;
- for (unsigned i = 0; i < FRetTypes.size(); i++)
- TyList.push_back(FRetTypes[i]);
-
- NewReturnTy =
- StructType::create(f->getContext(), TyList,
- Twine("struct.out." + f->getName()).str(), true);
- } else {
- NewReturnTy = getReturnTypeFromReturnInst(f);
- assert(NewReturnTy->isStructTy() && "Expecting a struct type!");
- }
-
- FunctionType *FTy =
- FunctionType::get(NewReturnTy, FArgTypes, f->isVarArg());
-
- // Change the function type
- Function *newF = cloneFunction(f, FTy, false);
- DEBUG(errs() << *newF << "\n");
-
- if (bind == mutateTypeCause::mtc_BIND) {
- // This is certainly an internal node, and hence just one BB with one
- // return terminator instruction. Change return statement
- ReturnInst *RI =
- cast<ReturnInst>(newF->getEntryBlock().getTerminator());
- ReturnInst *newRI = ReturnInst::Create(newF->getContext(),
- UndefValue::get(NewReturnTy));
- ReplaceInstWithInst(RI, newRI);
- }
- if (bind == mutateTypeCause::mtc_RETURN) {
- // Nothing
- }
- replaceNodeFunctionInIR(*f->getParent(), f, newF);
- DEBUG(errs() << "Function after fixing return type\n" << *newF << "\n");
- }
- }
- return false; // TODO: What does returning "false" mean?
-}
-
-// Generate Code for declaring a constant string [L x i8] and return a pointer
-// to the start of it.
-Value *GenHPVM::getStringPointer(const Twine &S, Instruction *IB,
- const Twine &Name) {
- Constant *SConstant =
- ConstantDataArray::getString(M->getContext(), S.str(), true);
- Value *SGlobal =
- new GlobalVariable(*M, SConstant->getType(), true,
- GlobalValue::InternalLinkage, SConstant, Name);
- Value *Zero = ConstantInt::get(Type::getInt64Ty(M->getContext()), 0);
- Value *GEPArgs[] = {Zero, Zero};
- GetElementPtrInst *SPtr = GetElementPtrInst::Create(
- nullptr, SGlobal, ArrayRef<Value *>(GEPArgs, 2), Name + "Ptr", IB);
- return SPtr;
-}
-
-void GenHPVM::initializeTimerSet(Instruction *InsertBefore) {
- Value *TimerSetAddr;
- StoreInst *SI;
- TIMER(TimerSet = new GlobalVariable(
- *M, Type::getInt8PtrTy(M->getContext()), false,
- GlobalValue::CommonLinkage,
- Constant::getNullValue(Type::getInt8PtrTy(M->getContext())),
- "hpvmTimerSet_GenHPVM"));
- DEBUG(errs() << "Inserting GV: " << *TimerSet->getType() << *TimerSet
- << "\n");
- // DEBUG(errs() << "Inserting call to: " << *llvm_hpvm_initializeTimerSet <<
- // "\n");
-
- TIMER(TimerSetAddr = CallInst::Create(llvm_hpvm_initializeTimerSet, None, "",
- InsertBefore));
- DEBUG(errs() << "TimerSetAddress = " << *TimerSetAddr << "\n");
- TIMER(SI = new StoreInst(TimerSetAddr, TimerSet, InsertBefore));
- DEBUG(errs() << "Store Timer Address in Global variable: " << *SI << "\n");
-}
-
-void GenHPVM::switchToTimer(enum hpvm_TimerID timer,
- Instruction *InsertBefore) {
- Value *switchArgs[] = {TimerSet, getTimerID(*M, timer)};
- TIMER(CallInst::Create(llvm_hpvm_switchToTimer,
- ArrayRef<Value *>(switchArgs, 2), "", InsertBefore));
-}
-
-void GenHPVM::printTimerSet(Instruction *InsertBefore) {
- Value *TimerName;
- TIMER(TimerName = getStringPointer("GenHPVM_Timer", InsertBefore));
- Value *printArgs[] = {TimerSet, TimerName};
- TIMER(CallInst::Create(llvm_hpvm_printTimerSet,
- ArrayRef<Value *>(printArgs, 2), "", InsertBefore));
-}
-
-static inline ConstantInt *getTimerID(Module &M, enum hpvm_TimerID timer) {
- return ConstantInt::get(Type::getInt32Ty(M.getContext()), timer);
-}
-
-static Function *transformReturnTypeToStruct(Function *F) {
- // Currently only works for void return types
- DEBUG(errs() << "Transforming return type of function to Struct: "
- << F->getName() << "\n");
-
- if (isa<StructType>(F->getReturnType())) {
- DEBUG(errs() << "Return type is already a Struct: " << F->getName() << ": "
- << *F->getReturnType() << "\n");
- return F;
- }
-
- assert(F->getReturnType()->isVoidTy() &&
- "Unhandled case - Only void return type handled\n");
-
- // Create the argument type list with added argument types
- std::vector<Type *> ArgTypes;
- for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end();
- ai != ae; ++ai) {
- ArgTypes.push_back(ai->getType());
- }
-
- StructType *RetTy =
- StructType::create(F->getContext(), None, "emptyStruct", true);
- FunctionType *FTy = FunctionType::get(RetTy, ArgTypes, F->isVarArg());
-
- SmallVector<ReturnInst *, 8> Returns;
- Function *newF = cloneFunction(F, FTy, false, &Returns);
- // Replace ret void instruction with ret %RetTy undef
- for (auto &RI : Returns) {
- DEBUG(errs() << "Found return inst: " << *RI << "\n");
- ReturnInst *newRI =
- ReturnInst::Create(newF->getContext(), UndefValue::get(RetTy));
- ReplaceInstWithInst(RI, newRI);
- }
-
- replaceNodeFunctionInIR(*F->getParent(), F, newF);
- return newF;
-}
-
-static Type *getReturnTypeFromReturnInst(Function *F) {
- for (BasicBlock &BB : *F) {
- if (ReturnInst *RI = dyn_cast<ReturnInst>(BB.getTerminator())) {
- DEBUG(errs() << "Return type value: " << *RI->getReturnValue()->getType()
- << "\n");
- return RI->getReturnValue()->getType();
- }
- }
-}
-
-char genhpvm::GenHPVM::ID = 0;
-static RegisterPass<genhpvm::GenHPVM>
- X("genhpvm",
- "Pass to generate HPVM IR from LLVM IR (with dummy function calls)",
- false, false);
-
-} // End of namespace genhpvm
diff --git a/hpvm/lib/Transforms/GenHPVM/CMakeLists.txt b/hpvm/lib/Transforms/GenVISC/CMakeLists.txt
similarity index 74%
rename from hpvm/lib/Transforms/GenHPVM/CMakeLists.txt
rename to hpvm/lib/Transforms/GenVISC/CMakeLists.txt
index 967766e705..ed087f63b4 100644
--- a/hpvm/lib/Transforms/GenHPVM/CMakeLists.txt
+++ b/hpvm/lib/Transforms/GenVISC/CMakeLists.txt
@@ -2,9 +2,9 @@ if(WIN32 OR CYGWIN)
set(LLVM_LINK_COMPONENTS Core Support)
endif()
-add_llvm_library( LLVMGenHPVM
+add_llvm_library( LLVMGenVISC
MODULE
- GenHPVM.cpp
+ GenVISC.cpp
DEPENDS
intrinsics_gen
diff --git a/hpvm/lib/Transforms/GenVISC/GenVISC.cpp b/hpvm/lib/Transforms/GenVISC/GenVISC.cpp
new file mode 100644
index 0000000000..cc50541539
--- /dev/null
+++ b/hpvm/lib/Transforms/GenVISC/GenVISC.cpp
@@ -0,0 +1,866 @@
+//=== GenVISC.cpp - Implements "Hierarchical Dataflow Graph Builder Pass" ===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "genvisc"
+#include "GenVISC/GenVISC.h"
+
+#include "llvm/ADT/Statistic.h"
+#include "llvm/IR/CallSite.h"
+#include "llvm/IR/InstIterator.h"
+#include "llvm/Transforms/Utils/BasicBlockUtils.h"
+#include "llvm/Support/SourceMgr.h"
+#include "llvm/IRReader/IRReader.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "SupportVISC/VISCHint.h"
+#include "SupportVISC/VISCUtils.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Transforms/Utils/ValueMapper.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/Transforms/Utils/Cloning.h"
+#include "SupportVISC/VISCUtils.h"
+
+
+#define TIMER(X) do { if (VISCTimer) { X; } } while (0)
+
+using namespace llvm;
+using namespace viscUtils;
+
+
+// VISC Command line option to use timer or not
+static cl::opt<bool>
+VISCTimer("visc-timers-gen", cl::desc("Enable GenVISC timer"));
+
+namespace genvisc {
+
+// Helper Functions
+
+static inline ConstantInt* getTimerID(Module&, enum visc_TimerID);
+static Function* transformReturnTypeToStruct(Function* F);
+static Type* getReturnTypeFromReturnInst(Function* F);
+
+// Check if the dummy function call is a __visc__node call
+#define IS_VISC_CALL(callName) \
+ static bool isVISCCall_##callName(Instruction* I) { \
+ if(!isa<CallInst>(I)) \
+ return false; \
+ CallInst* CI = cast<CallInst>(I); \
+ return (CI->getCalledValue()->stripPointerCasts()->getName()).equals("__visc__"#callName); \
+ }
+
+static void ReplaceCallWithIntrinsic(Instruction* I, Intrinsic::ID IntrinsicID, std::vector<Instruction*>* Erase) {
+ // Check if the instruction is Call Instruction
+ assert(isa<CallInst>(I) && "Expecting CallInst");
+ CallInst* CI = cast<CallInst>(I);
+ DEBUG(errs() << "Found call: " << *CI << "\n");
+
+ // Find the correct intrinsic call
+ Module* M = CI->getParent()->getParent()->getParent();
+ Function* F;
+ std::vector<Type*> ArgTypes;
+ std::vector<Value*> args;
+ if(Intrinsic::isOverloaded(IntrinsicID)) {
+ // This is an overloaded intrinsic. The types must exactly match. Get the
+ // argument types
+ for(unsigned i=0; i < CI->getNumArgOperands(); i++) {
+ ArgTypes.push_back(CI->getArgOperand(i)->getType());
+ args.push_back(CI->getArgOperand(i));
+ }
+ F = Intrinsic::getDeclaration(M, IntrinsicID, ArgTypes);
+ DEBUG(errs() << *F << "\n");
+ }
+ else { // Non-overloaded intrinsic
+ F = Intrinsic::getDeclaration(M, IntrinsicID);
+ FunctionType* FTy = F->getFunctionType();
+ DEBUG(errs() << *F << "\n");
+
+ // Create argument list
+ assert(CI->getNumArgOperands() == FTy->getNumParams()
+ && "Number of arguments of call do not match with Intrinsic");
+ for(unsigned i=0; i < CI->getNumArgOperands(); i++) {
+ Value* V = CI->getArgOperand(i);
+ // Either the type should match or both should be of pointer type
+ assert((V->getType() == FTy->getParamType(i) ||
+ (V->getType()->isPointerTy() && FTy->getParamType(i)->isPointerTy()))
+ && "Dummy function call argument does not match with Intrinsic argument!");
+ // If the types do not match, then both must be pointer type and pointer
+ // cast needs to be performed
+ if(V->getType() != FTy->getParamType(i)) {
+ V = CastInst::CreatePointerCast(V, FTy->getParamType(i), "", CI);
+ }
+ args.push_back(V);
+ }
+ }
+ // Insert call instruction
+ CallInst* Inst = CallInst::Create(F, args, F->getReturnType()->isVoidTy()? "" : CI->getName(), CI);
+
+ DEBUG(errs() << "\tSubstitute with: " << *Inst << "\n");
+
+ CI->replaceAllUsesWith(Inst);
+ // If the previous instruction needs to be erased, insert it in the vector
+ // Erased
+ if(Erase != NULL)
+ Erase->push_back(CI);
+}
+
+IS_VISC_CALL(launch) /* Exists but not required */
+IS_VISC_CALL(edge) /* Exists but not required */
+IS_VISC_CALL(createNodeND)
+//IS_VISC_CALL(createNode)
+//IS_VISC_CALL(createNode1D)
+//IS_VISC_CALL(createNode2D)
+//IS_VISC_CALL(createNode3D)
+IS_VISC_CALL(bindIn)
+IS_VISC_CALL(bindOut)
+IS_VISC_CALL(push)
+IS_VISC_CALL(pop)
+IS_VISC_CALL(getNode)
+IS_VISC_CALL(getParentNode)
+IS_VISC_CALL(barrier)
+IS_VISC_CALL(malloc)
+IS_VISC_CALL(return)
+IS_VISC_CALL(getNodeInstanceID_x)
+IS_VISC_CALL(getNodeInstanceID_y)
+IS_VISC_CALL(getNodeInstanceID_z)
+IS_VISC_CALL(getNumNodeInstances_x)
+IS_VISC_CALL(getNumNodeInstances_y)
+IS_VISC_CALL(getNumNodeInstances_z)
+// Atomics
+IS_VISC_CALL(atomic_cmpxchg)
+IS_VISC_CALL(atomic_add)
+IS_VISC_CALL(atomic_sub)
+IS_VISC_CALL(atomic_xchg)
+IS_VISC_CALL(atomic_inc)
+IS_VISC_CALL(atomic_dec)
+IS_VISC_CALL(atomic_min)
+IS_VISC_CALL(atomic_max)
+IS_VISC_CALL(atomic_umin)
+IS_VISC_CALL(atomic_umax)
+IS_VISC_CALL(atomic_and)
+IS_VISC_CALL(atomic_or)
+IS_VISC_CALL(atomic_xor)
+// Misc Fn
+IS_VISC_CALL(floor)
+IS_VISC_CALL(rsqrt)
+IS_VISC_CALL(sqrt)
+IS_VISC_CALL(sin)
+IS_VISC_CALL(cos)
+
+
+IS_VISC_CALL(init)
+IS_VISC_CALL(cleanup)
+IS_VISC_CALL(wait)
+IS_VISC_CALL(trackMemory)
+IS_VISC_CALL(untrackMemory)
+IS_VISC_CALL(requestMemory)
+IS_VISC_CALL(attributes)
+IS_VISC_CALL(hint)
+
+// Return the constant integer represented by value V
+static unsigned getNumericValue(Value* V) {
+ assert(isa<ConstantInt>(V)
+ && "Value indicating the number of arguments should be a constant integer");
+ return cast<ConstantInt>(V)->getZExtValue();
+}
+
+// Take the __visc__return instruction and generate code for combining the
+// values being returned into a struct and returning it.
+// The first operand is the number of returned values
+static Value* genCodeForReturn(CallInst* CI) {
+ LLVMContext& Ctx = CI->getContext();
+ assert(isVISCCall_return(CI)
+ && "__visc__return instruction expected!");
+
+ // Parse the dummy function call here
+ assert(CI->getNumArgOperands() > 0 && "Too few arguments for __visc_return call!\n");
+ unsigned numRetVals = getNumericValue(CI->getArgOperand(0));
+
+ assert(CI->getNumArgOperands()-1 == numRetVals &&
+ "Too few arguments for __visc_return call!\n");
+ DEBUG(errs() << "\tNum of return values = " << numRetVals << "\n");
+
+ std::vector<Type*> ArgTypes;
+ for(unsigned i=1; i < CI->getNumArgOperands(); i++) {
+ ArgTypes.push_back(CI->getArgOperand(i)->getType());
+ }
+ Twine outTyName = "struct.out." + CI->getParent()->getParent()->getName();
+ StructType* RetTy = StructType::create(Ctx, ArgTypes, outTyName.str(), true);
+
+ InsertValueInst* IV = InsertValueInst::Create(UndefValue::get(RetTy),
+ CI->getArgOperand(1),
+ 0,
+ "returnStruct",
+ CI);
+ DEBUG(errs() << "Code generation for return:\n");
+ DEBUG(errs() << *IV << "\n");
+
+ for(unsigned i=2; i < CI->getNumArgOperands(); i++) {
+ IV = InsertValueInst::Create(IV,
+ CI->getArgOperand(i),
+ i-1,
+ IV->getName(),
+ CI);
+ DEBUG(errs() << *IV << "\n");
+ }
+
+ return IV;
+}
+
+// Analyse the attribute call for this function. Add the in and out
+// attributes to pointer parameters.
+static void handleVISCAttributes(Function* F, CallInst* CI) {
+ DEBUG(errs() << "Kernel before adding In/Out VISC attributes:\n" << *F << "\n");
+ // Parse the dummy function call here
+ unsigned offset = 0;
+ // Find number of In pointers
+ assert(CI->getNumArgOperands() > offset
+ && "Too few arguments for __visc__attributes call!");
+ unsigned numInPtrs = getNumericValue(CI->getArgOperand(offset));
+ DEBUG(errs() << "\tNum of in pointers = " << numInPtrs << "\n");
+
+ for(unsigned i = offset+1; i< offset+1+numInPtrs; i++) {
+ Value* V = CI->getArgOperand(i);
+ if(Argument* arg = dyn_cast<Argument>(V)) {
+ F->addAttribute(1+arg->getArgNo(), Attribute::In);
+ }
+ else {
+ errs() << "Invalid argument to __visc__attribute: " << *V << "\n";
+ llvm_unreachable("Only pointer arguments can be passed to __visc__attributes call");
+ }
+ }
+ // Find number of Out Pointers
+ offset += 1 + numInPtrs;
+ assert(CI->getNumArgOperands() > offset
+ && "Too few arguments for __visc__attributes call!");
+ unsigned numOutPtrs = getNumericValue(CI->getOperand(offset));
+ DEBUG(errs() << "\tNum of out Pointers = " << numOutPtrs << "\n");
+ for(unsigned i = offset+1; i< offset+1+numOutPtrs; i++) {
+ Value* V = CI->getArgOperand(i);
+ if(Argument* arg = dyn_cast<Argument>(V)) {
+ F->addAttribute(1+arg->getArgNo(), Attribute::Out);
+ }
+ else {
+ errs() << "Invalid argument to __visc__attribute: " << *V << "\n";
+ llvm_unreachable("Only pointer arguments can be passed to __visc__attributes call");
+ }
+ }
+ DEBUG(errs() << "Kernel after adding In/Out VISC attributes:\n" << *F << "\n");
+}
+
+// Public Functions of GenVISC pass
+bool GenVISC::runOnModule(Module &M) {
+ errs() << "\nGENVISC PASS\n";
+ this->M = &M;
+
+ // Load Runtime API Module
+ SMDiagnostic Err;
+
+ char* LLVM_SRC_ROOT = getenv("LLVM_SRC_ROOT");
+ assert(LLVM_SRC_ROOT != NULL &&
+ "Define LLVM_SRC_ROOT environment variable!");
+
+ Twine llvmSrcRoot = LLVM_SRC_ROOT;
+ Twine runtimeAPI = llvmSrcRoot + "/../build/tools/hpvm/projects/visc-rt/visc-rt.bc";
+ errs() << llvmSrcRoot << "\n";
+
+ std::unique_ptr<Module> runtimeModule = parseIRFile(runtimeAPI.str(), Err, M.getContext());
+
+ if(runtimeModule == NULL) {
+ DEBUG(errs() << Err.getMessage() << " " << runtimeAPI << "\n");
+ assert(false && "couldn't parse runtime");
+ }
+ else
+ DEBUG(errs() << "Successfully loaded visc-rt API module\n");
+
+ llvm_visc_initializeTimerSet = M.getOrInsertFunction("llvm_visc_initializeTimerSet",
+ runtimeModule->getFunction("llvm_visc_initializeTimerSet")->getFunctionType());
+ //DEBUG(errs() << *llvm_visc_initializeTimerSet);
+
+ llvm_visc_switchToTimer = M.getOrInsertFunction("llvm_visc_switchToTimer",
+ runtimeModule->getFunction("llvm_visc_switchToTimer")->getFunctionType());
+ // DEBUG(errs() << *llvm_visc_switchToTimer);
+
+ llvm_visc_printTimerSet = M.getOrInsertFunction("llvm_visc_printTimerSet",
+ runtimeModule->getFunction("llvm_visc_printTimerSet")->getFunctionType());
+ //DEBUG(errs() << *llvm_visc_printTimerSet);
+
+ // Insert init context in main
+ DEBUG(errs() << "Locate __visc__init()\n");
+ Function* VI = M.getFunction("__visc__init");
+ assert(VI->getNumUses() == 1 && "__visc__init should only be used once");
+ Instruction* I = cast<Instruction>(*VI->user_begin());
+
+ DEBUG(errs() << "Initialize Timer Set\n");
+ initializeTimerSet(I);
+ switchToTimer(visc_TimerID_NONE, I);
+
+ // Insert print instruction at visc exit
+ DEBUG(errs() << "Locate __visc__cleanup()\n");
+ Function* VC = M.getFunction("__visc__cleanup");
+ assert(VC->getNumUses() == 1 && "__visc__cleanup should only be used once");
+ I = cast<Instruction>(*VC->user_begin());
+ printTimerSet(I);
+
+
+ DEBUG(errs() << "-------- Searching for launch sites ----------\n");
+
+ std::vector<Instruction*> toBeErased;
+ std::vector<Function*> functions;
+
+ for (auto &F : M)
+ functions.push_back(&F);
+
+ // Iterate over all functions in the module
+ for (Function *f : functions) {
+ DEBUG(errs() << "Function: " << f->getName() << "\n");
+
+ // List with the required additions in the function's return type
+ std::vector<Type*> FRetTypes;
+
+ enum mutateTypeCause {
+ mtc_None,
+ mtc_BIND,
+ mtc_RETURN,
+ mtc_NUM_CAUSES
+ } bind;
+ bind = mutateTypeCause::mtc_None;
+
+ // Iterate over all the instructions in this function
+ for (inst_iterator i = inst_begin(f), e = inst_end(f); i != e ; ++i) {
+ Instruction* I = &*i; // Grab pointer to Instruction
+ // If not a call instruction, move to next instruction
+ if(!isa<CallInst>(I))
+ continue;
+
+ CallInst* CI = cast<CallInst>(I);
+ LLVMContext& Ctx = CI->getContext();
+
+ if(isVISCCall_init(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_init, &toBeErased);
+ }
+ if(isVISCCall_cleanup(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_cleanup, &toBeErased);
+ }
+ if(isVISCCall_wait(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_wait, &toBeErased);
+ }
+ if(isVISCCall_trackMemory(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_trackMemory, &toBeErased);
+ }
+ if(isVISCCall_untrackMemory(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_untrackMemory, &toBeErased);
+ }
+ if(isVISCCall_requestMemory(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_requestMemory, &toBeErased);
+ }
+ if(isVISCCall_hint(I)) {
+ assert(isa<ConstantInt>(CI->getArgOperand(0))
+ && "Argument to hint must be constant integer!");
+ ConstantInt* hint = cast<ConstantInt>(CI->getArgOperand(0));
+
+ visc::Target t = (visc::Target) hint->getZExtValue();
+ addHint(CI->getParent()->getParent(), t);
+ DEBUG(errs() << "Found visc hint call: " << *CI << "\n");
+ toBeErased.push_back(CI);
+ }
+ if(isVISCCall_launch(I)) {
+ Function* LaunchF = Intrinsic::getDeclaration(&M, Intrinsic::visc_launch);
+ DEBUG(errs() << *LaunchF << "\n");
+ // Get i8* cast to function pointer
+ Function* graphFunc = cast<Function>(CI->getArgOperand(1));
+ graphFunc = transformReturnTypeToStruct(graphFunc);
+ Constant* F = ConstantExpr::getPointerCast(graphFunc, Type::getInt8PtrTy(Ctx));
+ assert(F && "Function invoked by VISC launch has to be define and constant.");
+
+ ConstantInt* Op = cast<ConstantInt>(CI->getArgOperand(0));
+ assert(Op && "VISC launch's streaming argument is a constant value.");
+ Value* isStreaming = Op->isZero()? ConstantInt::getFalse(Ctx)
+ : ConstantInt::getTrue(Ctx);
+
+ auto *ArgTy = dyn_cast<PointerType>(CI->getArgOperand(2)->getType());
+ assert(ArgTy && "VISC launch argument should be pointer type.");
+ Value *Arg = CI->getArgOperand(2);
+ if(!ArgTy->getElementType()->isIntegerTy(8))
+ Arg = BitCastInst::CreatePointerCast(CI->getArgOperand(2), Type::getInt8PtrTy(Ctx), "", CI);
+ Value* LaunchArgs[] = {F, Arg, isStreaming};
+ CallInst* LaunchInst = CallInst::Create(LaunchF,
+ ArrayRef<Value*>(LaunchArgs, 3),
+ "graphID", CI);
+ DEBUG(errs() << "Found visc launch call: " << *CI << "\n");
+ DEBUG(errs() << "\tSubstitute with: " << *LaunchInst << "\n");
+ CI->replaceAllUsesWith(LaunchInst);
+ toBeErased.push_back(CI);
+ }
+ if(isVISCCall_push(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_push, &toBeErased);
+ }
+ if(isVISCCall_pop(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_pop, &toBeErased);
+ }
+ if(isVISCCall_createNodeND(I)) {
+ assert(CI->getNumArgOperands() > 0 &&
+ "Too few arguments for __visc__createNodeND call");
+ unsigned numDims = getNumericValue(CI->getArgOperand(0));
+ // We need as meny dimension argments are there are dimensions
+ assert(CI->getNumArgOperands()-2 == numDims &&
+ "Too few arguments for __visc_createNodeND call!\n");
+
+ Function* CreateNodeF;
+ switch (numDims) {
+ case 0:
+ CreateNodeF = Intrinsic::getDeclaration(&M, Intrinsic::visc_createNode);
+ break;
+ case 1:
+ CreateNodeF = Intrinsic::getDeclaration(&M, Intrinsic::visc_createNode1D);
+ break;
+ case 2:
+ CreateNodeF = Intrinsic::getDeclaration(&M, Intrinsic::visc_createNode2D);
+ break;
+ case 3:
+ CreateNodeF = Intrinsic::getDeclaration(&M, Intrinsic::visc_createNode3D);
+ break;
+ default:
+ llvm_unreachable("Unsupported number of dimensions\n");
+ break;
+ }
+ DEBUG(errs() << *CreateNodeF << "\n");
+ DEBUG(errs() << *I << "\n");
+ DEBUG(errs() << "in " << I->getParent()->getParent()->getName() << "\n");
+
+ // Get i8* cast to function pointer
+ Function* graphFunc = cast<Function>(CI->getArgOperand(1));
+ graphFunc = transformReturnTypeToStruct(graphFunc);
+ Constant* F = ConstantExpr::getPointerCast(graphFunc, Type::getInt8PtrTy(Ctx));
+
+ CallInst* CreateNodeInst;
+ switch (numDims) {
+ case 0:
+ CreateNodeInst = CallInst::Create(CreateNodeF,
+ ArrayRef<Value*>(F),
+ graphFunc->getName()+".node", CI);
+ break;
+ case 1:
+ {
+ assert((CI->getArgOperand(2)->getType() == Type::getInt64Ty(Ctx)) &&
+ "CreateNodeND dimension argument, 2, expected to be i64\n");
+ Value* CreateNodeArgs[] = {F, CI->getArgOperand(2)};
+ CreateNodeInst = CallInst::Create(CreateNodeF,
+ ArrayRef<Value*>(CreateNodeArgs, 2),
+ graphFunc->getName()+".node", CI);
+ }
+ break;
+ case 2:
+ {
+ assert((CI->getArgOperand(2)->getType() == Type::getInt64Ty(Ctx)) &&
+ "CreateNodeND dimension argument, 2, expected to be i64\n");
+ assert((CI->getArgOperand(3)->getType() == Type::getInt64Ty(Ctx)) &&
+ "CreateNodeND dimension argument, 3, expected to be i64\n");
+ Value* CreateNodeArgs[] = {F,
+ CI->getArgOperand(2),
+ CI->getArgOperand(3)};
+ CreateNodeInst = CallInst::Create(CreateNodeF,
+ ArrayRef<Value*>(CreateNodeArgs, 3),
+ graphFunc->getName()+".node", CI);
+ }
+ break;
+ case 3:
+ {
+ assert((CI->getArgOperand(2)->getType() == Type::getInt64Ty(Ctx)) &&
+ "CreateNodeND dimension argument, 2, expected to be i64\n");
+ assert((CI->getArgOperand(3)->getType() == Type::getInt64Ty(Ctx)) &&
+ "CreateNodeND dimension argument, 3, expected to be i64\n");
+ assert((CI->getArgOperand(4)->getType() == Type::getInt64Ty(Ctx)) &&
+ "CreateNodeND dimension argument, 4, expected to be i64\n");
+ Value* CreateNodeArgs[] = {F,
+ CI->getArgOperand(2),
+ CI->getArgOperand(3),
+ CI->getArgOperand(4)};
+ CreateNodeInst = CallInst::Create(CreateNodeF,
+ ArrayRef<Value*>(CreateNodeArgs, 4),
+ graphFunc->getName()+".node", CI);
+ }
+ break;
+ default:
+ llvm_unreachable("Impossible path: number of dimensions is 0, 1, 2, 3\n");
+ break;
+ }
+
+ DEBUG(errs() << "Found visc createNode call: " << *CI << "\n");
+ DEBUG(errs() << "\tSubstitute with: " << *CreateNodeInst << "\n");
+ CI->replaceAllUsesWith(CreateNodeInst);
+ toBeErased.push_back(CI);
+ }
+
+ if(isVISCCall_edge(I)) {
+ Function* EdgeF = Intrinsic::getDeclaration(&M, Intrinsic::visc_createEdge);
+ DEBUG(errs() << *EdgeF << "\n");
+ ConstantInt* Op = cast<ConstantInt>(CI->getArgOperand(5));
+ ConstantInt* EdgeTypeOp = cast<ConstantInt>(CI->getArgOperand(2));
+ assert(Op && EdgeTypeOp && "Arguments of CreateEdge are not constant integers.");
+ Value* isStreaming = Op->isZero()? ConstantInt::getFalse(Ctx)
+ : ConstantInt::getTrue(Ctx);
+ Value* isAllToAll = EdgeTypeOp->isZero()? ConstantInt::getFalse(Ctx)
+ : ConstantInt::getTrue(Ctx);
+ Value* EdgeArgs[] = {CI->getArgOperand(0), CI->getArgOperand(1),
+ isAllToAll, CI->getArgOperand(3), CI->getArgOperand(4),
+ isStreaming
+ };
+ CallInst* EdgeInst = CallInst::Create(EdgeF,
+ ArrayRef<Value*>(EdgeArgs, 6),
+ "output", CI);
+ DEBUG(errs() << "Found visc edge call: " << *CI << "\n");
+ DEBUG(errs() << "\tSubstitute with: " << *EdgeInst << "\n");
+ CI->replaceAllUsesWith(EdgeInst);
+ toBeErased.push_back(CI);
+ }
+ if(isVISCCall_bindIn(I)) {
+ Function* BindInF = Intrinsic::getDeclaration(&M, Intrinsic::visc_bind_input);
+ DEBUG(errs() << *BindInF << "\n");
+ // Check if this is a streaming bind or not
+ ConstantInt* Op = cast<ConstantInt>(CI->getArgOperand(3));
+ assert(Op && "Streaming argument for bind in intrinsic should be a constant integer.");
+ Value* isStreaming = Op->isZero()? ConstantInt::getFalse(Ctx)
+ : ConstantInt::getTrue(Ctx);
+ Value* BindInArgs[] = {CI->getArgOperand(0), CI->getArgOperand(1),
+ CI->getArgOperand(2), isStreaming
+ };
+ CallInst* BindInInst = CallInst::Create(BindInF,
+ ArrayRef<Value*>(BindInArgs, 4),
+ "", CI);
+ DEBUG(errs() << "Found visc bindIn call: " << *CI << "\n");
+ DEBUG(errs() << "\tSubstitute with: " << *BindInInst << "\n");
+ CI->replaceAllUsesWith(BindInInst);
+ toBeErased.push_back(CI);
+ }
+ if(isVISCCall_bindOut(I)) {
+ Function* BindOutF = Intrinsic::getDeclaration(&M, Intrinsic::visc_bind_output);
+ DEBUG(errs() << *BindOutF << "\n");
+ // Check if this is a streaming bind or not
+ ConstantInt* Op = cast<ConstantInt>(CI->getArgOperand(3));
+ assert(Op && "Streaming argument for bind out intrinsic should be a constant integer.");
+ Value* isStreaming = Op->isZero()? ConstantInt::getFalse(Ctx)
+ : ConstantInt::getTrue(Ctx);
+ Value* BindOutArgs[] = {CI->getArgOperand(0), CI->getArgOperand(1),
+ CI->getArgOperand(2), isStreaming
+ };
+ CallInst* BindOutInst = CallInst::Create(BindOutF,
+ ArrayRef<Value*>(BindOutArgs, 4),
+ "", CI);
+ DEBUG(errs() << "Found visc bindOut call: " << *CI << "\n");
+ DEBUG(errs() << "\tSubstitute with: " << *BindOutInst << "\n");
+
+ DEBUG(errs() << "Fixing the return type of the function\n");
+ // FIXME: What if the child node function has not been visited already.
+ // i.e., it's return type has not been fixed.
+ Function* F = I->getParent()->getParent();
+ DEBUG(errs() << F->getName() << "\n";);
+ IntrinsicInst* NodeIntrinsic = cast<IntrinsicInst>(CI->getArgOperand(0));
+ assert(NodeIntrinsic && "Instruction value in bind out is not a create node intrinsic.");
+ DEBUG(errs() << "Node intrinsic: " << *NodeIntrinsic << "\n");
+ assert((NodeIntrinsic->getIntrinsicID() == Intrinsic::visc_createNode ||
+ NodeIntrinsic->getIntrinsicID() == Intrinsic::visc_createNode1D ||
+ NodeIntrinsic->getIntrinsicID() == Intrinsic::visc_createNode2D ||
+ NodeIntrinsic->getIntrinsicID() == Intrinsic::visc_createNode3D) &&
+ "Instruction value in bind out is not a create node intrinsic.");
+ Function* ChildF = cast<Function>(NodeIntrinsic->getArgOperand(0)->stripPointerCasts());
+ DEBUG(errs() << ChildF->getName() << "\n";);
+ int srcpos = cast<ConstantInt>(CI->getArgOperand(1))->getSExtValue();
+ int destpos = cast<ConstantInt>(CI->getArgOperand(2))->getSExtValue();
+ StructType* ChildReturnTy = cast<StructType>(ChildF->getReturnType());
+
+ Type* ReturnType = F->getReturnType();
+ DEBUG(errs() << *ReturnType << "\n";);
+ assert((ReturnType->isVoidTy() || isa<StructType>(ReturnType))
+ && "Return type should either be a struct or void type!");
+
+ FRetTypes.insert(FRetTypes.begin()+destpos, ChildReturnTy->getElementType(srcpos));
+ assert(((bind == mutateTypeCause::mtc_BIND) ||
+ (bind == mutateTypeCause::mtc_None)) &&
+ "Both bind_out and visc_return detected");
+ bind = mutateTypeCause::mtc_BIND;
+
+ CI->replaceAllUsesWith(BindOutInst);
+ toBeErased.push_back(CI);
+ }
+ if(isVISCCall_attributes(I)) {
+ Function* F = CI->getParent()->getParent();
+ handleVISCAttributes(F, CI);
+ toBeErased.push_back(CI);
+ }
+ if (isVISCCall_getNode(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_getNode, &toBeErased);
+ }
+ if (isVISCCall_getParentNode(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_getParentNode, &toBeErased);
+ }
+ if (isVISCCall_barrier(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_barrier, &toBeErased);
+ }
+ if (isVISCCall_malloc(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_malloc, &toBeErased);
+ }
+ if (isVISCCall_return(I)) {
+ DEBUG(errs() << "Function before visc return processing\n" << *I->getParent()->getParent() << "\n");
+ // The operands to this call are the values to be returned by the node
+ Value* ReturnVal = genCodeForReturn(CI);
+ DEBUG(errs() << *ReturnVal << "\n");
+ Type* ReturnType = ReturnVal->getType();
+ assert(isa<StructType>(ReturnType)
+ && "Return type should be a struct type!");
+
+ assert(((bind == mutateTypeCause::mtc_RETURN) ||
+ (bind == mutateTypeCause::mtc_None)) &&
+ "Both bind_out and visc_return detected");
+
+ if (bind == mutateTypeCause::mtc_None) {
+ // If this is None, this is the first __visc__return
+ // instruction we have come upon. Place the return type of the
+ // function in the return type vector
+ bind = mutateTypeCause::mtc_RETURN;
+ StructType* ReturnStructTy = cast<StructType>(ReturnType);
+ for (unsigned i = 0; i < ReturnStructTy->getNumElements(); i++)
+ FRetTypes.push_back(ReturnStructTy->getElementType(i));
+ } else { // bind == mutateTypeCause::mtc_RETURN
+ // This is not the first __visc__return
+ // instruction we have come upon.
+ // Check that the return types are the same
+ assert((ReturnType == FRetTypes[0])
+ && "Multiple returns with mismatching types");
+ }
+
+ ReturnInst* RetInst = ReturnInst::Create(Ctx, ReturnVal);
+ DEBUG(errs() << "Found visc return call: " << *CI << "\n");
+ Instruction* oldReturn = CI->getParent()->getTerminator();
+ assert(isa<ReturnInst>(oldReturn)
+ && "Expecting a return to be the terminator of this BB!");
+ DEBUG(errs() << "Found return statement of BB: " << *oldReturn << "\n");
+ DEBUG(errs() << "\tSubstitute return with: " << *RetInst << "\n");
+ //CI->replaceAllUsesWith(RetInst);
+ toBeErased.push_back(CI);
+ ReplaceInstWithInst(oldReturn, RetInst);
+ DEBUG(errs() << "Function after visc return processing\n" << *I->getParent()->getParent() << "\n");
+ }
+
+ if (isVISCCall_getNodeInstanceID_x(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_getNodeInstanceID_x, &toBeErased);
+ }
+ if (isVISCCall_getNodeInstanceID_y(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_getNodeInstanceID_y, &toBeErased);
+ }
+ if (isVISCCall_getNodeInstanceID_z(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_getNodeInstanceID_z, &toBeErased);
+ }
+ if (isVISCCall_getNumNodeInstances_x(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_getNumNodeInstances_x, &toBeErased);
+ }
+ if (isVISCCall_getNumNodeInstances_y(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_getNumNodeInstances_y, &toBeErased);
+ }
+ if (isVISCCall_getNumNodeInstances_z(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_getNumNodeInstances_z, &toBeErased);
+ }
+ if (isVISCCall_atomic_add(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_atomic_add, &toBeErased);
+ }
+ if (isVISCCall_atomic_sub(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_atomic_sub, &toBeErased);
+ }
+ if (isVISCCall_atomic_xchg(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_atomic_xchg, &toBeErased);
+ }
+ if (isVISCCall_atomic_min(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_atomic_min, &toBeErased);
+ }
+ if (isVISCCall_atomic_max(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_atomic_max, &toBeErased);
+ }
+ if (isVISCCall_atomic_and(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_atomic_and, &toBeErased);
+ }
+ if (isVISCCall_atomic_or(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_atomic_or, &toBeErased);
+ }
+ if (isVISCCall_atomic_xor(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::visc_atomic_xor, &toBeErased);
+ }
+ if (isVISCCall_sin(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::sin, &toBeErased);
+ }
+ if (isVISCCall_cos(I)) {
+ ReplaceCallWithIntrinsic(I, Intrinsic::cos, &toBeErased);
+ }
+ }
+
+ // Erase the __visc__node calls
+ DEBUG(errs() << "Erase " << toBeErased.size() << " Statements:\n");
+ for(auto I: toBeErased) {
+ DEBUG(errs() << *I << "\n");
+ }
+ while(!toBeErased.empty()) {
+ Instruction* I = toBeErased.back();
+ DEBUG(errs() << "\tErasing " << *I << "\n");
+ I->eraseFromParent();
+ toBeErased.pop_back();
+ }
+
+ if(bind == mutateTypeCause::mtc_BIND || bind == mutateTypeCause::mtc_RETURN) {
+ DEBUG(errs() << "Function before fixing return type\n" << *f << "\n");
+ // Argument type list.
+ std::vector<Type*> FArgTypes;
+ for(Function::const_arg_iterator ai = f->arg_begin(), ae = f->arg_end();
+ ai != ae; ++ai) {
+ FArgTypes.push_back(ai->getType());
+ }
+
+ // Find new return type of function
+ Type* NewReturnTy;
+ if(bind == mutateTypeCause::mtc_BIND) {
+
+ std::vector<Type*> TyList;
+ for (unsigned i = 0; i < FRetTypes.size(); i++)
+ TyList.push_back(FRetTypes[i]);
+
+ NewReturnTy = StructType::create(f->getContext(), TyList, Twine("struct.out."+f->getName()).str(), true);
+ }
+ else {
+ NewReturnTy = getReturnTypeFromReturnInst(f);
+ assert(NewReturnTy->isStructTy() && "Expecting a struct type!");
+ }
+
+ FunctionType* FTy = FunctionType::get(NewReturnTy, FArgTypes, f->isVarArg());
+
+ // Change the function type
+ Function* newF = cloneFunction(f, FTy, false);
+ DEBUG(errs() << *newF << "\n");
+
+ if (bind == mutateTypeCause::mtc_BIND) {
+ // This is certainly an internal node, and hence just one BB with one
+ // return terminator instruction. Change return statement
+ ReturnInst* RI = cast<ReturnInst>(newF->getEntryBlock().getTerminator());
+ ReturnInst* newRI = ReturnInst::Create(newF->getContext(), UndefValue::get(NewReturnTy));
+ ReplaceInstWithInst(RI, newRI);
+ }
+ if (bind == mutateTypeCause::mtc_RETURN) {
+ // Nothing
+ }
+ replaceNodeFunctionInIR(*f->getParent(), f, newF);
+ DEBUG(errs() << "Function after fixing return type\n" << *newF << "\n");
+ }
+
+
+ }
+ return false; //TODO: What does returning "false" mean?
+}
+
+// Generate Code for declaring a constant string [L x i8] and return a pointer
+// to the start of it.
+Value* GenVISC::getStringPointer(const Twine& S, Instruction* IB, const Twine& Name) {
+ Constant* SConstant = ConstantDataArray::getString(M->getContext(), S.str(), true);
+ Value* SGlobal = new GlobalVariable(*M, SConstant->getType(), true,
+ GlobalValue::InternalLinkage, SConstant, Name);
+ Value* Zero = ConstantInt::get(Type::getInt64Ty(M->getContext()), 0);
+ Value* GEPArgs[] = {Zero, Zero};
+ GetElementPtrInst* SPtr = GetElementPtrInst::Create(nullptr, SGlobal,
+ ArrayRef<Value*>(GEPArgs, 2), Name+"Ptr", IB);
+ return SPtr;
+}
+
+void GenVISC::initializeTimerSet(Instruction* InsertBefore) {
+ Value* TimerSetAddr;
+ StoreInst* SI;
+ TIMER(TimerSet = new GlobalVariable(*M,
+ Type::getInt8PtrTy(M->getContext()),
+ false,
+ GlobalValue::CommonLinkage,
+ Constant::getNullValue(Type::getInt8PtrTy(M->getContext())),
+ "viscTimerSet_GenVISC"));
+ DEBUG(errs() << "Inserting GV: " << *TimerSet->getType() << *TimerSet << "\n");
+ //DEBUG(errs() << "Inserting call to: " << *llvm_visc_initializeTimerSet << "\n");
+
+ TIMER(TimerSetAddr = CallInst::Create(llvm_visc_initializeTimerSet,
+ None,
+ "",
+ InsertBefore));
+ DEBUG(errs() << "TimerSetAddress = " << *TimerSetAddr << "\n");
+ TIMER(SI = new StoreInst(TimerSetAddr, TimerSet, InsertBefore));
+ DEBUG(errs() << "Store Timer Address in Global variable: " << *SI << "\n");
+}
+
+void GenVISC::switchToTimer(enum visc_TimerID timer, Instruction* InsertBefore) {
+ Value* switchArgs[] = {TimerSet, getTimerID(*M, timer)};
+ TIMER(CallInst::Create(llvm_visc_switchToTimer,
+ ArrayRef<Value*>(switchArgs, 2),
+ "",
+ InsertBefore));
+}
+
+void GenVISC::printTimerSet(Instruction* InsertBefore) {
+ Value* TimerName;
+ TIMER(TimerName = getStringPointer("GenVISC_Timer", InsertBefore));
+ Value* printArgs[] = {TimerSet, TimerName};
+ TIMER(CallInst::Create(llvm_visc_printTimerSet,
+ ArrayRef<Value*>(printArgs, 2),
+ "",
+ InsertBefore));
+}
+
+static inline ConstantInt* getTimerID(Module& M, enum visc_TimerID timer) {
+ return ConstantInt::get(Type::getInt32Ty(M.getContext()), timer);
+}
+
+static Function* transformReturnTypeToStruct(Function* F) {
+ // Currently only works for void return types
+ DEBUG(errs() << "Transforming return type of function to Struct: " << F->getName() << "\n");
+
+ if (isa<StructType>(F->getReturnType())) {
+ DEBUG(errs() << "Return type is already a Struct: " << F->getName() << ": " << *F->getReturnType() << "\n");
+ return F;
+ }
+
+ assert(F->getReturnType()->isVoidTy() && "Unhandled case - Only void return type handled\n");
+
+ // Create the argument type list with added argument types
+ std::vector<Type*> ArgTypes;
+ for(Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end();
+ ai != ae; ++ai) {
+ ArgTypes.push_back(ai->getType());
+ }
+
+ StructType* RetTy = StructType::create(F->getContext(), None, "emptyStruct", true);
+ FunctionType* FTy = FunctionType::get(RetTy, ArgTypes, F->isVarArg());
+
+ SmallVector<ReturnInst*, 8> Returns;
+ Function* newF = cloneFunction(F, FTy, false, &Returns);
+ // Replace ret void instruction with ret %RetTy undef
+ for(auto &RI: Returns) {
+ DEBUG(errs() << "Found return inst: "<< *RI << "\n");
+ ReturnInst* newRI = ReturnInst::Create(newF->getContext(), UndefValue::get(RetTy));
+ ReplaceInstWithInst(RI, newRI);
+ }
+
+ replaceNodeFunctionInIR(*F->getParent(), F, newF);
+ return newF;
+}
+
+static Type* getReturnTypeFromReturnInst(Function* F) {
+ for(BasicBlock &BB: *F) {
+ if(ReturnInst* RI = dyn_cast<ReturnInst>(BB.getTerminator())) {
+ DEBUG(errs() << "Return type value: " << *RI->getReturnValue()->getType() << "\n");
+ return RI->getReturnValue()->getType();
+ }
+ }
+}
+
+
+char genvisc::GenVISC::ID = 0;
+static RegisterPass<genvisc::GenVISC> X("genvisc", "Pass to generate VISC IR from LLVM IR (with dummy function calls)", false, false);
+
+} // End of namespace genvisc
+
+
diff --git a/hpvm/lib/Transforms/GenHPVM/GenHPVM.exports b/hpvm/lib/Transforms/GenVISC/GenVISC.exports
similarity index 100%
rename from hpvm/lib/Transforms/GenHPVM/GenHPVM.exports
rename to hpvm/lib/Transforms/GenVISC/GenVISC.exports
diff --git a/hpvm/lib/Transforms/GenHPVM/LLVMBuild.txt b/hpvm/lib/Transforms/GenVISC/LLVMBuild.txt
similarity index 88%
rename from hpvm/lib/Transforms/GenHPVM/LLVMBuild.txt
rename to hpvm/lib/Transforms/GenVISC/LLVMBuild.txt
index 94ef73ac07..9266b2c597 100644
--- a/hpvm/lib/Transforms/GenHPVM/LLVMBuild.txt
+++ b/hpvm/lib/Transforms/GenVISC/LLVMBuild.txt
@@ -1,4 +1,4 @@
-;===- ./lib/Transforms/GenHPVM/LLVMBuild.txt -------------------*- Conf -*--===;
+;===- ./lib/Transforms/GenVISC/LLVMBuild.txt -------------------*- Conf -*--===;
;
; The LLVM Compiler Infrastructure
;
@@ -17,5 +17,5 @@
[component_0]
type = Library
-name = GenHPVM
+name = GenVISC
parent = Transforms
diff --git a/hpvm/lib/Transforms/LocalMem/LocalMem.cpp b/hpvm/lib/Transforms/LocalMem/LocalMem.cpp
index fc33ebee71..7bd66b62c6 100644
--- a/hpvm/lib/Transforms/LocalMem/LocalMem.cpp
+++ b/hpvm/lib/Transforms/LocalMem/LocalMem.cpp
@@ -8,7 +8,7 @@
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "LocalMem"
-#include "SupportHPVM/DFG2LLVM.h"
+#include "SupportVISC/DFG2LLVM.h"
#include "llvm/IR/Constant.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/InstIterator.h"
@@ -134,7 +134,7 @@ void AT_OCL::codeGen(DFLeafNode *N) {
// Return pointer to property if this leaf node matches the conditions for being
// an allocation node. Conditions
// 1. No incoming memory pointer. No in/out attribute on a pointer argument
-// 2. Uses hpvm malloc intrinsic to allocate memory
+// 2. Uses visc malloc intrinsic to allocate memory
// 3. Sends it out
// 2. (TODO:) Whether the allocated pointer escapes the parent node
AllocationNodeProperty *isAllocationNode(DFLeafNode *N) {
@@ -148,18 +148,18 @@ AllocationNodeProperty *isAllocationNode(DFLeafNode *N) {
Function *F = N->getFuncPointer();
- // Allocation node must use hpvm malloc intrinsic
- bool usesHPVMMalloc = false;
+ // Allocation node must use visc malloc intrinsic
+ bool usesVISCMalloc = false;
for (inst_iterator i = inst_begin(F), e = inst_end(F); i != e; i++) {
Instruction *I = &*i;
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
- if (II->getIntrinsicID() == Intrinsic::hpvm_malloc) {
- usesHPVMMalloc = true;
+ if (II->getIntrinsicID() == Intrinsic::visc_malloc) {
+ usesVISCMalloc = true;
break;
}
}
}
- if (!usesHPVMMalloc)
+ if (!usesVISCMalloc)
return NULL;
// TODO: Check if allocated pointer leaves parent node
@@ -197,20 +197,20 @@ AllocationNodeProperty *isAllocationNode(DFLeafNode *N) {
assert(OutValues[i]->getType()->isPointerTy() &&
"Expected outgoing edge to be of pointer type");
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(OutValues[i])) {
- if (II->getIntrinsicID() == Intrinsic::hpvm_malloc) {
+ if (II->getIntrinsicID() == Intrinsic::visc_malloc) {
// Sanity check: Size passed to malloc intrinsic is same as the value
// going into the next outgoing edge
- DEBUG(errs() << "HPVM malloc size: " << *II->getArgOperand(0) << "\n");
+ DEBUG(errs() << "Visc malloc size: " << *II->getArgOperand(0) << "\n");
DEBUG(errs() << "Out edge value: " << *OutValues[i + 1] << "\n");
assert(II->getArgOperand(0) == OutValues[i + 1] &&
- "Sanity Check Failed: HPVM Malloc size argument != next "
+ "Sanity Check Failed: VISC Malloc size argument != next "
"outgoing edge");
ANP->insertAllocation(N->getOutDFEdgeAt(i), II->getArgOperand(0));
i = i + 2;
continue;
}
}
- llvm_unreachable("Expecting hpvm malloc intrinsic instruction!");
+ llvm_unreachable("Expecting visc malloc intrinsic instruction!");
}
return ANP;
}
diff --git a/hpvm/llvm_installer/llvm_installer.sh b/hpvm/llvm_installer/llvm_installer.sh
index e072d042b7..d7fcda4ac4 100755
--- a/hpvm/llvm_installer/llvm_installer.sh
+++ b/hpvm/llvm_installer/llvm_installer.sh
@@ -179,10 +179,10 @@ echo make -j$NUM_THREADS
make -j$NUM_THREADS
#make install
-# echo Building HPVM runtime
-# HPVM_RT_DIR=$HPVM_DIR/projects/hpvm-rt
-# cd $HPVM_RT_DIR
-# make
+#echo Building HPVM runtime
+#HPVM_RT_DIR=$HPVM_DIR/projects/visc-rt
+#cd $HPVM_RT_DIR
+#make
#cp -r $CURRENT_DIR/projects $HPVM_DIR/
#make -j$NUM_THREADS
diff --git a/hpvm/llvm_patches/apply_patch.sh b/hpvm/llvm_patches/apply_patch.sh
index 289e5c11e3..ea86575207 100644
--- a/hpvm/llvm_patches/apply_patch.sh
+++ b/hpvm/llvm_patches/apply_patch.sh
@@ -1,7 +1,7 @@
#!/bin/sh
### File Copies
-cp include/IR/IntrinsicsHPVM.td ${LLVM_SRC_ROOT}/include/llvm/IR/IntrinsicsHPVM.td
+cp include/IR/IntrinsicsVISC.td ${LLVM_SRC_ROOT}/include/llvm/IR/IntrinsicsVISC.td
## Header File Patches
diff --git a/hpvm/llvm_patches/include/IR/Attributes.td b/hpvm/llvm_patches/include/IR/Attributes.td
index c6ff8ef3c6..b644cdb30b 100644
--- a/hpvm/llvm_patches/include/IR/Attributes.td
+++ b/hpvm/llvm_patches/include/IR/Attributes.td
@@ -151,7 +151,7 @@ def ShadowCallStack : EnumAttr<"shadowcallstack">;
/// Sign extended before/after call.
def SExt : EnumAttr<"signext">;
-/// HPVM Attributes
+/// VISC Attributes
/// Pointer to read only memory
def In : EnumAttr<"in">;
diff --git a/hpvm/llvm_patches/include/IR/Intrinsics.td b/hpvm/llvm_patches/include/IR/Intrinsics.td
index 2e3f34eb1a..2f79964a2e 100644
--- a/hpvm/llvm_patches/include/IR/Intrinsics.td
+++ b/hpvm/llvm_patches/include/IR/Intrinsics.td
@@ -1249,4 +1249,4 @@ include "llvm/IR/IntrinsicsBPF.td"
include "llvm/IR/IntrinsicsSystemZ.td"
include "llvm/IR/IntrinsicsWebAssembly.td"
include "llvm/IR/IntrinsicsRISCV.td"
-include "llvm/IR/IntrinsicsHPVM.td"
+include "llvm/IR/IntrinsicsVISC.td"
diff --git a/hpvm/llvm_patches/include/IR/IntrinsicsHPVM.td b/hpvm/llvm_patches/include/IR/IntrinsicsHPVM.td
deleted file mode 100644
index 410e9c8d33..0000000000
--- a/hpvm/llvm_patches/include/IR/IntrinsicsHPVM.td
+++ /dev/null
@@ -1,208 +0,0 @@
-//===- IntrinsicsHPVM.td - Defines HPVM intrinsics ---------*- tablegen -*-===//
-//
-// The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file defines all of the HPVM-specific intrinsics.
-//
-//===----------------------------------------------------------------------===//
-
-let TargetPrefix = "hpvm" in {
- /* All intrinsics start with "llvm.hpvm."
- * As we do not want the compiler to mess with these intrinsics, we assume
- * worst memory behavior for all these intrinsics.
- */
-
- /* Initialization intrinsic -
- * i8* llvm.hpvm.setup(function*);
- */
- def int_hpvm_init : Intrinsic<[], [], []>;
-
- /* Launch intrinsic - with streaming argument
- * i8* llvm.hpvm.launch(i8*, ArgList*, i1);
- */
- def int_hpvm_launch : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty,
- llvm_ptr_ty, llvm_i1_ty], []>;
-
- /* Push intrinsic - push data on streaming pipeline
- * void llvm.hpvm.push(i8*, ArgList*);
- */
- def int_hpvm_push : Intrinsic<[], [llvm_ptr_ty, llvm_ptr_ty], []>;
-
- /* Pop intrinsic - pop data from streaming pipeline
- * i8* llvm.hpvm.pop(i8*);
- */
- def int_hpvm_pop : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty], []>;
-
- /* Cleanup intrinsic -
- * void llvm.hpvm.cleanup(i8*);
- */
- def int_hpvm_cleanup : Intrinsic<[], [], []>;
-
- /* Wait intrinsic -
- * void llvm.hpvm.wait(graphID*);
- */
- def int_hpvm_wait : Intrinsic<[], [llvm_ptr_ty], []>;
-
- /* Track memory intrinsic -
- * void llvm.hpvm.trackMemory(i8*, i64);
- */
- def int_hpvm_trackMemory : Intrinsic<[], [llvm_ptr_ty, llvm_i64_ty], []>;
-
- /* Track memory intrinsic -
- * void llvm.hpvm.untrackMemory(i8*);
- */
- def int_hpvm_untrackMemory : Intrinsic<[], [llvm_ptr_ty], []>;
-
- /* Request memory intrinsic -
- * void llvm.hpvm.requestMemory(i8*, i64);
- */
- def int_hpvm_requestMemory : Intrinsic<[], [llvm_ptr_ty, llvm_i64_ty], []>;
-
- /* Create Node intrinsic -
- * i8* llvm.hpvm.createNode(function*);
- */
- def int_hpvm_createNode : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty], []>;
-
- /* Create Node 1D array intrinsic -
- * i8* llvm.hpvm.createNode1D(function*, i64);
- */
- def int_hpvm_createNode1D : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty,
- llvm_i64_ty], []>;
-
- /* Create Node 2D array intrinsic -
- * i8* llvm.hpvm.createNode2D(function*, i64, i64);
- */
- def int_hpvm_createNode2D : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty,
- llvm_i64_ty, llvm_i64_ty], []>;
-
- /* Create Node 3D array intrinsic -
- * i8* llvm.hpvm.createNode2D(function*, i64, i64, i64);
- */
- def int_hpvm_createNode3D : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty,
- llvm_i64_ty, llvm_i64_ty, llvm_i64_ty],
- []>;
-
- /* Create dataflow edge intrinsic -
- * i8* llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32, i1);
- */
- def int_hpvm_createEdge : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty, llvm_ptr_ty,
- llvm_i1_ty, llvm_i32_ty, llvm_i32_ty,
- llvm_i1_ty],
- []>;
-
- /* Create bind input intrinsic -
- * void llvm.hpvm.bind.input(i8*, i32, i32);
- */
- def int_hpvm_bind_input : Intrinsic<[], [llvm_ptr_ty, llvm_i32_ty,
- llvm_i32_ty, llvm_i1_ty], []>;
-
- /* Create bind output intrinsic -
- * void llvm.hpvm.bind.output(i8*, i32, i32);
- */
- def int_hpvm_bind_output : Intrinsic<[], [llvm_ptr_ty, llvm_i32_ty,
- llvm_i32_ty, llvm_i1_ty], []>;
-
- /* Find associated dataflow node intrinsic -
- * i8* llvm.hpvm.getNode();
- */
- def int_hpvm_getNode : Intrinsic<[llvm_ptr_ty], [], [IntrNoMem]>;
-
- /* Find parent dataflow node intrinsic -
- * i8* llvm.hpvm.getParentNode(i8*);
- */
- def int_hpvm_getParentNode : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty], [IntrNoMem]>;
-
- /* Find the number of dimensions of a dataflow node intrinsic -
- * i32 llvm.hpvm.getNumDims(i8*);
- */
- def int_hpvm_getNumDims : Intrinsic<[llvm_i32_ty], [llvm_ptr_ty], [IntrNoMem]>;
-
- /* Find the unique indentifier of a dataflow node (with respect to his parent
- * node) in the specified dimension intrinsic -
- */
-
- /* i64 llvm.hpvm.getNodeInstanceID.[xyz](i8*);
- */
- def int_hpvm_getNodeInstanceID_x : Intrinsic<[llvm_i64_ty], [llvm_ptr_ty],
- [IntrNoMem]>;
-
- def int_hpvm_getNodeInstanceID_y : Intrinsic<[llvm_i64_ty], [llvm_ptr_ty],
- [IntrNoMem]>;
-
- def int_hpvm_getNodeInstanceID_z : Intrinsic<[llvm_i64_ty], [llvm_ptr_ty],
- [IntrNoMem]>;
-
- /* Find the number of instances of a dataflow node in the specified dimension
- * intrinsic -
- */
-
- /* i64 llvm.hpvm.getNumNodeInstances.[xyz](i8*);
- */
- def int_hpvm_getNumNodeInstances_x : Intrinsic<[llvm_i64_ty], [llvm_ptr_ty],
- [IntrNoMem]>;
-
- def int_hpvm_getNumNodeInstances_y : Intrinsic<[llvm_i64_ty], [llvm_ptr_ty],
- [IntrNoMem]>;
-
- def int_hpvm_getNumNodeInstances_z : Intrinsic<[llvm_i64_ty], [llvm_ptr_ty],
- [IntrNoMem]>;
-
- /* Local Barrier
- * void llvm.hpvm.barrier();
- */
- def int_hpvm_barrier : Intrinsic<[], [], []>;
-
- /* Memory allocation inside the graph
- * i8* llvm.hpvm.malloc();
- */
- def int_hpvm_malloc : Intrinsic<[llvm_ptr_ty], [llvm_i64_ty], []>;
-
- /* Find the vector length supported by target architecture
- * intrinsic -
- * i32 llvm.hpvm.getVectorLength();
- */
- def int_hpvm_getVectorLength : Intrinsic<[llvm_i32_ty], [], []>;
-
- /* ============ Atomic intrinsics ============= */
- // Atomic arithmetic operations
-
- /* i32 llvm.hpvm.atomic.add(i32*, i32)*/
- def int_hpvm_atomic_add: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
- []>;
-
- /* i32 llvm.hpvm.atomic.sub(i32*, i32)*/
- def int_hpvm_atomic_sub: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
- []>;
-
- /* i32 llvm.hpvm.atomic.xchg(i32*, i32)*/
- def int_hpvm_atomic_xchg: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
- []>;
-
- /* i32 llvm.hpvm.atomic.min(i32*, i32)*/
- def int_hpvm_atomic_min: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
- []>;
-
- /* i32 llvm.hpvm.atomic.maxi32*, i32)*/
- def int_hpvm_atomic_max: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
- []>;
-
- // Atomic bitwise operations
-
- /* i32 llvm.hpvm.atomic.and(i32*, i32)*/
- def int_hpvm_atomic_and: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
- []>;
-
- /* i32 llvm.hpvm.atomic.or(i32*, i32)*/
- def int_hpvm_atomic_or: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
- []>;
-
- /* i32 llvm.hpvm.atomic.xor(i32*, i32)*/
- def int_hpvm_atomic_xor: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
- []>;
-
-}
diff --git a/hpvm/llvm_patches/include/IR/IntrinsicsVISC.td b/hpvm/llvm_patches/include/IR/IntrinsicsVISC.td
new file mode 100644
index 0000000000..d5330175d8
--- /dev/null
+++ b/hpvm/llvm_patches/include/IR/IntrinsicsVISC.td
@@ -0,0 +1,208 @@
+//===- IntrinsicsVISC.td - Defines VISC intrinsics ---------*- tablegen -*-===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines all of the VISC-specific intrinsics.
+//
+//===----------------------------------------------------------------------===//
+
+let TargetPrefix = "visc" in {
+ /* All intrinsics start with "llvm.visc."
+ * As we do not want the compiler to mess with these intrinsics, we assume
+ * worst memory behavior for all these intrinsics.
+ */
+
+ /* Initialization intrinsic -
+ * i8* llvm.visc.setup(function*);
+ */
+ def int_visc_init : Intrinsic<[], [], []>;
+
+ /* Launch intrinsic - with streaming argument
+ * i8* llvm.visc.launch(i8*, ArgList*, i1);
+ */
+ def int_visc_launch : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty,
+ llvm_ptr_ty, llvm_i1_ty], []>;
+
+ /* Push intrinsic - push data on streaming pipeline
+ * void llvm.visc.push(i8*, ArgList*);
+ */
+ def int_visc_push : Intrinsic<[], [llvm_ptr_ty, llvm_ptr_ty], []>;
+
+ /* Pop intrinsic - pop data from streaming pipeline
+ * i8* llvm.visc.pop(i8*);
+ */
+ def int_visc_pop : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty], []>;
+
+ /* Cleanup intrinsic -
+ * void llvm.visc.cleanup(i8*);
+ */
+ def int_visc_cleanup : Intrinsic<[], [], []>;
+
+ /* Wait intrinsic -
+ * void llvm.visc.wait(graphID*);
+ */
+ def int_visc_wait : Intrinsic<[], [llvm_ptr_ty], []>;
+
+ /* Track memory intrinsic -
+ * void llvm.visc.trackMemory(i8*, i64);
+ */
+ def int_visc_trackMemory : Intrinsic<[], [llvm_ptr_ty, llvm_i64_ty], []>;
+
+ /* Track memory intrinsic -
+ * void llvm.visc.untrackMemory(i8*);
+ */
+ def int_visc_untrackMemory : Intrinsic<[], [llvm_ptr_ty], []>;
+
+ /* Request memory intrinsic -
+ * void llvm.visc.requestMemory(i8*, i64);
+ */
+ def int_visc_requestMemory : Intrinsic<[], [llvm_ptr_ty, llvm_i64_ty], []>;
+
+ /* Create Node intrinsic -
+ * i8* llvm.visc.createNode(function*);
+ */
+ def int_visc_createNode : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty], []>;
+
+ /* Create Node 1D array intrinsic -
+ * i8* llvm.visc.createNode1D(function*, i64);
+ */
+ def int_visc_createNode1D : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty,
+ llvm_i64_ty], []>;
+
+ /* Create Node 2D array intrinsic -
+ * i8* llvm.visc.createNode2D(function*, i64, i64);
+ */
+ def int_visc_createNode2D : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty,
+ llvm_i64_ty, llvm_i64_ty], []>;
+
+ /* Create Node 3D array intrinsic -
+ * i8* llvm.visc.createNode2D(function*, i64, i64, i64);
+ */
+ def int_visc_createNode3D : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty,
+ llvm_i64_ty, llvm_i64_ty, llvm_i64_ty],
+ []>;
+
+ /* Create dataflow edge intrinsic -
+ * i8* llvm.visc.createEdge(i8*, i8*, i1, i32, i32, i1);
+ */
+ def int_visc_createEdge : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty, llvm_ptr_ty,
+ llvm_i1_ty, llvm_i32_ty, llvm_i32_ty,
+ llvm_i1_ty],
+ []>;
+
+ /* Create bind input intrinsic -
+ * void llvm.visc.bind.input(i8*, i32, i32);
+ */
+ def int_visc_bind_input : Intrinsic<[], [llvm_ptr_ty, llvm_i32_ty,
+ llvm_i32_ty, llvm_i1_ty], []>;
+
+ /* Create bind output intrinsic -
+ * void llvm.visc.bind.output(i8*, i32, i32);
+ */
+ def int_visc_bind_output : Intrinsic<[], [llvm_ptr_ty, llvm_i32_ty,
+ llvm_i32_ty, llvm_i1_ty], []>;
+
+ /* Find associated dataflow node intrinsic -
+ * i8* llvm.visc.getNode();
+ */
+ def int_visc_getNode : Intrinsic<[llvm_ptr_ty], [], [IntrNoMem]>;
+
+ /* Find parent dataflow node intrinsic -
+ * i8* llvm.visc.getParentNode(i8*);
+ */
+ def int_visc_getParentNode : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty], [IntrNoMem]>;
+
+ /* Find the number of dimensions of a dataflow node intrinsic -
+ * i32 llvm.visc.getNumDims(i8*);
+ */
+ def int_visc_getNumDims : Intrinsic<[llvm_i32_ty], [llvm_ptr_ty], [IntrNoMem]>;
+
+ /* Find the unique indentifier of a dataflow node (with respect to his parent
+ * node) in the specified dimension intrinsic -
+ */
+
+ /* i64 llvm.visc.getNodeInstanceID.[xyz](i8*);
+ */
+ def int_visc_getNodeInstanceID_x : Intrinsic<[llvm_i64_ty], [llvm_ptr_ty],
+ [IntrNoMem]>;
+
+ def int_visc_getNodeInstanceID_y : Intrinsic<[llvm_i64_ty], [llvm_ptr_ty],
+ [IntrNoMem]>;
+
+ def int_visc_getNodeInstanceID_z : Intrinsic<[llvm_i64_ty], [llvm_ptr_ty],
+ [IntrNoMem]>;
+
+ /* Find the number of instances of a dataflow node in the specified dimension
+ * intrinsic -
+ */
+
+ /* i64 llvm.visc.getNumNodeInstances.[xyz](i8*);
+ */
+ def int_visc_getNumNodeInstances_x : Intrinsic<[llvm_i64_ty], [llvm_ptr_ty],
+ [IntrNoMem]>;
+
+ def int_visc_getNumNodeInstances_y : Intrinsic<[llvm_i64_ty], [llvm_ptr_ty],
+ [IntrNoMem]>;
+
+ def int_visc_getNumNodeInstances_z : Intrinsic<[llvm_i64_ty], [llvm_ptr_ty],
+ [IntrNoMem]>;
+
+ /* Local Barrier
+ * void llvm.visc.barrier();
+ */
+ def int_visc_barrier : Intrinsic<[], [], []>;
+
+ /* Memory allocation inside the graph
+ * i8* llvm.visc.malloc();
+ */
+ def int_visc_malloc : Intrinsic<[llvm_ptr_ty], [llvm_i64_ty], []>;
+
+ /* Find the vector length supported by target architecture
+ * intrinsic -
+ * i32 llvm.visc.getVectorLength();
+ */
+ def int_visc_getVectorLength : Intrinsic<[llvm_i32_ty], [], []>;
+
+ /* ============ Atomic intrinsics ============= */
+ // Atomic arithmetic operations
+
+ /* i32 llvm.visc.atomic.add(i32*, i32)*/
+ def int_visc_atomic_add: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
+ []>;
+
+ /* i32 llvm.visc.atomic.sub(i32*, i32)*/
+ def int_visc_atomic_sub: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
+ []>;
+
+ /* i32 llvm.visc.atomic.xchg(i32*, i32)*/
+ def int_visc_atomic_xchg: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
+ []>;
+
+ /* i32 llvm.visc.atomic.min(i32*, i32)*/
+ def int_visc_atomic_min: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
+ []>;
+
+ /* i32 llvm.visc.atomic.maxi32*, i32)*/
+ def int_visc_atomic_max: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
+ []>;
+
+ // Atomic bitwise operations
+
+ /* i32 llvm.visc.atomic.and(i32*, i32)*/
+ def int_visc_atomic_and: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
+ []>;
+
+ /* i32 llvm.visc.atomic.or(i32*, i32)*/
+ def int_visc_atomic_or: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
+ []>;
+
+ /* i32 llvm.visc.atomic.xor(i32*, i32)*/
+ def int_visc_atomic_xor: Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
+ []>;
+
+}
diff --git a/hpvm/llvm_patches/lib/AsmParser/LLLexer.cpp b/hpvm/llvm_patches/lib/AsmParser/LLLexer.cpp
index 2c54392f80..a924405a2c 100644
--- a/hpvm/llvm_patches/lib/AsmParser/LLLexer.cpp
+++ b/hpvm/llvm_patches/lib/AsmParser/LLLexer.cpp
@@ -855,7 +855,7 @@ lltok::Kind LLLexer::LexIdentifier() {
KEYWORD(bit);
KEYWORD(varFlags);
- // HPVM parameter attributes
+ // VISC parameter attributes
KEYWORD(in);
KEYWORD(out);
KEYWORD(inout);
diff --git a/hpvm/llvm_patches/lib/AsmParser/LLParser.cpp b/hpvm/llvm_patches/lib/AsmParser/LLParser.cpp
index 7446ff1e32..f5ce44e2a9 100644
--- a/hpvm/llvm_patches/lib/AsmParser/LLParser.cpp
+++ b/hpvm/llvm_patches/lib/AsmParser/LLParser.cpp
@@ -1470,7 +1470,7 @@ bool LLParser::ParseFnAttributeValuePairs(AttrBuilder &B,
case lltok::kw_swiftself:
case lltok::kw_immarg:
- // HPVM Parameter only attributes
+ // VISC Parameter only attributes
case lltok::kw_in:
case lltok::kw_out:
case lltok::kw_inout:
@@ -1808,7 +1808,7 @@ bool LLParser::ParseOptionalParamAttrs(AttrBuilder &B) {
B.addAttribute(Attribute::ImmArg);
break;
- // HPVM parameter attributes
+ // VISC parameter attributes
case lltok::kw_in:
B.addAttribute(Attribute::In);
break;
@@ -1927,7 +1927,7 @@ bool LLParser::ParseOptionalReturnAttrs(AttrBuilder &B) {
case lltok::kw_swiftself:
case lltok::kw_immarg:
- // HPVM Parameter only attributes
+ // VISC Parameter only attributes
case lltok::kw_in:
case lltok::kw_out:
case lltok::kw_inout:
diff --git a/hpvm/llvm_patches/lib/AsmParser/LLToken.h b/hpvm/llvm_patches/lib/AsmParser/LLToken.h
index cb0479b41c..7f9816965b 100644
--- a/hpvm/llvm_patches/lib/AsmParser/LLToken.h
+++ b/hpvm/llvm_patches/lib/AsmParser/LLToken.h
@@ -351,7 +351,7 @@ enum Kind {
kw_insertvalue,
kw_blockaddress,
- // HPVM parameter attributes
+ // VISC parameter attributes
kw_in,
kw_out,
kw_inout,
diff --git a/hpvm/llvm_patches/lib/Bitcode/Reader/BitcodeReader.cpp b/hpvm/llvm_patches/lib/Bitcode/Reader/BitcodeReader.cpp
index a1e6447285..7eb289d587 100644
--- a/hpvm/llvm_patches/lib/Bitcode/Reader/BitcodeReader.cpp
+++ b/hpvm/llvm_patches/lib/Bitcode/Reader/BitcodeReader.cpp
@@ -1395,7 +1395,7 @@ static uint64_t getRawAttributeMask(Attribute::AttrKind Val) {
case Attribute::NoFree:
return 1ULL << 63;
- // HPVM Attributes
+ // VISC Attributes
case Attribute::In:
return 3ULL << 0;
case Attribute::Out:
diff --git a/hpvm/llvm_patches/lib/Bitcode/Writer/BitcodeWriter.cpp b/hpvm/llvm_patches/lib/Bitcode/Writer/BitcodeWriter.cpp
index fd671c3975..55e7415efb 100644
--- a/hpvm/llvm_patches/lib/Bitcode/Writer/BitcodeWriter.cpp
+++ b/hpvm/llvm_patches/lib/Bitcode/Writer/BitcodeWriter.cpp
@@ -773,7 +773,7 @@ static uint64_t getAttrKindEncoding(Attribute::AttrKind Kind) {
case Attribute::SanitizeMemTag:
return bitc::ATTR_KIND_SANITIZE_MEMTAG;
- // HPVM Attributes
+ // VISC Attributes
case Attribute::In:
return bitc::ATTR_KIND_IN;
case Attribute::Out:
diff --git a/hpvm/llvm_patches/lib/IR/Attributes.cpp b/hpvm/llvm_patches/lib/IR/Attributes.cpp
index 29c47a9e11..3cc95b3102 100644
--- a/hpvm/llvm_patches/lib/IR/Attributes.cpp
+++ b/hpvm/llvm_patches/lib/IR/Attributes.cpp
@@ -404,7 +404,7 @@ std::string Attribute::getAsString(bool InAttrGrp) const {
if (hasAttribute(Attribute::ImmArg))
return "immarg";
- // HPVM attributes for arguments
+ // VISC attributes for arguments
if (hasAttribute(Attribute::In))
return "in";
if (hasAttribute(Attribute::Out))
diff --git a/hpvm/projects/hpvm-rt/CMakeLists.txt b/hpvm/projects/hpvm-rt/CMakeLists.txt
deleted file mode 100644
index be7f69c4bf..0000000000
--- a/hpvm/projects/hpvm-rt/CMakeLists.txt
+++ /dev/null
@@ -1,22 +0,0 @@
-add_definitions(-DNUM_CORES=8)
-
-SET(CMAKE_C_COMPILER ${CMAKE_BINARY_DIR}/bin/clang)
-SET(CMAKE_CXX_COMPILER ${CMAKE_BINARY_DIR}/bin/clang++)
-
-add_llvm_library(hpvm-rt.ll hpvm-rt.cpp
-
- DEPENDS
- clang
- llvm-dis
- )
-
-
-target_compile_options(hpvm-rt.ll PUBLIC -flto )
-target_compile_options(hpvm-rt.ll PUBLIC -std=c++11)
-
-add_custom_target(hpvm-rt.cpp.o ALL
- COMMAND ar -x ${CMAKE_BINARY_DIR}/lib/libhpvm-rt.ll.a
- COMMAND mv ${CMAKE_BINARY_DIR}/tools/hpvm/projects/hpvm-rt/hpvm-rt.cpp.o ${CMAKE_BINARY_DIR}/tools/hpvm/projects/hpvm-rt/hpvm-rt.bc
- COMMAND ${CMAKE_BINARY_DIR}/bin/llvm-dis ${CMAKE_BINARY_DIR}/tools/hpvm/projects/hpvm-rt/hpvm-rt.bc)
-
-add_dependencies(hpvm-rt.cpp.o hpvm-rt.ll)
diff --git a/hpvm/projects/visc-rt/CMakeLists.txt b/hpvm/projects/visc-rt/CMakeLists.txt
new file mode 100644
index 0000000000..5b9449bf2d
--- /dev/null
+++ b/hpvm/projects/visc-rt/CMakeLists.txt
@@ -0,0 +1,22 @@
+add_definitions(-DNUM_CORES=8)
+
+SET(CMAKE_C_COMPILER ${CMAKE_BINARY_DIR}/bin/clang)
+SET(CMAKE_CXX_COMPILER ${CMAKE_BINARY_DIR}/bin/clang++)
+
+add_llvm_library(visc-rt.ll visc-rt.cpp
+
+ DEPENDS
+ clang
+ llvm-dis
+ )
+
+
+target_compile_options(visc-rt.ll PUBLIC -flto )
+target_compile_options(visc-rt.ll PUBLIC -std=c++11)
+
+add_custom_target(visc-rt.cpp.o ALL
+ COMMAND ar -x ${CMAKE_BINARY_DIR}/lib/libvisc-rt.ll.a
+ COMMAND mv ${CMAKE_BINARY_DIR}/tools/hpvm/projects/visc-rt/visc-rt.cpp.o ${CMAKE_BINARY_DIR}/tools/hpvm/projects/visc-rt/visc-rt.bc
+ COMMAND ${CMAKE_BINARY_DIR}/bin/llvm-dis ${CMAKE_BINARY_DIR}/tools/hpvm/projects/visc-rt/visc-rt.bc)
+
+add_dependencies(visc-rt.cpp.o visc-rt.ll)
diff --git a/hpvm/projects/hpvm-rt/deviceStatusSwitchIntervals.txt b/hpvm/projects/visc-rt/deviceStatusSwitchIntervals.txt
similarity index 100%
rename from hpvm/projects/hpvm-rt/deviceStatusSwitchIntervals.txt
rename to hpvm/projects/visc-rt/deviceStatusSwitchIntervals.txt
diff --git a/hpvm/projects/hpvm-rt/device_abstraction.h b/hpvm/projects/visc-rt/device_abstraction.h
similarity index 96%
rename from hpvm/projects/hpvm-rt/device_abstraction.h
rename to hpvm/projects/visc-rt/device_abstraction.h
index 4948502ce8..7e77d100de 100644
--- a/hpvm/projects/hpvm-rt/device_abstraction.h
+++ b/hpvm/projects/visc-rt/device_abstraction.h
@@ -27,7 +27,7 @@ void initializeDeviceStatusIntervals() {
unsigned sz = 0;
unsigned tmp = 0;
- const char *fn = "/home/kotsifa2/HPVM/hpvm/build/projects/hpvm-rt/"
+ const char *fn = "/home/kotsifa2/HPVM/hpvm/build/projects/visc-rt/"
"deviceStatusSwitchIntervals.txt";
std::ifstream infile;
infile.open(fn);
diff --git a/hpvm/projects/hpvm-rt/makefile b/hpvm/projects/visc-rt/makefile
similarity index 97%
rename from hpvm/projects/hpvm-rt/makefile
rename to hpvm/projects/visc-rt/makefile
index 927e26e254..adcc632335 100644
--- a/hpvm/projects/hpvm-rt/makefile
+++ b/hpvm/projects/visc-rt/makefile
@@ -9,7 +9,7 @@ ifeq ($(NUM_CORES),)
endif
CPP_FLAGS = -I$(LLVM_SRC_ROOT)/include -I$(LLVM_BUILD_ROOT)/include -I$(CUDA_INC_PATH) -std=c++11 -D__STDC_CONSTANT_MACROS -D__STDC_LIMIT_MACROS
-TARGET:=hpvm-rt
+TARGET:=visc-rt
LLVM_CC:=$(LLVM_BUILD_ROOT)/bin/clang
LLVM_CXX:=$(LLVM_BUILD_ROOT)/bin/clang++
diff --git a/hpvm/projects/hpvm-rt/policy.h b/hpvm/projects/visc-rt/policy.h
similarity index 100%
rename from hpvm/projects/hpvm-rt/policy.h
rename to hpvm/projects/visc-rt/policy.h
diff --git a/hpvm/projects/hpvm-rt/hpvm-rt.cpp b/hpvm/projects/visc-rt/visc-rt.cpp
similarity index 82%
rename from hpvm/projects/hpvm-rt/hpvm-rt.cpp
rename to hpvm/projects/visc-rt/visc-rt.cpp
index ec2534cf43..53d3b516f2 100644
--- a/hpvm/projects/hpvm-rt/hpvm-rt.cpp
+++ b/hpvm/projects/visc-rt/visc-rt.cpp
@@ -13,7 +13,7 @@
#if _POSIX_VERSION >= 200112L
#include <sys/time.h>
#endif
-#include "hpvm-rt.h"
+#include "visc-rt.h"
#ifndef DEBUG_BUILD
#define DEBUG(s) \
@@ -59,7 +59,7 @@ vector<DFGDepth> DStack;
pthread_mutex_t ocl_mtx;
#define NUM_TESTS 1
-hpvm_TimerSet kernel_timer;
+visc_TimerSet kernel_timer;
static inline void checkErr(cl_int err, cl_int success, const char *name) {
if (err != success) {
@@ -70,7 +70,7 @@ static inline void checkErr(cl_int err, cl_int success, const char *name) {
}
/************************* Policies *************************************/
-void llvm_hpvm_policy_init() {
+void llvm_visc_policy_init() {
cout << "Initializing policy object ...\n";
// policy = new NodePolicy();
// policy = new IterationPolicy();
@@ -80,19 +80,19 @@ void llvm_hpvm_policy_init() {
cout << "DONE: Initializing policy object.\n";
}
-void llvm_hpvm_policy_clear() {
+void llvm_visc_policy_clear() {
if (policy)
free(policy);
}
-int llvm_hpvm_policy_getVersion(const char *name, int64_t i) {
+int llvm_visc_policy_getVersion(const char *name, int64_t i) {
return policy->getVersion(name, i);
}
/******************** Device Abstraction ********************************/
std::thread deviceStatusThread;
-void llvm_hpvm_deviceAbstraction_start() {
+void llvm_visc_deviceAbstraction_start() {
cout << "Starting device status simulation ...\n";
// Initialize vector with points where ti switch device status
initializeDeviceStatusIntervals();
@@ -102,7 +102,7 @@ void llvm_hpvm_deviceAbstraction_start() {
return;
}
-void llvm_hpvm_deviceAbstraction_end() {
+void llvm_visc_deviceAbstraction_end() {
cout << "Ending device status simulation thread ...\n";
// Set the variable that allows the thread to know that execution has ended
executionEnd = true;
@@ -112,7 +112,7 @@ void llvm_hpvm_deviceAbstraction_end() {
return;
}
-void llvm_hpvm_deviceAbstraction_waitOnDeviceStatus() {
+void llvm_visc_deviceAbstraction_waitOnDeviceStatus() {
while (!deviceStatus) {
};
return;
@@ -120,7 +120,7 @@ void llvm_hpvm_deviceAbstraction_waitOnDeviceStatus() {
/************************* Depth Stack Routines ***************************/
-void llvm_hpvm_x86_dstack_push(unsigned n, uint64_t limitX, uint64_t iX,
+void llvm_visc_x86_dstack_push(unsigned n, uint64_t limitX, uint64_t iX,
uint64_t limitY, uint64_t iY, uint64_t limitZ,
uint64_t iZ) {
DEBUG(cout << "Pushing node information on stack:\n");
@@ -134,7 +134,7 @@ void llvm_hpvm_x86_dstack_push(unsigned n, uint64_t limitX, uint64_t iX,
pthread_mutex_unlock(&ocl_mtx);
}
-void llvm_hpvm_x86_dstack_pop() {
+void llvm_visc_x86_dstack_pop() {
DEBUG(cout << "Popping from depth stack\n");
pthread_mutex_lock(&ocl_mtx);
DStack.pop_back();
@@ -142,7 +142,7 @@ void llvm_hpvm_x86_dstack_pop() {
pthread_mutex_unlock(&ocl_mtx);
}
-uint64_t llvm_hpvm_x86_getDimLimit(unsigned level, unsigned dim) {
+uint64_t llvm_visc_x86_getDimLimit(unsigned level, unsigned dim) {
DEBUG(cout << "Request limit for dim " << dim << " of ancestor " << level
<< flush << "\n");
pthread_mutex_lock(&ocl_mtx);
@@ -154,7 +154,7 @@ uint64_t llvm_hpvm_x86_getDimLimit(unsigned level, unsigned dim) {
return result;
}
-uint64_t llvm_hpvm_x86_getDimInstance(unsigned level, unsigned dim) {
+uint64_t llvm_visc_x86_getDimInstance(unsigned level, unsigned dim) {
DEBUG(cout << "Request instance id for dim " << dim << " of ancestor "
<< level << flush << "\n");
pthread_mutex_lock(&ocl_mtx);
@@ -168,7 +168,7 @@ uint64_t llvm_hpvm_x86_getDimInstance(unsigned level, unsigned dim) {
/********************** Memory Tracking Routines **************************/
-void llvm_hpvm_track_mem(void *ptr, size_t size) {
+void llvm_visc_track_mem(void *ptr, size_t size) {
DEBUG(cout << "Start tracking memory: " << ptr << flush << "\n");
MemTrackerEntry *MTE = MTracker.lookup(ptr);
if (MTE != NULL) {
@@ -180,7 +180,7 @@ void llvm_hpvm_track_mem(void *ptr, size_t size) {
DEBUG(MTracker.print());
}
-void llvm_hpvm_untrack_mem(void *ptr) {
+void llvm_visc_untrack_mem(void *ptr) {
DEBUG(cout << "Stop tracking memory: " << ptr << flush << "\n");
MemTrackerEntry *MTE = MTracker.lookup(ptr);
if (MTE == NULL) {
@@ -195,7 +195,7 @@ void llvm_hpvm_untrack_mem(void *ptr) {
DEBUG(MTracker.print());
}
-static void *llvm_hpvm_ocl_request_mem(void *ptr, size_t size,
+static void *llvm_visc_ocl_request_mem(void *ptr, size_t size,
DFNodeContext_OCL *Context, bool isInput,
bool isOutput) {
pthread_mutex_lock(&ocl_mtx);
@@ -233,7 +233,7 @@ static void *llvm_hpvm_ocl_request_mem(void *ptr, size_t size,
else
clFlags = CL_MEM_READ_ONLY;
- hpvm_SwitchToTimer(&kernel_timer, hpvm_TimerID_COPY);
+ visc_SwitchToTimer(&kernel_timer, visc_TimerID_COPY);
// pthread_mutex_lock(&ocl_mtx);
cl_mem d_input =
clCreateBuffer(Context->clOCLContext, clFlags, size, NULL, &errcode);
@@ -249,7 +249,7 @@ static void *llvm_hpvm_ocl_request_mem(void *ptr, size_t size,
checkErr(errcode, CL_SUCCESS, "Failure to copy memory to device");
}
- hpvm_SwitchToTimer(&kernel_timer, hpvm_TimerID_NONE);
+ visc_SwitchToTimer(&kernel_timer, visc_TimerID_NONE);
DEBUG(cout << " done\n");
MTE->update(MemTrackerEntry::DEVICE, (void *)d_input, Context);
DEBUG(cout << "Updated Table\n");
@@ -258,11 +258,11 @@ static void *llvm_hpvm_ocl_request_mem(void *ptr, size_t size,
return d_input;
}
-void *llvm_hpvm_x86_argument_ptr(void *ptr, size_t size) {
- return llvm_hpvm_request_mem(ptr, size);
+void *llvm_visc_x86_argument_ptr(void *ptr, size_t size) {
+ return llvm_visc_request_mem(ptr, size);
}
-void *llvm_hpvm_request_mem(void *ptr, size_t size) {
+void *llvm_visc_request_mem(void *ptr, size_t size) {
pthread_mutex_lock(&ocl_mtx);
DEBUG(cout << "[X86] Request memory: " << ptr << flush << "\n");
MemTrackerEntry *MTE = MTracker.lookup(ptr);
@@ -283,13 +283,13 @@ void *llvm_hpvm_request_mem(void *ptr, size_t size) {
DEBUG(cout << "\tMemory found on device at: " << MTE->getAddress() << flush
<< "\n");
DEBUG(cout << "\tCopying ...");
- hpvm_SwitchToTimer(&kernel_timer, hpvm_TimerID_COPY);
+ visc_SwitchToTimer(&kernel_timer, visc_TimerID_COPY);
// pthread_mutex_lock(&ocl_mtx);
cl_int errcode = clEnqueueReadBuffer(
((DFNodeContext_OCL *)MTE->getContext())->clCommandQue,
(cl_mem)MTE->getAddress(), CL_TRUE, 0, size, ptr, 0, NULL, NULL);
// pthread_mutex_unlock(&ocl_mtx);
- hpvm_SwitchToTimer(&kernel_timer, hpvm_TimerID_NONE);
+ visc_SwitchToTimer(&kernel_timer, visc_TimerID_NONE);
DEBUG(cout << " done\n");
checkErr(errcode, CL_SUCCESS, "[request mem] Failure to read output");
DEBUG(cout << "Free mem object on device\n");
@@ -303,25 +303,25 @@ void *llvm_hpvm_request_mem(void *ptr, size_t size) {
/*************************** Timer Routines **********************************/
-static int is_async(enum hpvm_TimerID timer) {
- return (timer == hpvm_TimerID_KERNEL) || (timer == hpvm_TimerID_COPY_ASYNC);
+static int is_async(enum visc_TimerID timer) {
+ return (timer == visc_TimerID_KERNEL) || (timer == visc_TimerID_COPY_ASYNC);
}
-static int is_blocking(enum hpvm_TimerID timer) {
- return (timer == hpvm_TimerID_COPY) || (timer == hpvm_TimerID_NONE);
+static int is_blocking(enum visc_TimerID timer) {
+ return (timer == visc_TimerID_COPY) || (timer == visc_TimerID_NONE);
}
-#define INVALID_TIMERID hpvm_TimerID_LAST
+#define INVALID_TIMERID visc_TimerID_LAST
-static int asyncs_outstanding(struct hpvm_TimerSet *timers) {
+static int asyncs_outstanding(struct visc_TimerSet *timers) {
return (timers->async_markers != NULL) &&
(timers->async_markers->timerID != INVALID_TIMERID);
}
-static struct hpvm_async_time_marker_list *
-get_last_async(struct hpvm_TimerSet *timers) {
+static struct visc_async_time_marker_list *
+get_last_async(struct visc_TimerSet *timers) {
/* Find the last event recorded thus far */
- struct hpvm_async_time_marker_list *last_event = timers->async_markers;
+ struct visc_async_time_marker_list *last_event = timers->async_markers;
if (last_event != NULL && last_event->timerID != INVALID_TIMERID) {
while (last_event->next != NULL &&
last_event->next->timerID != INVALID_TIMERID)
@@ -331,17 +331,17 @@ get_last_async(struct hpvm_TimerSet *timers) {
return NULL;
}
-static void insert_marker(struct hpvm_TimerSet *tset, enum hpvm_TimerID timer) {
+static void insert_marker(struct visc_TimerSet *tset, enum visc_TimerID timer) {
cl_int ciErrNum = CL_SUCCESS;
- struct hpvm_async_time_marker_list **new_event = &(tset->async_markers);
+ struct visc_async_time_marker_list **new_event = &(tset->async_markers);
while (*new_event != NULL && (*new_event)->timerID != INVALID_TIMERID) {
new_event = &((*new_event)->next);
}
if (*new_event == NULL) {
- *new_event = (struct hpvm_async_time_marker_list *)malloc(
- sizeof(struct hpvm_async_time_marker_list));
+ *new_event = (struct visc_async_time_marker_list *)malloc(
+ sizeof(struct visc_async_time_marker_list));
(*new_event)->marker = calloc(1, sizeof(cl_event));
/*
// I don't think this is needed at all. I believe clEnqueueMarker 'creates'
@@ -372,18 +372,18 @@ Event Status!\n");
}
}
-static void insert_submarker(struct hpvm_TimerSet *tset, char *label,
- enum hpvm_TimerID timer) {
+static void insert_submarker(struct visc_TimerSet *tset, char *label,
+ enum visc_TimerID timer) {
cl_int ciErrNum = CL_SUCCESS;
- struct hpvm_async_time_marker_list **new_event = &(tset->async_markers);
+ struct visc_async_time_marker_list **new_event = &(tset->async_markers);
while (*new_event != NULL && (*new_event)->timerID != INVALID_TIMERID) {
new_event = &((*new_event)->next);
}
if (*new_event == NULL) {
- *new_event = (struct hpvm_async_time_marker_list *)malloc(
- sizeof(struct hpvm_async_time_marker_list));
+ *new_event = (struct visc_async_time_marker_list *)malloc(
+ sizeof(struct visc_async_time_marker_list));
(*new_event)->marker = calloc(1, sizeof(cl_event));
/*
#if ( __OPENCL_VERSION__ >= CL_VERSION_1_1 )
@@ -414,10 +414,10 @@ Event Status!\n");
}
/* Assumes that all recorded events have completed */
-static hpvm_Timestamp record_async_times(struct hpvm_TimerSet *tset) {
- struct hpvm_async_time_marker_list *next_interval = NULL;
- struct hpvm_async_time_marker_list *last_marker = get_last_async(tset);
- hpvm_Timestamp total_async_time = 0;
+static visc_Timestamp record_async_times(struct visc_TimerSet *tset) {
+ struct visc_async_time_marker_list *next_interval = NULL;
+ struct visc_async_time_marker_list *last_marker = get_last_async(tset);
+ visc_Timestamp total_async_time = 0;
for (next_interval = tset->async_markers; next_interval != last_marker;
next_interval = next_interval->next) {
@@ -439,11 +439,11 @@ static hpvm_Timestamp record_async_times(struct hpvm_TimerSet *tset) {
ciErrNum);
}
- hpvm_Timestamp interval =
- (hpvm_Timestamp)(((double)(command_end - command_start)));
+ visc_Timestamp interval =
+ (visc_Timestamp)(((double)(command_end - command_start)));
tset->timers[next_interval->timerID].elapsed += interval;
if (next_interval->label != NULL) {
- struct hpvm_SubTimer *subtimer =
+ struct visc_SubTimer *subtimer =
tset->sub_timer_list[next_interval->timerID]->subtimer_list;
while (subtimer != NULL) {
if (strcmp(subtimer->label, next_interval->label) == 0) {
@@ -463,8 +463,8 @@ static hpvm_Timestamp record_async_times(struct hpvm_TimerSet *tset) {
return total_async_time;
}
-static void accumulate_time(hpvm_Timestamp *accum, hpvm_Timestamp start,
- hpvm_Timestamp end) {
+static void accumulate_time(visc_Timestamp *accum, visc_Timestamp start,
+ visc_Timestamp end) {
#if _POSIX_VERSION >= 200112L
*accum += end - start;
#else
@@ -473,33 +473,33 @@ static void accumulate_time(hpvm_Timestamp *accum, hpvm_Timestamp start,
}
#if _POSIX_VERSION >= 200112L
-static hpvm_Timestamp get_time() {
+static visc_Timestamp get_time() {
struct timespec tv;
clock_gettime(CLOCK_MONOTONIC, &tv);
- return (hpvm_Timestamp)(tv.tv_sec * BILLION + tv.tv_nsec);
+ return (visc_Timestamp)(tv.tv_sec * BILLION + tv.tv_nsec);
}
#else
#error "no supported time libraries are available on this platform"
#endif
-void hpvm_ResetTimer(struct hpvm_Timer *timer) {
- timer->state = hpvm_Timer_STOPPED;
+void visc_ResetTimer(struct visc_Timer *timer) {
+ timer->state = visc_Timer_STOPPED;
#if _POSIX_VERSION >= 200112L
timer->elapsed = 0;
#else
-#error "hpvm_ResetTimer: not implemented for this system"
+#error "visc_ResetTimer: not implemented for this system"
#endif
}
-void hpvm_StartTimer(struct hpvm_Timer *timer) {
- if (timer->state != hpvm_Timer_STOPPED) {
+void visc_StartTimer(struct visc_Timer *timer) {
+ if (timer->state != visc_Timer_STOPPED) {
// FIXME: Removing warning statement to avoid printing this error
// fputs("Ignoring attempt to start a running timer\n", stderr);
return;
}
- timer->state = hpvm_Timer_RUNNING;
+ timer->state = visc_Timer_RUNNING;
#if _POSIX_VERSION >= 200112L
{
@@ -508,19 +508,19 @@ void hpvm_StartTimer(struct hpvm_Timer *timer) {
timer->init = tv.tv_sec * BILLION + tv.tv_nsec;
}
#else
-#error "hpvm_StartTimer: not implemented for this system"
+#error "visc_StartTimer: not implemented for this system"
#endif
}
-void hpvm_StartTimerAndSubTimer(struct hpvm_Timer *timer,
- struct hpvm_Timer *subtimer) {
+void visc_StartTimerAndSubTimer(struct visc_Timer *timer,
+ struct visc_Timer *subtimer) {
unsigned int numNotStopped = 0x3; // 11
- if (timer->state != hpvm_Timer_STOPPED) {
+ if (timer->state != visc_Timer_STOPPED) {
fputs("Warning: Timer was not stopped\n", stderr);
numNotStopped &= 0x1; // Zero out 2^1
}
- if (subtimer->state != hpvm_Timer_STOPPED) {
+ if (subtimer->state != visc_Timer_STOPPED) {
fputs("Warning: Subtimer was not stopped\n", stderr);
numNotStopped &= 0x2; // Zero out 2^0
}
@@ -529,8 +529,8 @@ void hpvm_StartTimerAndSubTimer(struct hpvm_Timer *timer,
return;
}
- timer->state = hpvm_Timer_RUNNING;
- subtimer->state = hpvm_Timer_RUNNING;
+ timer->state = visc_Timer_RUNNING;
+ subtimer->state = visc_Timer_RUNNING;
#if _POSIX_VERSION >= 200112L
{
@@ -546,19 +546,19 @@ void hpvm_StartTimerAndSubTimer(struct hpvm_Timer *timer,
}
}
#else
-#error "hpvm_StartTimer: not implemented for this system"
+#error "visc_StartTimer: not implemented for this system"
#endif
}
-void hpvm_StopTimer(struct hpvm_Timer *timer) {
- hpvm_Timestamp fini;
+void visc_StopTimer(struct visc_Timer *timer) {
+ visc_Timestamp fini;
- if (timer->state != hpvm_Timer_RUNNING) {
+ if (timer->state != visc_Timer_RUNNING) {
// fputs("Ignoring attempt to stop a stopped timer\n", stderr);
return;
}
- timer->state = hpvm_Timer_STOPPED;
+ timer->state = visc_Timer_STOPPED;
#if _POSIX_VERSION >= 200112L
{
@@ -567,24 +567,24 @@ void hpvm_StopTimer(struct hpvm_Timer *timer) {
fini = tv.tv_sec * BILLION + tv.tv_nsec;
}
#else
-#error "hpvm_StopTimer: not implemented for this system"
+#error "visc_StopTimer: not implemented for this system"
#endif
accumulate_time(&timer->elapsed, timer->init, fini);
timer->init = fini;
}
-void hpvm_StopTimerAndSubTimer(struct hpvm_Timer *timer,
- struct hpvm_Timer *subtimer) {
+void visc_StopTimerAndSubTimer(struct visc_Timer *timer,
+ struct visc_Timer *subtimer) {
- hpvm_Timestamp fini;
+ visc_Timestamp fini;
unsigned int numNotRunning = 0x3; // 11
- if (timer->state != hpvm_Timer_RUNNING) {
+ if (timer->state != visc_Timer_RUNNING) {
fputs("Warning: Timer was not running\n", stderr);
numNotRunning &= 0x1; // Zero out 2^1
}
- if (subtimer->state != hpvm_Timer_RUNNING) {
+ if (subtimer->state != visc_Timer_RUNNING) {
fputs("Warning: Subtimer was not running\n", stderr);
numNotRunning &= 0x2; // Zero out 2^0
}
@@ -593,8 +593,8 @@ void hpvm_StopTimerAndSubTimer(struct hpvm_Timer *timer,
return;
}
- timer->state = hpvm_Timer_STOPPED;
- subtimer->state = hpvm_Timer_STOPPED;
+ timer->state = visc_Timer_STOPPED;
+ subtimer->state = visc_Timer_STOPPED;
#if _POSIX_VERSION >= 200112L
{
@@ -603,7 +603,7 @@ void hpvm_StopTimerAndSubTimer(struct hpvm_Timer *timer,
fini = tv.tv_sec * BILLION + tv.tv_nsec;
}
#else
-#error "hpvm_StopTimer: not implemented for this system"
+#error "visc_StopTimer: not implemented for this system"
#endif
if (numNotRunning & 0x2) {
@@ -618,59 +618,59 @@ void hpvm_StopTimerAndSubTimer(struct hpvm_Timer *timer,
}
/* Get the elapsed time in seconds. */
-double hpvm_GetElapsedTime(struct hpvm_Timer *timer) {
+double visc_GetElapsedTime(struct visc_Timer *timer) {
double ret;
- if (timer->state != hpvm_Timer_STOPPED) {
+ if (timer->state != visc_Timer_STOPPED) {
fputs("Elapsed time from a running timer is inaccurate\n", stderr);
}
#if _POSIX_VERSION >= 200112L
ret = timer->elapsed / 1e9;
#else
-#error "hpvm_GetElapsedTime: not implemented for this system"
+#error "visc_GetElapsedTime: not implemented for this system"
#endif
return ret;
}
-void hpvm_InitializeTimerSet(struct hpvm_TimerSet *timers) {
+void visc_InitializeTimerSet(struct visc_TimerSet *timers) {
int n;
timers->wall_begin = get_time();
- timers->current = hpvm_TimerID_NONE;
+ timers->current = visc_TimerID_NONE;
timers->async_markers = NULL;
- for (n = 0; n < hpvm_TimerID_LAST; n++) {
- hpvm_ResetTimer(&timers->timers[n]);
+ for (n = 0; n < visc_TimerID_LAST; n++) {
+ visc_ResetTimer(&timers->timers[n]);
timers->sub_timer_list[n] = NULL;
}
}
-void hpvm_AddSubTimer(struct hpvm_TimerSet *timers, char *label,
- enum hpvm_TimerID hpvm_Category) {
+void visc_AddSubTimer(struct visc_TimerSet *timers, char *label,
+ enum visc_TimerID visc_Category) {
- struct hpvm_SubTimer *subtimer =
- (struct hpvm_SubTimer *)malloc(sizeof(struct hpvm_SubTimer));
+ struct visc_SubTimer *subtimer =
+ (struct visc_SubTimer *)malloc(sizeof(struct visc_SubTimer));
int len = strlen(label);
subtimer->label = (char *)malloc(sizeof(char) * (len + 1));
sprintf(subtimer->label, "%s", label);
- hpvm_ResetTimer(&subtimer->timer);
+ visc_ResetTimer(&subtimer->timer);
subtimer->next = NULL;
- struct hpvm_SubTimerList *subtimerlist =
- timers->sub_timer_list[hpvm_Category];
+ struct visc_SubTimerList *subtimerlist =
+ timers->sub_timer_list[visc_Category];
if (subtimerlist == NULL) {
subtimerlist =
- (struct hpvm_SubTimerList *)calloc(1, sizeof(struct hpvm_SubTimerList));
+ (struct visc_SubTimerList *)calloc(1, sizeof(struct visc_SubTimerList));
subtimerlist->subtimer_list = subtimer;
- timers->sub_timer_list[hpvm_Category] = subtimerlist;
+ timers->sub_timer_list[visc_Category] = subtimerlist;
} else {
// Append to list
- struct hpvm_SubTimer *element = subtimerlist->subtimer_list;
+ struct visc_SubTimer *element = subtimerlist->subtimer_list;
while (element->next != NULL) {
element = element->next;
}
@@ -678,37 +678,37 @@ void hpvm_AddSubTimer(struct hpvm_TimerSet *timers, char *label,
}
}
-void hpvm_SwitchToTimer(struct hpvm_TimerSet *timers, enum hpvm_TimerID timer) {
+void visc_SwitchToTimer(struct visc_TimerSet *timers, enum visc_TimerID timer) {
// cerr << "Switch to timer: " << timer << flush << "\n";
/* Stop the currently running timer */
- if (timers->current != hpvm_TimerID_NONE) {
- struct hpvm_SubTimerList *subtimerlist =
+ if (timers->current != visc_TimerID_NONE) {
+ struct visc_SubTimerList *subtimerlist =
timers->sub_timer_list[timers->current];
- struct hpvm_SubTimer *currSubTimer =
+ struct visc_SubTimer *currSubTimer =
(subtimerlist != NULL) ? subtimerlist->current : NULL;
if (!is_async(timers->current)) {
if (timers->current != timer) {
if (currSubTimer != NULL) {
- hpvm_StopTimerAndSubTimer(&timers->timers[timers->current],
+ visc_StopTimerAndSubTimer(&timers->timers[timers->current],
&currSubTimer->timer);
} else {
- hpvm_StopTimer(&timers->timers[timers->current]);
+ visc_StopTimer(&timers->timers[timers->current]);
}
} else {
if (currSubTimer != NULL) {
- hpvm_StopTimer(&currSubTimer->timer);
+ visc_StopTimer(&currSubTimer->timer);
}
}
} else {
insert_marker(timers, timer);
if (!is_async(timer)) { // if switching to async too, keep driver going
- hpvm_StopTimer(&timers->timers[hpvm_TimerID_DRIVER]);
+ visc_StopTimer(&timers->timers[visc_TimerID_DRIVER]);
}
}
}
- hpvm_Timestamp currentTime = get_time();
+ visc_Timestamp currentTime = get_time();
/* The only cases we check for asynchronous task completion is
* when an overlapping CPU operation completes, or the next
@@ -716,7 +716,7 @@ void hpvm_SwitchToTimer(struct hpvm_TimerSet *timers, enum hpvm_TimerID timer) {
if (asyncs_outstanding(timers) &&
(!is_async(timers->current) || is_blocking(timer))) {
- struct hpvm_async_time_marker_list *last_event = get_last_async(timers);
+ struct visc_async_time_marker_list *last_event = get_last_async(timers);
/* CL_COMPLETE if completed */
cl_int ciErrNum = CL_SUCCESS;
@@ -736,7 +736,7 @@ void hpvm_SwitchToTimer(struct hpvm_TimerSet *timers, enum hpvm_TimerID timer) {
// timer to switch to is COPY or NONE
if (async_done != CL_COMPLETE) {
- accumulate_time(&(timers->timers[hpvm_TimerID_OVERLAP].elapsed),
+ accumulate_time(&(timers->timers[visc_TimerID_OVERLAP].elapsed),
timers->async_begin, currentTime);
}
@@ -746,14 +746,14 @@ void hpvm_SwitchToTimer(struct hpvm_TimerSet *timers, enum hpvm_TimerID timer) {
fprintf(stderr, "Error Waiting for Events!\n");
}
- hpvm_Timestamp total_async_time = record_async_times(timers);
+ visc_Timestamp total_async_time = record_async_times(timers);
/* Async operations completed before previous CPU operations:
* overlapped time is the total async time */
if (async_done == CL_COMPLETE) {
// fprintf(stderr, "Async_done: total_async_type = %lld\n",
// total_async_time);
- timers->timers[hpvm_TimerID_OVERLAP].elapsed += total_async_time;
+ timers->timers[visc_TimerID_OVERLAP].elapsed += total_async_time;
}
} else
@@ -763,15 +763,15 @@ void hpvm_SwitchToTimer(struct hpvm_TimerSet *timers, enum hpvm_TimerID timer) {
if (async_done == CL_COMPLETE) {
/* Async operations completed before previous CPU operations:
* overlapped time is the total async time */
- timers->timers[hpvm_TimerID_OVERLAP].elapsed +=
+ timers->timers[visc_TimerID_OVERLAP].elapsed +=
record_async_times(timers);
}
}
/* Start the new timer */
- if (timer != hpvm_TimerID_NONE) {
+ if (timer != visc_TimerID_NONE) {
if (!is_async(timer)) {
- hpvm_StartTimer(&timers->timers[timer]);
+ visc_StartTimer(&timers->timers[timer]);
} else {
// toSwitchTo Is Async (KERNEL/COPY_ASYNC)
if (!asyncs_outstanding(timers)) {
@@ -785,48 +785,48 @@ void hpvm_SwitchToTimer(struct hpvm_TimerSet *timers, enum hpvm_TimerID timer) {
* so we can rename that marker as the beginning of this async
* operation */
- struct hpvm_async_time_marker_list *last_event = get_last_async(timers);
+ struct visc_async_time_marker_list *last_event = get_last_async(timers);
last_event->label = NULL;
last_event->timerID = timer;
}
if (!is_async(timers->current)) {
- hpvm_StartTimer(&timers->timers[hpvm_TimerID_DRIVER]);
+ visc_StartTimer(&timers->timers[visc_TimerID_DRIVER]);
}
}
}
timers->current = timer;
}
-void hpvm_SwitchToSubTimer(struct hpvm_TimerSet *timers, char *label,
- enum hpvm_TimerID category) {
- struct hpvm_SubTimerList *subtimerlist =
+void visc_SwitchToSubTimer(struct visc_TimerSet *timers, char *label,
+ enum visc_TimerID category) {
+ struct visc_SubTimerList *subtimerlist =
timers->sub_timer_list[timers->current];
- struct hpvm_SubTimer *curr =
+ struct visc_SubTimer *curr =
(subtimerlist != NULL) ? subtimerlist->current : NULL;
- if (timers->current != hpvm_TimerID_NONE) {
+ if (timers->current != visc_TimerID_NONE) {
if (!is_async(timers->current)) {
if (timers->current != category) {
if (curr != NULL) {
- hpvm_StopTimerAndSubTimer(&timers->timers[timers->current],
+ visc_StopTimerAndSubTimer(&timers->timers[timers->current],
&curr->timer);
} else {
- hpvm_StopTimer(&timers->timers[timers->current]);
+ visc_StopTimer(&timers->timers[timers->current]);
}
} else {
if (curr != NULL) {
- hpvm_StopTimer(&curr->timer);
+ visc_StopTimer(&curr->timer);
}
}
} else {
insert_submarker(timers, label, category);
if (!is_async(category)) { // if switching to async too, keep driver going
- hpvm_StopTimer(&timers->timers[hpvm_TimerID_DRIVER]);
+ visc_StopTimer(&timers->timers[visc_TimerID_DRIVER]);
}
}
}
- hpvm_Timestamp currentTime = get_time();
+ visc_Timestamp currentTime = get_time();
/* The only cases we check for asynchronous task completion is
* when an overlapping CPU operation completes, or the next
@@ -834,7 +834,7 @@ void hpvm_SwitchToSubTimer(struct hpvm_TimerSet *timers, char *label,
if (asyncs_outstanding(timers) &&
(!is_async(timers->current) || is_blocking(category))) {
- struct hpvm_async_time_marker_list *last_event = get_last_async(timers);
+ struct visc_async_time_marker_list *last_event = get_last_async(timers);
/* CL_COMPLETE if completed */
cl_int ciErrNum = CL_SUCCESS;
@@ -858,7 +858,7 @@ void hpvm_SwitchToSubTimer(struct hpvm_TimerSet *timers, char *label,
// because everything is being stopped to wait for synchronization it
// seems that the extra sync wall time isn't being recorded anywhere
if (async_done != CL_COMPLETE)
- accumulate_time(&(timers->timers[hpvm_TimerID_OVERLAP].elapsed),
+ accumulate_time(&(timers->timers[visc_TimerID_OVERLAP].elapsed),
timers->async_begin, currentTime);
/* Wait on async operation completion */
@@ -866,7 +866,7 @@ void hpvm_SwitchToSubTimer(struct hpvm_TimerSet *timers, char *label,
if (ciErrNum != CL_SUCCESS) {
fprintf(stderr, "Error Waiting for Events!\n");
}
- hpvm_Timestamp total_async_time = record_async_times(timers);
+ visc_Timestamp total_async_time = record_async_times(timers);
/* Async operations completed before previous CPU operations:
* overlapped time is the total async time */
@@ -874,7 +874,7 @@ void hpvm_SwitchToSubTimer(struct hpvm_TimerSet *timers, char *label,
// into OVERLAP the immediately preceding EventSynchronize theoretically
// didn't have any effect since it was already completed.
if (async_done == CL_COMPLETE /*cudaSuccess*/)
- timers->timers[hpvm_TimerID_OVERLAP].elapsed += total_async_time;
+ timers->timers[visc_TimerID_OVERLAP].elapsed += total_async_time;
} else
/* implies (!is_async(timers->current) && asyncs_outstanding(timers)) */
@@ -883,14 +883,14 @@ void hpvm_SwitchToSubTimer(struct hpvm_TimerSet *timers, char *label,
if (async_done == CL_COMPLETE /*cudaSuccess*/) {
/* Async operations completed before previous CPU operations:
* overlapped time is the total async time */
- timers->timers[hpvm_TimerID_OVERLAP].elapsed +=
+ timers->timers[visc_TimerID_OVERLAP].elapsed +=
record_async_times(timers);
}
// else, this isn't blocking, so just check the next time around
}
subtimerlist = timers->sub_timer_list[category];
- struct hpvm_SubTimer *subtimer = NULL;
+ struct visc_SubTimer *subtimer = NULL;
if (label != NULL) {
subtimer = subtimerlist->subtimer_list;
@@ -904,18 +904,18 @@ void hpvm_SwitchToSubTimer(struct hpvm_TimerSet *timers, char *label,
}
/* Start the new timer */
- if (category != hpvm_TimerID_NONE) {
+ if (category != visc_TimerID_NONE) {
if (!is_async(category)) {
if (subtimerlist != NULL) {
subtimerlist->current = subtimer;
}
if (category != timers->current && subtimer != NULL) {
- hpvm_StartTimerAndSubTimer(&timers->timers[category], &subtimer->timer);
+ visc_StartTimerAndSubTimer(&timers->timers[category], &subtimer->timer);
} else if (subtimer != NULL) {
- hpvm_StartTimer(&subtimer->timer);
+ visc_StartTimer(&subtimer->timer);
} else {
- hpvm_StartTimer(&timers->timers[category]);
+ visc_StartTimer(&timers->timers[category]);
}
} else {
if (subtimerlist != NULL) {
@@ -933,7 +933,7 @@ void hpvm_SwitchToSubTimer(struct hpvm_TimerSet *timers, char *label,
* so we can rename that marker as the beginning of this async
* operation */
- struct hpvm_async_time_marker_list *last_event = get_last_async(timers);
+ struct visc_async_time_marker_list *last_event = get_last_async(timers);
last_event->timerID = category;
last_event->label = label;
} // else, marker for switchToThis was already inserted
@@ -941,7 +941,7 @@ void hpvm_SwitchToSubTimer(struct hpvm_TimerSet *timers, char *label,
// toSwitchto is already asynchronous, but if current/prev state is async
// too, then DRIVER is already running
if (!is_async(timers->current)) {
- hpvm_StartTimer(&timers->timers[hpvm_TimerID_DRIVER]);
+ visc_StartTimer(&timers->timers[visc_TimerID_DRIVER]);
}
}
}
@@ -949,11 +949,11 @@ void hpvm_SwitchToSubTimer(struct hpvm_TimerSet *timers, char *label,
timers->current = category;
}
-void hpvm_PrintTimerSet(struct hpvm_TimerSet *timers) {
- hpvm_Timestamp wall_end = get_time();
+void visc_PrintTimerSet(struct visc_TimerSet *timers) {
+ visc_Timestamp wall_end = get_time();
- struct hpvm_Timer *t = timers->timers;
- struct hpvm_SubTimer *sub = NULL;
+ struct visc_Timer *t = timers->timers;
+ struct visc_SubTimer *sub = NULL;
int maxSubLength;
@@ -970,13 +970,13 @@ void hpvm_PrintTimerSet(struct hpvm_TimerSet *timers) {
const int maxCategoryLength = 20;
int i;
- for (i = 1; i < hpvm_TimerID_LAST;
+ for (i = 1; i < visc_TimerID_LAST;
++i) { // exclude NONE and OVRELAP from this format
- if (hpvm_GetElapsedTime(&t[i]) != 0 || true) {
+ if (visc_GetElapsedTime(&t[i]) != 0 || true) {
// Print Category Timer
printf("%-*s: %.9f\n", maxCategoryLength, categories[i - 1],
- hpvm_GetElapsedTime(&t[i]));
+ visc_GetElapsedTime(&t[i]));
if (timers->sub_timer_list[i] != NULL) {
sub = timers->sub_timer_list[i]->subtimer_list;
@@ -999,24 +999,24 @@ void hpvm_PrintTimerSet(struct hpvm_TimerSet *timers) {
// Print SubTimers
while (sub != NULL) {
printf(" -%-*s: %.9f\n", maxSubLength, sub->label,
- hpvm_GetElapsedTime(&sub->timer));
+ visc_GetElapsedTime(&sub->timer));
sub = sub->next;
}
}
}
}
- if (hpvm_GetElapsedTime(&t[hpvm_TimerID_OVERLAP]) != 0)
+ if (visc_GetElapsedTime(&t[visc_TimerID_OVERLAP]) != 0)
printf("CPU/Kernel Overlap: %.9f\n",
- hpvm_GetElapsedTime(&t[hpvm_TimerID_OVERLAP]));
+ visc_GetElapsedTime(&t[visc_TimerID_OVERLAP]));
float walltime = (wall_end - timers->wall_begin) / 1e9;
printf("Timer Wall Time: %.9f\n", walltime);
}
-void hpvm_DestroyTimerSet(struct hpvm_TimerSet *timers) {
+void visc_DestroyTimerSet(struct visc_TimerSet *timers) {
/* clean up all of the async event markers */
- struct hpvm_async_time_marker_list *event = timers->async_markers;
+ struct visc_async_time_marker_list *event = timers->async_markers;
while (event != NULL) {
cl_int ciErrNum = CL_SUCCESS;
@@ -1031,7 +1031,7 @@ void hpvm_DestroyTimerSet(struct hpvm_TimerSet *timers) {
}
free((event)->marker);
- struct hpvm_async_time_marker_list *next = ((event)->next);
+ struct visc_async_time_marker_list *next = ((event)->next);
free(event);
@@ -1040,10 +1040,10 @@ void hpvm_DestroyTimerSet(struct hpvm_TimerSet *timers) {
}
int i = 0;
- for (i = 0; i < hpvm_TimerID_LAST; ++i) {
+ for (i = 0; i < visc_TimerID_LAST; ++i) {
if (timers->sub_timer_list[i] != NULL) {
- struct hpvm_SubTimer *subtimer = timers->sub_timer_list[i]->subtimer_list;
- struct hpvm_SubTimer *prev = NULL;
+ struct visc_SubTimer *subtimer = timers->sub_timer_list[i]->subtimer_list;
+ struct visc_SubTimer *prev = NULL;
while (subtimer != NULL) {
free(subtimer->label);
prev = subtimer;
@@ -1059,7 +1059,7 @@ void hpvm_DestroyTimerSet(struct hpvm_TimerSet *timers) {
#define BUFFER_SIZE 1
// Launch API for a streaming dataflow graph
-void *llvm_hpvm_streamLaunch(void (*LaunchFunc)(void *, void *), void *args) {
+void *llvm_visc_streamLaunch(void (*LaunchFunc)(void *, void *), void *args) {
DFNodeContext_X86 *Context =
(DFNodeContext_X86 *)malloc(sizeof(DFNodeContext_X86));
@@ -1081,7 +1081,7 @@ void *llvm_hpvm_streamLaunch(void (*LaunchFunc)(void *, void *), void *args) {
}
// Push API for a streaming dataflow graph
-void llvm_hpvm_streamPush(void *graphID, void *args) {
+void llvm_visc_streamPush(void *graphID, void *args) {
DEBUG(cout << "StreamPush -- Graph: " << graphID << ", Arguments: " << args
<< flush << "\n");
DFNodeContext_X86 *Ctx = (DFNodeContext_X86 *)graphID;
@@ -1094,17 +1094,17 @@ void llvm_hpvm_streamPush(void *graphID, void *args) {
if (Ctx->BindInSourcePort->at(j) == i) {
// Push to all bind buffers connected to parent node at this port
// DEBUG(cout << "\tPushing Value " << element << " to buffer\n");
- llvm_hpvm_bufferPush(Ctx->BindInputBuffers->at(j), element);
+ llvm_visc_bufferPush(Ctx->BindInputBuffers->at(j), element);
}
}
}
// Push 0 in isLastInput buffers of all child nodes
for (CircularBuffer<uint64_t> *buffer : *(Ctx->isLastInputBuffers))
- llvm_hpvm_bufferPush(buffer, 0);
+ llvm_visc_bufferPush(buffer, 0);
}
// Pop API for a streaming dataflow graph
-void *llvm_hpvm_streamPop(void *graphID) {
+void *llvm_visc_streamPop(void *graphID) {
DEBUG(cout << "StreamPop -- Graph: " << graphID << flush << "\n");
DFNodeContext_X86 *Ctx = (DFNodeContext_X86 *)graphID;
unsigned totalBytes = 0;
@@ -1113,7 +1113,7 @@ void *llvm_hpvm_streamPop(void *graphID) {
void *output = malloc(totalBytes);
unsigned offset = 0;
for (unsigned i = 0; i < Ctx->BindOutputBuffers->size(); i++) {
- uint64_t element = llvm_hpvm_bufferPop(Ctx->BindOutputBuffers->at(i));
+ uint64_t element = llvm_visc_bufferPop(Ctx->BindOutputBuffers->at(i));
// DEBUG(cout << "\tPopped Value " << element << " from buffer\n");
memcpy((char *)output + offset, &element, Ctx->BindOutSizes->at(i));
offset += Ctx->BindOutSizes->at(i);
@@ -1122,24 +1122,24 @@ void *llvm_hpvm_streamPop(void *graphID) {
}
// Wait API for a streaming dataflow graph
-void llvm_hpvm_streamWait(void *graphID) {
+void llvm_visc_streamWait(void *graphID) {
DEBUG(cout << "StreamWait -- Graph: " << graphID << flush << "\n");
DFNodeContext_X86 *Ctx = (DFNodeContext_X86 *)graphID;
// Push garbage to all other input buffers
for (unsigned i = 0; i < Ctx->BindInputBuffers->size(); i++) {
uint64_t element = 0;
// DEBUG(cout << "\tPushing Value " << element << " to buffer\n");
- llvm_hpvm_bufferPush(Ctx->BindInputBuffers->at(i), element);
+ llvm_visc_bufferPush(Ctx->BindInputBuffers->at(i), element);
}
// Push 1 in isLastInput buffers of all child nodes
for (unsigned i = 0; i < Ctx->isLastInputBuffers->size(); i++)
- llvm_hpvm_bufferPush(Ctx->isLastInputBuffers->at(i), 1);
+ llvm_visc_bufferPush(Ctx->isLastInputBuffers->at(i), 1);
- llvm_hpvm_freeThreads(graphID);
+ llvm_visc_freeThreads(graphID);
}
// Create a buffer and return the bufferID
-void *llvm_hpvm_createBindInBuffer(void *graphID, uint64_t size,
+void *llvm_visc_createBindInBuffer(void *graphID, uint64_t size,
unsigned inArgPort) {
DEBUG(cout << "Create BindInBuffer -- Graph: " << graphID
<< ", Size: " << size << flush << "\n");
@@ -1154,7 +1154,7 @@ void *llvm_hpvm_createBindInBuffer(void *graphID, uint64_t size,
return bufferID;
}
-void *llvm_hpvm_createBindOutBuffer(void *graphID, uint64_t size) {
+void *llvm_visc_createBindOutBuffer(void *graphID, uint64_t size) {
DEBUG(cout << "Create BindOutBuffer -- Graph: " << graphID
<< ", Size: " << size << flush << "\n");
DFNodeContext_X86 *Context = (DFNodeContext_X86 *)graphID;
@@ -1166,7 +1166,7 @@ void *llvm_hpvm_createBindOutBuffer(void *graphID, uint64_t size) {
Context->BindOutSizes->push_back(size);
return bufferID;
}
-void *llvm_hpvm_createEdgeBuffer(void *graphID, uint64_t size) {
+void *llvm_visc_createEdgeBuffer(void *graphID, uint64_t size) {
DEBUG(cout << "Create EdgeBuffer -- Graph: " << graphID << ", Size: " << size
<< flush << "\n");
DFNodeContext_X86 *Context = (DFNodeContext_X86 *)graphID;
@@ -1179,7 +1179,7 @@ void *llvm_hpvm_createEdgeBuffer(void *graphID, uint64_t size) {
return bufferID;
}
-void *llvm_hpvm_createLastInputBuffer(void *graphID, uint64_t size) {
+void *llvm_visc_createLastInputBuffer(void *graphID, uint64_t size) {
DEBUG(cout << "Create isLastInputBuffer -- Graph: " << graphID
<< ", Size: " << size << flush << "\n");
DFNodeContext_X86 *Context = (DFNodeContext_X86 *)graphID;
@@ -1192,7 +1192,7 @@ void *llvm_hpvm_createLastInputBuffer(void *graphID, uint64_t size) {
}
// Free buffers
-void llvm_hpvm_freeBuffers(void *graphID) {
+void llvm_visc_freeBuffers(void *graphID) {
DEBUG(cout << "Free all buffers -- Graph: " << graphID << flush << "\n");
DFNodeContext_X86 *Context = (DFNodeContext_X86 *)graphID;
for (CircularBuffer<uint64_t> *bufferID : *(Context->BindInputBuffers))
@@ -1206,19 +1206,19 @@ void llvm_hpvm_freeBuffers(void *graphID) {
}
// Pop an element from the buffer
-uint64_t llvm_hpvm_bufferPop(void *bufferID) {
+uint64_t llvm_visc_bufferPop(void *bufferID) {
CircularBuffer<uint64_t> *buffer = (CircularBuffer<uint64_t> *)bufferID;
return buffer->pop();
}
// Push an element into the buffer
-void llvm_hpvm_bufferPush(void *bufferID, uint64_t element) {
+void llvm_visc_bufferPush(void *bufferID, uint64_t element) {
CircularBuffer<uint64_t> *buffer = (CircularBuffer<uint64_t> *)bufferID;
buffer->push(element);
}
// Create a thread
-void llvm_hpvm_createThread(void *graphID, void *(*Func)(void *),
+void llvm_visc_createThread(void *graphID, void *(*Func)(void *),
void *arguments) {
DEBUG(cout << "Create Thread -- Graph: " << graphID << ", Func: " << Func
<< ", Args: " << arguments << flush << "\n");
@@ -1232,7 +1232,7 @@ void llvm_hpvm_createThread(void *graphID, void *(*Func)(void *),
}
// Wait for thread to finish
-void llvm_hpvm_freeThreads(void *graphID) {
+void llvm_visc_freeThreads(void *graphID) {
DEBUG(cout << "Free Threads -- Graph: " << graphID << flush << "\n");
DFNodeContext_X86 *Ctx = (DFNodeContext_X86 *)graphID;
for (pthread_t thread : *(Ctx->threads))
@@ -1241,7 +1241,7 @@ void llvm_hpvm_freeThreads(void *graphID) {
/************************ OPENCL & PTHREAD API ********************************/
-void *llvm_hpvm_x86_launch(void *(*rootFunc)(void *), void *arguments) {
+void *llvm_visc_x86_launch(void *(*rootFunc)(void *), void *arguments) {
DFNodeContext_X86 *Context =
(DFNodeContext_X86 *)malloc(sizeof(DFNodeContext_X86));
// int err;
@@ -1252,7 +1252,7 @@ void *llvm_hpvm_x86_launch(void *(*rootFunc)(void *), void *arguments) {
return Context;
}
-void llvm_hpvm_x86_wait(void *graphID) {
+void llvm_visc_x86_wait(void *graphID) {
DEBUG(cout << "Waiting for pthread to finish ...\n");
// DFNodeContext_X86* Context = (DFNodeContext_X86*) graphID;
// pthread_join(Context->threadID, NULL);
@@ -1260,9 +1260,9 @@ void llvm_hpvm_x86_wait(void *graphID) {
DEBUG(cout << "\t... pthread Done!\n");
}
-void *llvm_hpvm_ocl_initContext(enum hpvm::Target T) {
+void *llvm_visc_ocl_initContext(enum visc::Target T) {
pthread_mutex_lock(&ocl_mtx);
- DEBUG(std::string Target = T == hpvm::GPU_TARGET ? "GPU" : "SPIR");
+ DEBUG(std::string Target = T == visc::GPU_TARGET ? "GPU" : "SPIR");
DEBUG(cout << "Initializing Context for " << Target << " device\n");
cl_uint numPlatforms;
cl_int errcode;
@@ -1299,10 +1299,10 @@ void *llvm_hpvm_ocl_initContext(enum hpvm::Target T) {
// assert(numPlatforms >= 2 && "Expecting two OpenCL platforms");
// Choose second one which is X86 AVX
cl_context_properties properties[] = {
- CL_CONTEXT_PLATFORM, (long)platforms[T == hpvm::GPU_TARGET ? 0 : 1], 0};
+ CL_CONTEXT_PLATFORM, (long)platforms[T == visc::GPU_TARGET ? 0 : 1], 0};
globalOCLContext = clCreateContextFromType(
properties,
- T == hpvm::GPU_TARGET ? CL_DEVICE_TYPE_GPU : CL_DEVICE_TYPE_CPU, NULL,
+ T == visc::GPU_TARGET ? CL_DEVICE_TYPE_GPU : CL_DEVICE_TYPE_CPU, NULL,
NULL, &errcode);
// get the list of OCL devices associated with context
size_t dataBytes;
@@ -1314,7 +1314,7 @@ void *llvm_hpvm_ocl_initContext(enum hpvm::Target T) {
errcode |= clGetContextInfo(globalOCLContext, CL_CONTEXT_DEVICES, dataBytes,
clDevices, NULL);
checkErr(errcode, CL_SUCCESS, "Failure to get context info");
- if (false && T == hpvm::SPIR_TARGET) {
+ if (false && T == visc::SPIR_TARGET) {
cl_device_partition_property props[4];
props[0] = CL_DEVICE_PARTITION_BY_COUNTS;
props[1] = NUM_CORES;
@@ -1340,13 +1340,13 @@ void *llvm_hpvm_ocl_initContext(enum hpvm::Target T) {
checkErr(errcode, CL_SUCCESS, "Failure to create OCL context");
DEBUG(cout << "Initialize Kernel Timer\n");
- hpvm_InitializeTimerSet(&kernel_timer);
+ visc_InitializeTimerSet(&kernel_timer);
pthread_mutex_unlock(&ocl_mtx);
return globalOCLContext;
}
-void llvm_hpvm_ocl_clearContext(void *graphID) {
+void llvm_visc_ocl_clearContext(void *graphID) {
pthread_mutex_lock(&ocl_mtx);
DEBUG(cout << "Clear Context\n");
DFNodeContext_OCL *Context = (DFNodeContext_OCL *)graphID;
@@ -1359,12 +1359,12 @@ void llvm_hpvm_ocl_clearContext(void *graphID) {
// DEBUG(cout << "Released context at: " << globalOCLContext);
free(Context);
DEBUG(cout << "Done with OCL kernel\n");
- cout << "Printing HPVM Timer: KernelTimer\n";
- hpvm_PrintTimerSet(&kernel_timer);
+ cout << "Printing VISC Timer: KernelTimer\n";
+ visc_PrintTimerSet(&kernel_timer);
pthread_mutex_unlock(&ocl_mtx);
}
-void llvm_hpvm_ocl_argument_shared(void *graphID, int arg_index, size_t size) {
+void llvm_visc_ocl_argument_shared(void *graphID, int arg_index, size_t size) {
pthread_mutex_lock(&ocl_mtx);
DEBUG(cout << "Set Shared Memory Input:");
DEBUG(cout << "\tArgument Index = " << arg_index << ", Size = " << size
@@ -1379,7 +1379,7 @@ void llvm_hpvm_ocl_argument_shared(void *graphID, int arg_index, size_t size) {
pthread_mutex_unlock(&ocl_mtx);
}
-void llvm_hpvm_ocl_argument_scalar(void *graphID, void *input, int arg_index,
+void llvm_visc_ocl_argument_scalar(void *graphID, void *input, int arg_index,
size_t size) {
pthread_mutex_lock(&ocl_mtx);
DEBUG(cout << "Set Scalar Input:");
@@ -1395,7 +1395,7 @@ void llvm_hpvm_ocl_argument_scalar(void *graphID, void *input, int arg_index,
pthread_mutex_unlock(&ocl_mtx);
}
-void *llvm_hpvm_ocl_argument_ptr(void *graphID, void *input, int arg_index,
+void *llvm_visc_ocl_argument_ptr(void *graphID, void *input, int arg_index,
size_t size, bool isInput, bool isOutput) {
pthread_mutex_lock(&ocl_mtx);
DEBUG(cout << "Set Pointer Input:");
@@ -1409,7 +1409,7 @@ void *llvm_hpvm_ocl_argument_ptr(void *graphID, void *input, int arg_index,
pthread_mutex_unlock(&ocl_mtx);
// Check with runtime the location of this memory
- cl_mem d_input = (cl_mem)llvm_hpvm_ocl_request_mem(input, size, Context,
+ cl_mem d_input = (cl_mem)llvm_visc_ocl_request_mem(input, size, Context,
isInput, isOutput);
pthread_mutex_lock(&ocl_mtx);
@@ -1424,7 +1424,7 @@ void *llvm_hpvm_ocl_argument_ptr(void *graphID, void *input, int arg_index,
return d_input;
}
-void *llvm_hpvm_ocl_output_ptr(void *graphID, int arg_index, size_t size) {
+void *llvm_visc_ocl_output_ptr(void *graphID, int arg_index, size_t size) {
pthread_mutex_lock(&ocl_mtx);
DEBUG(cout << "Set device memory for Output Struct:");
DEBUG(cout << "\tArgument Index = " << arg_index << ", Size = " << size
@@ -1446,13 +1446,13 @@ void *llvm_hpvm_ocl_output_ptr(void *graphID, int arg_index, size_t size) {
return d_output;
}
-void llvm_hpvm_ocl_free(void *ptr) {
+void llvm_visc_ocl_free(void *ptr) {
// DEBUG(cout << "Release Device Pointer: " << ptr << flush << "\n");
// cl_mem d_ptr = (cl_mem) ptr;
// clReleaseMemObject(d_ptr);
}
-void *llvm_hpvm_ocl_getOutput(void *graphID, void *h_output, void *d_output,
+void *llvm_visc_ocl_getOutput(void *graphID, void *h_output, void *d_output,
size_t size) {
pthread_mutex_lock(&ocl_mtx);
DEBUG(cout << "Get Output:\n");
@@ -1471,7 +1471,7 @@ void *llvm_hpvm_ocl_getOutput(void *graphID, void *h_output, void *d_output,
return h_output;
}
-void *llvm_hpvm_ocl_executeNode(void *graphID, unsigned workDim,
+void *llvm_visc_ocl_executeNode(void *graphID, unsigned workDim,
const size_t *localWorkSize,
const size_t *globalWorkSize) {
pthread_mutex_lock(&ocl_mtx);
@@ -1517,7 +1517,7 @@ void *llvm_hpvm_ocl_executeNode(void *graphID, unsigned workDim,
// pthread_mutex_lock(&ocl_mtx);
clFinish(Context->clCommandQue);
// pthread_mutex_unlock(&ocl_mtx);
- hpvm_SwitchToTimer(&kernel_timer, hpvm_TimerID_COMPUTATION);
+ visc_SwitchToTimer(&kernel_timer, visc_TimerID_COMPUTATION);
// for(int i=0 ;i < NUM_TESTS; i++) {
// cout << "Iteration = " << i << flush << "\n";
// pthread_mutex_lock(&ocl_mtx);
@@ -1530,7 +1530,7 @@ void *llvm_hpvm_ocl_executeNode(void *graphID, unsigned workDim,
// pthread_mutex_lock(&ocl_mtx);
clFinish(Context->clCommandQue);
// pthread_mutex_unlock(&ocl_mtx);
- hpvm_SwitchToTimer(&kernel_timer, hpvm_TimerID_NONE);
+ visc_SwitchToTimer(&kernel_timer, visc_TimerID_NONE);
pthread_mutex_unlock(&ocl_mtx);
return event;
@@ -1579,7 +1579,7 @@ static char *LoadProgSource(const char *Filename, size_t *szFinalLength) {
return cSourceString;
}
-void *llvm_hpvm_ocl_launch(const char *FileName, const char *KernelName) {
+void *llvm_visc_ocl_launch(const char *FileName, const char *KernelName) {
pthread_mutex_lock(&ocl_mtx);
DEBUG(cout << "Launch OCL Kernel\n");
// Initialize OpenCL
@@ -1649,7 +1649,7 @@ void *llvm_hpvm_ocl_launch(const char *FileName, const char *KernelName) {
return Context;
}
-void llvm_hpvm_ocl_wait(void *graphID) {
+void llvm_visc_ocl_wait(void *graphID) {
pthread_mutex_lock(&ocl_mtx);
DEBUG(cout << "Wait\n");
DFNodeContext_OCL *Context = (DFNodeContext_OCL *)graphID;
@@ -1659,27 +1659,27 @@ void llvm_hpvm_ocl_wait(void *graphID) {
pthread_mutex_unlock(&ocl_mtx);
}
-void llvm_hpvm_switchToTimer(void **timerSet, enum hpvm_TimerID timer) {
+void llvm_visc_switchToTimer(void **timerSet, enum visc_TimerID timer) {
// cout << "Switching to timer " << timer << flush << "\n";
pthread_mutex_lock(&ocl_mtx);
- // hpvm_SwitchToTimer((hpvm_TimerSet*)(*timerSet), timer);
+ // visc_SwitchToTimer((visc_TimerSet*)(*timerSet), timer);
pthread_mutex_unlock(&ocl_mtx);
}
-void llvm_hpvm_printTimerSet(void **timerSet, char *timerName) {
+void llvm_visc_printTimerSet(void **timerSet, char *timerName) {
pthread_mutex_lock(&ocl_mtx);
- cout << "Printing HPVM Timer: ";
+ cout << "Printing VISC Timer: ";
if (timerName != NULL)
cout << timerName << flush << "\n";
else
cout << "Anonymous\n";
- hpvm_PrintTimerSet((hpvm_TimerSet *)(*timerSet));
+ visc_PrintTimerSet((visc_TimerSet *)(*timerSet));
pthread_mutex_unlock(&ocl_mtx);
}
-void *llvm_hpvm_initializeTimerSet() {
+void *llvm_visc_initializeTimerSet() {
pthread_mutex_lock(&ocl_mtx);
- hpvm_TimerSet *TS = (hpvm_TimerSet *)malloc(sizeof(hpvm_TimerSet));
- hpvm_InitializeTimerSet(TS);
+ visc_TimerSet *TS = (visc_TimerSet *)malloc(sizeof(visc_TimerSet));
+ visc_InitializeTimerSet(TS);
pthread_mutex_unlock(&ocl_mtx);
return TS;
}
diff --git a/hpvm/projects/hpvm-rt/hpvm-rt.h b/hpvm/projects/visc-rt/visc-rt.h
similarity index 72%
rename from hpvm/projects/hpvm-rt/hpvm-rt.h
rename to hpvm/projects/visc-rt/visc-rt.h
index 2b6dafba96..3ad315768b 100644
--- a/hpvm/projects/hpvm-rt/hpvm-rt.h
+++ b/hpvm/projects/visc-rt/visc-rt.h
@@ -2,8 +2,8 @@
*
* (c) 2010 The Board of Trustees of the University of Illinois.
*/
-#ifndef HPVM_RT_HEADER
-#define HPVM_RT_HEADER
+#ifndef VISC_RT_HEADER
+#define VISC_RT_HEADER
#include <ctime>
#include <iostream>
@@ -13,8 +13,8 @@
#include <vector>
//#include <condition_variable>
-#include "../../include/SupportHPVM/HPVMHint.h"
-#include "../../include/SupportHPVM/HPVMTimer.h"
+#include "../../include/SupportVISC/VISCHint.h"
+#include "../../include/SupportVISC/VISCTimer.h"
#include "device_abstraction.h"
#include "policy.h"
@@ -31,14 +31,14 @@ extern "C" {
/************************* Policies *************************************/
-void llvm_hpvm_policy_init();
-void llvm_hpvm_policy_clear();
-int llvm_hpvm_policy_getVersion(const char *, int64_t);
+void llvm_visc_policy_init();
+void llvm_visc_policy_clear();
+int llvm_visc_policy_getVersion(const char *, int64_t);
/******************** Device Abstraction ********************************/
-void llvm_hpvm_deviceAbstraction_start();
-void llvm_hpvm_deviceAbstraction_end();
-void llvm_hpvm_deviceAbstraction_waitOnDeviceStatus();
+void llvm_visc_deviceAbstraction_start();
+void llvm_visc_deviceAbstraction_end();
+void llvm_visc_deviceAbstraction_waitOnDeviceStatus();
/********************* DFG Depth Stack **********************************/
class DFGDepth {
@@ -77,12 +77,12 @@ public:
unsigned getNumDim() const { return numDim; }
};
-void llvm_hpvm_x86_dstack_push(unsigned n, uint64_t limitX = 0, uint64_t iX = 0,
+void llvm_visc_x86_dstack_push(unsigned n, uint64_t limitX = 0, uint64_t iX = 0,
uint64_t limitY = 0, uint64_t iY = 0,
uint64_t limitZ = 0, uint64_t iZ = 0);
-void llvm_hpvm_x86_dstack_pop();
-uint64_t llvm_hpvm_x86_getDimLimit(unsigned level, unsigned dim);
-uint64_t llvm_hpvm_x86_getDimInstance(unsigned level, unsigned dim);
+void llvm_visc_x86_dstack_pop();
+uint64_t llvm_visc_x86_getDimLimit(unsigned level, unsigned dim);
+uint64_t llvm_visc_x86_getDimInstance(unsigned level, unsigned dim);
/********************* Memory Tracker **********************************/
class MemTrackerEntry {
@@ -156,32 +156,32 @@ public:
}
};
-void llvm_hpvm_track_mem(void *, size_t);
-void llvm_hpvm_untrack_mem(void *);
-void *llvm_hpvm_request_mem(void *, size_t);
+void llvm_visc_track_mem(void *, size_t);
+void llvm_visc_untrack_mem(void *);
+void *llvm_visc_request_mem(void *, size_t);
/*********************** OPENCL & PTHREAD API **************************/
-void *llvm_hpvm_x86_launch(void *(void *), void *);
-void llvm_hpvm_x86_wait(void *);
-void *llvm_hpvm_ocl_initContext(enum hpvm::Target);
-
-void *llvm_hpvm_x86_argument_ptr(void *, size_t);
-
-void llvm_hpvm_ocl_clearContext(void *);
-void llvm_hpvm_ocl_argument_shared(void *, int, size_t);
-void llvm_hpvm_ocl_argument_scalar(void *, void *, int, size_t);
-void *llvm_hpvm_ocl_argument_ptr(void *, void *, int, size_t, bool, bool);
-void *llvm_hpvm_ocl_output_ptr(void *, int, size_t);
-void llvm_hpvm_ocl_free(void *);
-void *llvm_hpvm_ocl_getOutput(void *, void *, void *, size_t);
-void *llvm_hpvm_ocl_executeNode(void *, unsigned, const size_t *,
+void *llvm_visc_x86_launch(void *(void *), void *);
+void llvm_visc_x86_wait(void *);
+void *llvm_visc_ocl_initContext(enum visc::Target);
+
+void *llvm_visc_x86_argument_ptr(void *, size_t);
+
+void llvm_visc_ocl_clearContext(void *);
+void llvm_visc_ocl_argument_shared(void *, int, size_t);
+void llvm_visc_ocl_argument_scalar(void *, void *, int, size_t);
+void *llvm_visc_ocl_argument_ptr(void *, void *, int, size_t, bool, bool);
+void *llvm_visc_ocl_output_ptr(void *, int, size_t);
+void llvm_visc_ocl_free(void *);
+void *llvm_visc_ocl_getOutput(void *, void *, void *, size_t);
+void *llvm_visc_ocl_executeNode(void *, unsigned, const size_t *,
const size_t *);
-void *llvm_hpvm_ocl_launch(const char *, const char *);
-void llvm_hpvm_ocl_wait(void *);
+void *llvm_visc_ocl_launch(const char *, const char *);
+void llvm_visc_ocl_wait(void *);
-void llvm_hpvm_switchToTimer(void **timerSet, enum hpvm_TimerID);
-void llvm_hpvm_printTimerSet(void **timerSet, char *timerName = NULL);
-void *llvm_hpvm_initializeTimerSet();
+void llvm_visc_switchToTimer(void **timerSet, enum visc_TimerID);
+void llvm_visc_printTimerSet(void **timerSet, char *timerName = NULL);
+void *llvm_visc_initializeTimerSet();
}
/*************************** Pipeline API ******************************/
@@ -262,30 +262,30 @@ template <class ElementType> ElementType CircularBuffer<ElementType>::pop() {
extern "C" {
// Functions to push and pop values from pipeline buffers
-uint64_t llvm_hpvm_bufferPop(void *);
-void llvm_hpvm_bufferPush(void *, uint64_t);
+uint64_t llvm_visc_bufferPop(void *);
+void llvm_visc_bufferPush(void *, uint64_t);
// Functions to create and destroy buffers
-void *llvm_hpvm_createBindInBuffer(void *, uint64_t, unsigned);
-void *llvm_hpvm_createBindOutBuffer(void *, uint64_t);
-void *llvm_hpvm_createEdgeBuffer(void *, uint64_t);
-void *llvm_hpvm_createLastInputBuffer(void *, uint64_t);
+void *llvm_visc_createBindInBuffer(void *, uint64_t, unsigned);
+void *llvm_visc_createBindOutBuffer(void *, uint64_t);
+void *llvm_visc_createEdgeBuffer(void *, uint64_t);
+void *llvm_visc_createLastInputBuffer(void *, uint64_t);
-void llvm_hpvm_freeBuffers(void *);
+void llvm_visc_freeBuffers(void *);
// Functions to create and destroy threads
-void llvm_hpvm_createThread(void *graphID, void *(*Func)(void *), void *);
-void llvm_hpvm_freeThreads(void *);
+void llvm_visc_createThread(void *graphID, void *(*Func)(void *), void *);
+void llvm_visc_freeThreads(void *);
// Launch API for a streaming graph.
// Arguments:
// (1) Launch Function: void* (void*, void*)
// (2) Push Function: void (void*, std::vector<uint64_t>**, unsgined)
// (3) Pop Function: void* (std::vector<uint64_t>**, unsigned)
-void *llvm_hpvm_streamLaunch(void (*LaunchFunc)(void *, void *), void *);
-void llvm_hpvm_streamPush(void *graphID, void *args);
-void *llvm_hpvm_streamPop(void *graphID);
-void llvm_hpvm_streamWait(void *graphID);
+void *llvm_visc_streamLaunch(void (*LaunchFunc)(void *, void *), void *);
+void llvm_visc_streamPush(void *graphID, void *args);
+void *llvm_visc_streamPop(void *graphID);
+void llvm_visc_streamWait(void *graphID);
}
-#endif // HPVM_RT_HEADER
+#endif // VISC_RT_HEADER
diff --git a/hpvm/test/CTestSuite/Makefile b/hpvm/test/CTestSuite/Makefile
index 1169e4e896..226a83287d 100644
--- a/hpvm/test/CTestSuite/Makefile
+++ b/hpvm/test/CTestSuite/Makefile
@@ -9,7 +9,7 @@ LLVM_CC:=$(LLVM_INSTALL)/bin/clang
LLVM_OPT:=$(LLVM_INSTALL)/bin/opt
BUILD_DIR:=build
-all: $(BUILD_DIR) $(HOST:%=$(BUILD_DIR)/%.ll) $(HOST:%=$(BUILD_DIR)/%.hpvm.ll)
+all: $(BUILD_DIR) $(HOST:%=$(BUILD_DIR)/%.ll) $(HOST:%=$(BUILD_DIR)/%.visc.ll)
$(BUILD_DIR):
mkdir -p $(BUILD_DIR)
@@ -17,10 +17,10 @@ $(BUILD_DIR):
$(HOST:%=$(BUILD_DIR)/%.ll):$(BUILD_DIR)/%.ll:%.c
$(LLVM_CC) -S -emit-llvm $< -O3 -o $@
-$(HOST:%=$(BUILD_DIR)/%.hpvm.ll):$(BUILD_DIR)/%.hpvm.ll:$(BUILD_DIR)/%.ll
- $(LLVM_OPT) -load $(LLVM_SRC_ROOT)/Release+Asserts/lib/LLVMGenHPVM.so -genhpvm -globaldce $< -S -o $@
+$(HOST:%=$(BUILD_DIR)/%.visc.ll):$(BUILD_DIR)/%.visc.ll:$(BUILD_DIR)/%.ll
+ $(LLVM_OPT) -load $(LLVM_SRC_ROOT)/Release+Asserts/lib/LLVMGenVISC.so -genvisc -globaldce $< -S -o $@
@cat RUN.script $@ > $@.tmp
@mv $@.tmp $@
clean :
- rm -f $(HOST:%=$(BUILD_DIR)/%.ll) $(HOST:%=$(BUILD_DIR)/%.hpvm.ll) $(HOST:%=$(BUILD_DIR)/%.hpvm.ll.kernels.ll) $(HOST:%=$(BUILD_DIR)/%.hpvm.ll.nvptx.s) $(BUILD_DIR)/DataflowGraph.dot*
+ rm -f $(HOST:%=$(BUILD_DIR)/%.ll) $(HOST:%=$(BUILD_DIR)/%.visc.ll) $(HOST:%=$(BUILD_DIR)/%.visc.ll.kernels.ll) $(HOST:%=$(BUILD_DIR)/%.visc.ll.nvptx.s) $(BUILD_DIR)/DataflowGraph.dot*
diff --git a/hpvm/test/CTestSuite/RUN.script b/hpvm/test/CTestSuite/RUN.script
index 23fa1694eb..10bf667818 100644
--- a/hpvm/test/CTestSuite/RUN.script
+++ b/hpvm/test/CTestSuite/RUN.script
@@ -1,6 +1,6 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG -o %t.ll -S %s
; RUN: llvm-link %llvm_src/../libclc/built_libs/nvptx--nvidiacl.bc %s.kernels.ll -o %t.ll.kernels.linked.bc
; RUN: clang -O3 -target nvptx %t.ll.kernels.linked.bc -S -o %s.nvptx.s
-; RUN: llvm-link %t.ll %llvm_src/projects/hpvm-rt/hpvm-rt.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll %llvm_src/projects/visc-rt/visc-rt.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -o %t.bin
; RUN: %t.bin
diff --git a/hpvm/test/CTestSuite/gemm.c b/hpvm/test/CTestSuite/gemm.c
index eb0a3c5e92..d0a69ba25c 100644
--- a/hpvm/test/CTestSuite/gemm.c
+++ b/hpvm/test/CTestSuite/gemm.c
@@ -54,14 +54,14 @@ __attribute__((noinline)) int checkResults(float *A, float *B, float *C) {
return 1; // Success
}
-// Dummy hpvm node execution call
-// void __hpvm__node(void kernel (float*, float*, float*, unsigned, unsigned),
+// Dummy visc node execution call
+// void __visc__node(void kernel (float*, float*, float*, unsigned, unsigned),
// int numDims, void* dims, int numInputs, void* inputs, int numOutputs, void*
// outputs);
void matrixMul(float *A, float *B, float *C, unsigned k, unsigned n) {
- __hpvm__attributes(2, A, B, 1, C);
+ __visc__attributes(2, A, B, 1, C);
// printf("Entered function\n");
int tx = get_local_id(0); // 2D Global Thread ID x
int ty = get_local_id(1); // 2D Global Thread ID y
@@ -130,10 +130,10 @@ int main(int argc, char **argv) {
// Compute using OpenCL
// matrixMul(h_A, h_B, h_C, WA, WB);
- //__hpvm__node(matrixMul, 2, WB, HA, 3, h_A, h_B, h_C, 0);
- unsigned graphMM = __hpvm__node(matrixMul, 1, 2, WB, HA, 8, h_A, bytes_A, h_B,
+ //__visc__node(matrixMul, 2, WB, HA, 3, h_A, h_B, h_C, 0);
+ unsigned graphMM = __visc__node(matrixMul, 1, 2, WB, HA, 8, h_A, bytes_A, h_B,
bytes_B, h_C, bytes_C, WA, WB, 0);
- __hpvm__wait(graphMM);
+ __visc__wait(graphMM);
if (checkResults(h_A, h_B, h_C))
printf("\nPass!\n");
else
diff --git a/hpvm/test/CTestSuite/gemm_2.c b/hpvm/test/CTestSuite/gemm_2.c
index df45559363..bd7ab27fc0 100644
--- a/hpvm/test/CTestSuite/gemm_2.c
+++ b/hpvm/test/CTestSuite/gemm_2.c
@@ -54,13 +54,13 @@ __attribute__((noinline)) int checkResults(float *A, float *B, float *C) {
return 1; // Success
}
-// Dummy hpvm node execution call
-// void __hpvm__node(void kernel (float*, float*, float*, unsigned, unsigned),
+// Dummy visc node execution call
+// void __visc__node(void kernel (float*, float*, float*, unsigned, unsigned),
// int numDims, void* dims, int numInputs, void* inputs, int numOutputs, void*
// outputs);
void matrixMul(float *A, float *B, float *C, unsigned k, unsigned n) {
- __hpvm__attributes(2, A, B, 1, C);
+ __visc__attributes(2, A, B, 1, C);
// printf("Entered function\n");
int tx = get_global_id(0); // 2D Global Thread ID x
@@ -130,11 +130,11 @@ int main(int argc, char **argv) {
// Compute using OpenCL
// matrixMul(h_A, h_B, h_C, WA, WB);
- //__hpvm__node(matrixMul, 2, WB, HA, 3, h_A, h_B, h_C, 0);
+ //__visc__node(matrixMul, 2, WB, HA, 3, h_A, h_B, h_C, 0);
unsigned graphMM =
- __hpvm__node(matrixMul, 2, 2, 16, 16, WB / 16, HA / 16, 8, h_A, bytes_A,
+ __visc__node(matrixMul, 2, 2, 16, 16, WB / 16, HA / 16, 8, h_A, bytes_A,
h_B, bytes_B, h_C, bytes_C, WA, WB, 0);
- __hpvm__wait(graphMM);
+ __visc__wait(graphMM);
if (checkResults(h_A, h_B, h_C))
printf("\nPass!\n");
else
diff --git a/hpvm/test/hpvm-cava/.gitignore b/hpvm/test/hpvm-cava/.gitignore
index f08b880bf9..2fc1b23564 100644
--- a/hpvm/test/hpvm-cava/.gitignore
+++ b/hpvm/test/hpvm-cava/.gitignore
@@ -1,5 +1,5 @@
build/
-cava-hpvm
+cava-visc
Makefile.config
example-face/*.bin
diff --git a/hpvm/test/hpvm-cava/Makefile b/hpvm/test/hpvm-cava/Makefile
index 7530477f3d..62219a1cb0 100644
--- a/hpvm/test/hpvm-cava/Makefile
+++ b/hpvm/test/hpvm-cava/Makefile
@@ -26,21 +26,21 @@ CURRENT_DIR := $(dir $(abspath $(lastword $(MAKEFILE_LIST))))
INCLUDES += -I$(SRC_DIR) -I$(CAM_PIPE_SRC_DIR)
-INCLUDES += -I$(LLVM_SRC_ROOT)/include -I../include -I$(HPVM_BUILD_DIR)/include
+INCLUDES += -I$(LLVM_SRC_ROOT)/include -I../include -I$(VISC_BUILD_DIR)/include
ifneq ($(CONFUSE_ROOT),)
INCLUDES += -I$(CONFUSE_ROOT)/include
LFLAGS += -L$(CONFUSE_ROOT)/lib
endif
-EXE = cava-hpvm-$(VERSION)-$(TARGET)
+EXE = cava-visc-$(VERSION)-$(TARGET)
LFLAGS += -pthread
## BEGIN HPVM MAKEFILE
-LANGUAGE=hpvm
+LANGUAGE=visc
SRCDIR_OBJS= load_cam_model.ll cam_pipe_utility.ll dma_interface.ll utility.ll
OBJS_SRC=src/cam_pipe.c src/pipe_stages.c src/load_cam_model.c src/cam_pipe_utility.c src/dma_interface.c src/utility.c
-HPVM_OBJS=main.hpvm.ll
+VISC_OBJS=main.visc.ll
APP = $(EXE)
APP_CUDALDFLAGS=-lm -lstdc++
APP_CFLAGS= $(INCLUDES) -DDMA_MODE -DDMA_INTERFACE_V3
@@ -52,23 +52,23 @@ OBJS_CFLAGS = -O1 $(APP_CFLAGS) $(PLATFORM_CFLAGS)
CXXFLAGS = $(APP_CXXFLAGS) $(PLATFORM_CXXFLAGS)
LDFLAGS= $(APP_LDFLAGS) $(PLATFORM_LDFLAGS)
-HPVM_RT_PATH = $(LLVM_SRC_ROOT)/tools/hpvm/projects/hpvm-rt
+VISC_RT_PATH = $(LLVM_SRC_ROOT)/tools/hpvm/projects/visc-rt
-HPVM_RT_LIB = $(HPVM_RT_PATH)/hpvm-rt.ll
+VISC_RT_LIB = $(VISC_RT_PATH)/visc-rt.ll
-TESTGEN_OPTFLAGS = -load LLVMGenHPVM.so -genhpvm -globaldce
+TESTGEN_OPTFLAGS = -load LLVMGenVISC.so -genvisc -globaldce
ifeq ($(TARGET),seq)
DEVICE = CPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG
- HPVM_OPTFLAGS += -hpvm-timers-x86
+ VISC_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG
+ VISC_OPTFLAGS += -visc-timers-x86
else
DEVICE = GPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG
- HPVM_OPTFLAGS += -hpvm-timers-x86 -hpvm-timers-ptx
+ VISC_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG
+ VISC_OPTFLAGS += -visc-timers-x86 -visc-timers-ptx
endif
- TESTGEN_OPTFLAGS += -hpvm-timers-gen
+ TESTGEN_OPTFLAGS += -visc-timers-gen
CFLAGS += -DDEVICE=$(DEVICE)
CXXFLAGS += -DDEVICE=$(DEVICE)
@@ -79,7 +79,7 @@ INBUILDDIR=$(addprefix $(BUILD_DIR)/,$(1))
.PRECIOUS: $(BUILD_DIR)/%.ll
OBJS = $(call INBUILDDIR,$(SRCDIR_OBJS))
-TEST_OBJS = $(call INBUILDDIR,$(HPVM_OBJS))
+TEST_OBJS = $(call INBUILDDIR,$(VISC_OBJS))
KERNEL = $(TEST_OBJS).kernels.ll
ifeq ($(TARGET),seq)
@@ -107,14 +107,14 @@ $(KERNEL_OCL) : $(KERNEL)
$(EXE) : $(HOST_LINKED)
$(CXX) -O3 $(LDFLAGS) $< -o $@
-$(HOST_LINKED) : $(HOST) $(OBJS) $(HPVM_RT_LIB)
+$(HOST_LINKED) : $(HOST) $(OBJS) $(VISC_RT_LIB)
$(LLVM_LINK) $^ -S -o $@
-$(HPVM_RT_LIB) : $(HPVM_RT_PATH)/hpvm-rt.cpp
+$(VISC_RT_LIB) : $(VISC_RT_PATH)/visc-rt.cpp
make -C $(LLVM_LIB_PATH)
-$(HOST) $(KERNEL): $(BUILD_DIR)/$(HPVM_OBJS)
- $(OPT) -debug $(HPVM_OPTFLAGS) -S $< -o $(HOST)
+$(HOST) $(KERNEL): $(BUILD_DIR)/$(VISC_OBJS)
+ $(OPT) -debug $(VISC_OPTFLAGS) -S $< -o $(HOST)
$(BUILD_DIR):
mkdir -p $(BUILD_DIR)
@@ -125,7 +125,7 @@ $(BUILD_DIR)/%.ll : $(SRC_DIR)/%.c
$(BUILD_DIR)/main.ll : $(SRC_DIR)/main.c
$(CC) $(CFLAGS) -emit-llvm -S -o $@ $<
-$(BUILD_DIR)/main.hpvm.ll : $(BUILD_DIR)/main.ll
- $(OPT) -debug-only=genhpvm $(TESTGEN_OPTFLAGS) $< -S -o $@
+$(BUILD_DIR)/main.visc.ll : $(BUILD_DIR)/main.ll
+ $(OPT) -debug-only=genvisc $(TESTGEN_OPTFLAGS) $< -S -o $@
## END HPVM MAKEFILE
diff --git a/hpvm/test/hpvm-cava/Makefile.config.example b/hpvm/test/hpvm-cava/Makefile.config.example
index 8cbe04af78..269f0b7df2 100644
--- a/hpvm/test/hpvm-cava/Makefile.config.example
+++ b/hpvm/test/hpvm-cava/Makefile.config.example
@@ -4,20 +4,20 @@ OPENCL_PATH=/opt/intelFPGA_pro/18.0/hld/host/linux64
OPENCL_LIB_PATH=$(OPENCL_PATH)/lib
# NOTE: You may need to configure this based on your root path.
-HPVM_SRC_ROOT=$(LLVM_SRC_ROOT)
+VISC_SRC_ROOT=$(LLVM_SRC_ROOT)
-HPVM_BUILD_DIR =$(HPVM_SRC_ROOT)/build
-CC = $(HPVM_BUILD_DIR)/bin/clang
-PLATFORM_CFLAGS = -I$(LLVM_SRC_ROOT)/include -I$(HPVM_BUILD_DIR)/include
+VISC_BUILD_DIR =$(VISC_SRC_ROOT)/build
+CC = $(VISC_BUILD_DIR)/bin/clang
+PLATFORM_CFLAGS = -I$(LLVM_SRC_ROOT)/include -I$(VISC_BUILD_DIR)/include
-CXX = $(HPVM_BUILD_DIR)/bin/clang++
-PLATFORM_CXXFLAGS = -I$(LLVM_SRC_ROOT)/include -I$(HPVM_BUILD_DIR)/include
+CXX = $(VISC_BUILD_DIR)/bin/clang++
+PLATFORM_CXXFLAGS = -I$(LLVM_SRC_ROOT)/include -I$(VISC_BUILD_DIR)/include
-LINKER = $(HPVM_BUILD_DIR)/bin/clang++
+LINKER = $(VISC_BUILD_DIR)/bin/clang++
PLATFORM_LDFLAGS = -lm -lpthread -lrt -lOpenCL -L$(OPENCL_LIB_PATH)
-LLVM_LIB_PATH = $(HPVM_BUILD_DIR)/lib
-LLVM_BIN_PATH = $(HPVM_BUILD_DIR)/bin
+LLVM_LIB_PATH = $(VISC_BUILD_DIR)/lib
+LLVM_BIN_PATH = $(VISC_BUILD_DIR)/bin
OPT = $(LLVM_BIN_PATH)/opt
LLVM_LINK = $(LLVM_BIN_PATH)/llvm-link
diff --git a/hpvm/test/hpvm-cava/README.md b/hpvm/test/hpvm-cava/README.md
index 1106c4781b..890b629d17 100644
--- a/hpvm/test/hpvm-cava/README.md
+++ b/hpvm/test/hpvm-cava/README.md
@@ -12,7 +12,7 @@ See the original camera/vision pipeline repo (repo: `yaoyuannnn/cava`) for detai
After building HPVM, the following steps are required to build and run the camera pipeline:
1. Build with `make TARGET=seq` for CPU and `make TARGET=gpu` for gpu.
-2. Run with `./cava-hpvm-<Target> example-tulip-small/raw_tulip-small.bin example-tulip-small/tulip-small`.
+2. Run with `./cava-visc-<Target> example-tulip-small/raw_tulip-small.bin example-tulip-small/tulip-small`.
* `<Target>` can be either `seq` or `gpu` depending on what target is used to build.
* This processes the raw image `example-tulip-small/raw_tulip-small.bin`. Note that raw images are different from bitmaps, so you might need to obtain them using special software.
* This generates: `tulip-small.bin` and `tulip-small-<stage>.bin` where `<stage>` represents the stage of the pipeline.
diff --git a/hpvm/test/hpvm-cava/src/main.c b/hpvm/test/hpvm-cava/src/main.c
index 4188c9e860..e43bbb4f25 100644
--- a/hpvm/test/hpvm-cava/src/main.c
+++ b/hpvm/test/hpvm-cava/src/main.c
@@ -1,154 +1,136 @@
-#include "utility.h"
#include <argp.h>
-#include <assert.h>
-#include <math.h>
#include <stdio.h>
#include <stdlib.h>
+#include <assert.h>
#include <string.h>
+#include <math.h>
+#include "utility.h"
#include "cam_pipe_utility.h"
-#include "load_cam_model.h"
#include "pipe_stages.h"
+#include "load_cam_model.h"
-#include "hpvm.h"
+#include "visc.h"
int NUM_TEST_CASES;
int NUM_CLASSES;
int INPUT_DIM;
int NUM_WORKER_THREADS;
-// Type of struct holding the return value from the last node.
-struct RetStruct {
- size_t bytesRet;
-};
-
// Type of struct that is used to pass arguments to the HPVM dataflow graph
// using the hpvm launch operation
typedef struct __attribute__((__packed__)) {
- uint8_t *input;
- size_t bytes_input;
- uint8_t *result;
- size_t bytes_result;
- float *input_scaled;
- size_t bytes_input_scaled;
- float *result_scaled;
- size_t bytes_result_scaled;
- float *demosaic_out;
- size_t bytes_demosaic_out;
- float *denoise_out;
- size_t bytes_denoise_out;
- float *transform_out;
- size_t bytes_transform_out;
- float *gamut_out;
- size_t bytes_gamut_out;
- float *TsTw;
- size_t bytes_TsTw;
- float *ctrl_pts;
- size_t bytes_ctrl_pts;
- float *weights;
- size_t bytes_weights;
- float *coefs;
- size_t bytes_coefs;
- float *l2_dist;
- size_t bytes_l2_dist;
- float *tone_map;
- size_t bytes_tone_map;
- int row_size;
- int col_size;
- struct RetStruct ret; // Instance of RetStruct holding the return value.
-} RootIn;
+ uint8_t *input; size_t bytes_input;
+ uint8_t *result; size_t bytes_result;
+ float *input_scaled; size_t bytes_input_scaled;
+ float *result_scaled; size_t bytes_result_scaled;
+ float *demosaic_out; size_t bytes_demosaic_out;
+ float *denoise_out; size_t bytes_denoise_out;
+ float *transform_out; size_t bytes_transform_out;
+ float *gamut_out;size_t bytes_gamut_out;
+ float *TsTw; size_t bytes_TsTw;
+ float *ctrl_pts; size_t bytes_ctrl_pts;
+ float *weights; size_t bytes_weights;
+ float*coefs; size_t bytes_coefs;
+ float *l2_dist; size_t bytes_l2_dist;
+ float *tone_map; size_t bytes_tone_map;
+ size_t row_size; size_t col_size;
+}
+RootIn;
typedef enum _argnum {
- RAW_IMAGE_BIN,
- OUTPUT_IMAGE_BIN,
- NUM_REQUIRED_ARGS,
- DATA_FILE = NUM_REQUIRED_ARGS,
- NUM_ARGS,
+ RAW_IMAGE_BIN,
+ OUTPUT_IMAGE_BIN,
+ NUM_REQUIRED_ARGS,
+ DATA_FILE = NUM_REQUIRED_ARGS,
+ NUM_ARGS,
} argnum;
typedef struct _arguments {
- char *args[NUM_ARGS];
- int num_inputs;
- int num_threads;
+ char* args[NUM_ARGS];
+ int num_inputs;
+ int num_threads;
} arguments;
static char prog_doc[] = "\nCamera pipeline on gem5-Aladdin.\n";
static char args_doc[] = "path/to/raw-image-binary path/to/output-image-binary";
static struct argp_option options[] = {
- {"num-inputs", 'n', "N", 0, "Number of input images"},
- {0},
- {"data-file", 'f', "F", 0,
- "File to read data and weights from (if data-init-mode == READ_FILE or "
- "save-params is true). *.txt files are decoded as text files, while "
- "*.bin files are decoded as binary files."},
+ { "num-inputs", 'n', "N", 0, "Number of input images" }, { 0 },
+ { "data-file", 'f', "F", 0,
+ "File to read data and weights from (if data-init-mode == READ_FILE or "
+ "save-params is true). *.txt files are decoded as text files, while "
+ "*.bin files are decoded as binary files." },
};
-static error_t parse_opt(int key, char *arg, struct argp_state *state) {
- arguments *args = (arguments *)(state->input);
- switch (key) {
- case 'n': {
- args->num_inputs = strtol(arg, NULL, 10);
- break;
- }
- case 'f': {
- args->args[DATA_FILE] = arg;
- break;
- }
- case 't': {
- args->num_threads = strtol(arg, NULL, 10);
- break;
- }
- case ARGP_KEY_ARG: {
- if (state->arg_num >= NUM_REQUIRED_ARGS)
- argp_usage(state);
- args->args[state->arg_num] = arg;
- break;
- }
- case ARGP_KEY_END: {
- if (state->arg_num < NUM_REQUIRED_ARGS) {
- fprintf(stderr, "Not enough arguments! Got %d, require %d.\n",
- state->arg_num, NUM_REQUIRED_ARGS);
- argp_usage(state);
+static error_t parse_opt(int key, char* arg, struct argp_state* state) {
+ arguments* args = (arguments*)(state->input);
+ switch (key) {
+ case 'n': {
+ args->num_inputs = strtol(arg, NULL, 10);
+ break;
+ }
+ case 'f': {
+ args->args[DATA_FILE] = arg;
+ break;
+ }
+ case 't': {
+ args->num_threads = strtol(arg, NULL, 10);
+ break;
+ }
+ case ARGP_KEY_ARG: {
+ if (state->arg_num >= NUM_REQUIRED_ARGS)
+ argp_usage(state);
+ args->args[state->arg_num] = arg;
+ break;
+ }
+ case ARGP_KEY_END: {
+ if (state->arg_num < NUM_REQUIRED_ARGS) {
+ fprintf(stderr,
+ "Not enough arguments! Got %d, require %d.\n",
+ state->arg_num,
+ NUM_REQUIRED_ARGS);
+ argp_usage(state);
+ }
+ break;
+ }
+ default:
+ return ARGP_ERR_UNKNOWN;
}
- break;
- }
- default:
- return ARGP_ERR_UNKNOWN;
- }
- return 0;
+ return 0;
}
-void set_default_args(arguments *args) {
- args->num_inputs = 1;
- args->num_threads = 0;
- for (int i = 0; i < NUM_ARGS; i++) {
- args->args[i] = NULL;
- }
+void set_default_args(arguments* args) {
+ args->num_inputs = 1;
+ args->num_threads = 0;
+ for (int i = 0; i < NUM_ARGS; i++) {
+ args->args[i] = NULL;
+ }
}
-static struct argp parser = {options, parse_opt, args_doc, prog_doc};
+static struct argp parser = { options, parse_opt, args_doc, prog_doc };
// Helper function for printing intermediate results
-void descale_cpu(float *input, size_t bytes_input, uint8_t *output,
- size_t bytes_result, size_t row_size, size_t col_size) {
-
+void descale_cpu(float *input, size_t bytes_input,
+ uint8_t *output, size_t bytes_result,
+ size_t row_size, size_t col_size) {
+
for (int chan = 0; chan < CHAN_SIZE; chan++)
for (int row = 0; row < row_size; row++)
for (int col = 0; col < col_size; col++) {
- int index = (chan * row_size + row) * col_size + col;
+ int index = (chan*row_size + row) * col_size + col;
output[index] = min(max(input[index] * 255, 0), 255);
}
}
static void sort(float arr[], int n) {
- int i, j;
- for (i = 0; i < n - 1; i++)
- for (j = 0; j < n - i - 1; j++)
- if (arr[j] > arr[j + 1]) {
- float temp = arr[j];
- arr[j] = arr[j + 1];
- arr[j + 1] = temp;
- }
+ int i, j;
+ for (i = 0; i < n - 1; i++)
+ for (j = 0; j < n - i - 1; j++)
+ if (arr[j] > arr[j + 1]) {
+ float temp = arr[j];
+ arr[j] = arr[j + 1];
+ arr[j + 1] = temp;
+ }
}
/**************************************************************/
@@ -158,259 +140,256 @@ static void sort(float arr[], int n) {
// In this benchmark, no use of HPVM query intrinsics in the leaf node functions
// Leaf HPVM node function for scale
-void scale_fxp(uint8_t *input, size_t bytes_input, float *output,
- size_t bytes_output, size_t row_size, size_t col_size) {
+void scale_fxp(uint8_t *input, size_t bytes_input,
+ float *output, size_t bytes_output,
+ size_t row_size, size_t col_size) {
- // Specifies compilation target for current node
- __hpvm__hint(CPU_TARGET);
+ //Specifies compilation target for current node
+ __visc__hint(CPU_TARGET);
// Specifies pointer arguments that will be used as "in" and "out" arguments
// - count of "in" arguments
// - list of "in" argument , and similar for "out"
- __hpvm__attributes(2, input, output, 1, output);
- void *thisNode = __hpvm__getNode();
- int row = __hpvm__getNodeInstanceID_x(thisNode);
+ __visc__attributes(2, input, output, 1, output);
+ void* thisNode = __visc__getNode();
+ int row = __visc__getNodeInstanceID_x(thisNode);
for (int chan = 0; chan < CHAN_SIZE; chan++)
- // for (int row = 0; row < row_size; row++)
- for (int col = 0; col < col_size; col++) {
- int index = (chan * row_size + row) * col_size + col;
- output[index] = input[index] * 1.0 / 255;
- }
- __hpvm__return(1, bytes_output);
+// for (int row = 0; row < row_size; row++)
+ for (int col = 0; col < col_size; col++){
+ int index = (chan*row_size + row) * col_size + col;
+ output[index] = input[index] * 1.0 / 255;
+ }
+ __visc__return(1, bytes_output);
}
// Leaf HPVM node function for descale
-void descale_fxp(float *input, size_t bytes_input, uint8_t *output,
- size_t bytes_result, size_t row_size, size_t col_size) {
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(2, input, output, 1, output);
-
+void descale_fxp(float *input, size_t bytes_input,
+ uint8_t *output, size_t bytes_result,
+ size_t row_size, size_t col_size) {
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(2, input, output, 1, output);
+
for (int chan = 0; chan < CHAN_SIZE; chan++)
for (int row = 0; row < row_size; row++)
for (int col = 0; col < col_size; col++) {
- int index = (chan * row_size + row) * col_size + col;
+ int index = (chan*row_size + row) * col_size + col;
output[index] = min(max(input[index] * 255, 0), 255);
}
- __hpvm__return(1, bytes_result);
+ __visc__return(1, bytes_result);
}
// Leaf HPVM node function for demosaicing
-void demosaic_fxp(float *input, size_t bytes_input, float *result,
- size_t bytes_result, size_t row_size, size_t col_size) {
- __hpvm__hint(DEVICE);
- __hpvm__attributes(2, input, result, 1, result);
-
- void *thisNode = __hpvm__getNode();
- int row = __hpvm__getNodeInstanceID_x(thisNode);
- // for (int row = 1; row < row_size - 1; row++)
- for (int col = 1; col < col_size - 1; col++) {
- int index_0 = (0 * row_size + row) * col_size + col;
- int index_1 = (1 * row_size + row) * col_size + col;
- int index_2 = (2 * row_size + row) * col_size + col;
- if (row % 2 == 0 && col % 2 == 0) {
- // Green pixel
- // Getting the R values
- float R1 = input[index_0 - 1];
- float R2 = input[index_0 + 1];
- // Getting the B values
- float B1 = input[index_2 - col_size];
- float B2 = input[index_2 + col_size];
- // R
- result[index_0] = (R1 + R2) / 2;
- // G
- result[index_1] = input[index_1] * 2;
- // B
- result[index_2] = (B1 + B2) / 2;
- } else if (row % 2 == 0 && col % 2 == 1) {
- // Red pixel
- // Getting the G values
- float G1 = input[index_1 - col_size];
- float G2 = input[index_1 + col_size];
- float G3 = input[index_1 - 1];
- float G4 = input[index_1 + 1];
- // Getting the B values
- float B1 = input[index_2 - col_size - 1];
- float B2 = input[index_2 - col_size + 1];
- float B3 = input[index_2 + col_size - 1];
- float B4 = input[index_2 + col_size + 1];
- // R
- result[index_0] = input[index_0];
- // G
- result[index_1] = (G1 + G2 + G3 + G4) / 2;
- // B (center pixel)
- result[index_2] = (B1 + B2 + B3 + B4) / 4;
- } else if (row % 2 == 1 && col % 2 == 0) {
- // Blue pixel
- // Getting the R values
- float R1 = input[index_0 - col_size - 1];
- float R2 = input[index_0 + col_size - 1];
- float R3 = input[index_0 - col_size + 1];
- float R4 = input[index_0 + col_size + 1];
- // Getting the G values
- float G1 = input[index_1 - col_size];
- float G2 = input[index_1 + col_size];
- float G3 = input[index_1 - 1];
- float G4 = input[index_1 + 1];
- // R
- result[index_0] = (R1 + R2 + R3 + R4) / 4;
- // G
- result[index_1] = (G1 + G2 + G3 + G4) / 2;
- // B
- result[index_2] = input[index_2];
- } else {
- // Bottom Green pixel
- // Getting the R values
- float R1 = input[index_0 - col_size];
- float R2 = input[index_0 + col_size];
- // Getting the B values
- float B1 = input[index_2 - 1];
- float B2 = input[index_2 + 1];
- // R
- result[index_0] = (R1 + R2) / 2;
- // G
- result[index_1] = input[index_1] * 2;
- // B
- result[index_2] = (B1 + B2) / 2;
- }
- }
- __hpvm__return(1, bytes_result);
+void demosaic_fxp(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ size_t row_size, size_t col_size) {
+ __visc__hint(DEVICE);
+ __visc__attributes(2, input, result, 1, result);
+
+ void* thisNode = __visc__getNode();
+ int row = __visc__getNodeInstanceID_x(thisNode);
+// for (int row = 1; row < row_size - 1; row++)
+ for (int col = 1; col < col_size - 1; col++) {
+ int index_0 = (0 * row_size + row) * col_size + col;
+ int index_1 = (1 * row_size + row) * col_size + col;
+ int index_2 = (2 * row_size + row) * col_size + col;
+ if (row % 2 == 0 && col % 2 == 0) {
+ // Green pixel
+ // Getting the R values
+ float R1 = input[index_0 - 1];
+ float R2 = input[index_0 + 1];
+ // Getting the B values
+ float B1 = input[index_2 - col_size];
+ float B2 = input[index_2 + col_size];
+ // R
+ result[index_0] = (R1 + R2) / 2;
+ // G
+ result[index_1] = input[index_1] * 2;
+ // B
+ result[index_2] = (B1 + B2) / 2;
+ } else if (row % 2 == 0 && col % 2 == 1) {
+ // Red pixel
+ // Getting the G values
+ float G1 = input[index_1 - col_size];
+ float G2 = input[index_1 + col_size];
+ float G3 = input[index_1 - 1];
+ float G4 = input[index_1 + 1];
+ // Getting the B values
+ float B1 = input[index_2 - col_size - 1];
+ float B2 = input[index_2 - col_size + 1];
+ float B3 = input[index_2 + col_size - 1];
+ float B4 = input[index_2 + col_size + 1];
+ // R
+ result[index_0] = input[index_0];
+ // G
+ result[index_1] = (G1 + G2 + G3 + G4) / 2;
+ // B (center pixel)
+ result[index_2] = (B1 + B2 + B3 + B4) / 4;
+ } else if (row % 2 == 1 && col % 2 == 0) {
+ // Blue pixel
+ // Getting the R values
+ float R1 = input[index_0 - col_size - 1];
+ float R2 = input[index_0 + col_size - 1];
+ float R3 = input[index_0 - col_size + 1];
+ float R4 = input[index_0 + col_size + 1];
+ // Getting the G values
+ float G1 = input[index_1 - col_size];
+ float G2 = input[index_1 + col_size];
+ float G3 = input[index_1 - 1];
+ float G4 = input[index_1 + 1];
+ // R
+ result[index_0] = (R1 + R2 + R3 + R4) / 4;
+ // G
+ result[index_1] = (G1 + G2 + G3 + G4) / 2;
+ // B
+ result[index_2] = input[index_2];
+ } else {
+ // Bottom Green pixel
+ // Getting the R values
+ float R1 = input[index_0 - col_size];
+ float R2 = input[index_0 + col_size];
+ // Getting the B values
+ float B1 = input[index_2 - 1];
+ float B2 = input[index_2 + 1];
+ // R
+ result[index_0] = (R1 + R2) / 2;
+ // G
+ result[index_1] = input[index_1] * 2;
+ // B
+ result[index_2] = (B1 + B2) / 2;
+ }
+ }
+ __visc__return(1, bytes_result);
}
// Leaf HPVM node function for denoise
-void denoise_fxp(float *input, size_t bytes_input, float *result,
- size_t bytes_result, size_t row_size, size_t col_size) {
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(2, input, result, 1, result);
-
- void *thisNode = __hpvm__getNode();
- int row = __hpvm__getNodeInstanceID_x(thisNode);
+void denoise_fxp(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ size_t row_size, size_t col_size) {
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(2, input, result, 1, result);
+
+ void* thisNode = __visc__getNode();
+ int row = __visc__getNodeInstanceID_x(thisNode);
for (int chan = 0; chan < CHAN_SIZE; chan++)
- // for (int row = 0; row < row_size; row++)
- for (int col = 0; col < col_size; col++)
- if (row >= 1 && row < row_size - 1 && col >= 1 && col < col_size - 1) {
- float filter[9];
- for (int i = -1; i < 2; i++)
- for (int j = -1; j < 2; j++) {
- int index = ((i + row) - row + 1) * 3 + (j + col) - col + 1;
- filter[index] =
- input[(chan * row_size + (i + row)) * col_size + (j + col)];
- }
- sort(filter, 9);
- result[(chan * row_size + row) * col_size + col] = filter[4];
- } else {
- result[(chan * row_size + row) * col_size + col] =
- input[(chan * row_size + row) * col_size + col];
- }
- __hpvm__return(1, bytes_result);
+// for (int row = 0; row < row_size; row++)
+ for (int col = 0; col < col_size; col++)
+ if (row >= 1 && row < row_size - 1 && col >= 1 && col < col_size - 1) {
+ float filter[9];
+ for (int i = -1; i < 2; i++)
+ for (int j = -1; j < 2; j++) {
+ int index = ((i+row) - row + 1) * 3 + (j+col) - col + 1;
+ filter[index] = input[(chan * row_size + (i + row)) * col_size + (j + col)];
+ }
+ sort(filter, 9);
+ result[(chan * row_size + row) * col_size + col] = filter[4];
+ } else {
+ result[(chan * row_size + row) * col_size + col] = input[(chan * row_size + row) * col_size + col];
+ }
+ __visc__return(1, bytes_result);
}
// Leaf HPVM node function, for color map and white balance transform
-void transform_fxp(float *input, size_t bytes_input, float *result,
- size_t bytes_result, float *TsTw_tran, size_t bytes_TsTw,
+void transform_fxp(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ float *TsTw_tran, size_t bytes_TsTw,
size_t row_size, size_t col_size) {
- __hpvm__hint(DEVICE);
- __hpvm__attributes(3, input, result, TsTw_tran, 1, result);
-
- void *thisNode = __hpvm__getNode();
- int row = __hpvm__getNodeInstanceID_x(thisNode);
+ __visc__hint(DEVICE);
+ __visc__attributes(3, input, result, TsTw_tran, 1, result);
+
+ void* thisNode = __visc__getNode();
+ int row = __visc__getNodeInstanceID_x(thisNode);
for (int chan = 0; chan < CHAN_SIZE; chan++)
- // for (int row = 0; row < row_size; row++)
- for (int col = 0; col < col_size; col++) {
- int index = (chan * row_size + row) * col_size + col;
- int index_0 = (0 * row_size + row) * col_size + col;
- int index_1 = (1 * row_size + row) * col_size + col;
- int index_2 = (2 * row_size + row) * col_size + col;
- int index_2d_0 = 0 * CHAN_SIZE + chan;
- int index_2d_1 = 1 * CHAN_SIZE + chan;
- int index_2d_2 = 2 * CHAN_SIZE + chan;
- result[index] = max(input[index_0] * TsTw_tran[index_2d_0] +
- input[index_1] * TsTw_tran[index_2d_1] +
- input[index_2] * TsTw_tran[index_2d_2],
- 0);
- }
- __hpvm__return(1, bytes_result);
+// for (int row = 0; row < row_size; row++)
+ for (int col = 0; col < col_size; col++) {
+ int index = (chan * row_size + row) * col_size + col;
+ int index_0 = (0 * row_size + row) * col_size + col;
+ int index_1 = (1 * row_size + row) * col_size + col;
+ int index_2 = (2 * row_size + row) * col_size + col;
+ int index_2d_0 = 0 * CHAN_SIZE + chan;
+ int index_2d_1 = 1 * CHAN_SIZE + chan;
+ int index_2d_2 = 2 * CHAN_SIZE + chan;
+ result[index] =
+ max(input[index_0] * TsTw_tran[index_2d_0] +
+ input[index_1] * TsTw_tran[index_2d_1] +
+ input[index_2] * TsTw_tran[index_2d_2],
+ 0);
+ }
+ __visc__return(1, bytes_result);
}
// Leaf HPVM node function, for gamut mapping
-void gamut_map_fxp(float *input, size_t bytes_input, float *result,
- size_t bytes_result, float *ctrl_pts, size_t bytes_ctrl_pts,
- float *weights, size_t bytes_weights, float *coefs,
- size_t bytes_coefs, float *l2_dist, size_t bytes_l2_dist,
+void gamut_map_fxp(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ float *ctrl_pts, size_t bytes_ctrl_pts,
+ float *weights, size_t bytes_weights,
+ float *coefs, size_t bytes_coefs,
+ float *l2_dist, size_t bytes_l2_dist,
size_t row_size, size_t col_size) {
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(6, input, result, ctrl_pts, weights, coefs, l2_dist, 2,
- result, l2_dist);
-
- // First, get the L2 norm from every pixel to the control points,
- // Then, sum it and weight it. Finally, add the bias.
- void *thisNode = __hpvm__getNode();
- int row = __hpvm__getNodeInstanceID_x(thisNode);
- // for (int row = 0; row < row_size; row++)
- for (int col = 0; col < col_size; col++) {
- float chan_val_0 = 0.0;
- float chan_val_1 = 0.0;
- float chan_val_2 = 0.0;
- for (int cp = 0; cp < 3702; cp++) {
- int index_0 = (0 * row_size + row) * col_size + col;
- int index_1 = (1 * row_size + row) * col_size + col;
- int index_2 = (2 * row_size + row) * col_size + col;
- float val1 = (input[index_0] - ctrl_pts[cp * 3 + 0]);
- float val2 = (input[index_0] - ctrl_pts[cp * 3 + 0]);
- float val3 = (input[index_1] - ctrl_pts[cp * 3 + 1]);
- float val4 = (input[index_1] - ctrl_pts[cp * 3 + 1]);
- float val5 = (input[index_2] - ctrl_pts[cp * 3 + 2]);
- float val6 = (input[index_2] - ctrl_pts[cp * 3 + 2]);
- float val = val1 * val2 + val3 * val4 + val5 * val6;
- float sqrt_val = sqrt(val);
- chan_val_0 += sqrt_val * weights[cp * CHAN_SIZE + 0];
- chan_val_1 += sqrt_val * weights[cp * CHAN_SIZE + 1];
- chan_val_2 += sqrt_val * weights[cp * CHAN_SIZE + 2];
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(6, input, result, ctrl_pts, weights, coefs, l2_dist, 2, result, l2_dist);
+
+ // First, get the L2 norm from every pixel to the control points,
+ // Then, sum it and weight it. Finally, add the bias.
+ void* thisNode = __visc__getNode();
+ int row = __visc__getNodeInstanceID_x(thisNode);
+// for (int row = 0; row < row_size; row++)
+ for (int col = 0; col < col_size; col++) {
+ float chan_val_0 = 0.0;
+ float chan_val_1 = 0.0;
+ float chan_val_2 = 0.0;
+ for (int cp = 0; cp < 3702; cp++) {
+ int index_0 = (0 * row_size + row) * col_size + col;
+ int index_1 = (1 * row_size + row) * col_size + col;
+ int index_2 = (2 * row_size + row) * col_size + col;
+ float val1 = (input[index_0] - ctrl_pts[cp * 3 + 0]);
+ float val2 = (input[index_0] - ctrl_pts[cp * 3 + 0]);
+ float val3 = (input[index_1] - ctrl_pts[cp * 3 + 1]);
+ float val4 = (input[index_1] - ctrl_pts[cp * 3 + 1]);
+ float val5 = (input[index_2] - ctrl_pts[cp * 3 + 2]);
+ float val6 = (input[index_2] - ctrl_pts[cp * 3 + 2]);
+ float val = val1 * val2 + val3 * val4 + val5 * val6;
+ float sqrt_val = sqrt(val);
+ chan_val_0 += sqrt_val * weights[cp * CHAN_SIZE + 0];
+ chan_val_1 += sqrt_val * weights[cp * CHAN_SIZE + 1];
+ chan_val_2 += sqrt_val * weights[cp * CHAN_SIZE + 2];
+ }
+ chan_val_0 += coefs[0 * CHAN_SIZE + 0] +
+ coefs[1 * CHAN_SIZE + 0] * input[(0 * row_size + row) * col_size + col] +
+ coefs[2 * CHAN_SIZE + 0] * input[(1 * row_size + row) * col_size + col] +
+ coefs[3 * CHAN_SIZE + 0] * input[(2 * row_size + row) * col_size + col];
+ chan_val_1 += coefs[0 * CHAN_SIZE + 1] +
+ coefs[1 * CHAN_SIZE + 1] * input[(0 * row_size + row) * col_size + col] +
+ coefs[2 * CHAN_SIZE + 1] * input[(1 * row_size + row) * col_size + col] +
+ coefs[3 * CHAN_SIZE + 1] * input[(2 * row_size + row) * col_size + col];
+ chan_val_2 += coefs[0 * CHAN_SIZE + 2] +
+ coefs[1 * CHAN_SIZE + 2] * input[(0 * row_size + row) * col_size + col] +
+ coefs[2 * CHAN_SIZE + 2] * input[(1 * row_size + row) * col_size + col] +
+ coefs[3 * CHAN_SIZE + 2] * input[(2 * row_size + row) * col_size + col];
+ result[(0 * row_size + row) * col_size + col] = max(chan_val_0, 0);
+ result[(1 * row_size + row) * col_size + col] = max(chan_val_1, 0);
+ result[(2 * row_size + row) * col_size + col] = max(chan_val_2, 0);
}
- chan_val_0 +=
- coefs[0 * CHAN_SIZE + 0] +
- coefs[1 * CHAN_SIZE + 0] *
- input[(0 * row_size + row) * col_size + col] +
- coefs[2 * CHAN_SIZE + 0] *
- input[(1 * row_size + row) * col_size + col] +
- coefs[3 * CHAN_SIZE + 0] * input[(2 * row_size + row) * col_size + col];
- chan_val_1 +=
- coefs[0 * CHAN_SIZE + 1] +
- coefs[1 * CHAN_SIZE + 1] *
- input[(0 * row_size + row) * col_size + col] +
- coefs[2 * CHAN_SIZE + 1] *
- input[(1 * row_size + row) * col_size + col] +
- coefs[3 * CHAN_SIZE + 1] * input[(2 * row_size + row) * col_size + col];
- chan_val_2 +=
- coefs[0 * CHAN_SIZE + 2] +
- coefs[1 * CHAN_SIZE + 2] *
- input[(0 * row_size + row) * col_size + col] +
- coefs[2 * CHAN_SIZE + 2] *
- input[(1 * row_size + row) * col_size + col] +
- coefs[3 * CHAN_SIZE + 2] * input[(2 * row_size + row) * col_size + col];
- result[(0 * row_size + row) * col_size + col] = max(chan_val_0, 0);
- result[(1 * row_size + row) * col_size + col] = max(chan_val_1, 0);
- result[(2 * row_size + row) * col_size + col] = max(chan_val_2, 0);
- }
- __hpvm__return(1, bytes_result);
+ __visc__return(1, bytes_result);
}
// HPVM leaf node function, for tone mapping
-void tone_map_fxp(float *input, size_t bytes_input, float *result,
- size_t bytes_result, float *tone_map, size_t bytes_tone_map,
+void tone_map_fxp(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ float *tone_map, size_t bytes_tone_map,
size_t row_size, size_t col_size) {
- __hpvm__hint(DEVICE);
- __hpvm__attributes(3, input, result, tone_map, 1, result);
-
- void *thisNode = __hpvm__getNode();
- int row = __hpvm__getNodeInstanceID_x(thisNode);
+ __visc__hint(DEVICE);
+ __visc__attributes(3, input, result, tone_map, 1, result);
+
+ void* thisNode = __visc__getNode();
+ int row = __visc__getNodeInstanceID_x(thisNode);
for (int chan = 0; chan < CHAN_SIZE; chan++)
- // for (int row = 0; row < row_size; row++)
- for (int col = 0; col < col_size; col++) {
- int index = (chan * row_size + row) * col_size + col;
- uint8_t x = input[index] * 255;
- result[index] = tone_map[x * CHAN_SIZE + chan];
- }
- __hpvm__return(1, bytes_result);
+// for (int row = 0; row < row_size; row++)
+ for (int col = 0; col < col_size; col++) {
+ int index = (chan * row_size + row) * col_size + col;
+ uint8_t x = input[index] * 255;
+ result[index] = tone_map[x * CHAN_SIZE + chan];
+ }
+ __visc__return(1, bytes_result);
}
/********************************************************************/
@@ -421,184 +400,185 @@ void tone_map_fxp(float *input, size_t bytes_input, float *result,
// requirement for the FPGA backend . The CPU backend also supports this,
// so it does not cause a portability issue.
-void scale_fxp_wrapper(uint8_t *input, size_t bytes_input, float *result,
- size_t bytes_result, size_t row_size, size_t col_size) {
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(2, input, result, 1, result);
+void scale_fxp_wrapper(uint8_t *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ size_t row_size, size_t col_size) {
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(2, input, result, 1, result);
// Create an 1D (specified by 1st argument) HPVM node with 1 dynamic
// instance (last argument) associated with node function scale_fxp
- void *ScaleNode = __hpvm__createNodeND(1, scale_fxp, row_size);
+ void *ScaleNode = __visc__createNodeND(1, scale_fxp, row_size);
// Binds inputs of current node with specified node
// - destination node
// - argument position in argument list of function of source node
// - argument position in argument list of function of destination node
// - streaming (1) or non-streaming (0)
- __hpvm__bindIn(ScaleNode, 0, 0, 0); // bind input
- __hpvm__bindIn(ScaleNode, 1, 1, 0); // bind bytes_input
- __hpvm__bindIn(ScaleNode, 2, 2, 0); // bind result
- __hpvm__bindIn(ScaleNode, 3, 3, 0); // bind bytes_result
- __hpvm__bindIn(ScaleNode, 4, 4, 0); // bind row_size
- __hpvm__bindIn(ScaleNode, 5, 5, 0); // bind col_size
+ __visc__bindIn(ScaleNode, 0, 0, 0); // bind input
+ __visc__bindIn(ScaleNode, 1, 1, 0); // bind bytes_input
+ __visc__bindIn(ScaleNode, 2, 2, 0); // bind result
+ __visc__bindIn(ScaleNode, 3, 3, 0); // bind bytes_result
+ __visc__bindIn(ScaleNode, 4, 4, 0); // bind row_size
+ __visc__bindIn(ScaleNode, 5, 5, 0); // bind col_size
// Similar to bindIn, but for the output. Output of a node is a struct, and
// we consider the fields in increasing ordering.
- __hpvm__bindOut(ScaleNode, 0, 0, 0);
+ __visc__bindOut(ScaleNode, 0, 0, 0);
}
-void descale_fxp_wrapper(float *input, size_t bytes_input, uint8_t *result,
- size_t bytes_result, size_t row_size,
- size_t col_size) {
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(2, input, result, 1, result);
- void *DescaleNode = __hpvm__createNodeND(1, descale_fxp, row_size);
- __hpvm__bindIn(DescaleNode, 0, 0, 0); // bind input
- __hpvm__bindIn(DescaleNode, 1, 1, 0); // bind bytes_input
- __hpvm__bindIn(DescaleNode, 2, 2, 0); // bind result
- __hpvm__bindIn(DescaleNode, 3, 3, 0); // bind bytes_result
- __hpvm__bindIn(DescaleNode, 4, 4, 0); // bind row_size
- __hpvm__bindIn(DescaleNode, 5, 5, 0); // bind col_size
-
- __hpvm__bindOut(DescaleNode, 0, 0, 0);
+void descale_fxp_wrapper(float *input, size_t bytes_input,
+ uint8_t *result, size_t bytes_result,
+ size_t row_size, size_t col_size) {
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(2, input, result, 1, result);
+ void *DescaleNode = __visc__createNodeND(1, descale_fxp, row_size);
+ __visc__bindIn(DescaleNode, 0, 0, 0); // bind input
+ __visc__bindIn(DescaleNode, 1, 1, 0); // bind bytes_input
+ __visc__bindIn(DescaleNode, 2, 2, 0); // bind result
+ __visc__bindIn(DescaleNode, 3, 3, 0); // bind bytes_result
+ __visc__bindIn(DescaleNode, 4, 4, 0); // bind row_size
+ __visc__bindIn(DescaleNode, 5, 5, 0); // bind col_size
+
+ __visc__bindOut(DescaleNode, 0, 0, 0);
}
-void demosaic_fxp_wrapper(float *input, size_t bytes_input, float *result,
- size_t bytes_result, size_t row_size,
- size_t col_size) {
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(2, input, result, 1, result);
- void *DemosaicNode = __hpvm__createNodeND(1, demosaic_fxp, row_size);
- __hpvm__bindIn(DemosaicNode, 0, 0, 0); // bind input
- __hpvm__bindIn(DemosaicNode, 1, 1, 0); // bind bytes_input
- __hpvm__bindIn(DemosaicNode, 2, 2, 0); // bind result
- __hpvm__bindIn(DemosaicNode, 3, 3, 0); // bind bytes_result
- __hpvm__bindIn(DemosaicNode, 4, 4, 0); // bind row_size
- __hpvm__bindIn(DemosaicNode, 5, 5, 0); // bind col_size
-
- __hpvm__bindOut(DemosaicNode, 0, 0, 0);
+void demosaic_fxp_wrapper(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ size_t row_size, size_t col_size) {
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(2, input, result, 1, result);
+ void *DemosaicNode = __visc__createNodeND(1, demosaic_fxp, row_size);
+ __visc__bindIn(DemosaicNode, 0, 0, 0); // bind input
+ __visc__bindIn(DemosaicNode, 1, 1, 0); // bind bytes_input
+ __visc__bindIn(DemosaicNode, 2, 2, 0); // bind result
+ __visc__bindIn(DemosaicNode, 3, 3, 0); // bind bytes_result
+ __visc__bindIn(DemosaicNode, 4, 4, 0); // bind row_size
+ __visc__bindIn(DemosaicNode, 5, 5, 0); // bind col_size
+
+ __visc__bindOut(DemosaicNode, 0, 0, 0);
}
-void denoise_fxp_wrapper(float *input, size_t bytes_input, float *result,
- size_t bytes_result, size_t row_size,
- size_t col_size) {
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(2, input, result, 1, result);
- void *DenoiseNode = __hpvm__createNodeND(1, denoise_fxp, row_size);
- __hpvm__bindIn(DenoiseNode, 0, 0, 0); // bind input
- __hpvm__bindIn(DenoiseNode, 1, 1, 0); // bind bytes_input
- __hpvm__bindIn(DenoiseNode, 2, 2, 0); // bind result
- __hpvm__bindIn(DenoiseNode, 3, 3, 0); // bind bytes_result
- __hpvm__bindIn(DenoiseNode, 4, 4, 0); // bind row_size
- __hpvm__bindIn(DenoiseNode, 5, 5, 0); // bind col_size
-
- __hpvm__bindOut(DenoiseNode, 0, 0, 0);
+void denoise_fxp_wrapper(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ size_t row_size, size_t col_size) {
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(2, input, result, 1, result);
+ void *DenoiseNode = __visc__createNodeND(1, denoise_fxp, row_size);
+ __visc__bindIn(DenoiseNode, 0, 0, 0); // bind input
+ __visc__bindIn(DenoiseNode, 1, 1, 0); // bind bytes_input
+ __visc__bindIn(DenoiseNode, 2, 2, 0); // bind result
+ __visc__bindIn(DenoiseNode, 3, 3, 0); // bind bytes_result
+ __visc__bindIn(DenoiseNode, 4, 4, 0); // bind row_size
+ __visc__bindIn(DenoiseNode, 5, 5, 0); // bind col_size
+
+ __visc__bindOut(DenoiseNode, 0, 0, 0);
}
-void transform_fxp_wrapper(float *input, size_t bytes_input, float *result,
- size_t bytes_result, float *TsTw_tran,
- size_t bytes_TsTw, size_t row_size,
- size_t col_size) {
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(3, input, result, TsTw_tran, 1, result);
- void *TransformNode = __hpvm__createNodeND(1, transform_fxp, row_size);
- __hpvm__bindIn(TransformNode, 0, 0, 0); // bind input
- __hpvm__bindIn(TransformNode, 1, 1, 0); // bind bytes_input
- __hpvm__bindIn(TransformNode, 2, 2, 0); // bind result
- __hpvm__bindIn(TransformNode, 3, 3, 0); // bind bytes_result
- __hpvm__bindIn(TransformNode, 4, 4, 0); // bind tstw
- __hpvm__bindIn(TransformNode, 5, 5, 0); // bind bytes_tstw
- __hpvm__bindIn(TransformNode, 6, 6, 0); // bind row_size
- __hpvm__bindIn(TransformNode, 7, 7, 0); // bind col_size
-
- __hpvm__bindOut(TransformNode, 0, 0, 0);
+void transform_fxp_wrapper(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ float *TsTw_tran, size_t bytes_TsTw,
+ size_t row_size, size_t col_size) {
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(3, input, result, TsTw_tran, 1, result);
+ void *TransformNode = __visc__createNodeND(1, transform_fxp, row_size);
+ __visc__bindIn(TransformNode, 0, 0, 0); // bind input
+ __visc__bindIn(TransformNode, 1, 1, 0); // bind bytes_input
+ __visc__bindIn(TransformNode, 2, 2, 0); // bind result
+ __visc__bindIn(TransformNode, 3, 3, 0); // bind bytes_result
+ __visc__bindIn(TransformNode, 4, 4, 0); // bind tstw
+ __visc__bindIn(TransformNode, 5, 5, 0); // bind bytes_tstw
+ __visc__bindIn(TransformNode, 6, 6, 0); // bind row_size
+ __visc__bindIn(TransformNode, 7, 7, 0); // bind col_size
+
+ __visc__bindOut(TransformNode, 0, 0, 0);
}
-void gamut_fxp_wrapper(float *input, size_t bytes_input, float *result,
- size_t bytes_result, float *ctrl_pts,
- size_t bytes_ctrl_pts, float *weights,
- size_t bytes_weights, float *coefs, size_t bytes_coefs,
- float *l2_dist, size_t bytes_l2_dist, size_t row_size,
- size_t col_size) {
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(6, input, result, ctrl_pts, weights, coefs, l2_dist, 1,
- result);
- void *GamutNode = __hpvm__createNodeND(1, gamut_map_fxp, row_size);
- __hpvm__bindIn(GamutNode, 0, 0, 0); // bind input
- __hpvm__bindIn(GamutNode, 1, 1, 0); // bind bytes_input
- __hpvm__bindIn(GamutNode, 2, 2, 0); // bind result
- __hpvm__bindIn(GamutNode, 3, 3, 0); // bind bytes_result
- __hpvm__bindIn(GamutNode, 4, 4, 0); // bind ctrl_pts
- __hpvm__bindIn(GamutNode, 5, 5, 0); // bind bytes_ctrl_pts
- __hpvm__bindIn(GamutNode, 6, 6, 0); // bind weights
- __hpvm__bindIn(GamutNode, 7, 7, 0); // bind bytes_weights
- __hpvm__bindIn(GamutNode, 8, 8, 0); // bind coefs
- __hpvm__bindIn(GamutNode, 9, 9, 0); // bind bytes_coefs
- __hpvm__bindIn(GamutNode, 10, 10, 0); // bind l2_dist
- __hpvm__bindIn(GamutNode, 11, 11, 0); // bind bytes_l2_dist
- __hpvm__bindIn(GamutNode, 12, 12, 0); // bind row_size
- __hpvm__bindIn(GamutNode, 13, 13, 0); // bind col_size
-
- __hpvm__bindOut(GamutNode, 0, 0, 0);
+void gamut_fxp_wrapper(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ float *ctrl_pts, size_t bytes_ctrl_pts,
+ float *weights, size_t bytes_weights,
+ float *coefs, size_t bytes_coefs,
+ float *l2_dist, size_t bytes_l2_dist,
+ size_t row_size, size_t col_size) {
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(6, input, result, ctrl_pts, weights, coefs, l2_dist, 1, result);
+ void *GamutNode = __visc__createNodeND(1, gamut_map_fxp, row_size);
+ __visc__bindIn(GamutNode, 0, 0, 0); // bind input
+ __visc__bindIn(GamutNode, 1, 1, 0); // bind bytes_input
+ __visc__bindIn(GamutNode, 2, 2, 0); // bind result
+ __visc__bindIn(GamutNode, 3, 3, 0); // bind bytes_result
+ __visc__bindIn(GamutNode, 4, 4, 0); // bind ctrl_pts
+ __visc__bindIn(GamutNode, 5, 5, 0); // bind bytes_ctrl_pts
+ __visc__bindIn(GamutNode, 6, 6, 0); // bind weights
+ __visc__bindIn(GamutNode, 7, 7, 0); // bind bytes_weights
+ __visc__bindIn(GamutNode, 8, 8, 0); // bind coefs
+ __visc__bindIn(GamutNode, 9, 9, 0); // bind bytes_coefs
+ __visc__bindIn(GamutNode, 10, 10, 0); // bind l2_dist
+ __visc__bindIn(GamutNode, 11, 11, 0); // bind bytes_l2_dist
+ __visc__bindIn(GamutNode, 12, 12, 0); // bind row_size
+ __visc__bindIn(GamutNode, 13, 13, 0); // bind col_size
+
+ __visc__bindOut(GamutNode, 0, 0, 0);
}
-void tone_map_fxp_wrapper(float *input, size_t bytes_input, float *result,
- size_t bytes_result, float *tone_map,
- size_t bytes_tone_map, size_t row_size,
- size_t col_size) {
-
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(3, input, result, tone_map, 1, result);
- void *ToneMapNode = __hpvm__createNodeND(1, tone_map_fxp, row_size);
- __hpvm__bindIn(ToneMapNode, 0, 0, 0); // bind input
- __hpvm__bindIn(ToneMapNode, 1, 1, 0); // bind bytes_input
- __hpvm__bindIn(ToneMapNode, 2, 2, 0); // bind result
- __hpvm__bindIn(ToneMapNode, 3, 3, 0); // bind bytes_result
- __hpvm__bindIn(ToneMapNode, 4, 4, 0); // bind tone_map
- __hpvm__bindIn(ToneMapNode, 5, 5, 0); // bind bytes_tone_map
- __hpvm__bindIn(ToneMapNode, 6, 6, 0); // bind row_size
- __hpvm__bindIn(ToneMapNode, 7, 7, 0); // bind col_size
-
- __hpvm__bindOut(ToneMapNode, 0, 0, 0);
+void tone_map_fxp_wrapper(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ float *tone_map, size_t bytes_tone_map,
+ size_t row_size, size_t col_size) {
+
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(3, input, result, tone_map, 1, result);
+ void *ToneMapNode = __visc__createNodeND(1, tone_map_fxp, row_size);
+ __visc__bindIn(ToneMapNode, 0, 0, 0); // bind input
+ __visc__bindIn(ToneMapNode, 1, 1, 0); // bind bytes_input
+ __visc__bindIn(ToneMapNode, 2, 2, 0); // bind result
+ __visc__bindIn(ToneMapNode, 3, 3, 0); // bind bytes_result
+ __visc__bindIn(ToneMapNode, 4, 4, 0); // bind tone_map
+ __visc__bindIn(ToneMapNode, 5, 5, 0); // bind bytes_tone_map
+ __visc__bindIn(ToneMapNode, 6, 6, 0); // bind row_size
+ __visc__bindIn(ToneMapNode, 7, 7, 0); // bind col_size
+
+ __visc__bindOut(ToneMapNode, 0, 0, 0);
}
+
/*** ROOT Node - Top Level of the Graph Hierarchy ***/
-void CamPipeRoot(/*0*/ uint8_t *input, /*1*/ size_t bytes_input,
- /*2*/ uint8_t *result, /*3*/ size_t bytes_result,
- /*4*/ float *input_scaled, /*5*/ size_t bytes_input_scaled,
- /*6*/ float *result_scaled, /*7*/ size_t bytes_result_scaled,
- /*8*/ float *demosaic_out, /*9*/ size_t bytes_demosaic_out,
- /*10*/ float *denoise_out, /*11*/ size_t bytes_denoise_out,
- /*12*/ float *transform_out, /*13*/ size_t bytes_transform_out,
- /*14*/ float *gamut_out, /*15*/ size_t bytes_gamut_out,
- /*16*/ float *TsTw, /*17*/ size_t bytes_TsTw,
- /*18*/ float *ctrl_pts, /*19*/ size_t bytes_ctrl_pts,
- /*20*/ float *weights, /*21*/ size_t bytes_weights,
- /*22*/ float *coefs, /*23*/ size_t bytes_coefs,
- /*24*/ float *l2_dist, /*25*/ size_t bytes_l2_dist,
- /*26*/ float *tone_map, /*27*/ size_t bytes_tone_map,
- /*28*/ size_t row_size, /*29*/ size_t col_size) {
-
- // Specifies compilation target for current node
- __hpvm__hint(CPU_TARGET);
+void CamPipeRoot(/*0*/ uint8_t *input, /*1*/ size_t bytes_input,
+ /*2*/ uint8_t *result, /*3*/ size_t bytes_result,
+ /*4*/ float *input_scaled, /*5*/ size_t bytes_input_scaled,
+ /*6*/ float *result_scaled, /*7*/ size_t bytes_result_scaled,
+ /*8*/ float *demosaic_out, /*9*/ size_t bytes_demosaic_out,
+ /*10*/ float *denoise_out, /*11*/ size_t bytes_denoise_out,
+ /*12*/ float *transform_out, /*13*/ size_t bytes_transform_out,
+ /*14*/ float *gamut_out, /*15*/ size_t bytes_gamut_out,
+ /*16*/ float *TsTw, /*17*/ size_t bytes_TsTw,
+ /*18*/ float *ctrl_pts, /*19*/ size_t bytes_ctrl_pts,
+ /*20*/ float *weights, /*21*/ size_t bytes_weights,
+ /*22*/ float*coefs, /*23*/ size_t bytes_coefs,
+ /*24*/ float *l2_dist, /*25*/ size_t bytes_l2_dist,
+ /*26*/ float *tone_map, /*27*/ size_t bytes_tone_map,
+ /*28*/ size_t row_size, /*29*/ size_t col_size) {
+
+ //Specifies compilation target for current node
+ __visc__hint(CPU_TARGET);
// Specifies pointer arguments that will be used as "in" and "out" arguments
// - count of "in" arguments
// - list of "in" argument , and similar for "out"
- __hpvm__attributes(14, input, result, input_scaled, result_scaled,
- demosaic_out, denoise_out, transform_out, gamut_out, TsTw,
- ctrl_pts, weights, coefs, tone_map, l2_dist, 5, result,
- demosaic_out, denoise_out, transform_out, gamut_out);
+ __visc__attributes(14, input, result, input_scaled, result_scaled, demosaic_out, denoise_out,
+ transform_out, gamut_out, TsTw, ctrl_pts, weights, coefs, tone_map, l2_dist,
+ 5, result, demosaic_out, denoise_out, transform_out, gamut_out);
// Create an 0D (specified by 1st argument) HPVM node - so a single node
// associated with node function ---_fxp_wrapper
- void *ScNode = __hpvm__createNodeND(0, scale_fxp_wrapper);
- void *DmNode = __hpvm__createNodeND(0, demosaic_fxp_wrapper);
- void *DnNode = __hpvm__createNodeND(0, denoise_fxp_wrapper);
- void *TrNode = __hpvm__createNodeND(0, transform_fxp_wrapper);
- void *GmNode = __hpvm__createNodeND(0, gamut_fxp_wrapper);
- void *TnNode = __hpvm__createNodeND(0, tone_map_fxp_wrapper);
- void *DsNode = __hpvm__createNodeND(0, descale_fxp_wrapper);
-
+ void* ScNode = __visc__createNodeND(0, scale_fxp_wrapper);
+ void* DmNode = __visc__createNodeND(0, demosaic_fxp_wrapper);
+ void *DnNode = __visc__createNodeND(0, denoise_fxp_wrapper);
+ void *TrNode = __visc__createNodeND(0, transform_fxp_wrapper);
+ void *GmNode = __visc__createNodeND(0, gamut_fxp_wrapper);
+ void *TnNode = __visc__createNodeND(0, tone_map_fxp_wrapper);
+ void *DsNode = __visc__createNodeND(0, descale_fxp_wrapper);
+
// BindIn binds inputs of current node with specified node
// - destination node
// - argument position in argument list of function of source node
@@ -612,283 +592,268 @@ void CamPipeRoot(/*0*/ uint8_t *input, /*1*/ size_t bytes_input,
// - destination position (in argument list of destination node)
// - streaming (1) or non-streaming (0)
- // scale_fxp inputs
- __hpvm__bindIn(ScNode, 0, 0, 0); // input -> ScNode:input
- __hpvm__bindIn(ScNode, 1, 1, 0); // bytes_input -> ScNode:bytes_input
- __hpvm__bindIn(ScNode, 4, 2, 0); // input_scaled -> ScNode:result
- __hpvm__bindIn(ScNode, 5, 3, 0); // bytes_input_scaled -> ScNode:bytes_result
- __hpvm__bindIn(ScNode, 28, 4, 0); // row_size -> ScNode:row_size
- __hpvm__bindIn(ScNode, 29, 5, 0); // col_size -> ScNode:col_size
-
- // demosaic_fxp inputs
- __hpvm__bindIn(DmNode, 4, 0, 0); // input_scaled -> DmNode:input
- __hpvm__edge(ScNode, DmNode, 1, 0, 1,
- 0); // SCNode:bytes_result -> DmNode:bytes_input
- __hpvm__bindIn(DmNode, 8, 2, 0); // demosaic_out -> DmNode:result
- __hpvm__bindIn(DmNode, 9, 3, 0); // bytes_demosaic_out -> DmNode:bytes_result
- __hpvm__bindIn(DmNode, 28, 4, 0); // row_size -> DmNode:row_size
- __hpvm__bindIn(DmNode, 29, 5, 0); // col_size -> DmNode:col_size
-
- // denoise_fxp inputs
- __hpvm__bindIn(DnNode, 8, 0, 0); // demosaic_out -> DnNode:input
- __hpvm__edge(DmNode, DnNode, 1, 0, 1,
- 0); // DMNode:bytes_result -> DnNode:bytes_input
- __hpvm__bindIn(DnNode, 10, 2, 0); // denoise_out -> DnNode:result
- __hpvm__bindIn(DnNode, 11, 3, 0); // bytes_denoise_out -> DnNode:bytes_result
- __hpvm__bindIn(DnNode, 28, 4, 0); // row_size -> DnNode:row_size
- __hpvm__bindIn(DnNode, 29, 5, 0); // col_size -> DnNode:col_size
-
- // transform_fxp inputs
- __hpvm__bindIn(TrNode, 10, 0, 0); // denoise_out -> TrNode:input
- __hpvm__edge(DnNode, TrNode, 1, 0, 1,
- 0); // DnNode:bytes_result -> TrNode:bytes_input
- __hpvm__bindIn(TrNode, 12, 2, 0); // transform_out -> TrNode:result
- __hpvm__bindIn(TrNode, 13, 3,
- 0); // bytes_result_scaled -> TrNode:bytes_result
- __hpvm__bindIn(TrNode, 16, 4, 0); // TsTw -> TrNode:TsTw_trann
- __hpvm__bindIn(TrNode, 17, 5, 0); // bytes_TsTw -> TrNode:bytes_TsTw
- __hpvm__bindIn(TrNode, 28, 6, 0); // row_size -> TrNode:row_size
- __hpvm__bindIn(TrNode, 29, 7, 0); // col_size -> TrNode:col_size
-
- // gamut_fxp inputs
- __hpvm__bindIn(GmNode, 12, 0, 0); // transform_out -> GmNode:input
- __hpvm__edge(TrNode, GmNode, 1, 0, 1,
- 0); // TrNode:bytes_result -> GmNode:bytes_input
- __hpvm__bindIn(GmNode, 14, 2, 0); // gamut_out -> GmNode:result
- __hpvm__bindIn(GmNode, 15, 3, 0); // bytes_gamut_out -> GmNode:bytes_result
- __hpvm__bindIn(GmNode, 18, 4, 0); // ctrl_pts -> GmNode:ctrl_pts
- __hpvm__bindIn(GmNode, 19, 5, 0); // bytes_ctrl_pts -> GmNode:bytes_ctrl_pts
- __hpvm__bindIn(GmNode, 20, 6, 0); // weights -> GmNode:weights
- __hpvm__bindIn(GmNode, 21, 7, 0); // bytes_weights -> GmNode:bytes_weights
- __hpvm__bindIn(GmNode, 22, 8, 0); // coefs -> GmNode:coefs
- __hpvm__bindIn(GmNode, 23, 9, 0); // bytes_coefs -> GmNode:bytes_coefs
- __hpvm__bindIn(GmNode, 24, 10, 0); // l2_dist -> GmNode: l2_dist
- __hpvm__bindIn(GmNode, 25, 11, 0); // bytes_l2_dist -> GmNode:bytes_l2_dist
- __hpvm__bindIn(GmNode, 28, 12, 0); // row_size -> GmNode:row_size
- __hpvm__bindIn(GmNode, 29, 13, 0); // col_size -> GmNode:col_size
-
- // tone_map_fxp inputs
- __hpvm__bindIn(TnNode, 14, 0, 0); // gamut_out -> TnNode:input
- __hpvm__edge(GmNode, TnNode, 1, 0, 1,
- 0); // GmNode:bytes_result -> TnNode:bytes_input
- __hpvm__bindIn(TnNode, 6, 2, 0); // result_scaled -> TnNode:result
- __hpvm__bindIn(TnNode, 7, 3, 0); // bytes_result_scaled -> TnNode:bytes_result
- __hpvm__bindIn(TnNode, 26, 4, 0); // tone_map -> TnNode:tone_map
- __hpvm__bindIn(TnNode, 27, 5, 0); // bytes_tone_map -> TnNode:bytes_tone_map
- __hpvm__bindIn(TnNode, 28, 6, 0); // row_size -> TnNode:row_size
- __hpvm__bindIn(TnNode, 29, 7, 0); // col_size -> TnNode:col_size
-
- // descale_fxp inputs
- __hpvm__bindIn(DsNode, 6, 0, 0); // result_scaled -> DsNode:input
- __hpvm__edge(TnNode, DsNode, 1, 0, 1,
- 0); // TnNode:bytes_result -> DsNode:bytes_input
- __hpvm__bindIn(DsNode, 2, 2, 0); // result -> DsNode:result
- __hpvm__bindIn(DsNode, 3, 3, 0); // bytes_result -> DsNode:bytes_result
- __hpvm__bindIn(DsNode, 28, 4, 0); // row_size -> DsNode:row_size
- __hpvm__bindIn(DsNode, 29, 5, 0); // col_size -> DsNode:col_size
+ // scale_fxp inputs
+ __visc__bindIn(ScNode, 0, 0, 0); // input -> ScNode:input
+ __visc__bindIn(ScNode, 1, 1, 0); // bytes_input -> ScNode:bytes_input
+ __visc__bindIn(ScNode, 4, 2, 0); // input_scaled -> ScNode:result
+ __visc__bindIn(ScNode, 5, 3, 0); // bytes_input_scaled -> ScNode:bytes_result
+ __visc__bindIn(ScNode, 28, 4, 0); // row_size -> ScNode:row_size
+ __visc__bindIn(ScNode, 29, 5, 0); // col_size -> ScNode:col_size
+
+ // demosaic_fxp inputs
+ __visc__bindIn(DmNode, 4, 0, 0); // input_scaled -> DmNode:input
+ __visc__edge(ScNode, DmNode, 1, 0, 1, 0); // SCNode:bytes_result -> DmNode:bytes_input
+ __visc__bindIn(DmNode, 8, 2, 0); // demosaic_out -> DmNode:result
+ __visc__bindIn(DmNode, 9, 3, 0); // bytes_demosaic_out -> DmNode:bytes_result
+ __visc__bindIn(DmNode, 28, 4, 0); // row_size -> DmNode:row_size
+ __visc__bindIn(DmNode, 29, 5, 0); // col_size -> DmNode:col_size
+
+ // denoise_fxp inputs
+ __visc__bindIn(DnNode, 8, 0, 0); // demosaic_out -> DnNode:input
+ __visc__edge(DmNode, DnNode, 1, 0, 1, 0); // DMNode:bytes_result -> DnNode:bytes_input
+ __visc__bindIn(DnNode, 10, 2, 0); // denoise_out -> DnNode:result
+ __visc__bindIn(DnNode, 11, 3, 0); // bytes_denoise_out -> DnNode:bytes_result
+ __visc__bindIn(DnNode, 28, 4, 0); // row_size -> DnNode:row_size
+ __visc__bindIn(DnNode, 29, 5, 0); // col_size -> DnNode:col_size
+
+ // transform_fxp inputs
+ __visc__bindIn(TrNode, 10, 0, 0); // denoise_out -> TrNode:input
+ __visc__edge(DnNode, TrNode, 1, 0, 1, 0); // DnNode:bytes_result -> TrNode:bytes_input
+ __visc__bindIn(TrNode, 12, 2, 0); // transform_out -> TrNode:result
+ __visc__bindIn(TrNode, 13, 3, 0); // bytes_result_scaled -> TrNode:bytes_result
+ __visc__bindIn(TrNode, 16, 4, 0); // TsTw -> TrNode:TsTw_trann
+ __visc__bindIn(TrNode, 17, 5, 0); // bytes_TsTw -> TrNode:bytes_TsTw
+ __visc__bindIn(TrNode, 28, 6, 0); // row_size -> TrNode:row_size
+ __visc__bindIn(TrNode, 29, 7, 0); // col_size -> TrNode:col_size
+
+ // gamut_fxp inputs
+ __visc__bindIn(GmNode, 12, 0, 0); // transform_out -> GmNode:input
+ __visc__edge(TrNode, GmNode, 1, 0, 1, 0); // TrNode:bytes_result -> GmNode:bytes_input
+ __visc__bindIn(GmNode, 14, 2, 0); // gamut_out -> GmNode:result
+ __visc__bindIn(GmNode, 15, 3, 0); // bytes_gamut_out -> GmNode:bytes_result
+ __visc__bindIn(GmNode, 18, 4, 0); // ctrl_pts -> GmNode:ctrl_pts
+ __visc__bindIn(GmNode, 19, 5, 0); // bytes_ctrl_pts -> GmNode:bytes_ctrl_pts
+ __visc__bindIn(GmNode, 20, 6, 0); // weights -> GmNode:weights
+ __visc__bindIn(GmNode, 21, 7, 0); // bytes_weights -> GmNode:bytes_weights
+ __visc__bindIn(GmNode, 22, 8, 0); // coefs -> GmNode:coefs
+ __visc__bindIn(GmNode, 23, 9, 0); // bytes_coefs -> GmNode:bytes_coefs
+ __visc__bindIn(GmNode, 24, 10, 0); // l2_dist -> GmNode: l2_dist
+ __visc__bindIn(GmNode, 25, 11, 0); // bytes_l2_dist -> GmNode:bytes_l2_dist
+ __visc__bindIn(GmNode, 28, 12, 0); // row_size -> GmNode:row_size
+ __visc__bindIn(GmNode, 29, 13, 0); // col_size -> GmNode:col_size
+
+ // tone_map_fxp inputs
+ __visc__bindIn(TnNode, 14, 0, 0); // gamut_out -> TnNode:input
+ __visc__edge(GmNode, TnNode, 1, 0, 1, 0); // GmNode:bytes_result -> TnNode:bytes_input
+ __visc__bindIn(TnNode, 6, 2, 0); // result_scaled -> TnNode:result
+ __visc__bindIn(TnNode, 7, 3, 0); // bytes_result_scaled -> TnNode:bytes_result
+ __visc__bindIn(TnNode, 26, 4, 0); // tone_map -> TnNode:tone_map
+ __visc__bindIn(TnNode, 27, 5, 0); // bytes_tone_map -> TnNode:bytes_tone_map
+ __visc__bindIn(TnNode, 28, 6, 0); // row_size -> TnNode:row_size
+ __visc__bindIn(TnNode, 29, 7, 0); // col_size -> TnNode:col_size
+
+ // descale_fxp inputs
+ __visc__bindIn(DsNode, 6, 0, 0); // result_scaled -> DsNode:input
+ __visc__edge(TnNode, DsNode, 1, 0, 1, 0); // TnNode:bytes_result -> DsNode:bytes_input
+ __visc__bindIn(DsNode, 2, 2, 0); // result -> DsNode:result
+ __visc__bindIn(DsNode, 3, 3, 0); // bytes_result -> DsNode:bytes_result
+ __visc__bindIn(DsNode, 28, 4, 0); // row_size -> DsNode:row_size
+ __visc__bindIn(DsNode, 29, 5, 0); // col_size -> DsNode:col_size
// Similar to bindIn, but for the output. Output of a node is a struct, and
// we consider the fields in increasing ordering.
- __hpvm__bindOut(DsNode, 0, 0, 0);
+ __visc__bindOut(DsNode, 0, 0, 0);
+
}
-int main(int argc, char *argv[]) {
- // Parse the arguments.
- arguments args;
- set_default_args(&args);
- argp_parse(&parser, argc, argv, 0, 0, &args);
-
- // Read a raw image.
- // NOTE: We deliberately perform this file I/O outside of the kernel.
- printf("Reading a raw image from %s\n", args.args[RAW_IMAGE_BIN]);
- size_t row_size, col_size;
- uint8_t *image_in =
- read_image_from_binary(args.args[RAW_IMAGE_BIN], &row_size, &col_size);
-
- printf("Raw image shape: %d x %d x %d\n", row_size, col_size, CHAN_SIZE);
-
- // Allocate a buffer for storing the output image data.
- // (This is currently the same size as the input image data.)
- size_t bytes_image = sizeof(uint8_t) * row_size * col_size * CHAN_SIZE;
- size_t bytes_fimage = sizeof(float) * row_size * col_size * CHAN_SIZE;
- uint8_t *image_out = (uint8_t *)malloc_aligned(bytes_image);
- uint8_t *image_out_gamut = (uint8_t *)malloc_aligned(bytes_image);
- uint8_t *image_out_demosaic = (uint8_t *)malloc_aligned(bytes_image);
- uint8_t *image_out_denoise = (uint8_t *)malloc_aligned(bytes_image);
- uint8_t *image_out_transform = (uint8_t *)malloc_aligned(bytes_image);
-
- __hpvm__init();
-
- ///////////////////////////////////////////////////////////////
- // Camera Model Parameters
- ///////////////////////////////////////////////////////////////
- // Path to the camera model to be used
- // char cam_model_path[100];
- // char cam_model_path = "cam_models/NikonD7000/";
- // White balance index (select white balance from transform file)
- // The first white balance in the file has a wb_index of 1
- // For more information on model format see the readme
- int wb_index = 6;
-
- // Number of control points
- int num_ctrl_pts = 3702;
- uint8_t *input, *result;
- float *input_scaled, *result_scaled, *demosaic_out, *denoise_out,
- *transform_out, *gamut_out;
- float *TsTw, *ctrl_pts, *weights, *coefs, *tone_map, *l2_dist;
-
- TsTw = get_TsTw("cam_models/NikonD7000/", wb_index);
- float *trans = transpose_mat(TsTw, CHAN_SIZE, CHAN_SIZE);
- free(TsTw);
- TsTw = trans;
- ctrl_pts = get_ctrl_pts("cam_models/NikonD7000/", num_ctrl_pts);
- weights = get_weights("cam_models/NikonD7000/", num_ctrl_pts);
- coefs = get_coefs("cam_models/NikonD7000/", num_ctrl_pts);
- tone_map = get_tone_map("cam_models/NikonD7000/");
-
- input_scaled = (float *)malloc_aligned(bytes_fimage);
- result_scaled = (float *)malloc_aligned(bytes_fimage);
- demosaic_out = (float *)malloc_aligned(bytes_fimage);
- denoise_out = (float *)malloc_aligned(bytes_fimage);
- transform_out = (float *)malloc_aligned(bytes_fimage);
- gamut_out = (float *)malloc_aligned(bytes_fimage);
- l2_dist = (float *)malloc_aligned(sizeof(float) * num_ctrl_pts);
-
- // This is host_input in cam_pipe()
- input = (uint8_t *)malloc_aligned(bytes_image);
- convert_hwc_to_chw(image_in, row_size, col_size, &input);
-
- // This is host_result in cam_pipe()
- result = (uint8_t *)malloc_aligned(bytes_image);
-
- // Allocate struct to pass DFG inputs
- RootIn *rootArgs = (RootIn *)malloc(sizeof(RootIn));
-
- // Set up HPVM DFG inputs in the rootArgs struct.
- rootArgs->input = input;
- rootArgs->bytes_input = bytes_image;
-
- rootArgs->result = result;
- rootArgs->bytes_result = bytes_image;
-
- rootArgs->input_scaled = input_scaled;
- rootArgs->bytes_input_scaled = bytes_fimage;
-
- rootArgs->result_scaled = result_scaled;
- rootArgs->bytes_result_scaled = bytes_fimage;
-
- rootArgs->demosaic_out = demosaic_out;
- rootArgs->bytes_demosaic_out = bytes_fimage;
-
- rootArgs->denoise_out = denoise_out;
- rootArgs->bytes_denoise_out = bytes_fimage;
-
- rootArgs->transform_out = transform_out;
- rootArgs->bytes_transform_out = bytes_fimage;
-
- rootArgs->gamut_out = gamut_out;
- rootArgs->bytes_gamut_out = bytes_fimage;
-
- rootArgs->TsTw = TsTw;
- rootArgs->bytes_TsTw = CHAN_SIZE * CHAN_SIZE * sizeof(float);
-
- rootArgs->ctrl_pts = ctrl_pts;
- rootArgs->bytes_ctrl_pts = num_ctrl_pts * CHAN_SIZE * sizeof(float);
-
- rootArgs->weights = weights;
- rootArgs->bytes_weights = num_ctrl_pts * CHAN_SIZE * sizeof(float);
-
- rootArgs->coefs = coefs;
- rootArgs->bytes_coefs = 4 * CHAN_SIZE * sizeof(float);
-
- rootArgs->tone_map = tone_map;
- rootArgs->bytes_tone_map = 256 * CHAN_SIZE * sizeof(float);
-
- rootArgs->l2_dist = l2_dist;
- rootArgs->bytes_l2_dist = num_ctrl_pts * sizeof(float);
-
- rootArgs->row_size = row_size;
- rootArgs->col_size = col_size;
-
- // Memory tracking is required for pointer arguments.
- // Nodes can be scheduled on different targets, and
- // dataflow edge implementation needs to request data.
- // The pair (pointer, size) is inserted in memory tracker using this call
- llvm_hpvm_track_mem(input, bytes_image);
- llvm_hpvm_track_mem(result, bytes_image);
- llvm_hpvm_track_mem(input_scaled, bytes_fimage);
- llvm_hpvm_track_mem(result_scaled, bytes_fimage);
- llvm_hpvm_track_mem(demosaic_out, bytes_fimage);
- llvm_hpvm_track_mem(denoise_out, bytes_fimage);
- llvm_hpvm_track_mem(transform_out, bytes_fimage);
- llvm_hpvm_track_mem(gamut_out, bytes_fimage);
- llvm_hpvm_track_mem(TsTw, CHAN_SIZE * CHAN_SIZE * sizeof(float));
- llvm_hpvm_track_mem(ctrl_pts, num_ctrl_pts * CHAN_SIZE * sizeof(float));
- llvm_hpvm_track_mem(weights, num_ctrl_pts * CHAN_SIZE * sizeof(float));
- llvm_hpvm_track_mem(coefs, 4 * CHAN_SIZE * sizeof(float));
- llvm_hpvm_track_mem(tone_map, 256 * CHAN_SIZE * sizeof(float));
- llvm_hpvm_track_mem(l2_dist, num_ctrl_pts * sizeof(float));
-
- printf("\n\nLaunching CAVA pipeline!\n");
-
- void *camPipeDFG = __hpvm__launch(0, CamPipeRoot, (void *)rootArgs);
- __hpvm__wait(camPipeDFG);
-
- printf("\n\nPipeline execution completed!\n");
- printf("Pipeline final stage returned %lu; should be %lu\n",
- rootArgs->ret.bytesRet, bytes_image);
- printf("\n\nRequesting memory!\n");
-
- // Request data from graph.
- llvm_hpvm_request_mem(result, bytes_image);
- llvm_hpvm_request_mem(demosaic_out, bytes_fimage);
- llvm_hpvm_request_mem(denoise_out, bytes_fimage);
- llvm_hpvm_request_mem(transform_out, bytes_fimage);
- llvm_hpvm_request_mem(gamut_out, bytes_fimage);
- printf("\n\nDone requesting memory!\n");
-
- uint8_t *gamut_out_descaled = (uint8_t *)malloc_aligned(bytes_image);
- uint8_t *demosaic_out_descaled = (uint8_t *)malloc_aligned(bytes_image);
- uint8_t *transform_out_descaled = (uint8_t *)malloc_aligned(bytes_image);
- uint8_t *denoise_out_descaled = (uint8_t *)malloc_aligned(bytes_image);
-
- descale_cpu(demosaic_out, bytes_fimage, demosaic_out_descaled, bytes_image,
- row_size, col_size);
- descale_cpu(gamut_out, bytes_fimage, gamut_out_descaled, bytes_image,
- row_size, col_size);
- descale_cpu(denoise_out, bytes_fimage, denoise_out_descaled, bytes_image,
- row_size, col_size);
- descale_cpu(transform_out, bytes_fimage, transform_out_descaled, bytes_image,
- row_size, col_size);
-
- convert_chw_to_hwc(result, row_size, col_size, &image_out);
- convert_chw_to_hwc(gamut_out_descaled, row_size, col_size, &image_out_gamut);
- convert_chw_to_hwc(demosaic_out_descaled, row_size, col_size,
- &image_out_demosaic);
- convert_chw_to_hwc(denoise_out_descaled, row_size, col_size,
- &image_out_denoise);
- convert_chw_to_hwc(transform_out_descaled, row_size, col_size,
- &image_out_transform);
-
- // Remove tracked pointers.
- llvm_hpvm_untrack_mem(input);
- llvm_hpvm_untrack_mem(result);
- llvm_hpvm_untrack_mem(input_scaled);
- llvm_hpvm_untrack_mem(result_scaled);
- llvm_hpvm_untrack_mem(demosaic_out);
- llvm_hpvm_untrack_mem(denoise_out);
- llvm_hpvm_untrack_mem(transform_out);
- llvm_hpvm_untrack_mem(gamut_out);
-
- llvm_hpvm_untrack_mem(TsTw);
- llvm_hpvm_untrack_mem(ctrl_pts);
- llvm_hpvm_untrack_mem(weights);
- llvm_hpvm_untrack_mem(coefs);
- llvm_hpvm_untrack_mem(tone_map);
- llvm_hpvm_untrack_mem(l2_dist);
-
- // Output the image.
- // NOTE: We deliberately perform this file I/O outside of the kernel.
+int main(int argc, char* argv[]) {
+ // Parse the arguments.
+ arguments args;
+ set_default_args(&args);
+ argp_parse(&parser, argc, argv, 0, 0, &args);
+
+ // Read a raw image.
+ // NOTE: We deliberately perform this file I/O outside of the kernel.
+ printf("Reading a raw image from %s\n", args.args[RAW_IMAGE_BIN]);
+ size_t row_size, col_size;
+ uint8_t *image_in = read_image_from_binary(args.args[RAW_IMAGE_BIN], &row_size, &col_size);
+
+ printf("Raw image shape: %d x %d x %d\n", row_size, col_size, CHAN_SIZE);
+
+ // Allocate a buffer for storing the output image data.
+ // (This is currently the same size as the input image data.)
+ size_t bytes_image = sizeof(uint8_t) * row_size * col_size * CHAN_SIZE;
+ size_t bytes_fimage = sizeof(float) * row_size * col_size * CHAN_SIZE;
+ uint8_t *image_out = (uint8_t*) malloc_aligned(bytes_image);
+ uint8_t *image_out_gamut = (uint8_t*) malloc_aligned(bytes_image);
+ uint8_t *image_out_demosaic = (uint8_t*) malloc_aligned(bytes_image);
+ uint8_t *image_out_denoise = (uint8_t*) malloc_aligned(bytes_image);
+ uint8_t *image_out_transform = (uint8_t*) malloc_aligned(bytes_image);
+
+ __visc__init();
+
+ ///////////////////////////////////////////////////////////////
+ // Camera Model Parameters
+ ///////////////////////////////////////////////////////////////
+ // Path to the camera model to be used
+// char cam_model_path[100];
+// char cam_model_path = "cam_models/NikonD7000/";
+ // White balance index (select white balance from transform file)
+ // The first white balance in the file has a wb_index of 1
+ // For more information on model format see the readme
+ int wb_index = 6;
+
+ // Number of control points
+ int num_ctrl_pts = 3702;
+ uint8_t *input, *result;
+ float *input_scaled, *result_scaled, *demosaic_out, *denoise_out, *transform_out, *gamut_out;
+ float *TsTw, *ctrl_pts, *weights, *coefs, *tone_map, *l2_dist;
+
+ TsTw = get_TsTw("cam_models/NikonD7000/", wb_index);
+ float *trans = transpose_mat(TsTw, CHAN_SIZE, CHAN_SIZE);
+ free(TsTw);
+ TsTw = trans;
+ ctrl_pts = get_ctrl_pts("cam_models/NikonD7000/", num_ctrl_pts);
+ weights = get_weights("cam_models/NikonD7000/", num_ctrl_pts);
+ coefs = get_coefs("cam_models/NikonD7000/", num_ctrl_pts);
+ tone_map = get_tone_map("cam_models/NikonD7000/");
+
+ input_scaled = (float*) malloc_aligned(bytes_fimage);
+ result_scaled = (float*) malloc_aligned(bytes_fimage);
+ demosaic_out = (float*) malloc_aligned(bytes_fimage);
+ denoise_out = (float*) malloc_aligned(bytes_fimage);
+ transform_out = (float*) malloc_aligned(bytes_fimage);
+ gamut_out = (float*) malloc_aligned(bytes_fimage);
+ l2_dist = (float*) malloc_aligned(sizeof(float) * num_ctrl_pts);
+
+ // This is host_input in cam_pipe()
+ input = (uint8_t*) malloc_aligned(bytes_image);
+ convert_hwc_to_chw(image_in, row_size, col_size, &input);
+
+ // This is host_result in cam_pipe()
+ result = (uint8_t*) malloc_aligned(bytes_image);
+
+ // Allocate struct to pass DFG inputs
+ RootIn* rootArgs = (RootIn*) malloc(sizeof(RootIn));
+
+ // Set up HPVM DFG inputs in the rootArgs struct.
+ rootArgs->input = input;
+ rootArgs->bytes_input = bytes_image;
+
+ rootArgs->result = result;
+ rootArgs->bytes_result = bytes_image;
+
+ rootArgs->input_scaled = input_scaled;
+ rootArgs->bytes_input_scaled = bytes_fimage;
+
+ rootArgs->result_scaled = result_scaled;
+ rootArgs->bytes_result_scaled = bytes_fimage;
+
+ rootArgs->demosaic_out = demosaic_out;
+ rootArgs->bytes_demosaic_out = bytes_fimage;
+
+ rootArgs->denoise_out = denoise_out;
+ rootArgs->bytes_denoise_out = bytes_fimage;
+
+ rootArgs->transform_out = transform_out;
+ rootArgs->bytes_transform_out = bytes_fimage;
+
+ rootArgs->gamut_out = gamut_out;
+ rootArgs->bytes_gamut_out = bytes_fimage;
+
+ rootArgs->TsTw = TsTw;
+ rootArgs->bytes_TsTw = CHAN_SIZE * CHAN_SIZE * sizeof(float);
+
+ rootArgs->ctrl_pts = ctrl_pts;
+ rootArgs->bytes_ctrl_pts = num_ctrl_pts * CHAN_SIZE * sizeof(float);
+
+ rootArgs->weights = weights;
+ rootArgs->bytes_weights = num_ctrl_pts * CHAN_SIZE * sizeof(float);
+
+ rootArgs->coefs = coefs;
+ rootArgs->bytes_coefs = 4 * CHAN_SIZE * sizeof(float);
+
+ rootArgs->tone_map = tone_map;
+ rootArgs->bytes_tone_map = 256 * CHAN_SIZE * sizeof(float);
+
+ rootArgs->l2_dist = l2_dist;
+ rootArgs->bytes_l2_dist = num_ctrl_pts * sizeof(float);
+
+ rootArgs->row_size = row_size;
+ rootArgs->col_size = col_size;
+
+ // Memory tracking is required for pointer arguments.
+ // Nodes can be scheduled on different targets, and
+ // dataflow edge implementation needs to request data.
+ // The pair (pointer, size) is inserted in memory tracker using this call
+ llvm_visc_track_mem(input, bytes_image);
+ llvm_visc_track_mem(result, bytes_image);
+ llvm_visc_track_mem(input_scaled, bytes_fimage);
+ llvm_visc_track_mem(result_scaled, bytes_fimage);
+ llvm_visc_track_mem(demosaic_out, bytes_fimage);
+ llvm_visc_track_mem(denoise_out, bytes_fimage);
+ llvm_visc_track_mem(transform_out, bytes_fimage);
+ llvm_visc_track_mem(gamut_out, bytes_fimage);
+ llvm_visc_track_mem(TsTw, CHAN_SIZE * CHAN_SIZE * sizeof(float));
+ llvm_visc_track_mem(ctrl_pts, num_ctrl_pts * CHAN_SIZE * sizeof(float));
+ llvm_visc_track_mem(weights, num_ctrl_pts * CHAN_SIZE * sizeof(float));
+ llvm_visc_track_mem(coefs, 4 * CHAN_SIZE *sizeof(float));
+ llvm_visc_track_mem(tone_map, 256 * CHAN_SIZE * sizeof(float));
+ llvm_visc_track_mem(l2_dist, num_ctrl_pts * sizeof(float));
+
+ printf("\n\nLaunching CAVA pipeline!\n");
+
+ void* camPipeDFG = __visc__launch(0, CamPipeRoot, (void*) rootArgs);
+ __visc__wait(camPipeDFG);
+
+ printf("\n\nPipeline execution completed!\n");
+ printf("\n\nRequesting memory!\n");
+
+ // Request data from graph.
+ llvm_visc_request_mem(result, bytes_image);
+ llvm_visc_request_mem(demosaic_out, bytes_fimage);
+ llvm_visc_request_mem(denoise_out, bytes_fimage);
+ llvm_visc_request_mem(transform_out, bytes_fimage);
+ llvm_visc_request_mem(gamut_out, bytes_fimage);
+ printf("\n\nDone requesting memory!\n");
+
+
+ uint8_t* gamut_out_descaled = (uint8_t*) malloc_aligned(bytes_image);
+ uint8_t* demosaic_out_descaled = (uint8_t*) malloc_aligned(bytes_image);
+ uint8_t* transform_out_descaled = (uint8_t*) malloc_aligned(bytes_image);
+ uint8_t* denoise_out_descaled = (uint8_t*) malloc_aligned(bytes_image);
+
+ descale_cpu(demosaic_out, bytes_fimage, demosaic_out_descaled, bytes_image, row_size, col_size);
+ descale_cpu(gamut_out, bytes_fimage, gamut_out_descaled, bytes_image, row_size, col_size);
+ descale_cpu(denoise_out, bytes_fimage, denoise_out_descaled, bytes_image, row_size, col_size);
+ descale_cpu(transform_out, bytes_fimage, transform_out_descaled, bytes_image, row_size, col_size);
+
+ convert_chw_to_hwc(result, row_size, col_size, &image_out);
+ convert_chw_to_hwc(gamut_out_descaled, row_size, col_size, &image_out_gamut);
+ convert_chw_to_hwc(demosaic_out_descaled, row_size, col_size, &image_out_demosaic);
+ convert_chw_to_hwc(denoise_out_descaled, row_size, col_size, &image_out_denoise);
+ convert_chw_to_hwc(transform_out_descaled, row_size, col_size, &image_out_transform);
+
+
+ // Remove tracked pointers.
+ llvm_visc_untrack_mem(input);
+ llvm_visc_untrack_mem(result);
+ llvm_visc_untrack_mem(input_scaled);
+ llvm_visc_untrack_mem(result_scaled);
+ llvm_visc_untrack_mem(demosaic_out);
+ llvm_visc_untrack_mem(denoise_out);
+ llvm_visc_untrack_mem(transform_out);
+ llvm_visc_untrack_mem(gamut_out);
+
+ llvm_visc_untrack_mem(TsTw);
+ llvm_visc_untrack_mem(ctrl_pts);
+ llvm_visc_untrack_mem(weights);
+ llvm_visc_untrack_mem(coefs);
+ llvm_visc_untrack_mem(tone_map);
+ llvm_visc_untrack_mem(l2_dist);
+
+ // Output the image.
+ // NOTE: We deliberately perform this file I/O outside of the kernel.
char str[50], base_str[50];
strcpy(base_str, args.args[OUTPUT_IMAGE_BIN]);
strcpy(str, base_str);
@@ -912,7 +877,8 @@ int main(int argc, char *argv[]) {
printf("Writing output image to %s\n", str);
write_image_to_binary(str, image_out_transform, row_size, col_size);
- __hpvm__cleanup();
+ __visc__cleanup();
- return 0;
+ return 0;
}
+
diff --git a/hpvm/test/hpvm-cava/src/pipe_stages.c b/hpvm/test/hpvm-cava/src/pipe_stages.c
index 05bb06697f..2ebedec936 100644
--- a/hpvm/test/hpvm-cava/src/pipe_stages.c
+++ b/hpvm/test/hpvm-cava/src/pipe_stages.c
@@ -1,169 +1,172 @@
+#include <stdio.h>
+#include <math.h>
#include "pipe_stages.h"
#include "cam_pipe_utility.h"
-#include <math.h>
-#include <stdio.h>
-
-// void scale_fxp(uint8_t *input, int row_size, int col_size, float *output) {
-void scale_fxp(uint8_t *input, size_t bytes_input, float *output,
- size_t bytes_output, int row_size, int col_size) {
- __hpvm__hint(DEVICE);
- __hpvm__attributes(2, input, output, 1, output);
+//void scale_fxp(uint8_t *input, int row_size, int col_size, float *output) {
+void scale_fxp(uint8_t *input, size_t bytes_input,
+ float *output, size_t bytes_output,
+ int row_size, int col_size) {
+ __visc__hint(DEVICE);
+ __visc__attributes(2, input, output, 1, output);
+
ARRAY_3D(uint8_t, _input, input, row_size, col_size);
ARRAY_3D(float, _output, output, row_size, col_size);
-sl_chan:
+ sl_chan:
for (int chan = 0; chan < CHAN_SIZE; chan++)
- sl_row:
+ sl_row:
for (int row = 0; row < row_size; row++)
- sl_col:
+ sl_col:
for (int col = 0; col < col_size; col++)
_output[chan][row][col] = _input[chan][row][col] * 1.0 / 255;
- __hpvm__return(1, bytes_output);
+ __visc__return(1, bytes_output);
}
-// void descale_fxp(float *input, int row_size, int col_size, uint8_t *output) {
-void descale_fxp(float *input, size_t bytes_input, uint8_t *output,
- size_t bytes_result, int row_size, int col_size) {
- __hpvm__hint(DEVICE);
- __hpvm__attributes(2, input, output, 1, output);
-
+//void descale_fxp(float *input, int row_size, int col_size, uint8_t *output) {
+void descale_fxp(float *input, size_t bytes_input,
+ uint8_t *output, size_t bytes_result,
+ int row_size, int col_size) {
+ __visc__hint(DEVICE);
+ __visc__attributes(2, input, output, 1, output);
+
ARRAY_3D(float, _input, input, row_size, col_size);
ARRAY_3D(uint8_t, _output, output, row_size, col_size);
-dsl_chan:
+ dsl_chan:
for (int chan = 0; chan < CHAN_SIZE; chan++)
- dsl_row:
+ dsl_row:
for (int row = 0; row < row_size; row++)
- dsl_col:
+ dsl_col:
for (int col = 0; col < col_size; col++)
- _output[chan][row][col] =
- min(max(_input[chan][row][col] * 255, 0), 255);
+ _output[chan][row][col] = min(max(_input[chan][row][col] * 255, 0), 255);
- __hpvm__return(1, bytes_output);
+ __visc__return(1, bytes_output);
}
// Demosaicing stage
// G R
// B G
-// void demosaic_fxp(float *input, int row_size, int col_size, float *result) {
-void demosaic_fxp(float *input, size_t bytes_input, float *result,
- size_t bytes_result, int row_size, int col_size) {
- __hpvm__hint(DEVICE);
- __hpvm__attributes(2, input, result, 1, result);
-
+//void demosaic_fxp(float *input, int row_size, int col_size, float *result) {
+void demosaic_fxp(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ int row_size, int col_size) {
+ __visc__hint(DEVICE);
+ __visc__attributes(2, input, result, 1, result);
+
printf("Demosaicing.\n");
ARRAY_3D(float, _input, input, row_size, col_size);
ARRAY_3D(float, _result, result, row_size, col_size);
-dm_row:
+ dm_row:
for (int row = 1; row < row_size - 1; row++)
- dm_col:
+ dm_col:
for (int col = 1; col < col_size - 1; col++)
- if (row % 2 == 0 && col % 2 == 0) {
- // Green pixel
- // Getting the R values
- float R1 = _input[0][row][col - 1];
- float R2 = _input[0][row][col + 1];
- // Getting the B values
- float B1 = _input[2][row - 1][col];
- float B2 = _input[2][row + 1][col];
- // R
- _result[0][row][col] = (R1 + R2) / 2;
- // G
- _result[1][row][col] = _input[1][row][col] * 2;
- // B
- _result[2][row][col] = (B1 + B2) / 2;
- } else if (row % 2 == 0 && col % 2 == 1) {
- // Red pixel
- // Getting the G values
- float G1 = _input[1][row - 1][col];
- float G2 = _input[1][row + 1][col];
- float G3 = _input[1][row][col - 1];
- float G4 = _input[1][row][col + 1];
- // Getting the B values
- float B1 = _input[2][row - 1][col - 1];
- float B2 = _input[2][row - 1][col + 1];
- float B3 = _input[2][row + 1][col - 1];
- float B4 = _input[2][row + 1][col + 1];
- // R
- _result[0][row][col] = _input[0][row][col];
- // G
- _result[1][row][col] = (G1 + G2 + G3 + G4) / 2;
- // B (center pixel)
- _result[2][row][col] = (B1 + B2 + B3 + B4) / 4;
- } else if (row % 2 == 1 && col % 2 == 0) {
- // Blue pixel
- // Getting the R values
- float R1 = _input[0][row - 1][col - 1];
- float R2 = _input[0][row + 1][col - 1];
- float R3 = _input[0][row - 1][col + 1];
- float R4 = _input[0][row + 1][col + 1];
- // Getting the G values
- float G1 = _input[1][row - 1][col];
- float G2 = _input[1][row + 1][col];
- float G3 = _input[1][row][col - 1];
- float G4 = _input[1][row][col + 1];
- // R
- _result[0][row][col] = (R1 + R2 + R3 + R4) / 4;
- // G
- _result[1][row][col] = (G1 + G2 + G3 + G4) / 2;
- // B
- _result[2][row][col] = _input[2][row][col];
- } else {
- // Bottom Green pixel
- // Getting the R values
- float R1 = _input[0][row - 1][col];
- float R2 = _input[0][row + 1][col];
- // Getting the B values
- float B1 = _input[2][row][col - 1];
- float B2 = _input[2][row][col + 1];
- // R
- _result[0][row][col] = (R1 + R2) / 2;
- // G
- _result[1][row][col] = _input[1][row][col] * 2;
- // B
- _result[2][row][col] = (B1 + B2) / 2;
- }
+ if (row % 2 == 0 && col % 2 == 0) {
+ // Green pixel
+ // Getting the R values
+ float R1 = _input[0][row][col - 1];
+ float R2 = _input[0][row][col + 1];
+ // Getting the B values
+ float B1 = _input[2][row - 1][col];
+ float B2 = _input[2][row + 1][col];
+ // R
+ _result[0][row][col] = (R1 + R2) / 2;
+ // G
+ _result[1][row][col] = _input[1][row][col] * 2;
+ // B
+ _result[2][row][col] = (B1 + B2) / 2;
+ } else if (row % 2 == 0 && col % 2 == 1) {
+ // Red pixel
+ // Getting the G values
+ float G1 = _input[1][row - 1][col];
+ float G2 = _input[1][row + 1][col];
+ float G3 = _input[1][row][col - 1];
+ float G4 = _input[1][row][col + 1];
+ // Getting the B values
+ float B1 = _input[2][row - 1][col - 1];
+ float B2 = _input[2][row - 1][col + 1];
+ float B3 = _input[2][row + 1][col - 1];
+ float B4 = _input[2][row + 1][col + 1];
+ // R
+ _result[0][row][col] = _input[0][row][col];
+ // G
+ _result[1][row][col] = (G1 + G2 + G3 + G4) / 2;
+ // B (center pixel)
+ _result[2][row][col] = (B1 + B2 + B3 + B4) / 4;
+ } else if (row % 2 == 1 && col % 2 == 0) {
+ // Blue pixel
+ // Getting the R values
+ float R1 = _input[0][row - 1][col - 1];
+ float R2 = _input[0][row + 1][col - 1];
+ float R3 = _input[0][row - 1][col + 1];
+ float R4 = _input[0][row + 1][col + 1];
+ // Getting the G values
+ float G1 = _input[1][row - 1][col];
+ float G2 = _input[1][row + 1][col];
+ float G3 = _input[1][row][col - 1];
+ float G4 = _input[1][row][col + 1];
+ // R
+ _result[0][row][col] = (R1 + R2 + R3 + R4) / 4;
+ // G
+ _result[1][row][col] = (G1 + G2 + G3 + G4) / 2;
+ // B
+ _result[2][row][col] = _input[2][row][col];
+ } else {
+ // Bottom Green pixel
+ // Getting the R values
+ float R1 = _input[0][row - 1][col];
+ float R2 = _input[0][row + 1][col];
+ // Getting the B values
+ float B1 = _input[2][row][col - 1];
+ float B2 = _input[2][row][col + 1];
+ // R
+ _result[0][row][col] = (R1 + R2) / 2;
+ // G
+ _result[1][row][col] = _input[1][row][col] * 2;
+ // B
+ _result[2][row][col] = (B1 + B2) / 2;
+ }
- __hpvm__return(1, bytes_result);
+ __visc__return(1, bytes_result);
}
static void sort(float arr[], int n) {
- int i, j;
-dn_sort_i:
- for (i = 0; i < n - 1; i++)
- dn_sort_j:
- for (j = 0; j < n - i - 1; j++)
- if (arr[j] > arr[j + 1]) {
- float temp = arr[j];
- arr[j] = arr[j + 1];
- arr[j + 1] = temp;
- }
+ int i, j;
+ dn_sort_i:
+ for (i = 0; i < n - 1; i++)
+ dn_sort_j:
+ for (j = 0; j < n - i - 1; j++)
+ if (arr[j] > arr[j + 1]) {
+ float temp = arr[j];
+ arr[j] = arr[j + 1];
+ arr[j + 1] = temp;
+ }
}
// Simple denoise
-// void denoise_fxp(float *input, int row_size, int col_size, float *result) {
-void denoise_fxp(float *input, size_t bytes_input, float *result,
- size_t bytes_result, int row_size, int col_size) {
- __hpvm__hint(DEVICE);
- __hpvm__attributes(2, input, result, 1, result);
-
+//void denoise_fxp(float *input, int row_size, int col_size, float *result) {
+void denoise_fxp(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ int row_size, int col_size) {
+ __visc__hint(DEVICE);
+ __visc__attributes(2, input, result, 1, result);
+
printf("Denoising.\n");
ARRAY_3D(float, _input, input, row_size, col_size);
ARRAY_3D(float, _result, result, row_size, col_size);
-dn_chan:
+ dn_chan:
for (int chan = 0; chan < CHAN_SIZE; chan++)
- dn_row:
+ dn_row:
for (int row = 0; row < row_size; row++)
- dn_col:
+ dn_col:
for (int col = 0; col < col_size; col++)
if (row >= 1 && row < row_size - 1 && col >= 1 && col < col_size - 1) {
float filter[9];
- dn_slide_row:
- for (int i = row - 1; i < row + 2; i++)
- dn_slide_col:
- for (int j = col - 1; j < col + 2; j++) {
+ dn_slide_row:
+ for (int i = row-1; i < row+2; i++)
+ dn_slide_col:
+ for (int j = col-1; j < col+2; j++) {
int index = (i - row + 1) * 3 + j - col + 1;
filter[index] = _input[chan][i][j];
}
@@ -172,52 +175,53 @@ dn_chan:
} else {
_result[chan][row][col] = _input[chan][row][col];
}
- __hpvm__return(1, bytes_result);
+ __visc__return(1, bytes_result);
}
// Color map and white balance transform
-// void transform_fxp(float *input, int row_size, int col_size, float *result,
+//void transform_fxp(float *input, int row_size, int col_size, float *result,
// float *TsTw_tran) {
-void transform_fxp(float *input, size_t bytes_input, float *result,
- size_t bytes_result, float *TsTw_tran, size_t bytes_TsTw,
+void transform_fxp(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ float *TsTw_tran, size_t bytes_TsTw,
int row_size, int col_size) {
- __hpvm__hint(DEVICE);
- __hpvm__attributes(3, input, result, TsTw_tran, 1, result);
-
+ __visc__hint(DEVICE);
+ __visc__attributes(3, input, result, TsTw_tran, 1, result);
+
printf("Color mapping.\n");
ARRAY_3D(float, _input, input, row_size, col_size);
ARRAY_3D(float, _result, result, row_size, col_size);
ARRAY_2D(float, _TsTw_tran, TsTw_tran, 3);
-tr_chan:
+ tr_chan:
for (int chan = 0; chan < CHAN_SIZE; chan++)
- tr_row:
+ tr_row:
for (int row = 0; row < row_size; row++)
- tr_col:
+ tr_col:
for (int col = 0; col < col_size; col++)
_result[chan][row][col] =
max(_input[0][row][col] * _TsTw_tran[0][chan] +
_input[1][row][col] * _TsTw_tran[1][chan] +
_input[2][row][col] * _TsTw_tran[2][chan],
0);
- __hpvm__return(1, bytes_result);
+ __visc__return(1, bytes_result);
}
//
// Weighted radial basis function for gamut mapping
//
-// void gamut_map_fxp(float *input, int row_size, int col_size, float *result,
-// float *ctrl_pts, float *weights, float *coefs, float
-// *l2_dist) {
-void gamut_map_fxp(float *input, size_t bytes_input, float *result,
- size_t bytes_result, float *ctrl_pts, size_t bytes_ctrl_pts,
- float *weights, size_t bytes_weights, float *coefs,
- size_t bytes_coefs, float *l2_dist, size_t bytes_l2_dist,
+//void gamut_map_fxp(float *input, int row_size, int col_size, float *result,
+// float *ctrl_pts, float *weights, float *coefs, float *l2_dist) {
+void gamut_map_fxp(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ float *ctrl_pts, size_t bytes_ctrl_pts,
+ float *weights, size_t bytes_weights,
+ float *coefs, size_t bytes_coefs,
+ float *l2_dist, size_t bytes_l2_dist,
int row_size, int col_size) {
- __hpvm__hint(DEVICE);
- __hpvm__attributes(6, input, result, ctrl_pts, weights, coefs, l2_dist, 1,
- result);
-
+ __visc__hint(DEVICE);
+ __visc__attributes(6, input, result, ctrl_pts, weights, coefs, l2_dist, 1, result);
+
printf("Gamut mapping.\n");
ARRAY_3D(float, _input, input, row_size, col_size);
ARRAY_3D(float, _result, result, row_size, col_size);
@@ -225,25 +229,26 @@ void gamut_map_fxp(float *input, size_t bytes_input, float *result,
ARRAY_2D(float, _weights, weights, 3);
ARRAY_2D(float, _coefs, coefs, 3);
-// First, get the L2 norm from every pixel to the control points,
-// Then, sum it and weight it. Finally, add the bias.
-gm_rbf_row:
+ // First, get the L2 norm from every pixel to the control points,
+ // Then, sum it and weight it. Finally, add the bias.
+ gm_rbf_row:
for (int row = 0; row < row_size; row++)
- gm_rbf_col:
+ gm_rbf_col:
for (int col = 0; col < col_size; col++) {
- gm_rbf_cp0:
+ gm_rbf_cp0:
for (int cp = 0; cp < num_ctrl_pts; cp++) {
- l2_dist[cp] = sqrt((_input[0][row][col] - _ctrl_pts[cp][0]) *
- (_input[0][row][col] - _ctrl_pts[cp][0]) +
- (_input[1][row][col] - _ctrl_pts[cp][1]) *
- (_input[1][row][col] - _ctrl_pts[cp][1]) +
- (_input[2][row][col] - _ctrl_pts[cp][2]) *
- (_input[2][row][col] - _ctrl_pts[cp][2]));
+ l2_dist[cp] =
+ sqrt((_input[0][row][col] - _ctrl_pts[cp][0]) *
+ (_input[0][row][col] - _ctrl_pts[cp][0]) +
+ (_input[1][row][col] - _ctrl_pts[cp][1]) *
+ (_input[1][row][col] - _ctrl_pts[cp][1]) +
+ (_input[2][row][col] - _ctrl_pts[cp][2]) *
+ (_input[2][row][col] - _ctrl_pts[cp][2]));
}
- gm_rbf_chan:
+ gm_rbf_chan:
for (int chan = 0; chan < CHAN_SIZE; chan++) {
float chan_val = 0.0;
- gm_rbf_cp1:
+ gm_rbf_cp1:
for (int cp = 0; cp < num_ctrl_pts; cp++) {
chan_val += l2_dist[cp] * _weights[cp][chan];
}
@@ -254,31 +259,32 @@ gm_rbf_row:
_result[chan][row][col] = max(chan_val, 0);
}
}
- __hpvm__return(1, bytes_result);
+ __visc__return(1, bytes_result);
}
// Tone mapping
-// void tone_map_fxp(float *input, int row_size, int col_size, float *tone_map,
+//void tone_map_fxp(float *input, int row_size, int col_size, float *tone_map,
// float *result) {
-void tone_map_fxp(float *input, size_t bytes_input, float *result,
- size_t bytes_result, float *tone_map, size_t bytes_tone_map,
+void tone_map_fxp(float *input, size_t bytes_input,
+ float *result, size_t bytes_result,
+ float *tone_map, size_t bytes_tone_map,
int row_size, int col_size) {
- __hpvm__hint(DEVICE);
- __hpvm__attributes(3, input, result, tone_map, 1, result);
-
+ __visc__hint(DEVICE);
+ __visc__attributes(3, input, result, tone_map, 1, result);
+
printf("Tone mapping.\n");
ARRAY_3D(float, _input, input, row_size, col_size);
ARRAY_3D(float, _result, result, row_size, col_size);
ARRAY_2D(float, _tone_map, tone_map, 3);
-tm_chan:
+ tm_chan:
for (int chan = 0; chan < CHAN_SIZE; chan++)
- tm_row:
+ tm_row:
for (int row = 0; row < row_size; row++)
- tm_col:
+ tm_col:
for (int col = 0; col < col_size; col++) {
uint8_t x = _input[chan][row][col] * 255;
_result[chan][row][col] = _tone_map[x][chan];
}
- __hpvm__return(1, bytes_result);
+ __visc__return(1, bytes_result);
}
diff --git a/hpvm/test/include/hpvm.h b/hpvm/test/include/hpvm.h
deleted file mode 100644
index 1e31c98946..0000000000
--- a/hpvm/test/include/hpvm.h
+++ /dev/null
@@ -1,73 +0,0 @@
-/***************************************************************************
- *cr
- *cr (C) Copyright 2010 The Board of Trustees of the
- *cr University of Illinois
- *cr All Rights Reserved
- *cr
- ***************************************************************************/
-
-#ifndef DEVICE
-#define DEVICE GPU_TARGET
-#endif
-
-#include "../../include/SupportHPVM/HPVMHint.h"
-
-#ifndef __cplusplus
-#define noexcept
-#endif
-
-#ifdef __cplusplus
-extern "C" {
-void __hpvm__hint(hpvm::Target) noexcept;
-#else
-void __hpvm__hint(enum Target) noexcept;
-#endif
-
-void *__hpvm__createNodeND(unsigned, ...) noexcept;
-void __hpvm__return(unsigned, ...) noexcept;
-
-void __hpvm__attributes(unsigned, ...) noexcept;
-void __hpvm__init() noexcept;
-void __hpvm__cleanup() noexcept;
-
-void __hpvm__bindIn(void *, unsigned, unsigned, unsigned) noexcept;
-void __hpvm__bindOut(void *, unsigned, unsigned, unsigned) noexcept;
-void *__hpvm__edge(void *, void *, unsigned, unsigned, unsigned,
- unsigned) noexcept;
-
-void __hpvm__push(void *, void *) noexcept;
-void *__hpvm__pop(void *) noexcept;
-void *__hpvm__launch(unsigned, ...) noexcept;
-void __hpvm__wait(void *) noexcept;
-
-void *__hpvm__getNode() noexcept;
-void *__hpvm__getParentNode(void *) noexcept;
-void __hpvm__barrier() noexcept;
-void *__hpvm__malloc(long) noexcept;
-long __hpvm__getNodeInstanceID_x(void *) noexcept;
-long __hpvm__getNodeInstanceID_y(void *) noexcept;
-long __hpvm__getNodeInstanceID_z(void *) noexcept;
-long __hpvm__getNumNodeInstances_x(void *) noexcept;
-long __hpvm__getNumNodeInstances_y(void *) noexcept;
-long __hpvm__getNumNodeInstances_z(void *) noexcept;
-
-// Atomic
-// signed int
-int __hpvm__atomic_add(int *, int) noexcept;
-int __hpvm__atomic_sub(int *, int) noexcept;
-int __hpvm__atomic_xchg(int *, int) noexcept;
-int __hpvm__atomic_inc(int *) noexcept;
-int __hpvm__atomic_dec(int *) noexcept;
-int __hpvm__atomic_min(int *, int) noexcept;
-int __hpvm__atomic_max(int *, int) noexcept;
-int __hpvm__atomic_and(int *, int) noexcept;
-int __hpvm__atomic_or(int *, int) noexcept;
-int __hpvm__atomic_xor(int *, int) noexcept;
-
-void llvm_hpvm_track_mem(void *, size_t) noexcept;
-void llvm_hpvm_untrack_mem(void *) noexcept;
-void llvm_hpvm_request_mem(void *, size_t) noexcept;
-
-#ifdef __cplusplus
-}
-#endif
diff --git a/hpvm/test/include/visc.h b/hpvm/test/include/visc.h
new file mode 100644
index 0000000000..18b2950026
--- /dev/null
+++ b/hpvm/test/include/visc.h
@@ -0,0 +1,73 @@
+/***************************************************************************
+ *cr
+ *cr (C) Copyright 2010 The Board of Trustees of the
+ *cr University of Illinois
+ *cr All Rights Reserved
+ *cr
+ ***************************************************************************/
+
+#ifndef DEVICE
+#define DEVICE GPU_TARGET
+#endif
+
+#include "../../include/SupportVISC/VISCHint.h"
+
+#ifndef __cplusplus
+#define noexcept
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+void __visc__hint(visc::Target) noexcept;
+#else
+void __visc__hint(enum Target) noexcept;
+#endif
+
+void *__visc__createNodeND(unsigned, ...) noexcept;
+void __visc__return(unsigned, ...) noexcept;
+
+void __visc__attributes(unsigned, ...) noexcept;
+void __visc__init() noexcept;
+void __visc__cleanup() noexcept;
+
+void __visc__bindIn(void *, unsigned, unsigned, unsigned) noexcept;
+void __visc__bindOut(void *, unsigned, unsigned, unsigned) noexcept;
+void *__visc__edge(void *, void *, unsigned, unsigned, unsigned,
+ unsigned) noexcept;
+
+void __visc__push(void *, void *) noexcept;
+void *__visc__pop(void *) noexcept;
+void *__visc__launch(unsigned, ...) noexcept;
+void __visc__wait(void *) noexcept;
+
+void *__visc__getNode() noexcept;
+void *__visc__getParentNode(void *) noexcept;
+void __visc__barrier() noexcept;
+void *__visc__malloc(long) noexcept;
+long __visc__getNodeInstanceID_x(void *) noexcept;
+long __visc__getNodeInstanceID_y(void *) noexcept;
+long __visc__getNodeInstanceID_z(void *) noexcept;
+long __visc__getNumNodeInstances_x(void *) noexcept;
+long __visc__getNumNodeInstances_y(void *) noexcept;
+long __visc__getNumNodeInstances_z(void *) noexcept;
+
+// Atomic
+// signed int
+int __visc__atomic_add(int *, int) noexcept;
+int __visc__atomic_sub(int *, int) noexcept;
+int __visc__atomic_xchg(int *, int) noexcept;
+int __visc__atomic_inc(int *) noexcept;
+int __visc__atomic_dec(int *) noexcept;
+int __visc__atomic_min(int *, int) noexcept;
+int __visc__atomic_max(int *, int) noexcept;
+int __visc__atomic_and(int *, int) noexcept;
+int __visc__atomic_or(int *, int) noexcept;
+int __visc__atomic_xor(int *, int) noexcept;
+
+void llvm_visc_track_mem(void *, size_t) noexcept;
+void llvm_visc_untrack_mem(void *) noexcept;
+void llvm_visc_request_mem(void *, size_t) noexcept;
+
+#ifdef __cplusplus
+}
+#endif
diff --git a/hpvm/test/parboil/RUN.parboil.script b/hpvm/test/parboil/RUN.parboil.script
index 5cedcf480d..7f8c01ede7 100644
--- a/hpvm/test/parboil/RUN.parboil.script
+++ b/hpvm/test/parboil/RUN.parboil.script
@@ -1,5 +1,5 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG -o %t.ll -S %s
; RUN: llvm-link %llvm_src/../libclc/built_libs/nvptx--nvidiacl.bc %s.kernels.ll -o %t.ll.kernels.linked.bc
; RUN: clang -O3 -target nvptx %t.ll.kernels.linked.bc -S -o %s.nvptx.s
-; RUN: llvm-link %t.ll %llvm_src/projects/hpvm-rt/hpvm-rt.ll parboil.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll %llvm_src/projects/visc-rt/visc-rt.ll parboil.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -o %t.bin
diff --git a/hpvm/test/parboil/benchmarks/bfs/Makefile b/hpvm/test/parboil/benchmarks/bfs/Makefile
index e40a8484a3..cc6db67829 100644
--- a/hpvm/test/parboil/benchmarks/bfs/Makefile
+++ b/hpvm/test/parboil/benchmarks/bfs/Makefile
@@ -1,9 +1,9 @@
PARBOIL_ROOT = $(LLVM_SRC_ROOT)/tools/hpvm/test/parboil
APP = bfs
-# Default compile hpvm
+# Default compile visc
ifeq ($(VERSION),)
- VERSION = hpvm
+ VERSION = visc
endif
# Default use small test case
diff --git a/hpvm/test/parboil/benchmarks/bfs/src/opencl_cpu_baseline/kernel-spir64.ll b/hpvm/test/parboil/benchmarks/bfs/src/opencl_cpu_baseline/kernel-spir64.ll
index aca5667b70..9abdb29a3c 100644
--- a/hpvm/test/parboil/benchmarks/bfs/src/opencl_cpu_baseline/kernel-spir64.ll
+++ b/hpvm/test/parboil/benchmarks/bfs/src/opencl_cpu_baseline/kernel-spir64.ll
@@ -1,4 +1,4 @@
-; ModuleID = '/home/psrivas2.hpvm.llvm/test/HPVM/parboil/benchmarks/bfs/src/opencl_base/kernel.cl'
+; ModuleID = '/home/psrivas2/visc/llvm/test/VISC/parboil/benchmarks/bfs/src/opencl_base/kernel.cl'
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024"
target triple = "spir64-unknown-unknown"
diff --git a/hpvm/test/parboil/benchmarks/bfs/src/opencl_cpu_baseline/main.cpp b/hpvm/test/parboil/benchmarks/bfs/src/opencl_cpu_baseline/main.cpp
index 8e0d34c4b8..9b8b502688 100644
--- a/hpvm/test/parboil/benchmarks/bfs/src/opencl_cpu_baseline/main.cpp
+++ b/hpvm/test/parboil/benchmarks/bfs/src/opencl_cpu_baseline/main.cpp
@@ -237,7 +237,7 @@ int main(int argc, char **argv) {
NULL, NULL));
printf("Starting GPU kernel\n");
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
int num_of_blocks;
int num_of_threads_per_block;
@@ -272,7 +272,7 @@ int main(int argc, char **argv) {
OCL_ERRCK_RETVAL(clEnqueueWriteBuffer(clCommandQueue, tail, CL_TRUE, 0,
sizeof(int), &zero, 0, NULL, NULL));
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
if (num_t == 0) { // frontier is empty
break;
}
diff --git a/hpvm/test/parboil/benchmarks/bfs/src/opencl_nvidia/main.cpp b/hpvm/test/parboil/benchmarks/bfs/src/opencl_nvidia/main.cpp
index cfd0bf870a..3f9bc77557 100644
--- a/hpvm/test/parboil/benchmarks/bfs/src/opencl_nvidia/main.cpp
+++ b/hpvm/test/parboil/benchmarks/bfs/src/opencl_nvidia/main.cpp
@@ -428,7 +428,7 @@ int main(int argc, char **argv) {
OCL_ERRCK_RETVAL(clSetKernelArg(
BFS_kernel_S, 14, MAX_THREADS_PER_BLOCK * sizeof(int), NULL));
OCL_ERRCK_RETVAL(clSetKernelArg(BFS_kernel_S, 15, sizeof(int), NULL));
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
OCL_ERRCK_RETVAL(clEnqueueNDRangeKernel(clCommandQueue, BFS_kernel_S, 1,
0, grid, block, 0, 0, 0));
OCL_ERRCK_RETVAL(clFinish(clCommandQueue));
@@ -458,7 +458,7 @@ int main(int argc, char **argv) {
OCL_ERRCK_RETVAL(clSetKernelArg(BFS_kernel_M, 17, sizeof(int), NULL));
OCL_ERRCK_RETVAL(clSetKernelArg(BFS_kernel_M, 18, sizeof(int), NULL));
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
OCL_ERRCK_RETVAL(clEnqueueNDRangeKernel(clCommandQueue, BFS_kernel_M, 1,
0, grid, block, 0, 0, 0));
@@ -490,7 +490,7 @@ int main(int argc, char **argv) {
OCL_ERRCK_RETVAL(
clSetKernelArg(BFS_kernel_L, 13, NUM_BIN * sizeof(int), NULL));
OCL_ERRCK_RETVAL(clSetKernelArg(BFS_kernel_L, 14, sizeof(int), NULL));
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
OCL_ERRCK_RETVAL(clEnqueueNDRangeKernel(clCommandQueue, BFS_kernel_L, 1,
0, grid, block, 0, 0, 0));
OCL_ERRCK_RETVAL(clFinish(clCommandQueue));
@@ -542,7 +542,7 @@ int main(int argc, char **argv) {
OCL_ERRCK_RETVAL(clSetKernelArg(BFS_kernel_M, 16, sizeof(int), NULL));
OCL_ERRCK_RETVAL(clSetKernelArg(BFS_kernel_M, 17, sizeof(int), NULL));
OCL_ERRCK_RETVAL(clSetKernelArg(BFS_kernel_M, 18, sizeof(int), NULL));
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
OCL_ERRCK_RETVAL(clEnqueueNDRangeKernel(clCommandQueue, BFS_kernel_M, 1,
0, grid, block, 0, 0, 0));
OCL_ERRCK_RETVAL(clFinish(clCommandQueue));
@@ -572,7 +572,7 @@ int main(int argc, char **argv) {
clSetKernelArg(BFS_kernel_L, 13, NUM_BIN * sizeof(int), NULL));
OCL_ERRCK_RETVAL(clSetKernelArg(BFS_kernel_L, 14, sizeof(int), NULL));
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
OCL_ERRCK_RETVAL(clEnqueueNDRangeKernel(clCommandQueue, BFS_kernel_L, 1,
0, grid, block, 0, 0, 0));
OCL_ERRCK_RETVAL(clFinish(clCommandQueue));
diff --git a/hpvm/test/parboil/benchmarks/bfs/src/hpvm/Makefile b/hpvm/test/parboil/benchmarks/bfs/src/visc/Makefile
similarity index 81%
rename from hpvm/test/parboil/benchmarks/bfs/src/hpvm/Makefile
rename to hpvm/test/parboil/benchmarks/bfs/src/visc/Makefile
index 27cde148f7..a459707110 100644
--- a/hpvm/test/parboil/benchmarks/bfs/src/hpvm/Makefile
+++ b/hpvm/test/parboil/benchmarks/bfs/src/visc/Makefile
@@ -1,8 +1,8 @@
# (c) 2010 The Board of Trustees of the University of Illinois.
-LANGUAGE=hpvm
+LANGUAGE=visc
SRCDIR_OBJS=
-HPVM_OBJS=main.hpvm.ll
+VISC_OBJS=main.visc.ll
APP_CUDALDFLAGS=-lm -lstdc++
APP_CFLAGS=-ffast-math -O3
APP_CXXFLAGS=-ffast-math -O3
diff --git a/hpvm/test/parboil/benchmarks/bfs/src/hpvm/config.h b/hpvm/test/parboil/benchmarks/bfs/src/visc/config.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/bfs/src/hpvm/config.h
rename to hpvm/test/parboil/benchmarks/bfs/src/visc/config.h
diff --git a/hpvm/test/parboil/benchmarks/bfs/src/hpvm/main.cpp b/hpvm/test/parboil/benchmarks/bfs/src/visc/main.cpp
similarity index 70%
rename from hpvm/test/parboil/benchmarks/bfs/src/hpvm/main.cpp
rename to hpvm/test/parboil/benchmarks/bfs/src/visc/main.cpp
index 0fa9a60df8..9491218e5e 100644
--- a/hpvm/test/parboil/benchmarks/bfs/src/hpvm/main.cpp
+++ b/hpvm/test/parboil/benchmarks/bfs/src/visc/main.cpp
@@ -26,11 +26,11 @@
*/
#include "config.h"
#include "parboil.h"
-#include <hpvm.h>
#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
+#include <visc.h>
/**********
Define colors for BFS
@@ -113,11 +113,11 @@ void packData(RootIn *args, int *q1, size_t bytesq1, int *q2, size_t bytesq2,
void Allocation(long block) {
// Memory shared between threadblocks
- void *local_q_tail = __hpvm__malloc(sizeof(int));
- void *local_q = __hpvm__malloc(LOCAL_MEM_SIZE * sizeof(int));
- void *shift = __hpvm__malloc(sizeof(int));
+ void *local_q_tail = __visc__malloc(sizeof(int));
+ void *local_q = __visc__malloc(LOCAL_MEM_SIZE * sizeof(int));
+ void *shift = __visc__malloc(sizeof(int));
- __hpvm__return(6, local_q_tail, sizeof(int), local_q,
+ __visc__return(6, local_q_tail, sizeof(int), local_q,
LOCAL_MEM_SIZE * sizeof(int), shift, sizeof(int));
}
@@ -133,21 +133,21 @@ void BFSLeaf(int *q1, size_t bytesq1, int *q2, size_t bytesq2,
int *local_q_tail, size_t byteslocal_q_tail, int *local_q,
size_t byteslocal_q, int *shift, size_t bytesshift) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(6, q1, g_graph_nodes, g_graph_edges, g_color, g_cost, tail,
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(6, q1, g_graph_nodes, g_graph_edges, g_color, g_cost, tail,
4, q2, g_color, g_cost, tail);
- void *thisNode = __hpvm__getNode();
- void *parentNode = __hpvm__getParentNode(thisNode);
- int lx = __hpvm__getNodeInstanceID_x(thisNode);
- int gx = __hpvm__getNodeInstanceID_x(parentNode);
- int dimx = __hpvm__getNumNodeInstances_x(thisNode);
+ void *thisNode = __visc__getNode();
+ void *parentNode = __visc__getParentNode(thisNode);
+ int lx = __visc__getNodeInstanceID_x(thisNode);
+ int gx = __visc__getNodeInstanceID_x(parentNode);
+ int dimx = __visc__getNumNodeInstances_x(thisNode);
if (lx == 0) {
*local_q_tail = 0; // initialize the tail of w-queue
}
- __hpvm__barrier();
+ __visc__barrier();
// first, propagate and add the new frontier elements into w-queues
// int tid = get_group_id(0)*MAX_THREADS_PER_BLOCK + get_local_id(0);
@@ -170,16 +170,16 @@ void BFSLeaf(int *q1, size_t bytesq1, int *q2, size_t bytesq2,
int cost = cur_edge.y;
cost += cur_cost;
- int orig_cost = __hpvm__atomic_min(&g_cost[id], cost);
+ int orig_cost = __visc__atomic_min(&g_cost[id], cost);
if (orig_cost > cost) { // the node should be visited
if (g_color[id] > UP_LIMIT) {
- int old_color = __hpvm__atomic_xchg(&g_color[id], gray_shade);
+ int old_color = __visc__atomic_xchg(&g_color[id], gray_shade);
// this guarantees that only one thread will push this node
// into a queue
if (old_color != gray_shade) {
// atomic operation guarantees the correctness
// even if multiple warps are executing simultaneously
- int index = __hpvm__atomic_add(local_q_tail, 1);
+ int index = __visc__atomic_add(local_q_tail, 1);
local_q[index] = id;
}
}
@@ -187,16 +187,16 @@ void BFSLeaf(int *q1, size_t bytesq1, int *q2, size_t bytesq2,
}
}
- __hpvm__barrier();
+ __visc__barrier();
if (lx == 0) {
int tot_sum = *local_q_tail;
// the offset or "shift" of the block-level queue within the grid-level
// queue is determined by atomic operation
- *shift = __hpvm__atomic_add(tail, tot_sum);
+ *shift = __visc__atomic_add(tail, tot_sum);
}
- __hpvm__barrier();
+ __visc__barrier();
// shift within a w-queue
int local_shift = lx;
@@ -220,41 +220,41 @@ void BlockingBFS(int *q1, size_t bytesq1, int *q2, size_t bytesq2,
// ideally be placed in local memory
int *local_q_tail, size_t byteslocal_q_tail, int *local_q,
size_t byteslocal_q, int *shift, size_t bytesshift) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(6, q1, g_graph_nodes, g_graph_edges, g_color, g_cost, tail,
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(6, q1, g_graph_nodes, g_graph_edges, g_color, g_cost, tail,
4, q2, g_color, g_cost, tail);
- void *AllocationNode = __hpvm__createNodeND(0, Allocation);
- void *BFSLeafNode = __hpvm__createNodeND(1, BFSLeaf, block);
+ void *AllocationNode = __visc__createNodeND(0, Allocation);
+ void *BFSLeafNode = __visc__createNodeND(1, BFSLeaf, block);
// Bind edges
- __hpvm__bindIn(AllocationNode, 17, 0, 0); // Bind block
- __hpvm__bindIn(BFSLeafNode, 0, 0, 0); // Bind q1
- __hpvm__bindIn(BFSLeafNode, 1, 1, 0); // Bind bytes_q1
- __hpvm__bindIn(BFSLeafNode, 2, 2, 0); // Bind q2
- __hpvm__bindIn(BFSLeafNode, 3, 3, 0); // Bind bytes_q2
- __hpvm__bindIn(BFSLeafNode, 4, 4, 0); // Bind graph_nodes
- __hpvm__bindIn(BFSLeafNode, 5, 5, 0); // Bind bytes_graph_nodes
- __hpvm__bindIn(BFSLeafNode, 6, 6, 0); // Bind graph_edges
- __hpvm__bindIn(BFSLeafNode, 7, 7, 0); // Bind bytes_graph_edges
- __hpvm__bindIn(BFSLeafNode, 8, 8, 0); // Bind color
- __hpvm__bindIn(BFSLeafNode, 9, 9, 0); // Bind bytes_color
- __hpvm__bindIn(BFSLeafNode, 10, 10, 0); // Bind cost
- __hpvm__bindIn(BFSLeafNode, 11, 11, 0); // Bind bytes_cost
- __hpvm__bindIn(BFSLeafNode, 12, 12, 0); // Bind tail
- __hpvm__bindIn(BFSLeafNode, 13, 13, 0); // Bind bytes_tail
- __hpvm__bindIn(BFSLeafNode, 14, 14, 0); // Bind no_of_nodes
- __hpvm__bindIn(BFSLeafNode, 15, 15, 0); // Bind gray_shade
- __hpvm__bindIn(BFSLeafNode, 16, 16, 0); // Bind k
+ __visc__bindIn(AllocationNode, 17, 0, 0); // Bind block
+ __visc__bindIn(BFSLeafNode, 0, 0, 0); // Bind q1
+ __visc__bindIn(BFSLeafNode, 1, 1, 0); // Bind bytes_q1
+ __visc__bindIn(BFSLeafNode, 2, 2, 0); // Bind q2
+ __visc__bindIn(BFSLeafNode, 3, 3, 0); // Bind bytes_q2
+ __visc__bindIn(BFSLeafNode, 4, 4, 0); // Bind graph_nodes
+ __visc__bindIn(BFSLeafNode, 5, 5, 0); // Bind bytes_graph_nodes
+ __visc__bindIn(BFSLeafNode, 6, 6, 0); // Bind graph_edges
+ __visc__bindIn(BFSLeafNode, 7, 7, 0); // Bind bytes_graph_edges
+ __visc__bindIn(BFSLeafNode, 8, 8, 0); // Bind color
+ __visc__bindIn(BFSLeafNode, 9, 9, 0); // Bind bytes_color
+ __visc__bindIn(BFSLeafNode, 10, 10, 0); // Bind cost
+ __visc__bindIn(BFSLeafNode, 11, 11, 0); // Bind bytes_cost
+ __visc__bindIn(BFSLeafNode, 12, 12, 0); // Bind tail
+ __visc__bindIn(BFSLeafNode, 13, 13, 0); // Bind bytes_tail
+ __visc__bindIn(BFSLeafNode, 14, 14, 0); // Bind no_of_nodes
+ __visc__bindIn(BFSLeafNode, 15, 15, 0); // Bind gray_shade
+ __visc__bindIn(BFSLeafNode, 16, 16, 0); // Bind k
// Create Edges between AllocationNode and BFSLeafNodeNode
- __hpvm__edge(AllocationNode, BFSLeafNode, 1, 0, 17, 0); // Edge local_q_tail
- __hpvm__edge(AllocationNode, BFSLeafNode, 1, 1, 18,
+ __visc__edge(AllocationNode, BFSLeafNode, 1, 0, 17, 0); // Edge local_q_tail
+ __visc__edge(AllocationNode, BFSLeafNode, 1, 1, 18,
0); // Edge bytes_local_q_tail
- __hpvm__edge(AllocationNode, BFSLeafNode, 1, 2, 19, 0); // Edge local_q
- __hpvm__edge(AllocationNode, BFSLeafNode, 1, 3, 20, 0); // Edge bytes_local_q
- __hpvm__edge(AllocationNode, BFSLeafNode, 1, 4, 21, 0); // Edge shift
- __hpvm__edge(AllocationNode, BFSLeafNode, 1, 5, 22, 0); // Edge bytes_shift
+ __visc__edge(AllocationNode, BFSLeafNode, 1, 2, 19, 0); // Edge local_q
+ __visc__edge(AllocationNode, BFSLeafNode, 1, 3, 20, 0); // Edge bytes_local_q
+ __visc__edge(AllocationNode, BFSLeafNode, 1, 4, 21, 0); // Edge shift
+ __visc__edge(AllocationNode, BFSLeafNode, 1, 5, 22, 0); // Edge bytes_shift
}
// VoidRetTy
@@ -264,30 +264,30 @@ void BFS_Root(int *q1, size_t bytesq1, int *q2, size_t bytesq2,
int *g_color, size_t bytesg_color, int *g_cost,
size_t bytesg_cost, int *tail, size_t bytestail, int no_of_nodes,
int gray_shade, int k, long block, long grid) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(6, q1, g_graph_nodes, g_graph_edges, g_color, g_cost, tail,
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(6, q1, g_graph_nodes, g_graph_edges, g_color, g_cost, tail,
4, q2, g_color, g_cost, tail);
- void *BlockingBFSNode = __hpvm__createNodeND(1, BlockingBFS, grid);
+ void *BlockingBFSNode = __visc__createNodeND(1, BlockingBFS, grid);
// Bind edges
- __hpvm__bindIn(BlockingBFSNode, 0, 0, 0); // Bind q1
- __hpvm__bindIn(BlockingBFSNode, 1, 1, 0); // Bind bytes_q1
- __hpvm__bindIn(BlockingBFSNode, 2, 2, 0); // Bind q2
- __hpvm__bindIn(BlockingBFSNode, 3, 3, 0); // Bind bytes_q2
- __hpvm__bindIn(BlockingBFSNode, 4, 4, 0); // Bind graph_nodes
- __hpvm__bindIn(BlockingBFSNode, 5, 5, 0); // Bind bytes_graph_nodes
- __hpvm__bindIn(BlockingBFSNode, 6, 6, 0); // Bind graph_edges
- __hpvm__bindIn(BlockingBFSNode, 7, 7, 0); // Bind bytes_graph_edges
- __hpvm__bindIn(BlockingBFSNode, 8, 8, 0); // Bind color
- __hpvm__bindIn(BlockingBFSNode, 9, 9, 0); // Bind bytes_color
- __hpvm__bindIn(BlockingBFSNode, 10, 10, 0); // Bind cost
- __hpvm__bindIn(BlockingBFSNode, 11, 11, 0); // Bind bytes_cost
- __hpvm__bindIn(BlockingBFSNode, 12, 12, 0); // Bind tail
- __hpvm__bindIn(BlockingBFSNode, 13, 13, 0); // Bind bytes_tail
- __hpvm__bindIn(BlockingBFSNode, 14, 14, 0); // Bind no_of_nodes
- __hpvm__bindIn(BlockingBFSNode, 15, 15, 0); // Bind gray_shade
- __hpvm__bindIn(BlockingBFSNode, 16, 16, 0); // Bind k
- __hpvm__bindIn(BlockingBFSNode, 17, 17, 0); // Bind block
+ __visc__bindIn(BlockingBFSNode, 0, 0, 0); // Bind q1
+ __visc__bindIn(BlockingBFSNode, 1, 1, 0); // Bind bytes_q1
+ __visc__bindIn(BlockingBFSNode, 2, 2, 0); // Bind q2
+ __visc__bindIn(BlockingBFSNode, 3, 3, 0); // Bind bytes_q2
+ __visc__bindIn(BlockingBFSNode, 4, 4, 0); // Bind graph_nodes
+ __visc__bindIn(BlockingBFSNode, 5, 5, 0); // Bind bytes_graph_nodes
+ __visc__bindIn(BlockingBFSNode, 6, 6, 0); // Bind graph_edges
+ __visc__bindIn(BlockingBFSNode, 7, 7, 0); // Bind bytes_graph_edges
+ __visc__bindIn(BlockingBFSNode, 8, 8, 0); // Bind color
+ __visc__bindIn(BlockingBFSNode, 9, 9, 0); // Bind bytes_color
+ __visc__bindIn(BlockingBFSNode, 10, 10, 0); // Bind cost
+ __visc__bindIn(BlockingBFSNode, 11, 11, 0); // Bind bytes_cost
+ __visc__bindIn(BlockingBFSNode, 12, 12, 0); // Bind tail
+ __visc__bindIn(BlockingBFSNode, 13, 13, 0); // Bind bytes_tail
+ __visc__bindIn(BlockingBFSNode, 14, 14, 0); // Bind no_of_nodes
+ __visc__bindIn(BlockingBFSNode, 15, 15, 0); // Bind gray_shade
+ __visc__bindIn(BlockingBFSNode, 16, 16, 0); // Bind k
+ __visc__bindIn(BlockingBFSNode, 17, 17, 0); // Bind block
}
void BFS_Wrapper(int *q1, size_t bytesq1, // 0, 1
@@ -300,31 +300,31 @@ void BFS_Wrapper(int *q1, size_t bytesq1, // 0, 1
int no_of_nodes, int gray_shade, // 14, 15
int k, long block, long grid // 16 - 18
) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(6, q1, g_graph_nodes, g_graph_edges, g_color, g_cost, tail,
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(6, q1, g_graph_nodes, g_graph_edges, g_color, g_cost, tail,
4, q2, g_color, g_cost, tail);
- void *BlockingBFSNode = __hpvm__createNodeND(0, BFS_Root);
+ void *BlockingBFSNode = __visc__createNodeND(0, BFS_Root);
// Bind edges
- __hpvm__bindIn(BlockingBFSNode, 0, 0, 0); // Bind q1
- __hpvm__bindIn(BlockingBFSNode, 1, 1, 0); // Bind bytes_q1
- __hpvm__bindIn(BlockingBFSNode, 2, 2, 0); // Bind q2
- __hpvm__bindIn(BlockingBFSNode, 3, 3, 0); // Bind bytes_q2
- __hpvm__bindIn(BlockingBFSNode, 4, 4, 0); // Bind graph_nodes
- __hpvm__bindIn(BlockingBFSNode, 5, 5, 0); // Bind bytes_graph_nodes
- __hpvm__bindIn(BlockingBFSNode, 6, 6, 0); // Bind graph_edges
- __hpvm__bindIn(BlockingBFSNode, 7, 7, 0); // Bind bytes_graph_edges
- __hpvm__bindIn(BlockingBFSNode, 8, 8, 0); // Bind color
- __hpvm__bindIn(BlockingBFSNode, 9, 9, 0); // Bind bytes_color
- __hpvm__bindIn(BlockingBFSNode, 10, 10, 0); // Bind cost
- __hpvm__bindIn(BlockingBFSNode, 11, 11, 0); // Bind bytes_cost
- __hpvm__bindIn(BlockingBFSNode, 12, 12, 0); // Bind tail
- __hpvm__bindIn(BlockingBFSNode, 13, 13, 0); // Bind bytes_tail
- __hpvm__bindIn(BlockingBFSNode, 14, 14, 0); // Bind no_of_nodes
- __hpvm__bindIn(BlockingBFSNode, 15, 15, 0); // Bind gray_shade
- __hpvm__bindIn(BlockingBFSNode, 16, 16, 0); // Bind k
- __hpvm__bindIn(BlockingBFSNode, 17, 17, 0); // Bind block
- __hpvm__bindIn(BlockingBFSNode, 18, 18, 0); // Bind grid
+ __visc__bindIn(BlockingBFSNode, 0, 0, 0); // Bind q1
+ __visc__bindIn(BlockingBFSNode, 1, 1, 0); // Bind bytes_q1
+ __visc__bindIn(BlockingBFSNode, 2, 2, 0); // Bind q2
+ __visc__bindIn(BlockingBFSNode, 3, 3, 0); // Bind bytes_q2
+ __visc__bindIn(BlockingBFSNode, 4, 4, 0); // Bind graph_nodes
+ __visc__bindIn(BlockingBFSNode, 5, 5, 0); // Bind bytes_graph_nodes
+ __visc__bindIn(BlockingBFSNode, 6, 6, 0); // Bind graph_edges
+ __visc__bindIn(BlockingBFSNode, 7, 7, 0); // Bind bytes_graph_edges
+ __visc__bindIn(BlockingBFSNode, 8, 8, 0); // Bind color
+ __visc__bindIn(BlockingBFSNode, 9, 9, 0); // Bind bytes_color
+ __visc__bindIn(BlockingBFSNode, 10, 10, 0); // Bind cost
+ __visc__bindIn(BlockingBFSNode, 11, 11, 0); // Bind bytes_cost
+ __visc__bindIn(BlockingBFSNode, 12, 12, 0); // Bind tail
+ __visc__bindIn(BlockingBFSNode, 13, 13, 0); // Bind bytes_tail
+ __visc__bindIn(BlockingBFSNode, 14, 14, 0); // Bind no_of_nodes
+ __visc__bindIn(BlockingBFSNode, 15, 15, 0); // Bind gray_shade
+ __visc__bindIn(BlockingBFSNode, 16, 16, 0); // Bind k
+ __visc__bindIn(BlockingBFSNode, 17, 17, 0); // Bind block
+ __visc__bindIn(BlockingBFSNode, 18, 18, 0); // Bind grid
}
FILE *fp;
@@ -415,7 +415,7 @@ int main(int argc, char **argv) {
fclose(fp);
pb_InitializeTimerSet(&timers);
- __hpvm__init();
+ __visc__init();
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
// allocate mem for the result on host side
@@ -433,15 +433,15 @@ int main(int argc, char **argv) {
int *q2 = (int *)malloc(sizeof(int) * num_of_nodes);
int *tail = (int *)malloc(sizeof(int));
- llvm_hpvm_track_mem(graph_nodes, bytes_graph_nodes);
- llvm_hpvm_track_mem(graph_edges, bytes_graph_edges);
- llvm_hpvm_track_mem(cost, bytes_cost);
- llvm_hpvm_track_mem(color, bytes_cost);
+ llvm_visc_track_mem(graph_nodes, bytes_graph_nodes);
+ llvm_visc_track_mem(graph_edges, bytes_graph_edges);
+ llvm_visc_track_mem(cost, bytes_cost);
+ llvm_visc_track_mem(color, bytes_cost);
// Allocating stuff on host side, but these can also be allocated in the graph
- llvm_hpvm_track_mem(q1, bytes_cost);
- llvm_hpvm_track_mem(q2, bytes_cost);
+ llvm_visc_track_mem(q1, bytes_cost);
+ llvm_visc_track_mem(q2, bytes_cost);
// Scalar variable read/written by both graph and host.
- llvm_hpvm_track_mem(tail, sizeof(int));
+ llvm_visc_track_mem(tail, sizeof(int));
int num_of_blocks;
int num_of_threads_per_block;
@@ -466,9 +466,9 @@ int main(int argc, char **argv) {
graph_edges, bytes_graph_edges, color, bytes_cost, cost, bytes_cost,
tail, sizeof(int), num_of_nodes, gray, k, block, grid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
do {
- llvm_hpvm_request_mem(tail, sizeof(int));
+ llvm_visc_request_mem(tail, sizeof(int));
num_t = *tail;
// printf("tail for iteration %d = %d\n",k, num_t);
*tail = 0;
@@ -493,7 +493,7 @@ int main(int argc, char **argv) {
} else {
args->gray_shade = GRAY1;
}
- // void* bfsDFG = __hpvm__node(BFS_kernel, 2, 1, block, grid, 17,
+ // void* bfsDFG = __visc__node(BFS_kernel, 2, 1, block, grid, 17,
// q1, bytes_cost,
// q2, bytes_cost,
// graph_nodes, bytes_graph_nodes,
@@ -505,8 +505,8 @@ int main(int argc, char **argv) {
// gray,
// k,
// 0);
- void *bfsDFG = __hpvm__launch(0, BFS_Wrapper, (void *)args);
- __hpvm__wait(bfsDFG);
+ void *bfsDFG = __visc__launch(0, BFS_Wrapper, (void *)args);
+ __visc__wait(bfsDFG);
// Swap q1 and q2
// Swap q1 and q2
int *temp = args->q1;
@@ -518,22 +518,22 @@ int main(int argc, char **argv) {
// copy result from device to host
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
- llvm_hpvm_request_mem(cost, bytes_cost);
- llvm_hpvm_request_mem(color, bytes_cost);
+ llvm_visc_request_mem(cost, bytes_cost);
+ llvm_visc_request_mem(color, bytes_cost);
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
- llvm_hpvm_untrack_mem(graph_nodes);
- llvm_hpvm_untrack_mem(graph_edges);
- llvm_hpvm_untrack_mem(cost);
- llvm_hpvm_untrack_mem(color);
- llvm_hpvm_untrack_mem(q1);
- llvm_hpvm_untrack_mem(q2);
- llvm_hpvm_untrack_mem(tail);
+ llvm_visc_untrack_mem(graph_nodes);
+ llvm_visc_untrack_mem(graph_edges);
+ llvm_visc_untrack_mem(cost);
+ llvm_visc_untrack_mem(color);
+ llvm_visc_untrack_mem(q1);
+ llvm_visc_untrack_mem(q2);
+ llvm_visc_untrack_mem(tail);
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
pb_PrintTimerSet(&timers);
- __hpvm__cleanup();
+ __visc__cleanup();
// Store the result into a file
// FIXME: color is not even printed. Why are we reading it back??
diff --git a/hpvm/test/parboil/benchmarks/cutcp/Makefile b/hpvm/test/parboil/benchmarks/cutcp/Makefile
index e8edc6e731..5e56793360 100644
--- a/hpvm/test/parboil/benchmarks/cutcp/Makefile
+++ b/hpvm/test/parboil/benchmarks/cutcp/Makefile
@@ -1,9 +1,9 @@
PARBOIL_ROOT = $(LLVM_SRC_ROOT)/tools/hpvm/test/parboil
APP = cutcp
-# Default compile hpvm
+# Default compile visc
ifeq ($(VERSION),)
- VERSION = hpvm
+ VERSION = visc
endif
# Default use small test case
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/cutoff6overlap.c b/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/cutoff6overlap.c
index dfd7f1ff38..06f856c1a0 100644
--- a/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/cutoff6overlap.c
+++ b/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/cutoff6overlap.c
@@ -427,7 +427,7 @@ int gpu_compute_cutoff_potential_lattice6overlap(
/*CHECK_ERROR("clCreateCommandQueue")*/
/* loop over z-dimension, invoke OpenCL kernel for each x-y plane */
- pb_SwitchToTimer(timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(timers, visc_TimerID_COMPUTATION);
if (verbose)
printf("Invoking OpenCL kernel on %d region planes...\n", zRegionDim);
for (zRegionIndex = 0; zRegionIndex < zRegionDim; zRegionIndex++) {
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_hpvm.cl b/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_visc.cl
similarity index 100%
rename from hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_hpvm.cl
rename to hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_visc.cl
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_hpvm_x64.ll b/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_visc_x64.ll
similarity index 99%
rename from hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_hpvm_x64.ll
rename to hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_visc_x64.ll
index 85a73b291f..7f614e66ff 100644
--- a/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_hpvm_x64.ll
+++ b/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_visc_x64.ll
@@ -1,4 +1,4 @@
-; ModuleID = '/home/psrivas2.hpvm.llvm/test/HPVM/parboil/benchmarks/cutcp/src/opencl_nvidia/kernel_hpvm.cl'
+; ModuleID = '/home/psrivas2/visc/llvm/test/VISC/parboil/benchmarks/cutcp/src/opencl_nvidia/kernel_visc.cl'
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024"
target triple = "spir64-unknown-unknown"
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_hpvm_x64.spir b/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_visc_x64.spir
similarity index 100%
rename from hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_hpvm_x64.spir
rename to hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_visc_x64.spir
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_x64.ll b/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_x64.ll
index 5a3c1fcd5d..370e3c0f8f 100644
--- a/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_x64.ll
+++ b/hpvm/test/parboil/benchmarks/cutcp/src/opencl_cpu_baseline/kernel_x64.ll
@@ -1,4 +1,4 @@
-; ModuleID = '/home/psrivas2.hpvm.llvm/test/HPVM/parboil/benchmarks/cutcp/src/opencl_nvidia/kernel.cl'
+; ModuleID = '/home/psrivas2/visc/llvm/test/VISC/parboil/benchmarks/cutcp/src/opencl_nvidia/kernel.cl'
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024"
target triple = "spir64-unknown-unknown"
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/opencl_nvidia/cutoff6overlap.c b/hpvm/test/parboil/benchmarks/cutcp/src/opencl_nvidia/cutoff6overlap.c
index 076532b709..96ebeafbdf 100644
--- a/hpvm/test/parboil/benchmarks/cutcp/src/opencl_nvidia/cutoff6overlap.c
+++ b/hpvm/test/parboil/benchmarks/cutcp/src/opencl_nvidia/cutoff6overlap.c
@@ -423,7 +423,7 @@ int gpu_compute_cutoff_potential_lattice6overlap(
/*CHECK_ERROR("clCreateCommandQueue")*/
/* loop over z-dimension, invoke OpenCL kernel for each x-y plane */
- pb_SwitchToTimer(timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(timers, visc_TimerID_COMPUTATION);
if (verbose)
printf("Invoking OpenCL kernel on %d region planes...\n", zRegionDim);
for (zRegionIndex = 0; zRegionIndex < zRegionDim; zRegionIndex++) {
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/Makefile b/hpvm/test/parboil/benchmarks/cutcp/src/visc/Makefile
similarity index 85%
rename from hpvm/test/parboil/benchmarks/cutcp/src/hpvm/Makefile
rename to hpvm/test/parboil/benchmarks/cutcp/src/visc/Makefile
index 43a175b947..d4c650a17e 100644
--- a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/Makefile
+++ b/hpvm/test/parboil/benchmarks/cutcp/src/visc/Makefile
@@ -1,8 +1,8 @@
# (c) 2010 The Board of Trustees of the University of Illinois.
-LANGUAGE=hpvm
+LANGUAGE=visc
SRCDIR_OBJS=excl.ll cutcpu.ll cutoff6overlap.ll output.ll readatom.ll ocl.ll
-HPVM_OBJS=main.hpvm.ll
+VISC_OBJS=main.visc.ll
APP_CUDALDFLAGS=-lm
APP_CFLAGS=-ffast-math -O3
APP_CXXFLAGS=-ffast-math -O3
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/atom.h b/hpvm/test/parboil/benchmarks/cutcp/src/visc/atom.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/cutcp/src/hpvm/atom.h
rename to hpvm/test/parboil/benchmarks/cutcp/src/visc/atom.h
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/cutcpu.c b/hpvm/test/parboil/benchmarks/cutcp/src/visc/cutcpu.c
similarity index 100%
rename from hpvm/test/parboil/benchmarks/cutcp/src/hpvm/cutcpu.c
rename to hpvm/test/parboil/benchmarks/cutcp/src/visc/cutcpu.c
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/cutoff.h b/hpvm/test/parboil/benchmarks/cutcp/src/visc/cutoff.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/cutcp/src/hpvm/cutoff.h
rename to hpvm/test/parboil/benchmarks/cutcp/src/visc/cutoff.h
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/cutoff6overlap.c b/hpvm/test/parboil/benchmarks/cutcp/src/visc/cutoff6overlap.c
similarity index 100%
rename from hpvm/test/parboil/benchmarks/cutcp/src/hpvm/cutoff6overlap.c
rename to hpvm/test/parboil/benchmarks/cutcp/src/visc/cutoff6overlap.c
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/excl.c b/hpvm/test/parboil/benchmarks/cutcp/src/visc/excl.c
similarity index 100%
rename from hpvm/test/parboil/benchmarks/cutcp/src/hpvm/excl.c
rename to hpvm/test/parboil/benchmarks/cutcp/src/visc/excl.c
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/kernel.cl b/hpvm/test/parboil/benchmarks/cutcp/src/visc/kernel.cl
similarity index 100%
rename from hpvm/test/parboil/benchmarks/cutcp/src/hpvm/kernel.cl
rename to hpvm/test/parboil/benchmarks/cutcp/src/visc/kernel.cl
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/macros.h b/hpvm/test/parboil/benchmarks/cutcp/src/visc/macros.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/cutcp/src/hpvm/macros.h
rename to hpvm/test/parboil/benchmarks/cutcp/src/visc/macros.h
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/main.cpp b/hpvm/test/parboil/benchmarks/cutcp/src/visc/main.cpp
similarity index 82%
rename from hpvm/test/parboil/benchmarks/cutcp/src/hpvm/main.cpp
rename to hpvm/test/parboil/benchmarks/cutcp/src/visc/main.cpp
index 0a36196619..caf99a5b37 100644
--- a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/main.cpp
+++ b/hpvm/test/parboil/benchmarks/cutcp/src/visc/main.cpp
@@ -16,7 +16,7 @@
#include "cutoff.h"
#include "macros.h"
#include "output.h"
-#include <hpvm.h>
+#include <visc.h>
#define ERRTOL 1e-4f
@@ -54,11 +54,11 @@ extern float rsqrt(float x);
void Allocation(long block) {
// Memory shared between threadblocks
size_t bytes_AtomBinCache = sizeof(float) * BIN_CACHE_MAXLEN * BIN_DEPTH * 4;
- void *AtomBinCache = __hpvm__malloc(bytes_AtomBinCache);
+ void *AtomBinCache = __visc__malloc(bytes_AtomBinCache);
size_t bytes_myBinIndex = sizeof(xyz);
- void *myBinIndex = __hpvm__malloc(bytes_myBinIndex);
- __hpvm__return(4, AtomBinCache, bytes_AtomBinCache, myBinIndex,
+ void *myBinIndex = __visc__malloc(bytes_myBinIndex);
+ __visc__return(4, AtomBinCache, bytes_AtomBinCache, myBinIndex,
bytes_myBinIndex);
}
@@ -76,21 +76,21 @@ void CUTCPLeaf(int binDim_x, int binDim_y, float *binBaseAddr,
// local memory args
float *AtomBinCache, size_t bytes_AtomBinCache, int *myBinIndex,
size_t bytes_myBinIndex) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(4, binBaseAddr, regionZeroAddr, NbrListLen, NbrList, 1,
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(4, binBaseAddr, regionZeroAddr, NbrListLen, NbrList, 1,
regionZeroAddr);
- void *thisNode = __hpvm__getNode();
- void *parentNode = __hpvm__getParentNode(thisNode);
- int lx = __hpvm__getNodeInstanceID_x(thisNode);
- int ly = __hpvm__getNodeInstanceID_y(thisNode);
- int lz = __hpvm__getNodeInstanceID_z(thisNode);
- int gx = __hpvm__getNodeInstanceID_x(parentNode);
- int gy = __hpvm__getNodeInstanceID_y(parentNode);
- int dimx = __hpvm__getNumNodeInstances_x(thisNode);
- int dimy = __hpvm__getNumNodeInstances_y(thisNode);
- int gdimx = __hpvm__getNumNodeInstances_x(parentNode);
- int gdimy = __hpvm__getNumNodeInstances_y(parentNode);
+ void *thisNode = __visc__getNode();
+ void *parentNode = __visc__getParentNode(thisNode);
+ int lx = __visc__getNodeInstanceID_x(thisNode);
+ int ly = __visc__getNodeInstanceID_y(thisNode);
+ int lz = __visc__getNodeInstanceID_z(thisNode);
+ int gx = __visc__getNodeInstanceID_x(parentNode);
+ int gy = __visc__getNodeInstanceID_y(parentNode);
+ int dimx = __visc__getNumNodeInstances_x(thisNode);
+ int dimy = __visc__getNumNodeInstances_y(thisNode);
+ int gdimx = __visc__getNumNodeInstances_x(parentNode);
+ int gdimy = __visc__getNumNodeInstances_y(parentNode);
float *binZeroAddr = binBaseAddr + 4 * offset;
@@ -168,7 +168,7 @@ void CUTCPLeaf(int binDim_x, int binDim_y, float *binBaseAddr,
AtomBinCache[binIndex + tidmask + 16] = p_global[tidmask + 16];
}
- __hpvm__barrier();
+ __visc__barrier();
/* no warp divergence */
if (totalbins + BIN_CACHE_MAXLEN > *NbrListLen) {
numbins = *NbrListLen - totalbins;
@@ -196,7 +196,7 @@ void CUTCPLeaf(int binDim_x, int binDim_y, float *binBaseAddr,
if (r2 < cutoff2) {
float s = (1.f - r2 * inv_cutoff2);
energy0 += aq * rsqrt(r2) * s * s;
- // energy0 += aq * (1.0/__hpvm__sqrt(r2)) * s * s;
+ // energy0 += aq * (1.0/__visc__sqrt(r2)) * s * s;
}
#else
energy0 += (r2 < cutoff2);
@@ -208,7 +208,7 @@ void CUTCPLeaf(int binDim_x, int binDim_y, float *binBaseAddr,
if (r2 < cutoff2) {
float s = (1.f - r2 * inv_cutoff2);
energy1 += aq * rsqrt(r2) * s * s;
- // energy1 += aq * (1.0/__hpvm__sqrt(r2)) * s * s;
+ // energy1 += aq * (1.0/__visc__sqrt(r2)) * s * s;
}
#else
energy1 += (r2 < cutoff2);
@@ -219,7 +219,7 @@ void CUTCPLeaf(int binDim_x, int binDim_y, float *binBaseAddr,
if (r2 < cutoff2) {
float s = (1.f - r2 * inv_cutoff2);
energy2 += aq * rsqrt(r2) * s * s;
- // energy2 += aq * (1.0/__hpvm__sqrt(r2)) * s * s;
+ // energy2 += aq * (1.0/__visc__sqrt(r2)) * s * s;
}
#else
energy2 += (r2 < cutoff2);
@@ -237,7 +237,7 @@ void CUTCPLeaf(int binDim_x, int binDim_y, float *binBaseAddr,
#endif
} /* end loop over atoms in bin */
} /* end loop over cached atom bins */
- __hpvm__barrier();
+ __visc__barrier();
} /* end loop over neighbor list */
/* store into global memory */
@@ -260,38 +260,38 @@ void BlockingCUTCP(int binDim_x, int binDim_y, float4 *binBaseAddr,
size_t bytes_NbrList, long blockx, long blocky,
long blockz) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(4, binBaseAddr, regionZeroAddr, NbrListLen, NbrList, 1,
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(4, binBaseAddr, regionZeroAddr, NbrListLen, NbrList, 1,
regionZeroAddr);
- void *AllocationNode = __hpvm__createNodeND(0, Allocation);
+ void *AllocationNode = __visc__createNodeND(0, Allocation);
void *CUTCPLeafNode =
- __hpvm__createNodeND(3, CUTCPLeaf, blockx, blocky, blockz);
+ __visc__createNodeND(3, CUTCPLeaf, blockx, blocky, blockz);
// Bind Inputs
- __hpvm__bindIn(AllocationNode, 15, 0, 0); // Bind blockx
- __hpvm__bindIn(CUTCPLeafNode, 0, 0, 0); // Bind binDim_x
- __hpvm__bindIn(CUTCPLeafNode, 1, 1, 0); // Bind binDim_y
- __hpvm__bindIn(CUTCPLeafNode, 2, 2, 0); // Bind binBaseAddr
- __hpvm__bindIn(CUTCPLeafNode, 3, 3, 0); // Bind bytes_binBaseAddr
- __hpvm__bindIn(CUTCPLeafNode, 4, 4, 0); // Bind offset
- __hpvm__bindIn(CUTCPLeafNode, 5, 5, 0); // Bind h
- __hpvm__bindIn(CUTCPLeafNode, 6, 6, 0); // Bind cutoff2
- __hpvm__bindIn(CUTCPLeafNode, 7, 7, 0); // Bind inv_cutoff2
- __hpvm__bindIn(CUTCPLeafNode, 8, 8, 0); // Bind regionZeroAddr
- __hpvm__bindIn(CUTCPLeafNode, 9, 9, 0); // Bind bytes_regionZeroAddr
- __hpvm__bindIn(CUTCPLeafNode, 10, 10, 0); // Bind zRegionIndex
- __hpvm__bindIn(CUTCPLeafNode, 11, 11, 0); // Bind NbrListLen
- __hpvm__bindIn(CUTCPLeafNode, 12, 12, 0); // Bind bytes_NbrListLen
- __hpvm__bindIn(CUTCPLeafNode, 13, 13, 0); // Bind NbrList
- __hpvm__bindIn(CUTCPLeafNode, 14, 14, 0); // Bind bytes_NbrList
+ __visc__bindIn(AllocationNode, 15, 0, 0); // Bind blockx
+ __visc__bindIn(CUTCPLeafNode, 0, 0, 0); // Bind binDim_x
+ __visc__bindIn(CUTCPLeafNode, 1, 1, 0); // Bind binDim_y
+ __visc__bindIn(CUTCPLeafNode, 2, 2, 0); // Bind binBaseAddr
+ __visc__bindIn(CUTCPLeafNode, 3, 3, 0); // Bind bytes_binBaseAddr
+ __visc__bindIn(CUTCPLeafNode, 4, 4, 0); // Bind offset
+ __visc__bindIn(CUTCPLeafNode, 5, 5, 0); // Bind h
+ __visc__bindIn(CUTCPLeafNode, 6, 6, 0); // Bind cutoff2
+ __visc__bindIn(CUTCPLeafNode, 7, 7, 0); // Bind inv_cutoff2
+ __visc__bindIn(CUTCPLeafNode, 8, 8, 0); // Bind regionZeroAddr
+ __visc__bindIn(CUTCPLeafNode, 9, 9, 0); // Bind bytes_regionZeroAddr
+ __visc__bindIn(CUTCPLeafNode, 10, 10, 0); // Bind zRegionIndex
+ __visc__bindIn(CUTCPLeafNode, 11, 11, 0); // Bind NbrListLen
+ __visc__bindIn(CUTCPLeafNode, 12, 12, 0); // Bind bytes_NbrListLen
+ __visc__bindIn(CUTCPLeafNode, 13, 13, 0); // Bind NbrList
+ __visc__bindIn(CUTCPLeafNode, 14, 14, 0); // Bind bytes_NbrList
// Create Edges
- __hpvm__edge(AllocationNode, CUTCPLeafNode, 1, 0, 15, 0); // Edge AtomBinCache
- __hpvm__edge(AllocationNode, CUTCPLeafNode, 1, 1, 16,
+ __visc__edge(AllocationNode, CUTCPLeafNode, 1, 0, 15, 0); // Edge AtomBinCache
+ __visc__edge(AllocationNode, CUTCPLeafNode, 1, 1, 16,
0); // Edge bytes_AtomBinCache
- __hpvm__edge(AllocationNode, CUTCPLeafNode, 1, 2, 17, 0); // Edge myBinIndex
- __hpvm__edge(AllocationNode, CUTCPLeafNode, 1, 3, 18,
+ __visc__edge(AllocationNode, CUTCPLeafNode, 1, 2, 17, 0); // Edge myBinIndex
+ __visc__edge(AllocationNode, CUTCPLeafNode, 1, 3, 18,
0); // Edge bytes_myBinIndex
}
@@ -370,32 +370,32 @@ void CUTCPRoot(int binDim_x, int binDim_y, float4 *binBaseAddr,
int *NbrListLen, size_t bytes_NbrListLen, xyz *NbrList,
size_t bytes_NbrList, long blockx, long blocky, long blockz,
long gridx, long gridy, long gridz) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(4, binBaseAddr, regionZeroAddr, NbrListLen, NbrList, 1,
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(4, binBaseAddr, regionZeroAddr, NbrListLen, NbrList, 1,
regionZeroAddr);
void *BlockingCUTCPNode =
- __hpvm__createNodeND(3, BlockingCUTCP, gridx, gridy, gridz);
+ __visc__createNodeND(3, BlockingCUTCP, gridx, gridy, gridz);
// Bind Inputs
- __hpvm__bindIn(BlockingCUTCPNode, 0, 0, 0); // Bind binDim_x
- __hpvm__bindIn(BlockingCUTCPNode, 1, 1, 0); // Bind binDim_y
- __hpvm__bindIn(BlockingCUTCPNode, 2, 2, 0); // Bind binBaseAddr
- __hpvm__bindIn(BlockingCUTCPNode, 3, 3, 0); // Bind bytes_binBaseAddr
- __hpvm__bindIn(BlockingCUTCPNode, 4, 4, 0); // Bind offset
- __hpvm__bindIn(BlockingCUTCPNode, 5, 5, 0); // Bind h
- __hpvm__bindIn(BlockingCUTCPNode, 6, 6, 0); // Bind cutoff2
- __hpvm__bindIn(BlockingCUTCPNode, 7, 7, 0); // Bind inv_cutoff2
- __hpvm__bindIn(BlockingCUTCPNode, 8, 8, 0); // Bind regionZeroAddr
- __hpvm__bindIn(BlockingCUTCPNode, 9, 9, 0); // Bind bytes_regionZeroAddr
- __hpvm__bindIn(BlockingCUTCPNode, 10, 10, 0); // Bind zRegionIndex
- __hpvm__bindIn(BlockingCUTCPNode, 11, 11, 0); // Bind NbrListLen
- __hpvm__bindIn(BlockingCUTCPNode, 12, 12, 0); // Bind bytes_NbrListLen
- __hpvm__bindIn(BlockingCUTCPNode, 13, 13, 0); // Bind NbrList
- __hpvm__bindIn(BlockingCUTCPNode, 14, 14, 0); // Bind bytes_NbrList
- __hpvm__bindIn(BlockingCUTCPNode, 15, 15, 0); // Bind blockx
- __hpvm__bindIn(BlockingCUTCPNode, 16, 16, 0); // Bind blocky
- __hpvm__bindIn(BlockingCUTCPNode, 17, 17, 0); // Bind blockz
+ __visc__bindIn(BlockingCUTCPNode, 0, 0, 0); // Bind binDim_x
+ __visc__bindIn(BlockingCUTCPNode, 1, 1, 0); // Bind binDim_y
+ __visc__bindIn(BlockingCUTCPNode, 2, 2, 0); // Bind binBaseAddr
+ __visc__bindIn(BlockingCUTCPNode, 3, 3, 0); // Bind bytes_binBaseAddr
+ __visc__bindIn(BlockingCUTCPNode, 4, 4, 0); // Bind offset
+ __visc__bindIn(BlockingCUTCPNode, 5, 5, 0); // Bind h
+ __visc__bindIn(BlockingCUTCPNode, 6, 6, 0); // Bind cutoff2
+ __visc__bindIn(BlockingCUTCPNode, 7, 7, 0); // Bind inv_cutoff2
+ __visc__bindIn(BlockingCUTCPNode, 8, 8, 0); // Bind regionZeroAddr
+ __visc__bindIn(BlockingCUTCPNode, 9, 9, 0); // Bind bytes_regionZeroAddr
+ __visc__bindIn(BlockingCUTCPNode, 10, 10, 0); // Bind zRegionIndex
+ __visc__bindIn(BlockingCUTCPNode, 11, 11, 0); // Bind NbrListLen
+ __visc__bindIn(BlockingCUTCPNode, 12, 12, 0); // Bind bytes_NbrListLen
+ __visc__bindIn(BlockingCUTCPNode, 13, 13, 0); // Bind NbrList
+ __visc__bindIn(BlockingCUTCPNode, 14, 14, 0); // Bind bytes_NbrList
+ __visc__bindIn(BlockingCUTCPNode, 15, 15, 0); // Bind blockx
+ __visc__bindIn(BlockingCUTCPNode, 16, 16, 0); // Bind blocky
+ __visc__bindIn(BlockingCUTCPNode, 17, 17, 0); // Bind blockz
}
void CUTCPWrapper(int binDim_x, int binDim_y, float4 *binBaseAddr,
@@ -410,34 +410,34 @@ void CUTCPWrapper(int binDim_x, int binDim_y, float4 *binBaseAddr,
int *NbrListLen, size_t bytes_NbrListLen, xyz *NbrList,
size_t bytes_NbrList, long blockx, long blocky, long blockz,
long gridx, long gridy, long gridz) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(4, binBaseAddr, regionZeroAddr, NbrListLen, NbrList, 1,
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(4, binBaseAddr, regionZeroAddr, NbrListLen, NbrList, 1,
regionZeroAddr);
- void *BlockingCUTCPNode = __hpvm__createNodeND(0, CUTCPRoot);
+ void *BlockingCUTCPNode = __visc__createNodeND(0, CUTCPRoot);
// Bind Inputs
- __hpvm__bindIn(BlockingCUTCPNode, 0, 0, 0); // Bind binDim_x
- __hpvm__bindIn(BlockingCUTCPNode, 1, 1, 0); // Bind binDim_y
- __hpvm__bindIn(BlockingCUTCPNode, 2, 2, 0); // Bind binBaseAddr
- __hpvm__bindIn(BlockingCUTCPNode, 3, 3, 0); // Bind bytes_binBaseAddr
- __hpvm__bindIn(BlockingCUTCPNode, 4, 4, 0); // Bind offset
- __hpvm__bindIn(BlockingCUTCPNode, 5, 5, 0); // Bind h
- __hpvm__bindIn(BlockingCUTCPNode, 6, 6, 0); // Bind cutoff2
- __hpvm__bindIn(BlockingCUTCPNode, 7, 7, 0); // Bind inv_cutoff2
- __hpvm__bindIn(BlockingCUTCPNode, 8, 8, 0); // Bind regionZeroAddr
- __hpvm__bindIn(BlockingCUTCPNode, 9, 9, 0); // Bind bytes_regionZeroAddr
- __hpvm__bindIn(BlockingCUTCPNode, 10, 10, 0); // Bind zRegionIndex
- __hpvm__bindIn(BlockingCUTCPNode, 11, 11, 0); // Bind NbrListLen
- __hpvm__bindIn(BlockingCUTCPNode, 12, 12, 0); // Bind bytes_NbrListLen
- __hpvm__bindIn(BlockingCUTCPNode, 13, 13, 0); // Bind NbrList
- __hpvm__bindIn(BlockingCUTCPNode, 14, 14, 0); // Bind bytes_NbrList
- __hpvm__bindIn(BlockingCUTCPNode, 15, 15, 0); // Bind blockx
- __hpvm__bindIn(BlockingCUTCPNode, 16, 16, 0); // Bind blocky
- __hpvm__bindIn(BlockingCUTCPNode, 17, 17, 0); // Bind blockz
- __hpvm__bindIn(BlockingCUTCPNode, 18, 18, 0); // Bind gridx
- __hpvm__bindIn(BlockingCUTCPNode, 19, 19, 0); // Bind gridy
- __hpvm__bindIn(BlockingCUTCPNode, 20, 20, 0); // Bind gridz
+ __visc__bindIn(BlockingCUTCPNode, 0, 0, 0); // Bind binDim_x
+ __visc__bindIn(BlockingCUTCPNode, 1, 1, 0); // Bind binDim_y
+ __visc__bindIn(BlockingCUTCPNode, 2, 2, 0); // Bind binBaseAddr
+ __visc__bindIn(BlockingCUTCPNode, 3, 3, 0); // Bind bytes_binBaseAddr
+ __visc__bindIn(BlockingCUTCPNode, 4, 4, 0); // Bind offset
+ __visc__bindIn(BlockingCUTCPNode, 5, 5, 0); // Bind h
+ __visc__bindIn(BlockingCUTCPNode, 6, 6, 0); // Bind cutoff2
+ __visc__bindIn(BlockingCUTCPNode, 7, 7, 0); // Bind inv_cutoff2
+ __visc__bindIn(BlockingCUTCPNode, 8, 8, 0); // Bind regionZeroAddr
+ __visc__bindIn(BlockingCUTCPNode, 9, 9, 0); // Bind bytes_regionZeroAddr
+ __visc__bindIn(BlockingCUTCPNode, 10, 10, 0); // Bind zRegionIndex
+ __visc__bindIn(BlockingCUTCPNode, 11, 11, 0); // Bind NbrListLen
+ __visc__bindIn(BlockingCUTCPNode, 12, 12, 0); // Bind bytes_NbrListLen
+ __visc__bindIn(BlockingCUTCPNode, 13, 13, 0); // Bind NbrList
+ __visc__bindIn(BlockingCUTCPNode, 14, 14, 0); // Bind bytes_NbrList
+ __visc__bindIn(BlockingCUTCPNode, 15, 15, 0); // Bind blockx
+ __visc__bindIn(BlockingCUTCPNode, 16, 16, 0); // Bind blocky
+ __visc__bindIn(BlockingCUTCPNode, 17, 17, 0); // Bind blockz
+ __visc__bindIn(BlockingCUTCPNode, 18, 18, 0); // Bind gridx
+ __visc__bindIn(BlockingCUTCPNode, 19, 19, 0); // Bind gridy
+ __visc__bindIn(BlockingCUTCPNode, 20, 20, 0); // Bind gridz
}
// ==================== Host Code ==============================
@@ -546,7 +546,7 @@ int main(int argc, char *argv[]) {
}
pb_InitializeTimerSet(&timers);
- __hpvm__init();
+ __visc__init();
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
@@ -586,7 +586,7 @@ int main(int argc, char *argv[]) {
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
pb_PrintTimerSet(&timers);
- __hpvm__cleanup();
+ __visc__cleanup();
/* Print output */
// pb_SwitchToTimer(&timers, pb_TimerID_IO);
@@ -873,11 +873,11 @@ int gpu_compute_cutoff_potential_lattice6overlap(
printf("\n");
}
- // Track hpvm data
- llvm_hpvm_track_mem(regionZeroAddr, bytes_regionZeroAddr);
- llvm_hpvm_track_mem(binBaseAddr, bytes_binBaseAddr);
- llvm_hpvm_track_mem(nbrlistlen, sizeof(int));
- llvm_hpvm_track_mem(nbrlist, bytes_nbrlist);
+ // Track visc data
+ llvm_visc_track_mem(regionZeroAddr, bytes_regionZeroAddr);
+ llvm_visc_track_mem(binBaseAddr, bytes_binBaseAddr);
+ llvm_visc_track_mem(nbrlistlen, sizeof(int));
+ llvm_visc_track_mem(nbrlist, bytes_nbrlist);
/* setup OpenCL kernel parameters */
blockDim[0] = 8;
@@ -914,7 +914,7 @@ int gpu_compute_cutoff_potential_lattice6overlap(
gridDim[1], gridDim[2]);
/* loop over z-dimension, invoke OpenCL kernel for each x-y plane */
- pb_SwitchToTimer(timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(timers, visc_TimerID_COMPUTATION);
void *CUTCP_DFG;
if (verbose)
printf("Invoking OpenCL kernel on %d region planes...\n", zRegionDim);
@@ -926,9 +926,9 @@ int gpu_compute_cutoff_potential_lattice6overlap(
args->zRegionIndex = zRegionIndex;
- CUTCP_DFG = __hpvm__launch(0, CUTCPWrapper, (void *)args);
- __hpvm__wait(CUTCP_DFG);
- // llvm_hpvm_request_mem(regionZeroAddr, lnall*sizeof(ener_t));
+ CUTCP_DFG = __visc__launch(0, CUTCPWrapper, (void *)args);
+ __visc__wait(CUTCP_DFG);
+ // llvm_visc_request_mem(regionZeroAddr, lnall*sizeof(ener_t));
}
/*
@@ -941,14 +941,14 @@ int gpu_compute_cutoff_potential_lattice6overlap(
printf("computing extra atoms on CPU\n");
}
- pb_SwitchToTimer(timers, hpvm_TimerID_MISC);
+ pb_SwitchToTimer(timers, visc_TimerID_MISC);
if (cpu_compute_cutoff_potential_lattice(lattice, cutoff, extra)) {
fprintf(stderr, "cpu_compute_cutoff_potential_lattice() failed "
"for extra atoms\n");
return -1;
}
- pb_SwitchToTimer(timers, hpvm_TimerID_MISC);
+ pb_SwitchToTimer(timers, visc_TimerID_MISC);
printf("\n");
}
if (verbose)
@@ -957,7 +957,7 @@ int gpu_compute_cutoff_potential_lattice6overlap(
/* copy result regions from OpenCL device */
pb_SwitchToTimer(timers, pb_TimerID_COPY);
- llvm_hpvm_request_mem(regionZeroAddr, lnall * sizeof(ener_t));
+ llvm_visc_request_mem(regionZeroAddr, lnall * sizeof(ener_t));
/*
* transpose on CPU, updating, producing the final lattice
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/ocl.c b/hpvm/test/parboil/benchmarks/cutcp/src/visc/ocl.c
similarity index 100%
rename from hpvm/test/parboil/benchmarks/cutcp/src/hpvm/ocl.c
rename to hpvm/test/parboil/benchmarks/cutcp/src/visc/ocl.c
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/ocl.h b/hpvm/test/parboil/benchmarks/cutcp/src/visc/ocl.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/cutcp/src/hpvm/ocl.h
rename to hpvm/test/parboil/benchmarks/cutcp/src/visc/ocl.h
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/output.c b/hpvm/test/parboil/benchmarks/cutcp/src/visc/output.c
similarity index 100%
rename from hpvm/test/parboil/benchmarks/cutcp/src/hpvm/output.c
rename to hpvm/test/parboil/benchmarks/cutcp/src/visc/output.c
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/output.h b/hpvm/test/parboil/benchmarks/cutcp/src/visc/output.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/cutcp/src/hpvm/output.h
rename to hpvm/test/parboil/benchmarks/cutcp/src/visc/output.h
diff --git a/hpvm/test/parboil/benchmarks/cutcp/src/hpvm/readatom.c b/hpvm/test/parboil/benchmarks/cutcp/src/visc/readatom.c
similarity index 100%
rename from hpvm/test/parboil/benchmarks/cutcp/src/hpvm/readatom.c
rename to hpvm/test/parboil/benchmarks/cutcp/src/visc/readatom.c
diff --git a/hpvm/test/parboil/benchmarks/lbm/Makefile b/hpvm/test/parboil/benchmarks/lbm/Makefile
index af7215ff70..4ebf6fc0af 100644
--- a/hpvm/test/parboil/benchmarks/lbm/Makefile
+++ b/hpvm/test/parboil/benchmarks/lbm/Makefile
@@ -5,9 +5,9 @@ ifeq ($(NUM_CORES),)
NUM_CORES=8
endif
-# Default compile hpvm
+# Default compile visc
ifeq ($(VERSION),)
- VERSION = hpvm
+ VERSION = visc
endif
# Default use small test case
diff --git a/hpvm/test/parboil/benchmarks/lbm/src/opencl_cpu_baseline/main.c b/hpvm/test/parboil/benchmarks/lbm/src/opencl_cpu_baseline/main.c
index a55f0ce785..59aa8daf9a 100644
--- a/hpvm/test/parboil/benchmarks/lbm/src/opencl_cpu_baseline/main.c
+++ b/hpvm/test/parboil/benchmarks/lbm/src/opencl_cpu_baseline/main.c
@@ -79,7 +79,7 @@ int main(int nArgs, char *arg[]) {
OpenCL_LBM_initializeGrid(&prm, OpenCL_dstGrid, TEMP_dstGrid);
clFinish(prm.clCommandQueue);
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (t = 1; t <= param.nTimeSteps; t++) {
/*pb_SwitchToTimer(&timers, pb_TimerID_KERNEL);*/
@@ -101,7 +101,7 @@ int main(int nArgs, char *arg[]) {
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
OpenCL_LBM_getDeviceGrid(&prm, OpenCL_srcGrid, TEMP_srcGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCL_LBM_freeGrid(OpenCL_srcGrid);
OpenCL_LBM_freeGrid(OpenCL_dstGrid);
diff --git a/hpvm/test/parboil/benchmarks/lbm/src/opencl_cpu_long/main.c b/hpvm/test/parboil/benchmarks/lbm/src/opencl_cpu_long/main.c
index 64fe482b81..d93a919df3 100644
--- a/hpvm/test/parboil/benchmarks/lbm/src/opencl_cpu_long/main.c
+++ b/hpvm/test/parboil/benchmarks/lbm/src/opencl_cpu_long/main.c
@@ -65,7 +65,7 @@ int main(int nArgs, char *arg[]) {
LBM_showGridStatistics(TEMP_srcGrid);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCL_initialize(&prm);
@@ -78,7 +78,7 @@ int main(int nArgs, char *arg[]) {
OpenCL_LBM_initializeGrid(&prm, OpenCL_srcGrid, TEMP_srcGrid);
OpenCL_LBM_initializeGrid(&prm, OpenCL_dstGrid, TEMP_dstGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (t = 1; t <= param.nTimeSteps; t++) {
/*pb_SwitchToTimer(&timers, pb_TimerID_KERNEL);*/
@@ -100,7 +100,7 @@ int main(int nArgs, char *arg[]) {
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
OpenCL_LBM_getDeviceGrid(&prm, OpenCL_srcGrid, TEMP_srcGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCL_LBM_freeGrid(OpenCL_srcGrid);
OpenCL_LBM_freeGrid(OpenCL_dstGrid);
@@ -197,7 +197,7 @@ void MAIN_initialize(const MAIN_Param *param, const OpenCL_Param *prm) {
LBM_initializeSpecialCellsForLDC(TEMP_srcGrid);
LBM_initializeSpecialCellsForLDC(TEMP_dstGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
// Setup DEVICE datastructures
OpenCL_LBM_allocateGrid(prm, &OpenCL_srcGrid);
OpenCL_LBM_allocateGrid(prm, &OpenCL_dstGrid);
@@ -233,7 +233,7 @@ void MAIN_finalize(const MAIN_Param *param, const OpenCL_Param *prm) {
LBM_freeGrid((float **)&TEMP_srcGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCL_LBM_freeGrid(OpenCL_srcGrid);
OpenCL_LBM_freeGrid(OpenCL_dstGrid);
diff --git a/hpvm/test/parboil/benchmarks/lbm/src/opencl_cpu_short/main.c b/hpvm/test/parboil/benchmarks/lbm/src/opencl_cpu_short/main.c
index 64fe482b81..d93a919df3 100644
--- a/hpvm/test/parboil/benchmarks/lbm/src/opencl_cpu_short/main.c
+++ b/hpvm/test/parboil/benchmarks/lbm/src/opencl_cpu_short/main.c
@@ -65,7 +65,7 @@ int main(int nArgs, char *arg[]) {
LBM_showGridStatistics(TEMP_srcGrid);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCL_initialize(&prm);
@@ -78,7 +78,7 @@ int main(int nArgs, char *arg[]) {
OpenCL_LBM_initializeGrid(&prm, OpenCL_srcGrid, TEMP_srcGrid);
OpenCL_LBM_initializeGrid(&prm, OpenCL_dstGrid, TEMP_dstGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (t = 1; t <= param.nTimeSteps; t++) {
/*pb_SwitchToTimer(&timers, pb_TimerID_KERNEL);*/
@@ -100,7 +100,7 @@ int main(int nArgs, char *arg[]) {
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
OpenCL_LBM_getDeviceGrid(&prm, OpenCL_srcGrid, TEMP_srcGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCL_LBM_freeGrid(OpenCL_srcGrid);
OpenCL_LBM_freeGrid(OpenCL_dstGrid);
@@ -197,7 +197,7 @@ void MAIN_initialize(const MAIN_Param *param, const OpenCL_Param *prm) {
LBM_initializeSpecialCellsForLDC(TEMP_srcGrid);
LBM_initializeSpecialCellsForLDC(TEMP_dstGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
// Setup DEVICE datastructures
OpenCL_LBM_allocateGrid(prm, &OpenCL_srcGrid);
OpenCL_LBM_allocateGrid(prm, &OpenCL_dstGrid);
@@ -233,7 +233,7 @@ void MAIN_finalize(const MAIN_Param *param, const OpenCL_Param *prm) {
LBM_freeGrid((float **)&TEMP_srcGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCL_LBM_freeGrid(OpenCL_srcGrid);
OpenCL_LBM_freeGrid(OpenCL_dstGrid);
diff --git a/hpvm/test/parboil/benchmarks/lbm/src/opencl_nvidia/main.c b/hpvm/test/parboil/benchmarks/lbm/src/opencl_nvidia/main.c
index 54399ee119..18320b7394 100644
--- a/hpvm/test/parboil/benchmarks/lbm/src/opencl_nvidia/main.c
+++ b/hpvm/test/parboil/benchmarks/lbm/src/opencl_nvidia/main.c
@@ -79,7 +79,7 @@ int main(int nArgs, char *arg[]) {
OpenCL_LBM_initializeGrid(&prm, OpenCL_dstGrid, TEMP_dstGrid);
for (t = 1; t <= param.nTimeSteps; t++) {
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
OpenCL_LBM_performStreamCollide(&prm, OpenCL_srcGrid, OpenCL_dstGrid);
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
LBM_swapGrids(&OpenCL_srcGrid, &OpenCL_dstGrid);
diff --git a/hpvm/test/parboil/benchmarks/lbm/src/opencl_nvidia_long/main.c b/hpvm/test/parboil/benchmarks/lbm/src/opencl_nvidia_long/main.c
index 6d682e98e6..5e43b75427 100644
--- a/hpvm/test/parboil/benchmarks/lbm/src/opencl_nvidia_long/main.c
+++ b/hpvm/test/parboil/benchmarks/lbm/src/opencl_nvidia_long/main.c
@@ -65,7 +65,7 @@ int main(int nArgs, char *arg[]) {
LBM_showGridStatistics(TEMP_srcGrid);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCL_initialize(&prm);
@@ -78,7 +78,7 @@ int main(int nArgs, char *arg[]) {
OpenCL_LBM_initializeGrid(&prm, OpenCL_srcGrid, TEMP_srcGrid);
OpenCL_LBM_initializeGrid(&prm, OpenCL_dstGrid, TEMP_dstGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (int i = 0; i < 1; i++) {
for (t = 1; t <= param.nTimeSteps; t++) {
/*pb_SwitchToTimer(&timers, pb_TimerID_KERNEL);*/
@@ -100,7 +100,7 @@ int main(int nArgs, char *arg[]) {
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
OpenCL_LBM_getDeviceGrid(&prm, OpenCL_srcGrid, TEMP_srcGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCL_LBM_freeGrid(OpenCL_srcGrid);
OpenCL_LBM_freeGrid(OpenCL_dstGrid);
@@ -197,7 +197,7 @@ void MAIN_initialize(const MAIN_Param *param, const OpenCL_Param *prm) {
LBM_initializeSpecialCellsForLDC(TEMP_srcGrid);
LBM_initializeSpecialCellsForLDC(TEMP_dstGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
// Setup DEVICE datastructures
OpenCL_LBM_allocateGrid(prm, &OpenCL_srcGrid);
OpenCL_LBM_allocateGrid(prm, &OpenCL_dstGrid);
@@ -233,7 +233,7 @@ void MAIN_finalize(const MAIN_Param *param, const OpenCL_Param *prm) {
LBM_freeGrid((float **)&TEMP_srcGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCL_LBM_freeGrid(OpenCL_srcGrid);
OpenCL_LBM_freeGrid(OpenCL_dstGrid);
diff --git a/hpvm/test/parboil/benchmarks/lbm/src/opencl_nvidia_short/main.c b/hpvm/test/parboil/benchmarks/lbm/src/opencl_nvidia_short/main.c
index 9dc95e7d85..e66cb2c47c 100644
--- a/hpvm/test/parboil/benchmarks/lbm/src/opencl_nvidia_short/main.c
+++ b/hpvm/test/parboil/benchmarks/lbm/src/opencl_nvidia_short/main.c
@@ -65,7 +65,7 @@ int main(int nArgs, char *arg[]) {
LBM_showGridStatistics(TEMP_srcGrid);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCL_initialize(&prm);
@@ -78,7 +78,7 @@ int main(int nArgs, char *arg[]) {
OpenCL_LBM_initializeGrid(&prm, OpenCL_srcGrid, TEMP_srcGrid);
OpenCL_LBM_initializeGrid(&prm, OpenCL_dstGrid, TEMP_dstGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (int i = 0; i < 4; i++) {
for (t = 1; t <= param.nTimeSteps; t++) {
/*pb_SwitchToTimer(&timers, pb_TimerID_KERNEL);*/
@@ -100,7 +100,7 @@ int main(int nArgs, char *arg[]) {
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
OpenCL_LBM_getDeviceGrid(&prm, OpenCL_srcGrid, TEMP_srcGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCL_LBM_freeGrid(OpenCL_srcGrid);
OpenCL_LBM_freeGrid(OpenCL_dstGrid);
@@ -197,7 +197,7 @@ void MAIN_initialize(const MAIN_Param *param, const OpenCL_Param *prm) {
LBM_initializeSpecialCellsForLDC(TEMP_srcGrid);
LBM_initializeSpecialCellsForLDC(TEMP_dstGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
// Setup DEVICE datastructures
OpenCL_LBM_allocateGrid(prm, &OpenCL_srcGrid);
OpenCL_LBM_allocateGrid(prm, &OpenCL_dstGrid);
@@ -233,7 +233,7 @@ void MAIN_finalize(const MAIN_Param *param, const OpenCL_Param *prm) {
LBM_freeGrid((float **)&TEMP_srcGrid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCL_LBM_freeGrid(OpenCL_srcGrid);
OpenCL_LBM_freeGrid(OpenCL_dstGrid);
diff --git a/hpvm/test/parboil/benchmarks/lbm/src/hpvm/Makefile b/hpvm/test/parboil/benchmarks/lbm/src/visc/Makefile
similarity index 85%
rename from hpvm/test/parboil/benchmarks/lbm/src/hpvm/Makefile
rename to hpvm/test/parboil/benchmarks/lbm/src/visc/Makefile
index 5aa206f758..d1664ee988 100644
--- a/hpvm/test/parboil/benchmarks/lbm/src/hpvm/Makefile
+++ b/hpvm/test/parboil/benchmarks/lbm/src/visc/Makefile
@@ -1,8 +1,8 @@
# (c) 2010 The Board of Trustees of the University of Illinois.
-LANGUAGE=hpvm
+LANGUAGE=visc
SRCDIR_OBJS=lbm.ll
-HPVM_OBJS=main.hpvm.ll
+VISC_OBJS=main.visc.ll
APP_CUDALDFLAGS=-lm
APP_CFLAGS=-ffast-math -O3 -DNUM_CORES=$(NUM_CORES)
APP_CXXFLAGS=-ffast-math -O3 -DNUM_CORES=$(NUM_CORES)
diff --git a/hpvm/test/parboil/benchmarks/lbm/src/hpvm/layout_config.h b/hpvm/test/parboil/benchmarks/lbm/src/visc/layout_config.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/lbm/src/hpvm/layout_config.h
rename to hpvm/test/parboil/benchmarks/lbm/src/visc/layout_config.h
diff --git a/hpvm/test/parboil/benchmarks/lbm/src/hpvm/lbm.cpp b/hpvm/test/parboil/benchmarks/lbm/src/visc/lbm.cpp
similarity index 100%
rename from hpvm/test/parboil/benchmarks/lbm/src/hpvm/lbm.cpp
rename to hpvm/test/parboil/benchmarks/lbm/src/visc/lbm.cpp
diff --git a/hpvm/test/parboil/benchmarks/lbm/src/hpvm/lbm.h b/hpvm/test/parboil/benchmarks/lbm/src/visc/lbm.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/lbm/src/hpvm/lbm.h
rename to hpvm/test/parboil/benchmarks/lbm/src/visc/lbm.h
diff --git a/hpvm/test/parboil/benchmarks/lbm/src/hpvm/lbm_macros.h b/hpvm/test/parboil/benchmarks/lbm/src/visc/lbm_macros.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/lbm/src/hpvm/lbm_macros.h
rename to hpvm/test/parboil/benchmarks/lbm/src/visc/lbm_macros.h
diff --git a/hpvm/test/parboil/benchmarks/lbm/src/hpvm/main.cpp b/hpvm/test/parboil/benchmarks/lbm/src/visc/main.cpp
similarity index 86%
rename from hpvm/test/parboil/benchmarks/lbm/src/hpvm/main.cpp
rename to hpvm/test/parboil/benchmarks/lbm/src/visc/main.cpp
index 32db8e9b2c..b51864366b 100644
--- a/hpvm/test/parboil/benchmarks/lbm/src/hpvm/main.cpp
+++ b/hpvm/test/parboil/benchmarks/lbm/src/visc/main.cpp
@@ -8,11 +8,11 @@
/*############################################################################*/
-#include <hpvm.h>
#include <parboil.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/stat.h>
+#include <visc.h>
#include "layout_config.h"
#include "lbm.h"
@@ -92,18 +92,18 @@ typedef struct __attribute__((__packed__)) {
void performStreamCollide_kernel(float *srcG, size_t bytes_srcG, float *dstG,
size_t bytes_dstG) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(2, srcG, dstG, 1, dstG);
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(2, srcG, dstG, 1, dstG);
- void *thisNode = __hpvm__getNode();
- void *parentNode = __hpvm__getParentNode(thisNode);
+ void *thisNode = __visc__getNode();
+ void *parentNode = __visc__getParentNode(thisNode);
srcG += MARGIN;
dstG += MARGIN;
- int lx = __hpvm__getNodeInstanceID_x(thisNode);
- int gx = __hpvm__getNodeInstanceID_x(parentNode);
- int gy = __hpvm__getNodeInstanceID_y(parentNode);
+ int lx = __visc__getNodeInstanceID_x(thisNode);
+ int gx = __visc__getNodeInstanceID_x(parentNode);
+ int gy = __visc__getNodeInstanceID_y(parentNode);
// Using some predefined macros here. Consider this the declaration
// and initialization of the variables SWEEP_X, SWEEP_Y and SWEEP_Z
@@ -274,40 +274,40 @@ void performStreamCollide_kernel(float *srcG, size_t bytes_srcG, float *dstG,
void lbmLvl1(float *srcG, size_t bytes_srcG, float *dstG, size_t bytes_dstG,
size_t dim_X1) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(2, srcG, dstG, 1, dstG);
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(2, srcG, dstG, 1, dstG);
void *lbm_node =
- __hpvm__createNodeND(2, performStreamCollide_kernel, dim_X1, (size_t)1);
- __hpvm__bindIn(lbm_node, 0, 0, 0);
- __hpvm__bindIn(lbm_node, 1, 1, 0);
- __hpvm__bindIn(lbm_node, 2, 2, 0);
- __hpvm__bindIn(lbm_node, 3, 3, 0);
+ __visc__createNodeND(2, performStreamCollide_kernel, dim_X1, (size_t)1);
+ __visc__bindIn(lbm_node, 0, 0, 0);
+ __visc__bindIn(lbm_node, 1, 1, 0);
+ __visc__bindIn(lbm_node, 2, 2, 0);
+ __visc__bindIn(lbm_node, 3, 3, 0);
}
void lbmLvl2(float *srcG, size_t bytes_srcG, float *dstG, size_t bytes_dstG,
size_t dim_X1, size_t dim_X2, size_t dim_Y2) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(2, srcG, dstG, 1, dstG);
- void *lbm_node = __hpvm__createNodeND(2, lbmLvl1, dim_X2, dim_Y2);
- __hpvm__bindIn(lbm_node, 0, 0, 0);
- __hpvm__bindIn(lbm_node, 1, 1, 0);
- __hpvm__bindIn(lbm_node, 2, 2, 0);
- __hpvm__bindIn(lbm_node, 3, 3, 0);
- __hpvm__bindIn(lbm_node, 4, 4, 0);
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(2, srcG, dstG, 1, dstG);
+ void *lbm_node = __visc__createNodeND(2, lbmLvl1, dim_X2, dim_Y2);
+ __visc__bindIn(lbm_node, 0, 0, 0);
+ __visc__bindIn(lbm_node, 1, 1, 0);
+ __visc__bindIn(lbm_node, 2, 2, 0);
+ __visc__bindIn(lbm_node, 3, 3, 0);
+ __visc__bindIn(lbm_node, 4, 4, 0);
}
void lbmLvl3(float *srcG, size_t bytes_srcG, float *dstG, size_t bytes_dstG,
size_t dim_X1, size_t dim_X2, size_t dim_Y2) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(2, srcG, dstG, 1, dstG);
- void *lbm_node = __hpvm__createNodeND(0, lbmLvl2);
- __hpvm__bindIn(lbm_node, 0, 0, 0);
- __hpvm__bindIn(lbm_node, 1, 1, 0);
- __hpvm__bindIn(lbm_node, 2, 2, 0);
- __hpvm__bindIn(lbm_node, 3, 3, 0);
- __hpvm__bindIn(lbm_node, 4, 4, 0);
- __hpvm__bindIn(lbm_node, 5, 5, 0);
- __hpvm__bindIn(lbm_node, 6, 6, 0);
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(2, srcG, dstG, 1, dstG);
+ void *lbm_node = __visc__createNodeND(0, lbmLvl2);
+ __visc__bindIn(lbm_node, 0, 0, 0);
+ __visc__bindIn(lbm_node, 1, 1, 0);
+ __visc__bindIn(lbm_node, 2, 2, 0);
+ __visc__bindIn(lbm_node, 3, 3, 0);
+ __visc__bindIn(lbm_node, 4, 4, 0);
+ __visc__bindIn(lbm_node, 5, 5, 0);
+ __visc__bindIn(lbm_node, 6, 6, 0);
}
__attribute__((noinline)) void MAIN_performStreamCollide(LBM_Grid src,
@@ -321,9 +321,9 @@ __attribute__((noinline)) void MAIN_performStreamCollide(LBM_Grid src,
RootIn root_in_local = {src - MARGIN, size, dst - MARGIN, size,
SIZE_X, SIZE_Y, SIZE_Z};
*(RootIn *)root_in = root_in_local;
- void *lbmDFG = __hpvm__launch(0, lbmLvl3, root_in);
+ void *lbmDFG = __visc__launch(0, lbmLvl3, root_in);
- __hpvm__wait(lbmDFG);
+ __visc__wait(lbmDFG);
}
void MAIN_initialize(const MAIN_Param *param) {
@@ -379,12 +379,12 @@ int main(int nArgs, char *arg[]) {
MAIN_initialize(¶m);
pb_InitializeTimerSet(&timers);
- __hpvm__init();
+ __visc__init();
size_t size = TOTAL_PADDED_CELLS * N_CELL_ENTRIES * sizeof(float);
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_TRACK);
- llvm_hpvm_track_mem(srcGrid - MARGIN, size);
- llvm_hpvm_track_mem(dstGrid - MARGIN, size);
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_TRACK);
+ llvm_visc_track_mem(srcGrid - MARGIN, size);
+ llvm_visc_track_mem(dstGrid - MARGIN, size);
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
for (t = 1; t <= param.nTimeSteps; t++) {
@@ -404,15 +404,15 @@ int main(int nArgs, char *arg[]) {
}
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
- llvm_hpvm_request_mem(srcGrid - MARGIN, size);
+ llvm_visc_request_mem(srcGrid - MARGIN, size);
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_UNTRACK);
- llvm_hpvm_untrack_mem(srcGrid - MARGIN);
- llvm_hpvm_untrack_mem(dstGrid - MARGIN);
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_UNTRACK);
+ llvm_visc_untrack_mem(srcGrid - MARGIN);
+ llvm_visc_untrack_mem(dstGrid - MARGIN);
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
pb_PrintTimerSet(&timers);
- __hpvm__cleanup();
+ __visc__cleanup();
/*pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);*/
MAIN_finalize(¶m);
diff --git a/hpvm/test/parboil/benchmarks/lbm/src/hpvm/main.h b/hpvm/test/parboil/benchmarks/lbm/src/visc/main.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/lbm/src/hpvm/main.h
rename to hpvm/test/parboil/benchmarks/lbm/src/visc/main.h
diff --git a/hpvm/test/parboil/benchmarks/sgemm/Makefile b/hpvm/test/parboil/benchmarks/sgemm/Makefile
index 4757432d22..ace9ded22b 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/Makefile
+++ b/hpvm/test/parboil/benchmarks/sgemm/Makefile
@@ -1,9 +1,9 @@
PARBOIL_ROOT = $(LLVM_SRC_ROOT)/tools/hpvm/test/parboil
APP = sgemm
-# Default compile hpvm
+# Default compile visc
ifeq ($(VERSION),)
- VERSION = hpvm_sh
+ VERSION = visc_sh
endif
# Default use small test case
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec_opt/Makefile b/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec_opt/Makefile
deleted file mode 100644
index 2234bf54e1..0000000000
--- a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec_opt/Makefile
+++ /dev/null
@@ -1,8 +0,0 @@
-# (c) 2010 The Board of Trustees of the University of Illinois.
-
-LANGUAGE=hpvm
-SRCDIR_OBJS=io.ll #compute_gold.o
-HPVM_OBJS=main.hpvm.ll
-APP_CUDALDFLAGS=-lm -lstdc++
-APP_CFLAGS=-ffast-math -O3
-APP_CXXFLAGS=-ffast-math -O3
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_base/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_base/main.cc
index e8d1c69ec9..5489f6a55c 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_base/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_base/main.cc
@@ -109,7 +109,7 @@ int main(int argc, char *argv[]) {
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_platform_id clPlatform;
clStatus = clGetPlatformIDs(1, &clPlatform, NULL);
@@ -212,7 +212,7 @@ int main(int argc, char *argv[]) {
clEnqueueReadBuffer(clCommandQueue, dC, CL_TRUE, 0, C_sz, &matC.front(), 0,
NULL, NULL);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseKernel(clKernel);
clStatus = clReleaseProgram(clProgram);
clStatus = clReleaseMemObject(dA);
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_base_opt/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_base_opt/main.cc
index 4285a52a01..105baf590d 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_base_opt/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_base_opt/main.cc
@@ -120,7 +120,7 @@ int main(int argc, char *argv[]) {
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_platform_id clPlatform;
clStatus = clGetPlatformIDs(1, &clPlatform, NULL);
@@ -208,7 +208,7 @@ int main(int argc, char *argv[]) {
clEnqueueReadBuffer(clCommandQueue, dC, CL_TRUE, 0, C_sz, &matC.front(), 0,
NULL, NULL);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clReleaseKernel(clKernel);
clReleaseProgram(clProgram);
clReleaseMemObject(dA);
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_base_vec/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_base_vec/main.cc
index 7edbf05a4b..f72c18c293 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_base_vec/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_base_vec/main.cc
@@ -112,7 +112,7 @@ int main(int argc, char *argv[]) {
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_platform_id clPlatform;
clStatus = clGetPlatformIDs(1, &clPlatform, NULL);
@@ -204,7 +204,7 @@ int main(int argc, char *argv[]) {
clEnqueueReadBuffer(clCommandQueue, dC, CL_TRUE, 0, C_sz, &matC.front(), 0,
NULL, NULL);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseKernel(clKernel);
clStatus = clReleaseProgram(clProgram);
clStatus = clReleaseMemObject(dA);
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu/main.cc
index cccec04beb..744ee40966 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu/main.cc
@@ -110,7 +110,7 @@ int main(int argc, char *argv[]) {
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_uint numPlatforms;
@@ -254,7 +254,7 @@ int main(int argc, char *argv[]) {
clEnqueueReadBuffer(clCommandQueue, dC, CL_TRUE, 0, C_sz, &matC.front(), 0,
NULL, NULL);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseKernel(clKernel);
clStatus = clReleaseProgram(clProgram);
clStatus = clReleaseMemObject(dA);
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_4K/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_4K/main.cc
index 36e7b93571..45ed8e942a 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_4K/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_4K/main.cc
@@ -83,7 +83,7 @@ void basicSgemm(char transa, char transb, int m, int n, int k, float alpha,
clStatus = clSetKernelArg(clKernel, 8, sizeof(float), (void *)&beta);
CHECK_ERROR("clSetKernelArg")
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (int i = 0; i < 1; i++) {
clStatus = clEnqueueNDRangeKernel(clCommandQueue, clKernel, 2, NULL, dg, db,
@@ -123,7 +123,7 @@ int main(int argc, char *argv[]) {
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_uint numPlatforms;
@@ -247,7 +247,7 @@ int main(int argc, char *argv[]) {
clEnqueueReadBuffer(clCommandQueue, dC, CL_TRUE, 0, C_sz, &matC.front(), 0,
NULL, NULL);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseKernel(clKernel);
clStatus = clReleaseProgram(clProgram);
clStatus = clReleaseMemObject(dA);
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_baseline/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_baseline/main.cc
index 2cc311d1ef..d8275be777 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_baseline/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_baseline/main.cc
@@ -83,7 +83,7 @@ void basicSgemm(char transa, char transb, int m, int n, int k, float alpha,
clStatus = clSetKernelArg(clKernel, 8, sizeof(float), (void *)&beta);
CHECK_ERROR("clSetKernelArg")
- // pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ // pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
// for(int i=0; i<15; i++) {
clStatus = clEnqueueNDRangeKernel(clCommandQueue, clKernel, 2, NULL, dg, db,
@@ -123,7 +123,7 @@ int main(int argc, char *argv[]) {
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_uint numPlatforms;
@@ -212,7 +212,7 @@ int main(int argc, char *argv[]) {
clEnqueueReadBuffer(clCommandQueue, dC, CL_TRUE, 0, C_sz, &matC.front(), 0,
NULL, NULL);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseKernel(clKernel);
clStatus = clReleaseProgram(clProgram);
clStatus = clReleaseMemObject(dA);
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_medium/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_medium/main.cc
index 678b4d8131..b4e561ded6 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_medium/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_medium/main.cc
@@ -83,7 +83,7 @@ void basicSgemm(char transa, char transb, int m, int n, int k, float alpha,
clStatus = clSetKernelArg(clKernel, 8, sizeof(float), (void *)&beta);
CHECK_ERROR("clSetKernelArg")
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
// for(int i=0; i<15; i++) {
clStatus = clEnqueueNDRangeKernel(clCommandQueue, clKernel, 2, NULL, dg, db,
@@ -123,7 +123,7 @@ int main(int argc, char *argv[]) {
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_uint numPlatforms;
@@ -247,7 +247,7 @@ int main(int argc, char *argv[]) {
clEnqueueReadBuffer(clCommandQueue, dC, CL_TRUE, 0, C_sz, &matC.front(), 0,
NULL, NULL);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseKernel(clKernel);
clStatus = clReleaseProgram(clProgram);
clStatus = clReleaseMemObject(dA);
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_sm/kernel-spir64.ll b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_sm/kernel-spir64.ll
index 9b4cf7702d..ca9fcca060 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_sm/kernel-spir64.ll
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_sm/kernel-spir64.ll
@@ -1,4 +1,4 @@
-; ModuleID = '/home/psrivas2.hpvm.llvm/test/HPVM/parboil/benchmarks/sgemm/src/opencl_cpu_sm/kernel.cl'
+; ModuleID = '/home/psrivas2/visc/llvm/test/VISC/parboil/benchmarks/sgemm/src/opencl_cpu_sm/kernel.cl'
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024"
target triple = "spir64-unknown-unknown"
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_sm/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_sm/main.cc
index 79fecfb84b..8de437a4f8 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_sm/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_sm/main.cc
@@ -195,7 +195,7 @@ int main(int argc, char *argv[]) {
&matC.front(), 0, NULL, NULL);
CHECK_ERROR("clEnqueueWriteBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
// Use standard sgemm interface
regtileSgemm('N', 'T', matArow, matBcol, matAcol, 1.0f, dA, matArow, dB,
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_sm/test-spir64.ll b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_sm/test-spir64.ll
index 2f72a6ceba..908c7104bb 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_sm/test-spir64.ll
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_cpu_sm/test-spir64.ll
@@ -1,4 +1,4 @@
-; ModuleID = '/home/psrivas2.hpvm.llvm/test/HPVM/parboil/benchmarks/sgemm/src/opencl_cpu_sm/test.cl'
+; ModuleID = '/home/psrivas2/visc/llvm/test/VISC/parboil/benchmarks/sgemm/src/opencl_cpu_sm/test.cl'
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024"
target triple = "spir64-unknown-unknown"
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_nvidia/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_nvidia/main.cc
index 22f66ca0a8..06f5da5c31 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_nvidia/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_nvidia/main.cc
@@ -190,7 +190,7 @@ int main(int argc, char *argv[]) {
&matC.front(), 0, NULL, NULL);
CHECK_ERROR("clEnqueueWriteBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
// Use standard sgemm interface
regtileSgemm('N', 'T', matArow, matBcol, matAcol, 1.0f, dA, matArow, dB,
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8/main.cc
index 10e0445453..b22ebd8804 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8/main.cc
@@ -119,7 +119,7 @@ int main(int argc, char *argv[]) {
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_platform_id clPlatform;
clStatus = clGetPlatformIDs(1, &clPlatform, NULL);
@@ -214,7 +214,7 @@ int main(int argc, char *argv[]) {
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
pb_PrintTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clReleaseKernel(clKernel);
clReleaseProgram(clProgram);
clReleaseMemObject(dA);
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8_4K/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8_4K/main.cc
index 59da9562a1..a7cb9793e8 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8_4K/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8_4K/main.cc
@@ -79,7 +79,7 @@ void regtileSgemm(char transa, char transb, int m, int n, int k, float alpha,
clStatus = clSetKernelArg(clKernel, 8, sizeof(float), (void *)&beta);
CHECK_ERROR("clSetKernelArg")
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (int i = 0; i < 4; i++) {
clStatus = clEnqueueNDRangeKernel(clCommandQueue, clKernel, 2, NULL, dg, db,
0, NULL, NULL);
@@ -123,7 +123,7 @@ int main(int argc, char *argv[]) {
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_platform_id clPlatform;
clStatus = clGetPlatformIDs(1, &clPlatform, NULL);
@@ -211,7 +211,7 @@ int main(int argc, char *argv[]) {
clEnqueueReadBuffer(clCommandQueue, dC, CL_TRUE, 0, C_sz, &matC.front(), 0,
NULL, NULL);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clReleaseKernel(clKernel);
clReleaseProgram(clProgram);
clReleaseMemObject(dA);
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8_medium/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8_medium/main.cc
index 5069484492..713fd9e889 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8_medium/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8_medium/main.cc
@@ -79,7 +79,7 @@ void regtileSgemm(char transa, char transb, int m, int n, int k, float alpha,
clStatus = clSetKernelArg(clKernel, 8, sizeof(float), (void *)&beta);
CHECK_ERROR("clSetKernelArg")
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (int i = 0; i < 200; i++) {
clStatus = clEnqueueNDRangeKernel(clCommandQueue, clKernel, 2, NULL, dg, db,
0, NULL, NULL);
@@ -123,7 +123,7 @@ int main(int argc, char *argv[]) {
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_platform_id clPlatform;
clStatus = clGetPlatformIDs(1, &clPlatform, NULL);
@@ -211,7 +211,7 @@ int main(int argc, char *argv[]) {
clEnqueueReadBuffer(clCommandQueue, dC, CL_TRUE, 0, C_sz, &matC.front(), 0,
NULL, NULL);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clReleaseKernel(clKernel);
clReleaseProgram(clProgram);
clReleaseMemObject(dA);
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8_vec/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8_vec/main.cc
index bad8253870..7d5d75c533 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8_vec/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/opencl_opt_8_vec/main.cc
@@ -120,7 +120,7 @@ int main(int argc, char *argv[]) {
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_platform_id clPlatform;
clStatus = clGetPlatformIDs(1, &clPlatform, NULL);
@@ -208,7 +208,7 @@ int main(int argc, char *argv[]) {
clEnqueueReadBuffer(clCommandQueue, dC, CL_TRUE, 0, C_sz, &matC.front(), 0,
NULL, NULL);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clReleaseKernel(clKernel);
clReleaseProgram(clProgram);
clReleaseMemObject(dA);
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm/Makefile b/hpvm/test/parboil/benchmarks/sgemm/src/visc/Makefile
similarity index 83%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm/Makefile
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc/Makefile
index 6e63f83841..d1f6c96d0c 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm/Makefile
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc/Makefile
@@ -1,8 +1,8 @@
# (c) 2010 The Board of Trustees of the University of Illinois.
-LANGUAGE=hpvm
+LANGUAGE=visc
SRCDIR_OBJS=io.ll #compute_gold.o
-HPVM_OBJS=main.hpvm.ll
+VISC_OBJS=main.visc.ll
APP_CUDALDFLAGS=-lm -lstdc++
APP_CFLAGS=-ffast-math -O1
APP_CXXFLAGS=-ffast-math -O1
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm/io.cc b/hpvm/test/parboil/benchmarks/sgemm/src/visc/io.cc
similarity index 100%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm/io.cc
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc/io.cc
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm/kernel.cl b/hpvm/test/parboil/benchmarks/sgemm/src/visc/kernel.cl
similarity index 100%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm/kernel.cl
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc/kernel.cl
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/visc/main.cc
similarity index 69%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm/main.cc
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc/main.cc
index de36705707..627f5a8241 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc/main.cc
@@ -10,7 +10,6 @@
* Main entry of dense matrix-matrix multiplication kernel
*/
-#include <hpvm.h>
#include <iostream>
#include <malloc.h>
#include <math.h>
@@ -20,6 +19,7 @@
#include <string.h>
#include <sys/time.h>
#include <vector>
+#include <visc.h>
// I/O routines
extern bool readColMajorMatrixFile(const char *fn, int &nr_row, int &nr_col,
@@ -57,17 +57,17 @@ typedef struct __attribute__((__packed__)) {
void mysgemmNT(float *A, size_t bytes_A, int lda, float *B, size_t bytes_B,
int ldb, float *C, size_t bytes_C, int ldc, int k, float alpha,
float beta) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(3, A, B, C, 1, C);
-
- void *thisNode = __hpvm__getNode();
- void *parentNode = __hpvm__getParentNode(thisNode);
- int lx = __hpvm__getNodeInstanceID_x(thisNode);
- int ly = __hpvm__getNodeInstanceID_y(thisNode);
- int gx = __hpvm__getNodeInstanceID_x(parentNode);
- int gy = __hpvm__getNodeInstanceID_y(parentNode);
- int gridx = __hpvm__getNumNodeInstances_x(thisNode);
- int gridy = __hpvm__getNumNodeInstances_y(thisNode);
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(3, A, B, C, 1, C);
+
+ void *thisNode = __visc__getNode();
+ void *parentNode = __visc__getParentNode(thisNode);
+ int lx = __visc__getNodeInstanceID_x(thisNode);
+ int ly = __visc__getNodeInstanceID_y(thisNode);
+ int gx = __visc__getNodeInstanceID_x(parentNode);
+ int gy = __visc__getNodeInstanceID_y(parentNode);
+ int gridx = __visc__getNumNodeInstances_x(thisNode);
+ int gridy = __visc__getNumNodeInstances_y(thisNode);
int m = gx * gridx + lx;
int n = gy * gridy + ly;
@@ -83,46 +83,46 @@ void mysgemmNT(float *A, size_t bytes_A, int lda, float *B, size_t bytes_B,
void basicSgemmLvl1(float *A, size_t bytes_A, int lda, float *B, size_t bytes_B,
int ldb, float *C, size_t bytes_C, int ldc, int k,
float alpha, float beta, size_t dim_X1, size_t dim_Y1) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(3, A, B, C, 1, C);
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(3, A, B, C, 1, C);
void *sgemm_node =
- __hpvm__createNodeND(2, mysgemmNT, (size_t)dim_X1, (size_t)dim_Y1);
- __hpvm__bindIn(sgemm_node, 0, 0, 0);
- __hpvm__bindIn(sgemm_node, 1, 1, 0);
- __hpvm__bindIn(sgemm_node, 2, 2, 0);
- __hpvm__bindIn(sgemm_node, 3, 3, 0);
- __hpvm__bindIn(sgemm_node, 4, 4, 0);
- __hpvm__bindIn(sgemm_node, 5, 5, 0);
- __hpvm__bindIn(sgemm_node, 6, 6, 0);
- __hpvm__bindIn(sgemm_node, 7, 7, 0);
- __hpvm__bindIn(sgemm_node, 8, 8, 0);
- __hpvm__bindIn(sgemm_node, 9, 9, 0);
- __hpvm__bindIn(sgemm_node, 10, 10, 0);
- __hpvm__bindIn(sgemm_node, 11, 11, 0);
+ __visc__createNodeND(2, mysgemmNT, (size_t)dim_X1, (size_t)dim_Y1);
+ __visc__bindIn(sgemm_node, 0, 0, 0);
+ __visc__bindIn(sgemm_node, 1, 1, 0);
+ __visc__bindIn(sgemm_node, 2, 2, 0);
+ __visc__bindIn(sgemm_node, 3, 3, 0);
+ __visc__bindIn(sgemm_node, 4, 4, 0);
+ __visc__bindIn(sgemm_node, 5, 5, 0);
+ __visc__bindIn(sgemm_node, 6, 6, 0);
+ __visc__bindIn(sgemm_node, 7, 7, 0);
+ __visc__bindIn(sgemm_node, 8, 8, 0);
+ __visc__bindIn(sgemm_node, 9, 9, 0);
+ __visc__bindIn(sgemm_node, 10, 10, 0);
+ __visc__bindIn(sgemm_node, 11, 11, 0);
}
void basicSgemmLvl2(float *A, size_t bytes_A, int lda, float *B, size_t bytes_B,
int ldb, float *C, size_t bytes_C, int ldc, int k,
float alpha, float beta, size_t dim_X1, size_t dim_Y1,
size_t dim_X2, size_t dim_Y2) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(3, A, B, C, 1, C);
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(3, A, B, C, 1, C);
void *sgemm_node =
- __hpvm__createNodeND(2, basicSgemmLvl1, (size_t)dim_X2, (size_t)dim_Y2);
- __hpvm__bindIn(sgemm_node, 0, 0, 0);
- __hpvm__bindIn(sgemm_node, 1, 1, 0);
- __hpvm__bindIn(sgemm_node, 2, 2, 0);
- __hpvm__bindIn(sgemm_node, 3, 3, 0);
- __hpvm__bindIn(sgemm_node, 4, 4, 0);
- __hpvm__bindIn(sgemm_node, 5, 5, 0);
- __hpvm__bindIn(sgemm_node, 6, 6, 0);
- __hpvm__bindIn(sgemm_node, 7, 7, 0);
- __hpvm__bindIn(sgemm_node, 8, 8, 0);
- __hpvm__bindIn(sgemm_node, 9, 9, 0);
- __hpvm__bindIn(sgemm_node, 10, 10, 0);
- __hpvm__bindIn(sgemm_node, 11, 11, 0);
- __hpvm__bindIn(sgemm_node, 12, 12, 0);
- __hpvm__bindIn(sgemm_node, 13, 13, 0);
+ __visc__createNodeND(2, basicSgemmLvl1, (size_t)dim_X2, (size_t)dim_Y2);
+ __visc__bindIn(sgemm_node, 0, 0, 0);
+ __visc__bindIn(sgemm_node, 1, 1, 0);
+ __visc__bindIn(sgemm_node, 2, 2, 0);
+ __visc__bindIn(sgemm_node, 3, 3, 0);
+ __visc__bindIn(sgemm_node, 4, 4, 0);
+ __visc__bindIn(sgemm_node, 5, 5, 0);
+ __visc__bindIn(sgemm_node, 6, 6, 0);
+ __visc__bindIn(sgemm_node, 7, 7, 0);
+ __visc__bindIn(sgemm_node, 8, 8, 0);
+ __visc__bindIn(sgemm_node, 9, 9, 0);
+ __visc__bindIn(sgemm_node, 10, 10, 0);
+ __visc__bindIn(sgemm_node, 11, 11, 0);
+ __visc__bindIn(sgemm_node, 12, 12, 0);
+ __visc__bindIn(sgemm_node, 13, 13, 0);
}
// A wrapper level used in codegen for some backends
@@ -130,25 +130,25 @@ void basicSgemmLvl3(float *A, size_t bytes_A, int lda, float *B, size_t bytes_B,
int ldb, float *C, size_t bytes_C, int ldc, int k,
float alpha, float beta, size_t dim_X1, size_t dim_Y1,
size_t dim_X2, size_t dim_Y2) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(3, A, B, C, 1, C);
- void *sgemm_node = __hpvm__createNodeND(0, basicSgemmLvl2);
- __hpvm__bindIn(sgemm_node, 0, 0, 0);
- __hpvm__bindIn(sgemm_node, 1, 1, 0);
- __hpvm__bindIn(sgemm_node, 2, 2, 0);
- __hpvm__bindIn(sgemm_node, 3, 3, 0);
- __hpvm__bindIn(sgemm_node, 4, 4, 0);
- __hpvm__bindIn(sgemm_node, 5, 5, 0);
- __hpvm__bindIn(sgemm_node, 6, 6, 0);
- __hpvm__bindIn(sgemm_node, 7, 7, 0);
- __hpvm__bindIn(sgemm_node, 8, 8, 0);
- __hpvm__bindIn(sgemm_node, 9, 9, 0);
- __hpvm__bindIn(sgemm_node, 10, 10, 0);
- __hpvm__bindIn(sgemm_node, 11, 11, 0);
- __hpvm__bindIn(sgemm_node, 12, 12, 0);
- __hpvm__bindIn(sgemm_node, 13, 13, 0);
- __hpvm__bindIn(sgemm_node, 14, 14, 0);
- __hpvm__bindIn(sgemm_node, 15, 15, 0);
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(3, A, B, C, 1, C);
+ void *sgemm_node = __visc__createNodeND(0, basicSgemmLvl2);
+ __visc__bindIn(sgemm_node, 0, 0, 0);
+ __visc__bindIn(sgemm_node, 1, 1, 0);
+ __visc__bindIn(sgemm_node, 2, 2, 0);
+ __visc__bindIn(sgemm_node, 3, 3, 0);
+ __visc__bindIn(sgemm_node, 4, 4, 0);
+ __visc__bindIn(sgemm_node, 5, 5, 0);
+ __visc__bindIn(sgemm_node, 6, 6, 0);
+ __visc__bindIn(sgemm_node, 7, 7, 0);
+ __visc__bindIn(sgemm_node, 8, 8, 0);
+ __visc__bindIn(sgemm_node, 9, 9, 0);
+ __visc__bindIn(sgemm_node, 10, 10, 0);
+ __visc__bindIn(sgemm_node, 11, 11, 0);
+ __visc__bindIn(sgemm_node, 12, 12, 0);
+ __visc__bindIn(sgemm_node, 13, 13, 0);
+ __visc__bindIn(sgemm_node, 14, 14, 0);
+ __visc__bindIn(sgemm_node, 15, 15, 0);
}
__attribute__((noinline)) void basicSgemm(char transa, char transb, int m,
@@ -194,8 +194,8 @@ __attribute__((noinline)) void basicSgemm(char transa, char transb, int m,
dg[0] / db[0],
dg[1] / db[1]};
*(RootIn *)root_in = root_in_local;
- void *sgemmDFG = __hpvm__launch(0, basicSgemmLvl3, root_in);
- __hpvm__wait(sgemmDFG);
+ void *sgemmDFG = __visc__launch(0, basicSgemmLvl3, root_in);
+ __visc__wait(sgemmDFG);
}
int main(int argc, char *argv[]) {
@@ -233,7 +233,7 @@ int main(int argc, char *argv[]) {
readColMajorMatrixFile(params->inpFiles[2], matBcol, matBrow, matBT);
pb_InitializeTimerSet(&timers);
- __hpvm__init();
+ __visc__init();
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
// copy A to device memory
@@ -246,9 +246,9 @@ int main(int argc, char *argv[]) {
// OpenCL memory allocation
std::vector<float> matC(matArow * matBcol);
- llvm_hpvm_track_mem(&matA.front(), A_sz);
- llvm_hpvm_track_mem(&matBT.front(), B_sz);
- llvm_hpvm_track_mem(&matC.front(), C_sz);
+ llvm_visc_track_mem(&matA.front(), A_sz);
+ llvm_visc_track_mem(&matBT.front(), B_sz);
+ llvm_visc_track_mem(&matC.front(), C_sz);
// Copy A and B^T into device memory
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
@@ -263,16 +263,16 @@ int main(int argc, char *argv[]) {
matArow);
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
- llvm_hpvm_request_mem(&matC.front(), C_sz);
+ llvm_visc_request_mem(&matC.front(), C_sz);
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_UNTRACK);
- llvm_hpvm_untrack_mem(&matA.front());
- llvm_hpvm_untrack_mem(&matBT.front());
- llvm_hpvm_untrack_mem(&matC.front());
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_UNTRACK);
+ llvm_visc_untrack_mem(&matA.front());
+ llvm_visc_untrack_mem(&matBT.front());
+ llvm_visc_untrack_mem(&matC.front());
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
pb_PrintTimerSet(&timers);
- __hpvm__cleanup();
+ __visc__cleanup();
if (params->outFile) {
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc/Makefile b/hpvm/test/parboil/benchmarks/sgemm/src/visc_opt/Makefile
similarity index 83%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc/Makefile
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_opt/Makefile
index 2234bf54e1..f74ee8921a 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc/Makefile
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_opt/Makefile
@@ -1,8 +1,8 @@
# (c) 2010 The Board of Trustees of the University of Illinois.
-LANGUAGE=hpvm
+LANGUAGE=visc
SRCDIR_OBJS=io.ll #compute_gold.o
-HPVM_OBJS=main.hpvm.ll
+VISC_OBJS=main.visc.ll
APP_CUDALDFLAGS=-lm -lstdc++
APP_CFLAGS=-ffast-math -O3
APP_CXXFLAGS=-ffast-math -O3
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_opt/io.cc b/hpvm/test/parboil/benchmarks/sgemm/src/visc_opt/io.cc
similarity index 100%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_opt/io.cc
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_opt/io.cc
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_opt/kernel.cl b/hpvm/test/parboil/benchmarks/sgemm/src/visc_opt/kernel.cl
similarity index 100%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_opt/kernel.cl
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_opt/kernel.cl
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_opt/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/visc_opt/main.cc
similarity index 90%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_opt/main.cc
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_opt/main.cc
index a1db2e56a5..62f9285e8a 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_opt/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_opt/main.cc
@@ -10,7 +10,6 @@
* Main entry of dense matrix-matrix multiplication kernel
*/
-#include <hpvm.h>
#include <iostream>
#include <malloc.h>
#include <math.h>
@@ -20,6 +19,7 @@
#include <string.h>
#include <sys/time.h>
#include <vector>
+#include <visc.h>
// I/O routines
extern bool readColMajorMatrixFile(const char *fn, int &nr_row, int &nr_col,
@@ -42,8 +42,8 @@ extern char *readFile(const char *);
void mysgemmNT(float *A, int lda, float *B, int ldb, float *C, int ldc, int k,
float alpha, float beta) {
- __hpvm__hint(hpvm::GPU_TARGET);
- __hpvm__attributes(3, A, B, C, 1, C);
+ __visc__hint(visc::GPU_TARGET);
+ __visc__attributes(3, A, B, C, 1, C);
float c[TILE_N];
for (int i = 0; i < TILE_N; i++)
@@ -96,10 +96,10 @@ __attribute__((noinline)) void basicSgemm(char transa, char transb, int m,
// unsigned dg[2] = {m*TILE_N/TILE_M,n*TILE_TB_HEIGHT/TILE_N};
unsigned dg[2] = {m * db[0] / TILE_M, n * db[1] / TILE_N};
- unsigned sgemmDFG = __hpvm__node(mysgemmNT, 2, 2, db[0], db[1], dg[0] / db[0],
+ unsigned sgemmDFG = __visc__node(mysgemmNT, 2, 2, db[0], db[1], dg[0] / db[0],
dg[1] / db[1], 12, A, bytesA, lda, B, bytesB,
ldb, C, bytesC, ldc, k, alpha, beta, 0);
- __hpvm__wait(sgemmDFG);
+ __visc__wait(sgemmDFG);
}
int main(int argc, char *argv[]) {
@@ -129,7 +129,7 @@ int main(int argc, char *argv[]) {
readColMajorMatrixFile(params->inpFiles[2], matBcol, matBrow, matBT);
pb_InitializeTimerSet(&timers);
- __hpvm__init();
+ __visc__init();
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
// copy A to device memory
@@ -142,9 +142,9 @@ int main(int argc, char *argv[]) {
// OpenCL memory allocation
std::vector<float> matC(matArow * matBcol);
- llvm_hpvm_track_mem(&matA.front(), A_sz);
- llvm_hpvm_track_mem(&matBT.front(), B_sz);
- llvm_hpvm_track_mem(&matC.front(), C_sz);
+ llvm_visc_track_mem(&matA.front(), A_sz);
+ llvm_visc_track_mem(&matBT.front(), B_sz);
+ llvm_visc_track_mem(&matC.front(), C_sz);
// Copy A and B^T into device memory
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
@@ -159,16 +159,16 @@ int main(int argc, char *argv[]) {
matArow);
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
- llvm_hpvm_request_mem(&matC.front(), C_sz);
+ llvm_visc_request_mem(&matC.front(), C_sz);
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_UNTRACK);
- llvm_hpvm_untrack_mem(&matA.front());
- llvm_hpvm_untrack_mem(&matBT.front());
- llvm_hpvm_untrack_mem(&matC.front());
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_UNTRACK);
+ llvm_visc_untrack_mem(&matA.front());
+ llvm_visc_untrack_mem(&matBT.front());
+ llvm_visc_untrack_mem(&matC.front());
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
pb_PrintTimerSet(&timers);
- __hpvm__cleanup();
+ __visc__cleanup();
if (params->outFile) {
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_sh/Makefile b/hpvm/test/parboil/benchmarks/sgemm/src/visc_sh/Makefile
similarity index 86%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_sh/Makefile
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_sh/Makefile
index f81bac4707..a0fd0e9575 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_sh/Makefile
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_sh/Makefile
@@ -1,8 +1,8 @@
# (c) 2010 The Board of Trustees of the University of Illinois.
-LANGUAGE=hpvm
+LANGUAGE=visc
SRCDIR_OBJS=io.ll #compute_gold.o
-HPVM_OBJS=main.hpvm.ll
+VISC_OBJS=main.visc.ll
APP_CUDALDFLAGS=-lm -lstdc++
APP_CFLAGS=-ffast-math -O3
APP_CXXFLAGS=-ffast-math -O3
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_sh/io.cc b/hpvm/test/parboil/benchmarks/sgemm/src/visc_sh/io.cc
similarity index 100%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_sh/io.cc
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_sh/io.cc
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_sh/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/visc_sh/main.cc
similarity index 65%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_sh/main.cc
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_sh/main.cc
index de0d473ed6..05d143b588 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_sh/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_sh/main.cc
@@ -10,7 +10,6 @@
* Main entry of dense matrix-matrix multiplication kernel
*/
-#include <hpvm.h>
#include <iostream>
#include <malloc.h>
#include <math.h>
@@ -20,6 +19,7 @@
#include <string.h>
#include <sys/time.h>
#include <vector>
+#include <visc.h>
// I/O routines
extern bool readColMajorMatrixFile(const char *fn, int &nr_row, int &nr_col,
@@ -82,29 +82,29 @@ void packData(RootIn *args, float *A, size_t bytesA, int lda, float *B,
}
void Allocation(long block_x, long block_y) {
- void *shB = __hpvm__malloc(block_x * block_y * sizeof(float));
- __hpvm__return(2, shB, block_x * block_y * sizeof(float));
+ void *shB = __visc__malloc(block_x * block_y * sizeof(float));
+ __visc__return(2, shB, block_x * block_y * sizeof(float));
}
void SgemmLeaf(float *A, size_t bytesA, int lda, float *B, size_t bytesB,
int ldb, float *C, size_t bytesC, int ldc, int k, float alpha,
float beta, float *shB, size_t bytesshB) {
- __hpvm__hint(hpvm::DEVICE);
- //__hpvm__hint(hpvm::SPIR_TARGET);
- //__hpvm__hint(hpvm::GPU_TARGET);
+ __visc__hint(visc::DEVICE);
+ //__visc__hint(visc::SPIR_TARGET);
+ //__visc__hint(visc::GPU_TARGET);
- __hpvm__attributes(3, A, B, C, 1, C);
+ __visc__attributes(3, A, B, C, 1, C);
- void *thisNode = __hpvm__getNode();
- void *parentNode = __hpvm__getParentNode(thisNode);
+ void *thisNode = __visc__getNode();
+ void *parentNode = __visc__getParentNode(thisNode);
- long lx = __hpvm__getNodeInstanceID_x(thisNode);
- long ly = __hpvm__getNodeInstanceID_y(thisNode);
+ long lx = __visc__getNodeInstanceID_x(thisNode);
+ long ly = __visc__getNodeInstanceID_y(thisNode);
- long gx = __hpvm__getNodeInstanceID_x(parentNode);
- long gy = __hpvm__getNodeInstanceID_y(parentNode);
+ long gx = __visc__getNodeInstanceID_x(parentNode);
+ long gy = __visc__getNodeInstanceID_y(parentNode);
- long dimx = __hpvm__getNumNodeInstances_x(thisNode);
+ long dimx = __visc__getNumNodeInstances_x(thisNode);
float c[TILE_N];
for (int i = 0; i < TILE_N; i++)
@@ -119,7 +119,7 @@ void SgemmLeaf(float *A, size_t bytesA, int lda, float *B, size_t bytesB,
// shB[ly][lx] = B[n+(i+ly)*ldb];
shB[ly * dimx + lx] = B[n + (i + ly) * ldb];
- __hpvm__barrier();
+ __visc__barrier();
for (int j = 0; j < TILE_TB_HEIGHT; j++) {
a = A[m + (i + j) * lda];
for (int kk = 0; kk < TILE_N; kk++) {
@@ -127,7 +127,7 @@ void SgemmLeaf(float *A, size_t bytesA, int lda, float *B, size_t bytesB,
c[kk] += a * shB[j * dimx + kk];
}
}
- __hpvm__barrier();
+ __visc__barrier();
}
int t = ldc * gy * TILE_N + m;
@@ -140,31 +140,31 @@ void SgemmLeaf(float *A, size_t bytesA, int lda, float *B, size_t bytesB,
void SgemmTB(float *A, size_t bytesA, int lda, float *B, size_t bytesB, int ldb,
float *C, size_t bytesC, int ldc, int k, float alpha, float beta,
long block_x, long block_y) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(3, A, B, C, 1, C);
- void *AllocationNode = __hpvm__createNodeND(0, Allocation);
- void *SgemmLeafNode = __hpvm__createNodeND(2, SgemmLeaf, block_x, block_y);
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(3, A, B, C, 1, C);
+ void *AllocationNode = __visc__createNodeND(0, Allocation);
+ void *SgemmLeafNode = __visc__createNodeND(2, SgemmLeaf, block_x, block_y);
// Bind edges
- __hpvm__bindIn(SgemmLeafNode, 0, 0, 0); // Bind A
- __hpvm__bindIn(SgemmLeafNode, 1, 1, 0); // Bind bytesA
- __hpvm__bindIn(SgemmLeafNode, 2, 2, 0); // Bind lda
- __hpvm__bindIn(SgemmLeafNode, 3, 3, 0); // Bind B
- __hpvm__bindIn(SgemmLeafNode, 4, 4, 0); // Bind bytesB
- __hpvm__bindIn(SgemmLeafNode, 5, 5, 0); // Bind ldb
- __hpvm__bindIn(SgemmLeafNode, 6, 6, 0); // Bind C
- __hpvm__bindIn(SgemmLeafNode, 7, 7, 0); // Bind bytesC
- __hpvm__bindIn(SgemmLeafNode, 8, 8, 0); // Bind ldc
- __hpvm__bindIn(SgemmLeafNode, 9, 9, 0); // Bind k
- __hpvm__bindIn(SgemmLeafNode, 10, 10, 0); // Bind alpha
- __hpvm__bindIn(SgemmLeafNode, 11, 11, 0); // Bind beta
-
- __hpvm__bindIn(AllocationNode, 12, 0, 0); // Bind block_x
- __hpvm__bindIn(AllocationNode, 13, 1, 0); // Bind block_y
+ __visc__bindIn(SgemmLeafNode, 0, 0, 0); // Bind A
+ __visc__bindIn(SgemmLeafNode, 1, 1, 0); // Bind bytesA
+ __visc__bindIn(SgemmLeafNode, 2, 2, 0); // Bind lda
+ __visc__bindIn(SgemmLeafNode, 3, 3, 0); // Bind B
+ __visc__bindIn(SgemmLeafNode, 4, 4, 0); // Bind bytesB
+ __visc__bindIn(SgemmLeafNode, 5, 5, 0); // Bind ldb
+ __visc__bindIn(SgemmLeafNode, 6, 6, 0); // Bind C
+ __visc__bindIn(SgemmLeafNode, 7, 7, 0); // Bind bytesC
+ __visc__bindIn(SgemmLeafNode, 8, 8, 0); // Bind ldc
+ __visc__bindIn(SgemmLeafNode, 9, 9, 0); // Bind k
+ __visc__bindIn(SgemmLeafNode, 10, 10, 0); // Bind alpha
+ __visc__bindIn(SgemmLeafNode, 11, 11, 0); // Bind beta
+
+ __visc__bindIn(AllocationNode, 12, 0, 0); // Bind block_x
+ __visc__bindIn(AllocationNode, 13, 1, 0); // Bind block_y
// Create Edges between AllocationNode and BFSLeafNodeNode
- __hpvm__edge(AllocationNode, SgemmLeafNode, 1, 0, 12, 0); // Edge local_B
- __hpvm__edge(AllocationNode, SgemmLeafNode, 1, 1, 13,
+ __visc__edge(AllocationNode, SgemmLeafNode, 1, 0, 12, 0); // Edge local_B
+ __visc__edge(AllocationNode, SgemmLeafNode, 1, 1, 13,
0); // Edge bytes_local_B
}
@@ -175,25 +175,25 @@ void SgemmRoot(float *A, size_t bytesA, int lda, // 0-2
int k, float alpha, float beta, // 9-11
long block_x, long block_y, long grid_x, long grid_y // 12-15
) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(3, A, B, C, 1, C);
- void *SgemmTBNode = __hpvm__createNodeND(2, SgemmTB, grid_x, grid_y);
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(3, A, B, C, 1, C);
+ void *SgemmTBNode = __visc__createNodeND(2, SgemmTB, grid_x, grid_y);
// Bind edges
- __hpvm__bindIn(SgemmTBNode, 0, 0, 0); // Bind A
- __hpvm__bindIn(SgemmTBNode, 1, 1, 0); // Bind bytesA
- __hpvm__bindIn(SgemmTBNode, 2, 2, 0); // Bind lda
- __hpvm__bindIn(SgemmTBNode, 3, 3, 0); // Bind B
- __hpvm__bindIn(SgemmTBNode, 4, 4, 0); // Bind bytesB
- __hpvm__bindIn(SgemmTBNode, 5, 5, 0); // Bind ldb
- __hpvm__bindIn(SgemmTBNode, 6, 6, 0); // Bind C
- __hpvm__bindIn(SgemmTBNode, 7, 7, 0); // Bind bytesC
- __hpvm__bindIn(SgemmTBNode, 8, 8, 0); // Bind ldc
- __hpvm__bindIn(SgemmTBNode, 9, 9, 0); // Bind k
- __hpvm__bindIn(SgemmTBNode, 10, 10, 0); // Bind alpha
- __hpvm__bindIn(SgemmTBNode, 11, 11, 0); // Bind beta
- __hpvm__bindIn(SgemmTBNode, 12, 12, 0); // Bind block_x
- __hpvm__bindIn(SgemmTBNode, 13, 13, 0); // Bind block_y
+ __visc__bindIn(SgemmTBNode, 0, 0, 0); // Bind A
+ __visc__bindIn(SgemmTBNode, 1, 1, 0); // Bind bytesA
+ __visc__bindIn(SgemmTBNode, 2, 2, 0); // Bind lda
+ __visc__bindIn(SgemmTBNode, 3, 3, 0); // Bind B
+ __visc__bindIn(SgemmTBNode, 4, 4, 0); // Bind bytesB
+ __visc__bindIn(SgemmTBNode, 5, 5, 0); // Bind ldb
+ __visc__bindIn(SgemmTBNode, 6, 6, 0); // Bind C
+ __visc__bindIn(SgemmTBNode, 7, 7, 0); // Bind bytesC
+ __visc__bindIn(SgemmTBNode, 8, 8, 0); // Bind ldc
+ __visc__bindIn(SgemmTBNode, 9, 9, 0); // Bind k
+ __visc__bindIn(SgemmTBNode, 10, 10, 0); // Bind alpha
+ __visc__bindIn(SgemmTBNode, 11, 11, 0); // Bind beta
+ __visc__bindIn(SgemmTBNode, 12, 12, 0); // Bind block_x
+ __visc__bindIn(SgemmTBNode, 13, 13, 0); // Bind block_y
}
void SgemmWrapper(float *A, size_t bytesA, int lda, // 0-2
@@ -202,27 +202,27 @@ void SgemmWrapper(float *A, size_t bytesA, int lda, // 0-2
int k, float alpha, float beta, // 9-11
long block_x, long block_y, long grid_x, long grid_y // 12-15
) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(3, A, B, C, 1, C);
- void *SgemmRootNode = __hpvm__createNodeND(0, SgemmRoot);
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(3, A, B, C, 1, C);
+ void *SgemmRootNode = __visc__createNodeND(0, SgemmRoot);
// Bind edges
- __hpvm__bindIn(SgemmRootNode, 0, 0, 0); // Bind A
- __hpvm__bindIn(SgemmRootNode, 1, 1, 0); // Bind bytesA
- __hpvm__bindIn(SgemmRootNode, 2, 2, 0); // Bind lda
- __hpvm__bindIn(SgemmRootNode, 3, 3, 0); // Bind B
- __hpvm__bindIn(SgemmRootNode, 4, 4, 0); // Bind bytesB
- __hpvm__bindIn(SgemmRootNode, 5, 5, 0); // Bind ldb
- __hpvm__bindIn(SgemmRootNode, 6, 6, 0); // Bind C
- __hpvm__bindIn(SgemmRootNode, 7, 7, 0); // Bind bytesC
- __hpvm__bindIn(SgemmRootNode, 8, 8, 0); // Bind ldc
- __hpvm__bindIn(SgemmRootNode, 9, 9, 0); // Bind k
- __hpvm__bindIn(SgemmRootNode, 10, 10, 0); // Bind alpha
- __hpvm__bindIn(SgemmRootNode, 11, 11, 0); // Bind beta
- __hpvm__bindIn(SgemmRootNode, 12, 12, 0); // Bind block_x
- __hpvm__bindIn(SgemmRootNode, 13, 13, 0); // Bind block_y
- __hpvm__bindIn(SgemmRootNode, 14, 14, 0); // Bind grid_x
- __hpvm__bindIn(SgemmRootNode, 15, 15, 0); // Bind grid_y
+ __visc__bindIn(SgemmRootNode, 0, 0, 0); // Bind A
+ __visc__bindIn(SgemmRootNode, 1, 1, 0); // Bind bytesA
+ __visc__bindIn(SgemmRootNode, 2, 2, 0); // Bind lda
+ __visc__bindIn(SgemmRootNode, 3, 3, 0); // Bind B
+ __visc__bindIn(SgemmRootNode, 4, 4, 0); // Bind bytesB
+ __visc__bindIn(SgemmRootNode, 5, 5, 0); // Bind ldb
+ __visc__bindIn(SgemmRootNode, 6, 6, 0); // Bind C
+ __visc__bindIn(SgemmRootNode, 7, 7, 0); // Bind bytesC
+ __visc__bindIn(SgemmRootNode, 8, 8, 0); // Bind ldc
+ __visc__bindIn(SgemmRootNode, 9, 9, 0); // Bind k
+ __visc__bindIn(SgemmRootNode, 10, 10, 0); // Bind alpha
+ __visc__bindIn(SgemmRootNode, 11, 11, 0); // Bind beta
+ __visc__bindIn(SgemmRootNode, 12, 12, 0); // Bind block_x
+ __visc__bindIn(SgemmRootNode, 13, 13, 0); // Bind block_y
+ __visc__bindIn(SgemmRootNode, 14, 14, 0); // Bind grid_x
+ __visc__bindIn(SgemmRootNode, 15, 15, 0); // Bind grid_y
}
// Creates root node for sgemm
@@ -262,10 +262,10 @@ __attribute__((noinline)) void basicSgemm(struct pb_TimerSet *timers,
packData(args, A, bytesA, lda, B, bytesB, ldb, C, bytesC, ldc, k, alpha, beta,
block_x, block_y, grid_x, grid_y);
- pb_SwitchToTimer(timers, hpvm_TimerID_COMPUTATION);
- void *sgemmDFG = __hpvm__launch(0, SgemmWrapper, (void *)args);
+ pb_SwitchToTimer(timers, visc_TimerID_COMPUTATION);
+ void *sgemmDFG = __visc__launch(0, SgemmWrapper, (void *)args);
- __hpvm__wait(sgemmDFG);
+ __visc__wait(sgemmDFG);
pb_SwitchToTimer(timers, pb_TimerID_COMPUTE);
}
@@ -296,7 +296,7 @@ int main(int argc, char *argv[]) {
readColMajorMatrixFile(params->inpFiles[2], matBcol, matBrow, matBT);
pb_InitializeTimerSet(&timers);
- __hpvm__init();
+ __visc__init();
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
// copy A to device memory
@@ -309,9 +309,9 @@ int main(int argc, char *argv[]) {
// OpenCL memory allocation
std::vector<float> matC(matArow * matBcol);
- llvm_hpvm_track_mem(&matA.front(), A_sz);
- llvm_hpvm_track_mem(&matBT.front(), B_sz);
- llvm_hpvm_track_mem(&matC.front(), C_sz);
+ llvm_visc_track_mem(&matA.front(), A_sz);
+ llvm_visc_track_mem(&matBT.front(), B_sz);
+ llvm_visc_track_mem(&matC.front(), C_sz);
// Copy A and B^T into device memory
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
@@ -324,16 +324,16 @@ int main(int argc, char *argv[]) {
C_sz, matArow);
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
- llvm_hpvm_request_mem(&matC.front(), C_sz);
+ llvm_visc_request_mem(&matC.front(), C_sz);
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
- llvm_hpvm_untrack_mem(&matA.front());
- llvm_hpvm_untrack_mem(&matBT.front());
- llvm_hpvm_untrack_mem(&matC.front());
+ llvm_visc_untrack_mem(&matA.front());
+ llvm_visc_untrack_mem(&matBT.front());
+ llvm_visc_untrack_mem(&matC.front());
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
pb_PrintTimerSet(&timers);
- __hpvm__cleanup();
+ __visc__cleanup();
if (params->outFile) {
/* Write C to file */
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec/Makefile b/hpvm/test/parboil/benchmarks/sgemm/src/visc_tc/Makefile
similarity index 83%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec/Makefile
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_tc/Makefile
index 2234bf54e1..f74ee8921a 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec/Makefile
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_tc/Makefile
@@ -1,8 +1,8 @@
# (c) 2010 The Board of Trustees of the University of Illinois.
-LANGUAGE=hpvm
+LANGUAGE=visc
SRCDIR_OBJS=io.ll #compute_gold.o
-HPVM_OBJS=main.hpvm.ll
+VISC_OBJS=main.visc.ll
APP_CUDALDFLAGS=-lm -lstdc++
APP_CFLAGS=-ffast-math -O3
APP_CXXFLAGS=-ffast-math -O3
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc/io.cc b/hpvm/test/parboil/benchmarks/sgemm/src/visc_tc/io.cc
similarity index 100%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc/io.cc
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_tc/io.cc
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/visc_tc/main.cc
similarity index 90%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc/main.cc
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_tc/main.cc
index be39d713d5..0dfcdfb835 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_tc/main.cc
@@ -10,7 +10,6 @@
* Main entry of dense matrix-matrix multiplication kernel
*/
-#include <hpvm.h>
#include <iostream>
#include <malloc.h>
#include <math.h>
@@ -20,6 +19,7 @@
#include <string.h>
#include <sys/time.h>
#include <vector>
+#include <visc.h>
// I/O routines
extern bool readColMajorMatrixFile(const char *fn, int &nr_row, int &nr_col,
@@ -40,7 +40,7 @@ extern char *readFile(const char *);
void mysgemmNT(float *A, int lda, float *B, int ldb, float *C, int ldc, int k,
float alpha, float beta) {
- __hpvm__attributes(3, A, B, C, 1, C);
+ __visc__attributes(3, A, B, C, 1, C);
float c0, c1, c2, c3;
c0 = c1 = c2 = c3 = 0.0f;
int m = 4 * get_global_id(0);
@@ -90,10 +90,10 @@ __attribute__((noinline)) void basicSgemm(char transa, char transb, int m,
unsigned db[2] = {TILE_SZ / 4, TILE_SZ};
unsigned dg[2] = {m / TILE_SZ * db[0], n / TILE_SZ * db[1]};
- unsigned sgemmDFG = __hpvm__node(mysgemmNT, 2, 2, db[0], db[1], dg[0] / db[0],
+ unsigned sgemmDFG = __visc__node(mysgemmNT, 2, 2, db[0], db[1], dg[0] / db[0],
dg[1] / db[1], 12, A, bytesA, lda, B, bytesB,
ldb, C, bytesC, ldc, k, alpha, beta, 0);
- __hpvm__wait(sgemmDFG);
+ __visc__wait(sgemmDFG);
}
int main(int argc, char *argv[]) {
@@ -107,7 +107,7 @@ int main(int argc, char *argv[]) {
std::vector<float> matA, matBT;
pb_InitializeTimerSet(&timers);
- __hpvm__init();
+ __visc__init();
/* Read command line. Expect 3 inputs: A, B and B^T
in column-major layout*/
@@ -138,9 +138,9 @@ int main(int argc, char *argv[]) {
// OpenCL memory allocation
std::vector<float> matC(matArow * matBcol);
- llvm_hpvm_track_mem(&matA.front(), A_sz);
- llvm_hpvm_track_mem(&matBT.front(), B_sz);
- llvm_hpvm_track_mem(&matC.front(), C_sz);
+ llvm_visc_track_mem(&matA.front(), A_sz);
+ llvm_visc_track_mem(&matBT.front(), B_sz);
+ llvm_visc_track_mem(&matC.front(), C_sz);
// Copy A and B^T into device memory
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
@@ -158,22 +158,22 @@ int main(int argc, char *argv[]) {
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
/* Write C to file */
- llvm_hpvm_request_mem(&matC.front(), C_sz);
+ llvm_visc_request_mem(&matC.front(), C_sz);
pb_SwitchToTimer(&timers, pb_TimerID_IO);
writeColMajorMatrixFile(params->outFile, matArow, matBcol, matC);
}
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_UNTRACK);
- llvm_hpvm_untrack_mem(&matA.front());
- llvm_hpvm_untrack_mem(&matBT.front());
- llvm_hpvm_untrack_mem(&matC.front());
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_UNTRACK);
+ llvm_visc_untrack_mem(&matA.front());
+ llvm_visc_untrack_mem(&matBT.front());
+ llvm_visc_untrack_mem(&matC.front());
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
double GPUtime = pb_GetElapsedTime(&(timers.timers[pb_TimerID_KERNEL]));
std::cout << "GFLOPs = " << 2. * matArow * matBcol * matAcol / GPUtime / 1e9
<< std::endl;
pb_PrintTimerSet(&timers);
- __hpvm__cleanup();
+ __visc__cleanup();
pb_FreeParameters(params);
return 0;
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc_vec/Makefile b/hpvm/test/parboil/benchmarks/sgemm/src/visc_tc_vec/Makefile
similarity index 83%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc_vec/Makefile
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_tc_vec/Makefile
index 2234bf54e1..f74ee8921a 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc_vec/Makefile
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_tc_vec/Makefile
@@ -1,8 +1,8 @@
# (c) 2010 The Board of Trustees of the University of Illinois.
-LANGUAGE=hpvm
+LANGUAGE=visc
SRCDIR_OBJS=io.ll #compute_gold.o
-HPVM_OBJS=main.hpvm.ll
+VISC_OBJS=main.visc.ll
APP_CUDALDFLAGS=-lm -lstdc++
APP_CFLAGS=-ffast-math -O3
APP_CXXFLAGS=-ffast-math -O3
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc_vec/io.cc b/hpvm/test/parboil/benchmarks/sgemm/src/visc_tc_vec/io.cc
similarity index 100%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc_vec/io.cc
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_tc_vec/io.cc
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc_vec/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/visc_tc_vec/main.cc
similarity index 90%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc_vec/main.cc
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_tc_vec/main.cc
index be39d713d5..0dfcdfb835 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_tc_vec/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_tc_vec/main.cc
@@ -10,7 +10,6 @@
* Main entry of dense matrix-matrix multiplication kernel
*/
-#include <hpvm.h>
#include <iostream>
#include <malloc.h>
#include <math.h>
@@ -20,6 +19,7 @@
#include <string.h>
#include <sys/time.h>
#include <vector>
+#include <visc.h>
// I/O routines
extern bool readColMajorMatrixFile(const char *fn, int &nr_row, int &nr_col,
@@ -40,7 +40,7 @@ extern char *readFile(const char *);
void mysgemmNT(float *A, int lda, float *B, int ldb, float *C, int ldc, int k,
float alpha, float beta) {
- __hpvm__attributes(3, A, B, C, 1, C);
+ __visc__attributes(3, A, B, C, 1, C);
float c0, c1, c2, c3;
c0 = c1 = c2 = c3 = 0.0f;
int m = 4 * get_global_id(0);
@@ -90,10 +90,10 @@ __attribute__((noinline)) void basicSgemm(char transa, char transb, int m,
unsigned db[2] = {TILE_SZ / 4, TILE_SZ};
unsigned dg[2] = {m / TILE_SZ * db[0], n / TILE_SZ * db[1]};
- unsigned sgemmDFG = __hpvm__node(mysgemmNT, 2, 2, db[0], db[1], dg[0] / db[0],
+ unsigned sgemmDFG = __visc__node(mysgemmNT, 2, 2, db[0], db[1], dg[0] / db[0],
dg[1] / db[1], 12, A, bytesA, lda, B, bytesB,
ldb, C, bytesC, ldc, k, alpha, beta, 0);
- __hpvm__wait(sgemmDFG);
+ __visc__wait(sgemmDFG);
}
int main(int argc, char *argv[]) {
@@ -107,7 +107,7 @@ int main(int argc, char *argv[]) {
std::vector<float> matA, matBT;
pb_InitializeTimerSet(&timers);
- __hpvm__init();
+ __visc__init();
/* Read command line. Expect 3 inputs: A, B and B^T
in column-major layout*/
@@ -138,9 +138,9 @@ int main(int argc, char *argv[]) {
// OpenCL memory allocation
std::vector<float> matC(matArow * matBcol);
- llvm_hpvm_track_mem(&matA.front(), A_sz);
- llvm_hpvm_track_mem(&matBT.front(), B_sz);
- llvm_hpvm_track_mem(&matC.front(), C_sz);
+ llvm_visc_track_mem(&matA.front(), A_sz);
+ llvm_visc_track_mem(&matBT.front(), B_sz);
+ llvm_visc_track_mem(&matC.front(), C_sz);
// Copy A and B^T into device memory
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
@@ -158,22 +158,22 @@ int main(int argc, char *argv[]) {
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
/* Write C to file */
- llvm_hpvm_request_mem(&matC.front(), C_sz);
+ llvm_visc_request_mem(&matC.front(), C_sz);
pb_SwitchToTimer(&timers, pb_TimerID_IO);
writeColMajorMatrixFile(params->outFile, matArow, matBcol, matC);
}
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_UNTRACK);
- llvm_hpvm_untrack_mem(&matA.front());
- llvm_hpvm_untrack_mem(&matBT.front());
- llvm_hpvm_untrack_mem(&matC.front());
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_UNTRACK);
+ llvm_visc_untrack_mem(&matA.front());
+ llvm_visc_untrack_mem(&matBT.front());
+ llvm_visc_untrack_mem(&matC.front());
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
double GPUtime = pb_GetElapsedTime(&(timers.timers[pb_TimerID_KERNEL]));
std::cout << "GFLOPs = " << 2. * matArow * matBcol * matAcol / GPUtime / 1e9
<< std::endl;
pb_PrintTimerSet(&timers);
- __hpvm__cleanup();
+ __visc__cleanup();
pb_FreeParameters(params);
return 0;
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/visc_tc_vec/main.visc.ll b/hpvm/test/parboil/benchmarks/sgemm/src/visc_tc_vec/main.visc.ll
new file mode 100644
index 0000000000..ea1e7b3b7c
--- /dev/null
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_tc_vec/main.visc.ll
@@ -0,0 +1,894 @@
+; ModuleID = 'build/visc_tc_vec_default/main.ll'
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%"class.std::ios_base::Init" = type { i8 }
+%"class.std::basic_ostream" = type { i32 (...)**, %"class.std::basic_ios" }
+%"class.std::basic_ios" = type { %"class.std::ios_base", %"class.std::basic_ostream"*, i8, i8, %"class.std::basic_streambuf"*, %"class.std::ctype"*, %"class.std::num_put"*, %"class.std::num_get"* }
+%"class.std::ios_base" = type { i32 (...)**, i64, i64, i32, i32, i32, %"struct.std::ios_base::_Callback_list"*, %"struct.std::ios_base::_Words", [8 x %"struct.std::ios_base::_Words"], i32, %"struct.std::ios_base::_Words"*, %"class.std::locale" }
+%"struct.std::ios_base::_Callback_list" = type { %"struct.std::ios_base::_Callback_list"*, void (i32, %"class.std::ios_base"*, i32)*, i32, i32 }
+%"struct.std::ios_base::_Words" = type { i8*, i64 }
+%"class.std::locale" = type { %"class.std::locale::_Impl"* }
+%"class.std::locale::_Impl" = type { i32, %"class.std::locale::facet"**, i64, %"class.std::locale::facet"**, i8** }
+%"class.std::locale::facet" = type { i32 (...)**, i32 }
+%"class.std::basic_streambuf" = type { i32 (...)**, i8*, i8*, i8*, i8*, i8*, i8*, %"class.std::locale" }
+%"class.std::ctype" = type { %"class.std::locale::facet", %struct.__locale_struct*, i8, i32*, i32*, i16*, i8, [256 x i8], [256 x i8], i8 }
+%struct.__locale_struct = type { [13 x %struct.__locale_data*], i16*, i32*, i32*, [13 x i8*] }
+%struct.__locale_data = type opaque
+%"class.std::num_put" = type { %"class.std::locale::facet" }
+%"class.std::num_get" = type { %"class.std::locale::facet" }
+%struct._IO_FILE = type { i32, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, %struct._IO_marker*, %struct._IO_FILE*, i32, i32, i64, i16, i8, [1 x i8], i8*, i64, i8*, i8*, i8*, i8*, i64, i32, [20 x i8] }
+%struct._IO_marker = type { %struct._IO_marker*, %struct._IO_FILE*, i32 }
+%rtype = type {}
+%struct.arg = type <{ float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float, i32, i32, i32, i32, %rtype }>
+%struct.pb_TimerSet = type { i32, %struct.pb_async_time_marker_list*, i64, i64, [24 x %struct.pb_Timer], [24 x %struct.pb_SubTimerList*] }
+%struct.pb_async_time_marker_list = type { i8*, i32, i8*, %struct.pb_async_time_marker_list* }
+%struct.pb_Timer = type { i32, i64, i64 }
+%struct.pb_SubTimerList = type { %struct.pb_SubTimer*, %struct.pb_SubTimer* }
+%struct.pb_SubTimer = type { i8*, %struct.pb_Timer, %struct.pb_SubTimer* }
+%"class.std::vector" = type { %"struct.std::_Vector_base" }
+%"struct.std::_Vector_base" = type { %"struct.std::_Vector_base<float, std::allocator<float> >::_Vector_impl" }
+%"struct.std::_Vector_base<float, std::allocator<float> >::_Vector_impl" = type { float*, float*, float* }
+%struct.pb_Parameters = type { i8*, i8** }
+
+@_ZStL8__ioinit = internal global %"class.std::ios_base::Init" zeroinitializer, align 1
+@__dso_handle = external global i8
+@_ZSt4cerr = external global %"class.std::basic_ostream"
+@.str = private unnamed_addr constant [48 x i8] c"unsupported value of 'transa' in regtileSgemm()\00", align 1
+@.str1 = private unnamed_addr constant [48 x i8] c"unsupported value of 'transb' in regtileSgemm()\00", align 1
+@.str2 = private unnamed_addr constant [53 x i8] c"unsupported size of matrix. m should be multiple of \00", align 1
+@.str3 = private unnamed_addr constant [27 x i8] c"; n should be multiple of \00", align 1
+@stderr = external global %struct._IO_FILE*
+@.str4 = private unnamed_addr constant [33 x i8] c"Expecting three input filenames\0A\00", align 1
+@_ZSt4cout = external global %"class.std::basic_ostream"
+@.str5 = private unnamed_addr constant [10 x i8] c"GFLOPs = \00", align 1
+@llvm.global_ctors = appending global [1 x { i32, void ()* }] [{ i32, void ()* } { i32 65535, void ()* @_GLOBAL__I_a }]
+@viscTimerSet_GenVISC = common global i8* null
+@0 = internal constant [14 x i8] c"GenVISC_Timer\00"
+
+declare void @_ZNSt8ios_base4InitC1Ev(%"class.std::ios_base::Init"*) #0
+
+declare void @_ZNSt8ios_base4InitD1Ev(%"class.std::ios_base::Init"*) #0
+
+; Function Attrs: nounwind
+declare i32 @__cxa_atexit(void (i8*)*, i8*, i8*) #1
+
+; Function Attrs: nounwind uwtable
+define %rtype @_Z9mysgemmNTPfiS_iS_iiff(float* in %A, i64 %bytes_A, i32 %lda, float* in %B, i64 %bytes_B, i32 %ldb, float* in out %C, i64 %bytes_C, i32 %ldc, i32 %k, float %alpha, float %beta) #2 {
+entry:
+ %_Z9mysgemmNTPfiS_iS_iiff.node = call i8* @llvm.visc.getNode()
+ %_Z9mysgemmNTPfiS_iS_iiff.parentNode = call i8* @llvm.visc.getParentNode(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ %0 = call i32 @llvm.visc.getNodeInstanceID.x(i8* %_Z9mysgemmNTPfiS_iS_iiff.parentNode)
+ %1 = call i32 @llvm.visc.getNumNodeInstances.x(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ %2 = mul i32 %0, %1
+ %3 = call i32 @llvm.visc.getNodeInstanceID.x(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ %4 = add i32 %2, %3
+ %mul = shl nsw i32 %4, 2
+ %5 = call i32 @llvm.visc.getNodeInstanceID.y(i8* %_Z9mysgemmNTPfiS_iS_iiff.parentNode)
+ %6 = call i32 @llvm.visc.getNumNodeInstances.y(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ %7 = mul i32 %5, %6
+ %8 = call i32 @llvm.visc.getNodeInstanceID.y(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ %9 = add i32 %7, %8
+ %cmp147 = icmp sgt i32 %k, 0
+ %add3144 = or i32 %mul, 1
+ %add8145 = or i32 %mul, 2
+ %add13146 = or i32 %mul, 3
+
+ %mul.tmp1 = insertelement <4 x i32> < i32 0, i32 0, i32 0, i32 0 >, i32 %mul, i32 0
+ %mul.tmp2 = insertelement <4 x i32> %mul.tmp1, i32 %add3144, i32 1
+ %mul.tmp3 = insertelement <4 x i32> %mul.tmp2, i32 %add8145, i32 2
+ %mul.vector = insertelement <4 x i32> %mul.tmp2, i32 %add13146, i32 3
+
+ %lda.tmp = insertelement <1 x i32> < i32 0 >, i32 %lda, i32 0
+ %lda.vector = shufflevector <1 x i32> %lda.tmp, <1 x i32> undef, <4 x i32> < i32 0, i32 0, i32 0, i32 0 >
+
+ br i1 %cmp147, label %for.body, label %for.end
+
+for.body: ; preds = %for.body, %entry
+ %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
+
+; %c0.0152 = phi float [ %add23, %for.body ], [ 0.000000e+00, %entry ]
+; %c1.0151 = phi float [ %add25, %for.body ], [ 0.000000e+00, %entry ]
+; %c2.0150 = phi float [ %add27, %for.body ], [ 0.000000e+00, %entry ]
+; %c3.0149 = phi float [ %add29, %for.body ], [ 0.000000e+00, %entry ]
+ %c.vector = phi <4 x float> [ %add23, %for.body ], [ < float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00 >, %entry ]
+
+ %10 = trunc i64 %indvars.iv to i32
+ %mul2 = mul nsw i32 %10, %lda
+
+; %add = add nsw i32 %mul2, %mul
+; %idxprom = sext i32 %add to i64
+; %arrayidx = getelementptr inbounds float* %A, i64 %idxprom
+; %11 = load float* %arrayidx, align 4, !tbaa !0
+; %add5 = add nsw i32 %mul2, %add3144
+; %idxprom6 = sext i32 %add5 to i64
+; %arrayidx7 = getelementptr inbounds float* %A, i64 %idxprom6
+; %12 = load float* %arrayidx7, align 4, !tbaa !0
+; %add10 = add nsw i32 %mul2, %add8145
+; %idxprom11 = sext i32 %add10 to i64
+; %arrayidx12 = getelementptr inbounds float* %A, i64 %idxprom11
+; %13 = load float* %arrayidx12, align 4, !tbaa !0
+; %add15 = add nsw i32 %mul2, %add13146
+; %idxprom16 = sext i32 %add15 to i64
+; %arrayidx17 = getelementptr inbounds float* %A, i64 %idxprom16
+; %14 = load float* %arrayidx17, align 4, !tbaa !0
+ %add = add nsw i32 %mul2, %mul
+ %idxprom = sext i32 %add to i64
+ %arrayidx = getelementptr inbounds float* %A, i64 %idxprom
+ %arrayidx.cast = bitcast float* %arrayidx to <4 x float>*
+ %11 = load <4 x float>* %arrayidx.cast, align 4
+
+ %mul18 = mul nsw i32 %10, %ldb
+ %add19 = add nsw i32 %mul18, %9
+ %idxprom20 = sext i32 %add19 to i64
+ %arrayidx21 = getelementptr inbounds float* %B, i64 %idxprom20
+; %15 = load float* %arrayidx21, align 4, !tbaa !0
+ %12 = load float* %arrayidx21, align 4, !tbaa !0
+
+ %b.tmp = insertelement <1 x float> < float 0.000000e+00 >, float %12, i32 0
+ %b.vector = shufflevector <1 x float> %b.tmp, <1 x float> undef, <4 x i32> < i32 0, i32 0, i32 0, i32 0 >
+
+; %mul22 = fmul fast float %11, %15
+; %add23 = fadd fast float %c0.0152, %mul22
+; %mul24 = fmul fast float %12, %15
+; %add25 = fadd fast float %c1.0151, %mul24
+; %mul26 = fmul fast float %13, %15
+; %add27 = fadd fast float %c2.0150, %mul26
+; %mul28 = fmul fast float %14, %15
+; %add29 = fadd fast float %c3.0149, %mul28
+ %mul22 = fmul fast <4 x float> %11, %b.vector
+ %add23 = fadd fast <4 x float> %c.vector, %mul22
+
+ %indvars.iv.next = add i64 %indvars.iv, 1
+ %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+ %exitcond = icmp eq i32 %lftr.wideiv, %k
+ br i1 %exitcond, label %for.end, label %for.body
+
+for.end: ; preds = %for.body, %entry
+; %c0.0.lcssa = phi float [ %add23, %for.body ], [ 0.000000e+00, %entry ]
+; %c1.0.lcssa = phi float [ %add25, %for.body ], [ 0.000000e+00, %entry ]
+; %c2.0.lcssa = phi float [ %add27, %for.body ], [ 0.000000e+00, %entry ]
+; %c3.0.lcssa = phi float [ %add29, %for.body ], [ 0.000000e+00, %entry ]
+ %c.end.vector = phi <4 x float> [ %add23, %for.body ], [ < float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00 >, %entry ]
+
+ %c0.0.lcssa = extractelement <4 x float> %c.end.vector, i32 0
+ %c1.0.lcssa = extractelement <4 x float> %c.end.vector, i32 1
+ %c2.0.lcssa = extractelement <4 x float> %c.end.vector, i32 2
+ %c3.0.lcssa = extractelement <4 x float> %c.end.vector, i32 3
+
+ %mul30 = mul nsw i32 %9, %ldc
+ %add31 = add nsw i32 %mul30, %mul
+ %idxprom32 = sext i32 %add31 to i64
+ %arrayidx33 = getelementptr inbounds float* %C, i64 %idxprom32
+
+; %16 = load float* %arrayidx33, align 4, !tbaa !0
+; %mul34 = fmul fast float %16, %beta
+ %13 = load float* %arrayidx33, align 4, !tbaa !0
+ %mul34 = fmul fast float %13, %beta
+
+ %mul35 = fmul fast float %c0.0.lcssa, %alpha
+ %add36 = fadd fast float %mul35, %mul34
+ store float %add36, float* %arrayidx33, align 4, !tbaa !0
+ %add43 = add nsw i32 %add3144, %mul30
+ %idxprom44 = sext i32 %add43 to i64
+ %arrayidx45 = getelementptr inbounds float* %C, i64 %idxprom44
+
+; %17 = load float* %arrayidx45, align 4, !tbaa !0
+; %mul46 = fmul fast float %17, %beta
+ %14 = load float* %arrayidx45, align 4, !tbaa !0
+ %mul46 = fmul fast float %14, %beta
+
+ %mul47 = fmul fast float %c1.0.lcssa, %alpha
+ %add48 = fadd fast float %mul47, %mul46
+ store float %add48, float* %arrayidx45, align 4, !tbaa !0
+ %add56 = add nsw i32 %add8145, %mul30
+ %idxprom57 = sext i32 %add56 to i64
+ %arrayidx58 = getelementptr inbounds float* %C, i64 %idxprom57
+
+; %18 = load float* %arrayidx58, align 4, !tbaa !0
+; %mul59 = fmul fast float %18, %beta
+ %15 = load float* %arrayidx58, align 4, !tbaa !0
+ %mul59 = fmul fast float %15, %beta
+
+ %mul60 = fmul fast float %c2.0.lcssa, %alpha
+ %add61 = fadd fast float %mul60, %mul59
+ store float %add61, float* %arrayidx58, align 4, !tbaa !0
+ %add69 = add nsw i32 %add13146, %mul30
+ %idxprom70 = sext i32 %add69 to i64
+ %arrayidx71 = getelementptr inbounds float* %C, i64 %idxprom70
+
+; %19 = load float* %arrayidx71, align 4, !tbaa !0
+; %mul72 = fmul fast float %19, %beta
+ %16 = load float* %arrayidx71, align 4, !tbaa !0
+ %mul72 = fmul fast float %16, %beta
+
+ %mul73 = fmul fast float %c3.0.lcssa, %alpha
+ %add74 = fadd fast float %mul73, %mul72
+ store float %add74, float* %arrayidx71, align 4, !tbaa !0
+ ret %rtype undef
+}
+
+; Function Attrs: noinline nounwind uwtable
+define void @_Z10basicSgemmcciiifPfmiS_mifS_mi(i8 signext %transa, i8 signext %transb, i32 %m, i32 %n, i32 %k, float %alpha, float* %A, i64 %bytesA, i32 %lda, float* %B, i64 %bytesB, i32 %ldb, float %beta, float* %C, i64 %bytesC, i32 %ldc) #3 {
+entry:
+ switch i8 %transa, label %if.then [
+ i8 78, label %if.end
+ i8 110, label %if.end
+ ]
+
+if.then: ; preds = %entry
+ %call1.i = tail call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* @_ZSt4cerr, i8* getelementptr inbounds ([48 x i8]* @.str, i64 0, i64 0), i64 47) #1
+ %vtable.i = load i8** bitcast (%"class.std::basic_ostream"* @_ZSt4cerr to i8**), align 8, !tbaa !3
+ %vbase.offset.ptr.i = getelementptr i8* %vtable.i, i64 -24
+ %0 = bitcast i8* %vbase.offset.ptr.i to i64*
+ %vbase.offset.i = load i64* %0, align 8
+ %add.ptr.i.sum = add i64 %vbase.offset.i, 240
+ %_M_ctype.i = getelementptr inbounds i8* bitcast (%"class.std::basic_ostream"* @_ZSt4cerr to i8*), i64 %add.ptr.i.sum
+ %1 = bitcast i8* %_M_ctype.i to %"class.std::ctype"**
+ %2 = load %"class.std::ctype"** %1, align 8, !tbaa !4
+ %tobool.i97 = icmp eq %"class.std::ctype"* %2, null
+ br i1 %tobool.i97, label %if.then.i98, label %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit
+
+if.then.i98: ; preds = %if.then
+ tail call void @_ZSt16__throw_bad_castv() #7
+ unreachable
+
+_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit: ; preds = %if.then
+ %_M_widen_ok.i = getelementptr inbounds %"class.std::ctype"* %2, i64 0, i32 6
+ %3 = load i8* %_M_widen_ok.i, align 1, !tbaa !1
+ %tobool.i = icmp eq i8 %3, 0
+ br i1 %tobool.i, label %if.end.i, label %if.then.i
+
+if.then.i: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit
+ %arrayidx.i = getelementptr inbounds %"class.std::ctype"* %2, i64 0, i32 7, i64 10
+ %4 = load i8* %arrayidx.i, align 1, !tbaa !1
+ br label %_ZNKSt5ctypeIcE5widenEc.exit
+
+if.end.i: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit
+ tail call void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"* %2) #1
+ %5 = bitcast %"class.std::ctype"* %2 to i8 (%"class.std::ctype"*, i8)***
+ %vtable.i71 = load i8 (%"class.std::ctype"*, i8)*** %5, align 8, !tbaa !3
+ %vfn.i = getelementptr inbounds i8 (%"class.std::ctype"*, i8)** %vtable.i71, i64 6
+ %6 = load i8 (%"class.std::ctype"*, i8)** %vfn.i, align 8
+ %call.i72 = tail call signext i8 %6(%"class.std::ctype"* %2, i8 signext 10) #1
+ br label %_ZNKSt5ctypeIcE5widenEc.exit
+
+_ZNKSt5ctypeIcE5widenEc.exit: ; preds = %if.end.i, %if.then.i
+ %retval.0.i = phi i8 [ %4, %if.then.i ], [ %call.i72, %if.end.i ]
+ %call1.i47 = tail call %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"* @_ZSt4cerr, i8 signext %retval.0.i) #1
+ %call.i = tail call %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"* %call1.i47) #1
+ br label %return
+
+if.end: ; preds = %entry, %entry
+ switch i8 %transb, label %if.then9 [
+ i8 84, label %if.end12
+ i8 116, label %if.end12
+ ]
+
+if.then9: ; preds = %if.end
+ %call1.i49 = tail call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* @_ZSt4cerr, i8* getelementptr inbounds ([48 x i8]* @.str1, i64 0, i64 0), i64 47) #1
+ %vtable.i51 = load i8** bitcast (%"class.std::basic_ostream"* @_ZSt4cerr to i8**), align 8, !tbaa !3
+ %vbase.offset.ptr.i52 = getelementptr i8* %vtable.i51, i64 -24
+ %7 = bitcast i8* %vbase.offset.ptr.i52 to i64*
+ %vbase.offset.i53 = load i64* %7, align 8
+ %add.ptr.i54.sum = add i64 %vbase.offset.i53, 240
+ %_M_ctype.i73 = getelementptr inbounds i8* bitcast (%"class.std::basic_ostream"* @_ZSt4cerr to i8*), i64 %add.ptr.i54.sum
+ %8 = bitcast i8* %_M_ctype.i73 to %"class.std::ctype"**
+ %9 = load %"class.std::ctype"** %8, align 8, !tbaa !4
+ %tobool.i100 = icmp eq %"class.std::ctype"* %9, null
+ br i1 %tobool.i100, label %if.then.i101, label %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit103
+
+if.then.i101: ; preds = %if.then9
+ tail call void @_ZSt16__throw_bad_castv() #7
+ unreachable
+
+_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit103: ; preds = %if.then9
+ %_M_widen_ok.i75 = getelementptr inbounds %"class.std::ctype"* %9, i64 0, i32 6
+ %10 = load i8* %_M_widen_ok.i75, align 1, !tbaa !1
+ %tobool.i76 = icmp eq i8 %10, 0
+ br i1 %tobool.i76, label %if.end.i82, label %if.then.i78
+
+if.then.i78: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit103
+ %arrayidx.i77 = getelementptr inbounds %"class.std::ctype"* %9, i64 0, i32 7, i64 10
+ %11 = load i8* %arrayidx.i77, align 1, !tbaa !1
+ br label %_ZNKSt5ctypeIcE5widenEc.exit84
+
+if.end.i82: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit103
+ tail call void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"* %9) #1
+ %12 = bitcast %"class.std::ctype"* %9 to i8 (%"class.std::ctype"*, i8)***
+ %vtable.i79 = load i8 (%"class.std::ctype"*, i8)*** %12, align 8, !tbaa !3
+ %vfn.i80 = getelementptr inbounds i8 (%"class.std::ctype"*, i8)** %vtable.i79, i64 6
+ %13 = load i8 (%"class.std::ctype"*, i8)** %vfn.i80, align 8
+ %call.i81 = tail call signext i8 %13(%"class.std::ctype"* %9, i8 signext 10) #1
+ br label %_ZNKSt5ctypeIcE5widenEc.exit84
+
+_ZNKSt5ctypeIcE5widenEc.exit84: ; preds = %if.end.i82, %if.then.i78
+ %retval.0.i83 = phi i8 [ %11, %if.then.i78 ], [ %call.i81, %if.end.i82 ]
+ %call1.i56 = tail call %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"* @_ZSt4cerr, i8 signext %retval.0.i83) #1
+ %call.i57 = tail call %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"* %call1.i56) #1
+ br label %return
+
+if.end12: ; preds = %if.end, %if.end
+ %rem44 = and i32 %m, 15
+ %tobool = icmp eq i32 %rem44, 0
+ br i1 %tobool, label %lor.lhs.false, label %if.then15
+
+lor.lhs.false: ; preds = %if.end12
+ %rem1345 = and i32 %n, 15
+ %tobool14 = icmp eq i32 %rem1345, 0
+ br i1 %tobool14, label %if.end21, label %if.then15
+
+if.then15: ; preds = %lor.lhs.false, %if.end12
+ %call1.i59 = tail call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* @_ZSt4cerr, i8* getelementptr inbounds ([53 x i8]* @.str2, i64 0, i64 0), i64 52) #1
+ %call17 = tail call %"class.std::basic_ostream"* @_ZNSolsEi(%"class.std::basic_ostream"* @_ZSt4cerr, i32 16) #1
+ %call1.i61 = tail call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* %call17, i8* getelementptr inbounds ([27 x i8]* @.str3, i64 0, i64 0), i64 26) #1
+ %call19 = tail call %"class.std::basic_ostream"* @_ZNSolsEi(%"class.std::basic_ostream"* %call17, i32 16) #1
+ %14 = bitcast %"class.std::basic_ostream"* %call19 to i8**
+ %vtable.i63 = load i8** %14, align 8, !tbaa !3
+ %vbase.offset.ptr.i64 = getelementptr i8* %vtable.i63, i64 -24
+ %15 = bitcast i8* %vbase.offset.ptr.i64 to i64*
+ %vbase.offset.i65 = load i64* %15, align 8
+ %16 = bitcast %"class.std::basic_ostream"* %call19 to i8*
+ %add.ptr.i66.sum = add i64 %vbase.offset.i65, 240
+ %_M_ctype.i85 = getelementptr inbounds i8* %16, i64 %add.ptr.i66.sum
+ %17 = bitcast i8* %_M_ctype.i85 to %"class.std::ctype"**
+ %18 = load %"class.std::ctype"** %17, align 8, !tbaa !4
+ %tobool.i104 = icmp eq %"class.std::ctype"* %18, null
+ br i1 %tobool.i104, label %if.then.i105, label %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit107
+
+if.then.i105: ; preds = %if.then15
+ tail call void @_ZSt16__throw_bad_castv() #7
+ unreachable
+
+_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit107: ; preds = %if.then15
+ %_M_widen_ok.i87 = getelementptr inbounds %"class.std::ctype"* %18, i64 0, i32 6
+ %19 = load i8* %_M_widen_ok.i87, align 1, !tbaa !1
+ %tobool.i88 = icmp eq i8 %19, 0
+ br i1 %tobool.i88, label %if.end.i94, label %if.then.i90
+
+if.then.i90: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit107
+ %arrayidx.i89 = getelementptr inbounds %"class.std::ctype"* %18, i64 0, i32 7, i64 10
+ %20 = load i8* %arrayidx.i89, align 1, !tbaa !1
+ br label %_ZNKSt5ctypeIcE5widenEc.exit96
+
+if.end.i94: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit107
+ tail call void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"* %18) #1
+ %21 = bitcast %"class.std::ctype"* %18 to i8 (%"class.std::ctype"*, i8)***
+ %vtable.i91 = load i8 (%"class.std::ctype"*, i8)*** %21, align 8, !tbaa !3
+ %vfn.i92 = getelementptr inbounds i8 (%"class.std::ctype"*, i8)** %vtable.i91, i64 6
+ %22 = load i8 (%"class.std::ctype"*, i8)** %vfn.i92, align 8
+ %call.i93 = tail call signext i8 %22(%"class.std::ctype"* %18, i8 signext 10) #1
+ br label %_ZNKSt5ctypeIcE5widenEc.exit96
+
+_ZNKSt5ctypeIcE5widenEc.exit96: ; preds = %if.end.i94, %if.then.i90
+ %retval.0.i95 = phi i8 [ %20, %if.then.i90 ], [ %call.i93, %if.end.i94 ]
+ %call1.i68 = tail call %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"* %call19, i8 signext %retval.0.i95) #1
+ %call.i69 = tail call %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"* %call1.i68) #1
+ br label %if.end21
+
+if.end21: ; preds = %_ZNKSt5ctypeIcE5widenEc.exit96, %lor.lhs.false
+ %div = sdiv i32 %m, 16
+ %mul = and i32 %div, 1073741823
+ %div22 = sdiv i32 %n, 16
+ %mul24 = and i32 %div22, 268435455
+ %conv33 = fpext float %alpha to double
+ %conv34 = fpext float %beta to double
+ call void @llvm_visc_switchToTimer(i8** @viscTimerSet_GenVISC, i32 19)
+ %in.addr = alloca %struct.arg
+ %in.addr.A = getelementptr %struct.arg* %in.addr, i32 0, i32 0
+ store float* %A, float** %in.addr.A
+ %in.addr.bytes_A = getelementptr %struct.arg* %in.addr, i32 0, i32 1
+ store i64 %bytesA, i64* %in.addr.bytes_A
+ %in.addr.lda = getelementptr %struct.arg* %in.addr, i32 0, i32 2
+ store i32 %lda, i32* %in.addr.lda
+ %in.addr.B = getelementptr %struct.arg* %in.addr, i32 0, i32 3
+ store float* %B, float** %in.addr.B
+ %in.addr.bytes_B = getelementptr %struct.arg* %in.addr, i32 0, i32 4
+ store i64 %bytesB, i64* %in.addr.bytes_B
+ %in.addr.ldb = getelementptr %struct.arg* %in.addr, i32 0, i32 5
+ store i32 %ldb, i32* %in.addr.ldb
+ %in.addr.C = getelementptr %struct.arg* %in.addr, i32 0, i32 6
+ store float* %C, float** %in.addr.C
+ %in.addr.bytes_C = getelementptr %struct.arg* %in.addr, i32 0, i32 7
+ store i64 %bytesC, i64* %in.addr.bytes_C
+ %in.addr.ldc = getelementptr %struct.arg* %in.addr, i32 0, i32 8
+ store i32 %ldc, i32* %in.addr.ldc
+ %in.addr.k = getelementptr %struct.arg* %in.addr, i32 0, i32 9
+ store i32 %k, i32* %in.addr.k
+ %in.addr.alpha = getelementptr %struct.arg* %in.addr, i32 0, i32 10
+ %in.addr.alpha.cast = fptrunc double %conv33 to float
+ store float %in.addr.alpha.cast, float* %in.addr.alpha
+ %in.addr.beta = getelementptr %struct.arg* %in.addr, i32 0, i32 11
+ %in.addr.beta.cast = fptrunc double %conv34 to float
+ store float %in.addr.beta.cast, float* %in.addr.beta
+ %in.addr.dimX0 = getelementptr %struct.arg* %in.addr, i32 0, i32 12
+ store i32 4, i32* %in.addr.dimX0
+ %in.addr.dimY0 = getelementptr %struct.arg* %in.addr, i32 0, i32 13
+ store i32 16, i32* %in.addr.dimY0
+ %in.addr.dimX1 = getelementptr %struct.arg* %in.addr, i32 0, i32 14
+ store i32 %mul, i32* %in.addr.dimX1
+ %in.addr.dimY1 = getelementptr %struct.arg* %in.addr, i32 0, i32 15
+ store i32 %mul24, i32* %in.addr.dimY1
+ %args = bitcast %struct.arg* %in.addr to i8*
+ call void @llvm_visc_switchToTimer(i8** @viscTimerSet_GenVISC, i32 0)
+ %graph_Z9mysgemmNTPfiS_iS_iiffInternal_level2 = call i8* @llvm.visc.launch(i8* bitcast (%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float, i32, i32, i32, i32)* @_Z9mysgemmNTPfiS_iS_iiffInternal_level2 to i8*), i8* %args)
+ call void @llvm.visc.wait(i8* %graph_Z9mysgemmNTPfiS_iS_iiffInternal_level2)
+ br label %return
+
+return: ; preds = %if.end21, %_ZNKSt5ctypeIcE5widenEc.exit84, %_ZNKSt5ctypeIcE5widenEc.exit
+ ret void
+}
+
+declare %"class.std::basic_ostream"* @_ZNSolsEi(%"class.std::basic_ostream"*, i32) #0
+
+; Function Attrs: nounwind uwtable
+define i32 @main(i32 %argc, i8** %argv) #2 {
+entry:
+ %argc.addr = alloca i32, align 4
+ %timers = alloca %struct.pb_TimerSet, align 8
+ %matArow = alloca i32, align 4
+ %matAcol = alloca i32, align 4
+ %matBrow = alloca i32, align 4
+ %matBcol = alloca i32, align 4
+ %matA = alloca %"class.std::vector", align 8
+ %matBT = alloca %"class.std::vector", align 8
+ %matC = alloca %"class.std::vector", align 8
+ store i32 %argc, i32* %argc.addr, align 4, !tbaa !5
+ %0 = bitcast %struct.pb_TimerSet* %timers to i8*
+ call void @llvm.lifetime.start(i64 800, i8* %0) #1
+ %1 = bitcast %"class.std::vector"* %matA to i8*
+ call void @llvm.memset.p0i8.i64(i8* %1, i8 0, i64 24, i32 8, i1 false) #1
+ %2 = bitcast %"class.std::vector"* %matBT to i8*
+ call void @llvm.memset.p0i8.i64(i8* %2, i8 0, i64 24, i32 8, i1 false) #1
+ call void @pb_InitializeTimerSet(%struct.pb_TimerSet* %timers) #1
+ %3 = call i8* @llvm_visc_initializeTimerSet()
+ store i8* %3, i8** @viscTimerSet_GenVISC
+ call void @llvm_visc_switchToTimer(i8** @viscTimerSet_GenVISC, i32 0)
+ call void @llvm.visc.init()
+ %call = call %struct.pb_Parameters* @pb_ReadParameters(i32* %argc.addr, i8** %argv) #1
+ %inpFiles = getelementptr inbounds %struct.pb_Parameters* %call, i64 0, i32 1
+ %4 = load i8*** %inpFiles, align 8, !tbaa !4
+ %5 = load i8** %4, align 8, !tbaa !4
+ %cmp = icmp eq i8* %5, null
+ br i1 %cmp, label %if.then, label %lor.lhs.false
+
+lor.lhs.false: ; preds = %entry
+ %arrayidx2 = getelementptr inbounds i8** %4, i64 1
+ %6 = load i8** %arrayidx2, align 8, !tbaa !4
+ %cmp3 = icmp eq i8* %6, null
+ br i1 %cmp3, label %if.then, label %lor.lhs.false4
+
+lor.lhs.false4: ; preds = %lor.lhs.false
+ %arrayidx6 = getelementptr inbounds i8** %4, i64 2
+ %7 = load i8** %arrayidx6, align 8, !tbaa !4
+ %cmp7 = icmp eq i8* %7, null
+ br i1 %cmp7, label %if.then, label %lor.lhs.false8
+
+lor.lhs.false8: ; preds = %lor.lhs.false4
+ %arrayidx10 = getelementptr inbounds i8** %4, i64 3
+ %8 = load i8** %arrayidx10, align 8, !tbaa !4
+ %cmp11 = icmp eq i8* %8, null
+ br i1 %cmp11, label %if.end, label %if.then
+
+if.then: ; preds = %lor.lhs.false8, %lor.lhs.false4, %lor.lhs.false, %entry
+ %9 = load %struct._IO_FILE** @stderr, align 8, !tbaa !4
+ %10 = call i64 @fwrite(i8* getelementptr inbounds ([33 x i8]* @.str4, i64 0, i64 0), i64 32, i64 1, %struct._IO_FILE* %9)
+ call void @exit(i32 -1) #7
+ unreachable
+
+if.end: ; preds = %lor.lhs.false8
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 1) #1
+ %11 = load i8*** %inpFiles, align 8, !tbaa !4
+ %12 = load i8** %11, align 8, !tbaa !4
+ %call15 = call zeroext i1 @_Z22readColMajorMatrixFilePKcRiS1_RSt6vectorIfSaIfEE(i8* %12, i32* %matArow, i32* %matAcol, %"class.std::vector"* %matA) #1
+ %13 = load i8*** %inpFiles, align 8, !tbaa !4
+ %arrayidx17 = getelementptr inbounds i8** %13, i64 2
+ %14 = load i8** %arrayidx17, align 8, !tbaa !4
+ %call18 = call zeroext i1 @_Z22readColMajorMatrixFilePKcRiS1_RSt6vectorIfSaIfEE(i8* %14, i32* %matBcol, i32* %matBrow, %"class.std::vector"* %matBT) #1
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 6) #1
+ %15 = load i32* %matArow, align 4, !tbaa !5
+ %16 = load i32* %matAcol, align 4, !tbaa !5
+ %mul = mul nsw i32 %16, %15
+ %conv = sext i32 %mul to i64
+ %mul19 = shl nsw i64 %conv, 2
+ %17 = load i32* %matBrow, align 4, !tbaa !5
+ %18 = load i32* %matBcol, align 4, !tbaa !5
+ %mul20 = mul nsw i32 %18, %17
+ %conv21 = sext i32 %mul20 to i64
+ %mul22 = shl nsw i64 %conv21, 2
+ %mul23 = mul nsw i32 %18, %15
+ %conv24 = sext i32 %mul23 to i64
+ %mul25 = shl nsw i64 %conv24, 2
+ %19 = bitcast %"class.std::vector"* %matC to i8*
+ call void @llvm.memset.p0i8.i64(i8* %19, i8 0, i64 24, i32 8, i1 false) #1
+ %cmp.i.i.i.i = icmp eq i32 %mul23, 0
+ br i1 %cmp.i.i.i.i, label %_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i, label %cond.true.i.i.i.i
+
+cond.true.i.i.i.i: ; preds = %if.end
+ %cmp.i.i.i.i.i = icmp slt i32 %mul23, 0
+ br i1 %cmp.i.i.i.i.i, label %if.then.i.i.i.i.i, label %_ZN9__gnu_cxx13new_allocatorIfE8allocateEmPKv.exit.i.i.i.i, !prof !6
+
+if.then.i.i.i.i.i: ; preds = %cond.true.i.i.i.i
+ call void @_ZSt17__throw_bad_allocv() #7
+ unreachable
+
+_ZN9__gnu_cxx13new_allocatorIfE8allocateEmPKv.exit.i.i.i.i: ; preds = %cond.true.i.i.i.i
+ %call2.i.i.i.i.i = call noalias i8* @_Znwm(i64 %mul25) #1
+ %20 = bitcast i8* %call2.i.i.i.i.i to float*
+ br label %_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i
+
+_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i: ; preds = %_ZN9__gnu_cxx13new_allocatorIfE8allocateEmPKv.exit.i.i.i.i, %if.end
+ %cond.i.i.i.i = phi float* [ %20, %_ZN9__gnu_cxx13new_allocatorIfE8allocateEmPKv.exit.i.i.i.i ], [ null, %if.end ]
+ %_M_start.i.i.i81 = getelementptr inbounds %"class.std::vector"* %matC, i64 0, i32 0, i32 0, i32 0
+ store float* %cond.i.i.i.i, float** %_M_start.i.i.i81, align 8, !tbaa !4
+ %_M_finish.i.i.i = getelementptr inbounds %"class.std::vector"* %matC, i64 0, i32 0, i32 0, i32 1
+ store float* %cond.i.i.i.i, float** %_M_finish.i.i.i, align 8, !tbaa !4
+ %add.ptr.i.i.i = getelementptr inbounds float* %cond.i.i.i.i, i64 %conv24
+ %_M_end_of_storage.i.i.i = getelementptr inbounds %"class.std::vector"* %matC, i64 0, i32 0, i32 0, i32 2
+ store float* %add.ptr.i.i.i, float** %_M_end_of_storage.i.i.i, align 8, !tbaa !4
+ br i1 %cmp.i.i.i.i, label %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit, label %for.body.lr.ph.i.i.i.i.i.i.i.i
+
+for.body.lr.ph.i.i.i.i.i.i.i.i: ; preds = %_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i
+ %n.mod.vf.i.i.i.i.i.i.i.i = and i64 %conv24, 7
+ %n.vec.i.i.i.i.i.i.i.i = sub i64 %conv24, %n.mod.vf.i.i.i.i.i.i.i.i
+ %cmp.zero.i.i.i.i.i.i.i.i = icmp eq i64 %n.mod.vf.i.i.i.i.i.i.i.i, %conv24
+ %ptr.ind.end.i.i.i.i.i.i.i.i = getelementptr float* %cond.i.i.i.i, i64 %n.vec.i.i.i.i.i.i.i.i
+ br i1 %cmp.zero.i.i.i.i.i.i.i.i, label %middle.block.i.i.i.i.i.i.i.i, label %vector.body.i.i.i.i.i.i.i.i
+
+vector.body.i.i.i.i.i.i.i.i: ; preds = %vector.body.i.i.i.i.i.i.i.i, %for.body.lr.ph.i.i.i.i.i.i.i.i
+ %index.i.i.i.i.i.i.i.i = phi i64 [ %index.next.i.i.i.i.i.i.i.i, %vector.body.i.i.i.i.i.i.i.i ], [ 0, %for.body.lr.ph.i.i.i.i.i.i.i.i ]
+ %next.gep.i.i.i.i.i.i.i.i = getelementptr float* %cond.i.i.i.i, i64 %index.i.i.i.i.i.i.i.i
+ %21 = bitcast float* %next.gep.i.i.i.i.i.i.i.i to <4 x float>*
+ store <4 x float> zeroinitializer, <4 x float>* %21, align 4
+ %next.gep.sum41.i.i.i.i.i.i.i.i = or i64 %index.i.i.i.i.i.i.i.i, 4
+ %22 = getelementptr float* %cond.i.i.i.i, i64 %next.gep.sum41.i.i.i.i.i.i.i.i
+ %23 = bitcast float* %22 to <4 x float>*
+ store <4 x float> zeroinitializer, <4 x float>* %23, align 4
+ %index.next.i.i.i.i.i.i.i.i = add i64 %index.i.i.i.i.i.i.i.i, 8
+ %24 = icmp eq i64 %index.next.i.i.i.i.i.i.i.i, %n.vec.i.i.i.i.i.i.i.i
+ br i1 %24, label %middle.block.i.i.i.i.i.i.i.i, label %vector.body.i.i.i.i.i.i.i.i
+
+middle.block.i.i.i.i.i.i.i.i: ; preds = %vector.body.i.i.i.i.i.i.i.i, %for.body.lr.ph.i.i.i.i.i.i.i.i
+ %resume.val.i.i.i.i.i.i.i.i = phi float* [ %cond.i.i.i.i, %for.body.lr.ph.i.i.i.i.i.i.i.i ], [ %ptr.ind.end.i.i.i.i.i.i.i.i, %vector.body.i.i.i.i.i.i.i.i ]
+ %resume.val7.i.i.i.i.i.i.i.i = phi i64 [ %conv24, %for.body.lr.ph.i.i.i.i.i.i.i.i ], [ %n.mod.vf.i.i.i.i.i.i.i.i, %vector.body.i.i.i.i.i.i.i.i ]
+ %new.indc.resume.val.i.i.i.i.i.i.i.i = phi i64 [ 0, %for.body.lr.ph.i.i.i.i.i.i.i.i ], [ %n.vec.i.i.i.i.i.i.i.i, %vector.body.i.i.i.i.i.i.i.i ]
+ %cmp.n.i.i.i.i.i.i.i.i = icmp eq i64 %new.indc.resume.val.i.i.i.i.i.i.i.i, %conv24
+ br i1 %cmp.n.i.i.i.i.i.i.i.i, label %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit, label %for.body.i.i.i.i.i.i.i.i.preheader
+
+for.body.i.i.i.i.i.i.i.i.preheader: ; preds = %middle.block.i.i.i.i.i.i.i.i
+ %resume.val.i.i.i.i.i.i.i.i101 = bitcast float* %resume.val.i.i.i.i.i.i.i.i to i8*
+ %25 = shl nsw i64 %resume.val7.i.i.i.i.i.i.i.i, 2
+ call void @llvm.memset.p0i8.i64(i8* %resume.val.i.i.i.i.i.i.i.i101, i8 0, i64 %25, i32 4, i1 false)
+ br label %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit
+
+_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit: ; preds = %for.body.i.i.i.i.i.i.i.i.preheader, %middle.block.i.i.i.i.i.i.i.i, %_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i
+ store float* %add.ptr.i.i.i, float** %_M_finish.i.i.i, align 8, !tbaa !4
+ %_M_start.i.i = getelementptr inbounds %"class.std::vector"* %matA, i64 0, i32 0, i32 0, i32 0
+ %26 = load float** %_M_start.i.i, align 8, !tbaa !4
+ %27 = bitcast float* %26 to i8*
+ call void @llvm_visc_track_mem(i8* %27, i64 %mul19) #1
+ %_M_start.i.i82 = getelementptr inbounds %"class.std::vector"* %matBT, i64 0, i32 0, i32 0, i32 0
+ %28 = load float** %_M_start.i.i82, align 8, !tbaa !4
+ %29 = bitcast float* %28 to i8*
+ call void @llvm_visc_track_mem(i8* %29, i64 %mul22) #1
+ %30 = load float** %_M_start.i.i.i81, align 8, !tbaa !4
+ %31 = bitcast float* %30 to i8*
+ call void @llvm_visc_track_mem(i8* %31, i64 %mul25) #1
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 6) #1
+ %32 = load float** %_M_finish.i.i.i, align 8, !tbaa !4
+ %33 = load float** %_M_start.i.i.i81, align 8, !tbaa !4
+ %cmp3399 = icmp eq float* %32, %33
+ br i1 %cmp3399, label %for.end, label %for.body.lr.ph
+
+for.body.lr.ph: ; preds = %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit
+ %sub.ptr.lhs.cast.i = ptrtoint float* %32 to i64
+ %sub.ptr.rhs.cast.i = ptrtoint float* %33 to i64
+ %sub.ptr.sub.i = sub i64 %sub.ptr.lhs.cast.i, %sub.ptr.rhs.cast.i
+ %sub.ptr.div.i = ashr exact i64 %sub.ptr.sub.i, 2
+ br label %for.body
+
+for.body: ; preds = %for.body, %for.body.lr.ph
+ %i.0100 = phi i64 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
+ %add.ptr.i = getelementptr inbounds float* %33, i64 %i.0100
+ store float 0.000000e+00, float* %add.ptr.i, align 4, !tbaa !0
+ %inc = add i64 %i.0100, 1
+ %cmp33 = icmp ult i64 %inc, %sub.ptr.div.i
+ br i1 %cmp33, label %for.body, label %for.end
+
+for.end: ; preds = %for.body, %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 0) #1
+ %34 = load i32* %matArow, align 4, !tbaa !5
+ %35 = load i32* %matBcol, align 4, !tbaa !5
+ %36 = load i32* %matAcol, align 4, !tbaa !5
+ %37 = load float** %_M_start.i.i, align 8, !tbaa !4
+ %38 = load float** %_M_start.i.i82, align 8, !tbaa !4
+ %39 = load float** %_M_start.i.i.i81, align 8, !tbaa !4
+ call void @_Z10basicSgemmcciiifPfmiS_mifS_mi(i8 signext 78, i8 signext 84, i32 %34, i32 %35, i32 %36, float 1.000000e+00, float* %37, i64 %mul19, i32 %34, float* %38, i64 %mul22, i32 %35, float 0.000000e+00, float* %39, i64 %mul25, i32 %34)
+ %outFile = getelementptr inbounds %struct.pb_Parameters* %call, i64 0, i32 0
+ %40 = load i8** %outFile, align 8, !tbaa !4
+ %tobool = icmp eq i8* %40, null
+ br i1 %tobool, label %if.end42, label %if.then38
+
+if.then38: ; preds = %for.end
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 3) #1
+ %41 = load float** %_M_start.i.i.i81, align 8, !tbaa !4
+ %42 = bitcast float* %41 to i8*
+ call void @llvm_visc_request_mem(i8* %42, i64 %mul25) #1
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 1) #1
+ %43 = load i8** %outFile, align 8, !tbaa !4
+ %44 = load i32* %matArow, align 4, !tbaa !5
+ %45 = load i32* %matBcol, align 4, !tbaa !5
+ %call41 = call zeroext i1 @_Z23writeColMajorMatrixFilePKciiRSt6vectorIfSaIfEE(i8* %43, i32 %44, i32 %45, %"class.std::vector"* %matC) #1
+ br label %if.end42
+
+if.end42: ; preds = %if.then38, %for.end
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 16) #1
+ %46 = load float** %_M_start.i.i, align 8, !tbaa !4
+ %47 = bitcast float* %46 to i8*
+ call void @llvm_visc_untrack_mem(i8* %47) #1
+ %48 = load float** %_M_start.i.i82, align 8, !tbaa !4
+ %49 = bitcast float* %48 to i8*
+ call void @llvm_visc_untrack_mem(i8* %49) #1
+ %50 = load float** %_M_start.i.i.i81, align 8, !tbaa !4
+ %51 = bitcast float* %50 to i8*
+ call void @llvm_visc_untrack_mem(i8* %51) #1
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 0) #1
+ %arrayidx47 = getelementptr inbounds %struct.pb_TimerSet* %timers, i64 0, i32 4, i64 2
+ %call48 = call double @pb_GetElapsedTime(%struct.pb_Timer* %arrayidx47) #1
+ %call1.i88 = call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* @_ZSt4cout, i8* getelementptr inbounds ([10 x i8]* @.str5, i64 0, i64 0), i64 9) #1
+ %52 = load i32* %matArow, align 4, !tbaa !5
+ %conv50 = sitofp i32 %52 to double
+ %mul51 = fmul fast double %conv50, 2.000000e+00
+ %53 = load i32* %matBcol, align 4, !tbaa !5
+ %conv52 = sitofp i32 %53 to double
+ %mul53 = fmul fast double %mul51, %conv52
+ %54 = load i32* %matAcol, align 4, !tbaa !5
+ %conv54 = sitofp i32 %54 to double
+ %mul55 = fmul fast double %mul53, %conv54
+ %div = fdiv fast double %mul55, %call48
+ %div56 = fmul double %div, 1.000000e-09
+ %call.i = call %"class.std::basic_ostream"* @_ZNSo9_M_insertIdEERSoT_(%"class.std::basic_ostream"* @_ZSt4cout, double %div56) #1
+ %55 = bitcast %"class.std::basic_ostream"* %call.i to i8**
+ %vtable.i = load i8** %55, align 8, !tbaa !3
+ %vbase.offset.ptr.i = getelementptr i8* %vtable.i, i64 -24
+ %56 = bitcast i8* %vbase.offset.ptr.i to i64*
+ %vbase.offset.i = load i64* %56, align 8
+ %57 = bitcast %"class.std::basic_ostream"* %call.i to i8*
+ %add.ptr.sum.i = add i64 %vbase.offset.i, 240
+ %_M_ctype.i.i = getelementptr inbounds i8* %57, i64 %add.ptr.sum.i
+ %58 = bitcast i8* %_M_ctype.i.i to %"class.std::ctype"**
+ %59 = load %"class.std::ctype"** %58, align 8, !tbaa !4
+ %tobool.i.i.i = icmp eq %"class.std::ctype"* %59, null
+ br i1 %tobool.i.i.i, label %if.then.i.i.i, label %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit.i.i
+
+if.then.i.i.i: ; preds = %if.end42
+ call void @_ZSt16__throw_bad_castv() #7
+ unreachable
+
+_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit.i.i: ; preds = %if.end42
+ %_M_widen_ok.i.i.i = getelementptr inbounds %"class.std::ctype"* %59, i64 0, i32 6
+ %60 = load i8* %_M_widen_ok.i.i.i, align 1, !tbaa !1
+ %tobool.i3.i.i = icmp eq i8 %60, 0
+ br i1 %tobool.i3.i.i, label %if.end.i.i.i, label %if.then.i4.i.i
+
+if.then.i4.i.i: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit.i.i
+ %arrayidx.i.i.i = getelementptr inbounds %"class.std::ctype"* %59, i64 0, i32 7, i64 10
+ %61 = load i8* %arrayidx.i.i.i, align 1, !tbaa !1
+ br label %_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit
+
+if.end.i.i.i: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit.i.i
+ call void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"* %59) #1
+ %62 = bitcast %"class.std::ctype"* %59 to i8 (%"class.std::ctype"*, i8)***
+ %vtable.i.i.i = load i8 (%"class.std::ctype"*, i8)*** %62, align 8, !tbaa !3
+ %vfn.i.i.i = getelementptr inbounds i8 (%"class.std::ctype"*, i8)** %vtable.i.i.i, i64 6
+ %63 = load i8 (%"class.std::ctype"*, i8)** %vfn.i.i.i, align 8
+ %call.i.i.i = call signext i8 %63(%"class.std::ctype"* %59, i8 signext 10) #1
+ br label %_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit
+
+_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit: ; preds = %if.end.i.i.i, %if.then.i4.i.i
+ %retval.0.i.i.i = phi i8 [ %61, %if.then.i4.i.i ], [ %call.i.i.i, %if.end.i.i.i ]
+ %call1.i = call %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"* %call.i, i8 signext %retval.0.i.i.i) #1
+ %call.i.i = call %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"* %call1.i) #1
+ call void @pb_PrintTimerSet(%struct.pb_TimerSet* %timers) #1
+ %Ptr = getelementptr [14 x i8]* @0, i64 0, i64 0
+ call void @llvm_visc_printTimerSet(i8** @viscTimerSet_GenVISC, i8* %Ptr)
+ call void @llvm.visc.cleanup()
+ call void @pb_FreeParameters(%struct.pb_Parameters* %call) #1
+ %64 = load float** %_M_start.i.i.i81, align 8, !tbaa !4
+ %tobool.i.i.i.i78 = icmp eq float* %64, null
+ br i1 %tobool.i.i.i.i78, label %_ZNSt6vectorIfSaIfEED1Ev.exit80, label %if.then.i.i.i.i79
+
+if.then.i.i.i.i79: ; preds = %_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit
+ %65 = bitcast float* %64 to i8*
+ call void @_ZdlPv(i8* %65) #1
+ br label %_ZNSt6vectorIfSaIfEED1Ev.exit80
+
+_ZNSt6vectorIfSaIfEED1Ev.exit80: ; preds = %if.then.i.i.i.i79, %_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit
+ %66 = load float** %_M_start.i.i82, align 8, !tbaa !4
+ %tobool.i.i.i.i74 = icmp eq float* %66, null
+ br i1 %tobool.i.i.i.i74, label %_ZNSt6vectorIfSaIfEED1Ev.exit76, label %if.then.i.i.i.i75
+
+if.then.i.i.i.i75: ; preds = %_ZNSt6vectorIfSaIfEED1Ev.exit80
+ %67 = bitcast float* %66 to i8*
+ call void @_ZdlPv(i8* %67) #1
+ br label %_ZNSt6vectorIfSaIfEED1Ev.exit76
+
+_ZNSt6vectorIfSaIfEED1Ev.exit76: ; preds = %if.then.i.i.i.i75, %_ZNSt6vectorIfSaIfEED1Ev.exit80
+ %68 = load float** %_M_start.i.i, align 8, !tbaa !4
+ %tobool.i.i.i.i = icmp eq float* %68, null
+ br i1 %tobool.i.i.i.i, label %_ZNSt6vectorIfSaIfEED1Ev.exit, label %if.then.i.i.i.i
+
+if.then.i.i.i.i: ; preds = %_ZNSt6vectorIfSaIfEED1Ev.exit76
+ %69 = bitcast float* %68 to i8*
+ call void @_ZdlPv(i8* %69) #1
+ br label %_ZNSt6vectorIfSaIfEED1Ev.exit
+
+_ZNSt6vectorIfSaIfEED1Ev.exit: ; preds = %if.then.i.i.i.i, %_ZNSt6vectorIfSaIfEED1Ev.exit76
+ call void @llvm.lifetime.end(i64 800, i8* %0) #1
+ ret i32 0
+}
+
+; Function Attrs: nounwind
+declare void @llvm.lifetime.start(i64, i8* nocapture) #1
+
+declare void @pb_InitializeTimerSet(%struct.pb_TimerSet*) #0
+
+declare %struct.pb_Parameters* @pb_ReadParameters(i32*, i8**) #0
+
+; Function Attrs: noreturn nounwind
+declare void @exit(i32) #4
+
+declare void @pb_SwitchToTimer(%struct.pb_TimerSet*, i32) #0
+
+declare zeroext i1 @_Z22readColMajorMatrixFilePKcRiS1_RSt6vectorIfSaIfEE(i8*, i32*, i32*, %"class.std::vector"*) #0
+
+declare void @llvm_visc_track_mem(i8*, i64) #0
+
+declare void @llvm_visc_request_mem(i8*, i64) #0
+
+declare zeroext i1 @_Z23writeColMajorMatrixFilePKciiRSt6vectorIfSaIfEE(i8*, i32, i32, %"class.std::vector"*) #0
+
+declare void @llvm_visc_untrack_mem(i8*) #0
+
+declare double @pb_GetElapsedTime(%struct.pb_Timer*) #0
+
+declare void @pb_PrintTimerSet(%struct.pb_TimerSet*) #0
+
+declare void @pb_FreeParameters(%struct.pb_Parameters*) #0
+
+; Function Attrs: nounwind
+declare void @llvm.lifetime.end(i64, i8* nocapture) #1
+
+declare %"class.std::basic_ostream"* @_ZNSo9_M_insertIdEERSoT_(%"class.std::basic_ostream"*, double) #0
+
+; Function Attrs: noreturn
+declare void @_ZSt17__throw_bad_allocv() #5
+
+declare noalias i8* @_Znwm(i64) #0
+
+; Function Attrs: nounwind
+declare void @_ZdlPv(i8*) #6
+
+declare %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"*, i8 signext) #0
+
+declare void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"*) #0
+
+; Function Attrs: noreturn
+declare void @_ZSt16__throw_bad_castv() #5
+
+declare %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"*) #0
+
+declare %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"*, i8*, i64) #0
+
+; Function Attrs: nounwind
+define internal void @_GLOBAL__I_a() #1 section ".text.startup" {
+entry:
+ tail call void @_ZNSt8ios_base4InitC1Ev(%"class.std::ios_base::Init"* @_ZStL8__ioinit) #1
+ %0 = tail call i32 @__cxa_atexit(void (i8*)* bitcast (void (%"class.std::ios_base::Init"*)* @_ZNSt8ios_base4InitD1Ev to void (i8*)*), i8* getelementptr inbounds (%"class.std::ios_base::Init"* @_ZStL8__ioinit, i64 0, i32 0), i8* @__dso_handle) #1
+ ret void
+}
+
+; Function Attrs: nounwind
+declare i64 @fwrite(i8* nocapture, i64, i64, %struct._IO_FILE* nocapture) #1
+
+; Function Attrs: nounwind
+declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i32, i1) #1
+
+declare i8* @llvm_visc_initializeTimerSet()
+
+declare void @llvm_visc_switchToTimer(i8**, i32)
+
+declare void @llvm_visc_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.getNode() #1
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.getParentNode(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNodeInstanceID.x(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNumNodeInstances.x(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNodeInstanceID.y(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNumNodeInstances.y(i8*) #1
+
+; Function Attrs: nounwind uwtable
+define %rtype @_Z9mysgemmNTPfiS_iS_iiffInternal_level1(float* in %A, i64 %bytes_A, i32 %lda, float* in %B, i64 %bytes_B, i32 %ldb, float* in out %C, i64 %bytes_C, i32 %ldc, i32 %k, float %alpha, float %beta, i32 %dimX, i32 %dimY) #2 {
+entry:
+ %_Z9mysgemmNTPfiS_iS_iiff.node = call i8* @llvm.visc.createNode2D(i8* bitcast (%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float)* @_Z9mysgemmNTPfiS_iS_iiff to i8*), i32 %dimX, i32 %dimY)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 1, i32 1)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 2, i32 2)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 3, i32 3)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 4, i32 4)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 5, i32 5)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 6, i32 6)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 7, i32 7)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 8, i32 8)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 9, i32 9)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 10, i32 10)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 11, i32 11)
+ ret %rtype undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.createNode2D(i8*, i32, i32) #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.bind.input(i8*, i32, i32) #1
+
+; Function Attrs: nounwind uwtable
+define %rtype @_Z9mysgemmNTPfiS_iS_iiffInternal_level2(float* in %A, i64 %bytes_A, i32 %lda, float* in %B, i64 %bytes_B, i32 %ldb, float* in out %C, i64 %bytes_C, i32 %ldc, i32 %k, float %alpha, float %beta, i32 %dimX, i32 %dimY, i32 %dimX1, i32 %dimY2) #2 {
+entry:
+ %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node = call i8* @llvm.visc.createNode2D(i8* bitcast (%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float, i32, i32)* @_Z9mysgemmNTPfiS_iS_iiffInternal_level1 to i8*), i32 %dimX1, i32 %dimY2)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 1, i32 1)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 2, i32 2)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 3, i32 3)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 4, i32 4)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 5, i32 5)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 6, i32 6)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 7, i32 7)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 8, i32 8)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 9, i32 9)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 10, i32 10)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 11, i32 11)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 12, i32 12)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 13, i32 13)
+ ret %rtype undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.launch(i8*, i8*) #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.wait(i8*) #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.init() #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.cleanup() #1
+
+attributes #0 = { "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { nounwind }
+attributes #2 = { nounwind uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { noinline nounwind uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #4 = { noreturn nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #5 = { noreturn "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #6 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #7 = { noreturn nounwind }
+
+!0 = metadata !{metadata !"float", metadata !1}
+!1 = metadata !{metadata !"omnipotent char", metadata !2}
+!2 = metadata !{metadata !"Simple C/C++ TBAA"}
+!3 = metadata !{metadata !"vtable pointer", metadata !2}
+!4 = metadata !{metadata !"any pointer", metadata !1}
+!5 = metadata !{metadata !"int", metadata !1}
+!6 = metadata !{metadata !"branch_weights", i32 4, i32 64}
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_opt/Makefile b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec/Makefile
similarity index 83%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_opt/Makefile
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_vec/Makefile
index 2234bf54e1..f74ee8921a 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_opt/Makefile
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec/Makefile
@@ -1,8 +1,8 @@
# (c) 2010 The Board of Trustees of the University of Illinois.
-LANGUAGE=hpvm
+LANGUAGE=visc
SRCDIR_OBJS=io.ll #compute_gold.o
-HPVM_OBJS=main.hpvm.ll
+VISC_OBJS=main.visc.ll
APP_CUDALDFLAGS=-lm -lstdc++
APP_CFLAGS=-ffast-math -O3
APP_CXXFLAGS=-ffast-math -O3
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec/io.cc b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec/io.cc
similarity index 100%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec/io.cc
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_vec/io.cc
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec/kernel.cl b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec/kernel.cl
similarity index 100%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec/kernel.cl
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_vec/kernel.cl
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec/main.cc
similarity index 90%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec/main.cc
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_vec/main.cc
index 286297d6fe..76d0cefc81 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec/main.cc
@@ -10,7 +10,6 @@
* Main entry of dense matrix-matrix multiplication kernel
*/
-#include <hpvm.h>
#include <iostream>
#include <malloc.h>
#include <math.h>
@@ -20,6 +19,7 @@
#include <string.h>
#include <sys/time.h>
#include <vector>
+#include <visc.h>
// I/O routines
extern bool readColMajorMatrixFile(const char *fn, int &nr_row, int &nr_col,
@@ -41,8 +41,8 @@ extern char *readFile(const char *);
void mysgemmNT(float *A, int lda, float *B, int ldb, float *C, int ldc, int k,
float alpha, float beta) {
- __hpvm__hint(hpvm::GPU_TARGET);
- __hpvm__attributes(3, A, B, C, 1, C);
+ __visc__hint(visc::GPU_TARGET);
+ __visc__attributes(3, A, B, C, 1, C);
float c = 0.0f;
int m = get_global_id(0);
@@ -99,10 +99,10 @@ __attribute__((noinline)) void basicSgemm(char transa, char transb, int m,
unsigned db[2] = {TILE_SZ / VEC_SZ, TILE_SZ};
unsigned dg[2] = {m / TILE_SZ * db[0], n / TILE_SZ * db[1]};
- unsigned sgemmDFG = __hpvm__node(mysgemmNT, 2, 2, db[0], db[1], dg[0] / db[0],
+ unsigned sgemmDFG = __visc__node(mysgemmNT, 2, 2, db[0], db[1], dg[0] / db[0],
dg[1] / db[1], 12, A, bytesA, lda, B, bytesB,
ldb, C, bytesC, ldc, k, alpha, beta, 0);
- __hpvm__wait(sgemmDFG);
+ __visc__wait(sgemmDFG);
}
int main(int argc, char *argv[]) {
@@ -132,7 +132,7 @@ int main(int argc, char *argv[]) {
readColMajorMatrixFile(params->inpFiles[2], matBcol, matBrow, matBT);
pb_InitializeTimerSet(&timers);
- __hpvm__init();
+ __visc__init();
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
// copy A to device memory
@@ -145,9 +145,9 @@ int main(int argc, char *argv[]) {
// OpenCL memory allocation
std::vector<float> matC(matArow * matBcol);
- llvm_hpvm_track_mem(&matA.front(), A_sz);
- llvm_hpvm_track_mem(&matBT.front(), B_sz);
- llvm_hpvm_track_mem(&matC.front(), C_sz);
+ llvm_visc_track_mem(&matA.front(), A_sz);
+ llvm_visc_track_mem(&matBT.front(), B_sz);
+ llvm_visc_track_mem(&matC.front(), C_sz);
// Copy A and B^T into device memory
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
@@ -162,16 +162,16 @@ int main(int argc, char *argv[]) {
matArow);
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
- llvm_hpvm_request_mem(&matC.front(), C_sz);
+ llvm_visc_request_mem(&matC.front(), C_sz);
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_UNTRACK);
- llvm_hpvm_untrack_mem(&matA.front());
- llvm_hpvm_untrack_mem(&matBT.front());
- llvm_hpvm_untrack_mem(&matC.front());
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_UNTRACK);
+ llvm_visc_untrack_mem(&matA.front());
+ llvm_visc_untrack_mem(&matBT.front());
+ llvm_visc_untrack_mem(&matC.front());
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
pb_PrintTimerSet(&timers);
- __hpvm__cleanup();
+ __visc__cleanup();
if (params->outFile) {
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec/main.visc.ll b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec/main.visc.ll
new file mode 100644
index 0000000000..b6e9e3818e
--- /dev/null
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec/main.visc.ll
@@ -0,0 +1,869 @@
+; ModuleID = 'build/visc_vec_default/main.ll'
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%"class.std::ios_base::Init" = type { i8 }
+%"class.std::basic_ostream" = type { i32 (...)**, %"class.std::basic_ios" }
+%"class.std::basic_ios" = type { %"class.std::ios_base", %"class.std::basic_ostream"*, i8, i8, %"class.std::basic_streambuf"*, %"class.std::ctype"*, %"class.std::num_put"*, %"class.std::num_get"* }
+%"class.std::ios_base" = type { i32 (...)**, i64, i64, i32, i32, i32, %"struct.std::ios_base::_Callback_list"*, %"struct.std::ios_base::_Words", [8 x %"struct.std::ios_base::_Words"], i32, %"struct.std::ios_base::_Words"*, %"class.std::locale" }
+%"struct.std::ios_base::_Callback_list" = type { %"struct.std::ios_base::_Callback_list"*, void (i32, %"class.std::ios_base"*, i32)*, i32, i32 }
+%"struct.std::ios_base::_Words" = type { i8*, i64 }
+%"class.std::locale" = type { %"class.std::locale::_Impl"* }
+%"class.std::locale::_Impl" = type { i32, %"class.std::locale::facet"**, i64, %"class.std::locale::facet"**, i8** }
+%"class.std::locale::facet" = type { i32 (...)**, i32 }
+%"class.std::basic_streambuf" = type { i32 (...)**, i8*, i8*, i8*, i8*, i8*, i8*, %"class.std::locale" }
+%"class.std::ctype" = type { %"class.std::locale::facet", %struct.__locale_struct*, i8, i32*, i32*, i16*, i8, [256 x i8], [256 x i8], i8 }
+%struct.__locale_struct = type { [13 x %struct.__locale_data*], i16*, i32*, i32*, [13 x i8*] }
+%struct.__locale_data = type opaque
+%"class.std::num_put" = type { %"class.std::locale::facet" }
+%"class.std::num_get" = type { %"class.std::locale::facet" }
+%struct._IO_FILE = type { i32, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, %struct._IO_marker*, %struct._IO_FILE*, i32, i32, i64, i16, i8, [1 x i8], i8*, i64, i8*, i8*, i8*, i8*, i64, i32, [20 x i8] }
+%struct._IO_marker = type { %struct._IO_marker*, %struct._IO_FILE*, i32 }
+%rtype = type {}
+%struct.arg = type <{ float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float, i32, i32, i32, i32, %rtype }>
+%struct.pb_TimerSet = type { i32, %struct.pb_async_time_marker_list*, i64, i64, [24 x %struct.pb_Timer], [24 x %struct.pb_SubTimerList*] }
+%struct.pb_async_time_marker_list = type { i8*, i32, i8*, %struct.pb_async_time_marker_list* }
+%struct.pb_Timer = type { i32, i64, i64 }
+%struct.pb_SubTimerList = type { %struct.pb_SubTimer*, %struct.pb_SubTimer* }
+%struct.pb_SubTimer = type { i8*, %struct.pb_Timer, %struct.pb_SubTimer* }
+%"class.std::vector" = type { %"struct.std::_Vector_base" }
+%"struct.std::_Vector_base" = type { %"struct.std::_Vector_base<float, std::allocator<float> >::_Vector_impl" }
+%"struct.std::_Vector_base<float, std::allocator<float> >::_Vector_impl" = type { float*, float*, float* }
+%struct.pb_Parameters = type { i8*, i8** }
+
+@_ZStL8__ioinit = internal global %"class.std::ios_base::Init" zeroinitializer, align 1
+@__dso_handle = external global i8
+@_ZSt4cerr = external global %"class.std::basic_ostream"
+@.str = private unnamed_addr constant [48 x i8] c"unsupported value of 'transa' in regtileSgemm()\00", align 1
+@.str1 = private unnamed_addr constant [48 x i8] c"unsupported value of 'transb' in regtileSgemm()\00", align 1
+@.str2 = private unnamed_addr constant [53 x i8] c"unsupported size of matrix. m should be multiple of \00", align 1
+@.str3 = private unnamed_addr constant [27 x i8] c"; n should be multiple of \00", align 1
+@stderr = external global %struct._IO_FILE*
+@.str4 = private unnamed_addr constant [33 x i8] c"Expecting three input filenames\0A\00", align 1
+@_ZSt4cout = external global %"class.std::basic_ostream"
+@.str5 = private unnamed_addr constant [10 x i8] c"GFLOPs = \00", align 1
+@llvm.global_ctors = appending global [1 x { i32, void ()* }] [{ i32, void ()* } { i32 65535, void ()* @_GLOBAL__I_a }]
+@viscTimerSet_GenVISC = common global i8* null
+@0 = internal constant [14 x i8] c"GenVISC_Timer\00"
+
+declare void @_ZNSt8ios_base4InitC1Ev(%"class.std::ios_base::Init"*) #0
+
+declare void @_ZNSt8ios_base4InitD1Ev(%"class.std::ios_base::Init"*) #0
+
+; Function Attrs: nounwind
+declare i32 @__cxa_atexit(void (i8*)*, i8*, i8*) #1
+
+; Function Attrs: nounwind readnone
+declare <8 x float> @llvm.fmuladd.v8f32(<8 x float>, <8 x float>, <8 x float>) #1
+
+; Function Attrs: nounwind uwtable
+define %rtype @_Z9mysgemmNTPfiS_iS_iiff(float* in %A, i64 %bytes_A, i32 %lda, float* in %B, i64 %bytes_B, i32 %ldb, float* in out %C, i64 %bytes_C, i32 %ldc, i32 %k, float %alpha, float %beta) #2 {
+ %_Z9mysgemmNTPfiS_iS_iiff.node = call i8* @llvm.visc.getNode()
+ %_Z9mysgemmNTPfiS_iS_iiff.parentNode = call i8* @llvm.visc.getParentNode(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ %a0 = call i32 @llvm.visc.getNodeInstanceID.x(i8* %_Z9mysgemmNTPfiS_iS_iiff.parentNode)
+ %1 = call i32 @llvm.visc.getNumNodeInstances.x(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ %2 = mul i32 %a0, %1
+ %3 = call i32 @llvm.visc.getNodeInstanceID.x(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ %4 = add i32 %2, %3
+ %5 = call i32 @llvm.visc.getNodeInstanceID.y(i8* %_Z9mysgemmNTPfiS_iS_iiff.parentNode)
+ %a6 = call i32 @llvm.visc.getNumNodeInstances.y(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ %a7 = mul i32 %5, %a6
+ %a8 = call i32 @llvm.visc.getNodeInstanceID.y(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ %a9 = add i32 %a7, %a8
+
+ %a10 = shl i32 %4, 3
+
+
+ ;a10 = %3, a9 = %5
+ ;%1 = tail call i64 @_Z13get_global_idj(i32 0) #1
+ ;%2 = shl i64 %1, 3
+ ;%3 = trunc i64 %2 to i32
+ ;%4 = tail call i64 @_Z13get_global_idj(i32 1) #1
+ ;%5 = trunc i64 %4 to i32
+
+
+
+
+ %6 = icmp sgt i32 %k, 0
+ br i1 %6, label %.lr.ph, label %._crit_edge
+
+.lr.ph: ; preds = %.lr.ph, %0
+ %cp.021 = phi <8 x float> [ %20, %.lr.ph ], [ zeroinitializer, %0 ]
+ %i.020 = phi i32 [ %21, %.lr.ph ], [ 0, %0 ]
+ %7 = mul nsw i32 %i.020, %lda
+ %8 = add nsw i32 %7, %a10
+ %9 = sext i32 %8 to i64
+ %10 = getelementptr inbounds float* %A, i64 %9
+ %v10 = bitcast float* %10 to <8 x float>*
+ %11 = load <8 x float>* %v10
+ %12 = mul nsw i32 %i.020, %ldb
+ %13 = add nsw i32 %12, %a9
+ %14 = sext i32 %13 to i64
+ %15 = getelementptr inbounds float* %B, i64 %14
+ %16 = load float* %15, align 4, !tbaa !9
+ %17 = insertelement <8 x float> undef, float %16, i32 0
+ %18 = shufflevector <8 x float> %17, <8 x float> undef, <8 x i32> zeroinitializer
+ %19 = fmul <8 x float> %11, %18
+ %20 = fadd <8 x float> %cp.021, %19
+ %21 = add nsw i32 %i.020, 1
+ %22 = icmp slt i32 %21, %k
+ br i1 %22, label %.lr.ph, label %._crit_edge
+
+._crit_edge: ; preds = %.lr.ph, %0
+ %cp.0.lcssa = phi <8 x float> [ zeroinitializer, %0 ], [ %20, %.lr.ph ]
+ %23 = mul nsw i32 %a9, %ldc
+ %24 = add nsw i32 %23, %a10
+ %25 = sext i32 %24 to i64
+ %26 = getelementptr inbounds float* %C, i64 %25
+ %v26 = bitcast float* %26 to <8 x float>*
+ %27 = load <8 x float>* %v26
+ %28 = insertelement <8 x float> undef, float %beta, i32 0
+ %29 = shufflevector <8 x float> %28, <8 x float> undef, <8 x i32> zeroinitializer
+ %30 = insertelement <8 x float> undef, float %alpha, i32 0
+ %31 = shufflevector <8 x float> %30, <8 x float> undef, <8 x i32> zeroinitializer
+ %32 = fmul <8 x float> %31, %cp.0.lcssa
+
+ ;%33 = tail call <8 x float> @llvm.fmuladd.v8f32(<8 x float> %27, <8 x float> %29, <8 x float> %32)
+
+ %mul = fmul <8 x float> %27, %29
+ %33 = fadd <8 x float> %mul, %32
+ store <8 x float> %33, <8 x float>* %v26
+
+;entry:
+ ;%_Z9mysgemmNTPfiS_iS_iiff.node = call i8* @llvm.visc.getNode()
+ ;%_Z9mysgemmNTPfiS_iS_iiff.parentNode = call i8* @llvm.visc.getParentNode(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ ;%0 = call i32 @llvm.visc.getNodeInstanceID.x(i8* %_Z9mysgemmNTPfiS_iS_iiff.parentNode)
+ ;%1 = call i32 @llvm.visc.getNumNodeInstances.x(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ ;%2 = mul i32 %0, %1
+ ;%3 = call i32 @llvm.visc.getNodeInstanceID.x(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ ;%4 = add i32 %2, %3
+ ;%5 = call i32 @llvm.visc.getNodeInstanceID.y(i8* %_Z9mysgemmNTPfiS_iS_iiff.parentNode)
+ ;%6 = call i32 @llvm.visc.getNumNodeInstances.y(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ ;%7 = mul i32 %5, %6
+ ;%8 = call i32 @llvm.visc.getNodeInstanceID.y(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+ ;%9 = add i32 %7, %8
+ ;%cmp32 = icmp sgt i32 %k, 0
+ ;br i1 %cmp32, label %for.body, label %for.end
+
+;for.body: ; preds = %for.body, %entry
+ ;%indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
+ ;%c.034 = phi float [ %add7, %for.body ], [ 0.000000e+00, %entry ]
+ ;%10 = trunc i64 %indvars.iv to i32
+ ;%mul = mul nsw i32 %10, %lda
+ ;%add = add nsw i32 %mul, %4
+ ;%idxprom = sext i32 %add to i64
+ ;%arrayidx = getelementptr inbounds float* %A, i64 %idxprom
+ ;%11 = load float* %arrayidx, align 4, !tbaa !3
+ ;%mul2 = mul nsw i32 %10, %ldb
+ ;%add3 = add nsw i32 %mul2, %9
+ ;%idxprom4 = sext i32 %add3 to i64
+ ;%arrayidx5 = getelementptr inbounds float* %B, i64 %idxprom4
+ ;%12 = load float* %arrayidx5, align 4, !tbaa !3
+ ;%mul6 = fmul fast float %11, %12
+ ;%add7 = fadd fast float %c.034, %mul6
+ ;%indvars.iv.next = add i64 %indvars.iv, 1
+ ;%lftr.wideiv = trunc i64 %indvars.iv.next to i32
+ ;%exitcond = icmp eq i32 %lftr.wideiv, %k
+ ;br i1 %exitcond, label %for.end, label %for.body
+
+;for.end: ; preds = %for.body, %entry
+ ;%c.0.lcssa = phi float [ 0.000000e+00, %entry ], [ %add7, %for.body ]
+ ;%mul8 = mul nsw i32 %9, %ldc
+ ;%add9 = add nsw i32 %mul8, %4
+ ;%idxprom10 = sext i32 %add9 to i64
+ ;%arrayidx11 = getelementptr inbounds float* %C, i64 %idxprom10
+ ;%13 = load float* %arrayidx11, align 4, !tbaa !3
+ ;%mul12 = fmul fast float %13, %beta
+ ;%mul13 = fmul fast float %c.0.lcssa, %alpha
+ ;%add14 = fadd fast float %mul13, %mul12
+ ;store float %add14, float* %arrayidx11, align 4, !tbaa !3
+ ret %rtype undef
+
+
+}
+
+; Function Attrs: noinline nounwind uwtable
+define void @_Z10basicSgemmcciiifPfmiS_mifS_mi(i8 signext %transa, i8 signext %transb, i32 %m, i32 %n, i32 %k, float %alpha, float* %A, i64 %bytesA, i32 %lda, float* %B, i64 %bytesB, i32 %ldb, float %beta, float* %C, i64 %bytesC, i32 %ldc) #3 {
+entry:
+ switch i8 %transa, label %if.then [
+ i8 78, label %if.end
+ i8 110, label %if.end
+ ]
+
+if.then: ; preds = %entry
+ %call1.i = tail call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* @_ZSt4cerr, i8* getelementptr inbounds ([48 x i8]* @.str, i64 0, i64 0), i64 47) #1
+ %vtable.i = load i8** bitcast (%"class.std::basic_ostream"* @_ZSt4cerr to i8**), align 8, !tbaa !6
+ %vbase.offset.ptr.i = getelementptr i8* %vtable.i, i64 -24
+ %0 = bitcast i8* %vbase.offset.ptr.i to i64*
+ %vbase.offset.i = load i64* %0, align 8
+ %add.ptr.i.sum = add i64 %vbase.offset.i, 240
+ %_M_ctype.i = getelementptr inbounds i8* bitcast (%"class.std::basic_ostream"* @_ZSt4cerr to i8*), i64 %add.ptr.i.sum
+ %1 = bitcast i8* %_M_ctype.i to %"class.std::ctype"**
+ %2 = load %"class.std::ctype"** %1, align 8, !tbaa !7
+ %tobool.i97 = icmp eq %"class.std::ctype"* %2, null
+ br i1 %tobool.i97, label %if.then.i98, label %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit
+
+if.then.i98: ; preds = %if.then
+ tail call void @_ZSt16__throw_bad_castv() #7
+ unreachable
+
+_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit: ; preds = %if.then
+ %_M_widen_ok.i = getelementptr inbounds %"class.std::ctype"* %2, i64 0, i32 6
+ %3 = load i8* %_M_widen_ok.i, align 1, !tbaa !4
+ %tobool.i = icmp eq i8 %3, 0
+ br i1 %tobool.i, label %if.end.i, label %if.then.i
+
+if.then.i: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit
+ %arrayidx.i = getelementptr inbounds %"class.std::ctype"* %2, i64 0, i32 7, i64 10
+ %4 = load i8* %arrayidx.i, align 1, !tbaa !4
+ br label %_ZNKSt5ctypeIcE5widenEc.exit
+
+if.end.i: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit
+ tail call void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"* %2) #1
+ %5 = bitcast %"class.std::ctype"* %2 to i8 (%"class.std::ctype"*, i8)***
+ %vtable.i71 = load i8 (%"class.std::ctype"*, i8)*** %5, align 8, !tbaa !6
+ %vfn.i = getelementptr inbounds i8 (%"class.std::ctype"*, i8)** %vtable.i71, i64 6
+ %6 = load i8 (%"class.std::ctype"*, i8)** %vfn.i, align 8
+ %call.i72 = tail call signext i8 %6(%"class.std::ctype"* %2, i8 signext 10) #1
+ br label %_ZNKSt5ctypeIcE5widenEc.exit
+
+_ZNKSt5ctypeIcE5widenEc.exit: ; preds = %if.end.i, %if.then.i
+ %retval.0.i = phi i8 [ %4, %if.then.i ], [ %call.i72, %if.end.i ]
+ %call1.i47 = tail call %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"* @_ZSt4cerr, i8 signext %retval.0.i) #1
+ %call.i = tail call %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"* %call1.i47) #1
+ br label %return
+
+if.end: ; preds = %entry, %entry
+ switch i8 %transb, label %if.then9 [
+ i8 84, label %if.end12
+ i8 116, label %if.end12
+ ]
+
+if.then9: ; preds = %if.end
+ %call1.i49 = tail call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* @_ZSt4cerr, i8* getelementptr inbounds ([48 x i8]* @.str1, i64 0, i64 0), i64 47) #1
+ %vtable.i51 = load i8** bitcast (%"class.std::basic_ostream"* @_ZSt4cerr to i8**), align 8, !tbaa !6
+ %vbase.offset.ptr.i52 = getelementptr i8* %vtable.i51, i64 -24
+ %7 = bitcast i8* %vbase.offset.ptr.i52 to i64*
+ %vbase.offset.i53 = load i64* %7, align 8
+ %add.ptr.i54.sum = add i64 %vbase.offset.i53, 240
+ %_M_ctype.i73 = getelementptr inbounds i8* bitcast (%"class.std::basic_ostream"* @_ZSt4cerr to i8*), i64 %add.ptr.i54.sum
+ %8 = bitcast i8* %_M_ctype.i73 to %"class.std::ctype"**
+ %9 = load %"class.std::ctype"** %8, align 8, !tbaa !7
+ %tobool.i100 = icmp eq %"class.std::ctype"* %9, null
+ br i1 %tobool.i100, label %if.then.i101, label %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit103
+
+if.then.i101: ; preds = %if.then9
+ tail call void @_ZSt16__throw_bad_castv() #7
+ unreachable
+
+_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit103: ; preds = %if.then9
+ %_M_widen_ok.i75 = getelementptr inbounds %"class.std::ctype"* %9, i64 0, i32 6
+ %10 = load i8* %_M_widen_ok.i75, align 1, !tbaa !4
+ %tobool.i76 = icmp eq i8 %10, 0
+ br i1 %tobool.i76, label %if.end.i82, label %if.then.i78
+
+if.then.i78: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit103
+ %arrayidx.i77 = getelementptr inbounds %"class.std::ctype"* %9, i64 0, i32 7, i64 10
+ %11 = load i8* %arrayidx.i77, align 1, !tbaa !4
+ br label %_ZNKSt5ctypeIcE5widenEc.exit84
+
+if.end.i82: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit103
+ tail call void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"* %9) #1
+ %12 = bitcast %"class.std::ctype"* %9 to i8 (%"class.std::ctype"*, i8)***
+ %vtable.i79 = load i8 (%"class.std::ctype"*, i8)*** %12, align 8, !tbaa !6
+ %vfn.i80 = getelementptr inbounds i8 (%"class.std::ctype"*, i8)** %vtable.i79, i64 6
+ %13 = load i8 (%"class.std::ctype"*, i8)** %vfn.i80, align 8
+ %call.i81 = tail call signext i8 %13(%"class.std::ctype"* %9, i8 signext 10) #1
+ br label %_ZNKSt5ctypeIcE5widenEc.exit84
+
+_ZNKSt5ctypeIcE5widenEc.exit84: ; preds = %if.end.i82, %if.then.i78
+ %retval.0.i83 = phi i8 [ %11, %if.then.i78 ], [ %call.i81, %if.end.i82 ]
+ %call1.i56 = tail call %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"* @_ZSt4cerr, i8 signext %retval.0.i83) #1
+ %call.i57 = tail call %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"* %call1.i56) #1
+ br label %return
+
+if.end12: ; preds = %if.end, %if.end
+ %rem44 = and i32 %m, 15
+ %tobool = icmp eq i32 %rem44, 0
+ br i1 %tobool, label %lor.lhs.false, label %if.then15
+
+lor.lhs.false: ; preds = %if.end12
+ %rem1345 = and i32 %n, 15
+ %tobool14 = icmp eq i32 %rem1345, 0
+ br i1 %tobool14, label %if.end21, label %if.then15
+
+if.then15: ; preds = %lor.lhs.false, %if.end12
+ %call1.i59 = tail call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* @_ZSt4cerr, i8* getelementptr inbounds ([53 x i8]* @.str2, i64 0, i64 0), i64 52) #1
+ %call17 = tail call %"class.std::basic_ostream"* @_ZNSolsEi(%"class.std::basic_ostream"* @_ZSt4cerr, i32 16) #1
+ %call1.i61 = tail call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* %call17, i8* getelementptr inbounds ([27 x i8]* @.str3, i64 0, i64 0), i64 26) #1
+ %call19 = tail call %"class.std::basic_ostream"* @_ZNSolsEi(%"class.std::basic_ostream"* %call17, i32 16) #1
+ %14 = bitcast %"class.std::basic_ostream"* %call19 to i8**
+ %vtable.i63 = load i8** %14, align 8, !tbaa !6
+ %vbase.offset.ptr.i64 = getelementptr i8* %vtable.i63, i64 -24
+ %15 = bitcast i8* %vbase.offset.ptr.i64 to i64*
+ %vbase.offset.i65 = load i64* %15, align 8
+ %16 = bitcast %"class.std::basic_ostream"* %call19 to i8*
+ %add.ptr.i66.sum = add i64 %vbase.offset.i65, 240
+ %_M_ctype.i85 = getelementptr inbounds i8* %16, i64 %add.ptr.i66.sum
+ %17 = bitcast i8* %_M_ctype.i85 to %"class.std::ctype"**
+ %18 = load %"class.std::ctype"** %17, align 8, !tbaa !7
+ %tobool.i104 = icmp eq %"class.std::ctype"* %18, null
+ br i1 %tobool.i104, label %if.then.i105, label %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit107
+
+if.then.i105: ; preds = %if.then15
+ tail call void @_ZSt16__throw_bad_castv() #7
+ unreachable
+
+_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit107: ; preds = %if.then15
+ %_M_widen_ok.i87 = getelementptr inbounds %"class.std::ctype"* %18, i64 0, i32 6
+ %19 = load i8* %_M_widen_ok.i87, align 1, !tbaa !4
+ %tobool.i88 = icmp eq i8 %19, 0
+ br i1 %tobool.i88, label %if.end.i94, label %if.then.i90
+
+if.then.i90: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit107
+ %arrayidx.i89 = getelementptr inbounds %"class.std::ctype"* %18, i64 0, i32 7, i64 10
+ %20 = load i8* %arrayidx.i89, align 1, !tbaa !4
+ br label %_ZNKSt5ctypeIcE5widenEc.exit96
+
+if.end.i94: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit107
+ tail call void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"* %18) #1
+ %21 = bitcast %"class.std::ctype"* %18 to i8 (%"class.std::ctype"*, i8)***
+ %vtable.i91 = load i8 (%"class.std::ctype"*, i8)*** %21, align 8, !tbaa !6
+ %vfn.i92 = getelementptr inbounds i8 (%"class.std::ctype"*, i8)** %vtable.i91, i64 6
+ %22 = load i8 (%"class.std::ctype"*, i8)** %vfn.i92, align 8
+ %call.i93 = tail call signext i8 %22(%"class.std::ctype"* %18, i8 signext 10) #1
+ br label %_ZNKSt5ctypeIcE5widenEc.exit96
+
+_ZNKSt5ctypeIcE5widenEc.exit96: ; preds = %if.end.i94, %if.then.i90
+ %retval.0.i95 = phi i8 [ %20, %if.then.i90 ], [ %call.i93, %if.end.i94 ]
+ %call1.i68 = tail call %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"* %call19, i8 signext %retval.0.i95) #1
+ %call.i69 = tail call %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"* %call1.i68) #1
+ br label %if.end21
+
+if.end21: ; preds = %_ZNKSt5ctypeIcE5widenEc.exit96, %lor.lhs.false
+ %div = sdiv i32 %m, 16
+ %mul = and i32 %div, 2147483647
+ %div22 = sdiv i32 %n, 16
+ %mul24 = and i32 %div22, 268435455
+ %conv33 = fpext float %alpha to double
+ %conv34 = fpext float %beta to double
+ call void @llvm_visc_switchToTimer(i8** @viscTimerSet_GenVISC, i32 19)
+ %in.addr = alloca %struct.arg
+ %in.addr.A = getelementptr %struct.arg* %in.addr, i32 0, i32 0
+ store float* %A, float** %in.addr.A
+ %in.addr.bytes_A = getelementptr %struct.arg* %in.addr, i32 0, i32 1
+ store i64 %bytesA, i64* %in.addr.bytes_A
+ %in.addr.lda = getelementptr %struct.arg* %in.addr, i32 0, i32 2
+ store i32 %lda, i32* %in.addr.lda
+ %in.addr.B = getelementptr %struct.arg* %in.addr, i32 0, i32 3
+ store float* %B, float** %in.addr.B
+ %in.addr.bytes_B = getelementptr %struct.arg* %in.addr, i32 0, i32 4
+ store i64 %bytesB, i64* %in.addr.bytes_B
+ %in.addr.ldb = getelementptr %struct.arg* %in.addr, i32 0, i32 5
+ store i32 %ldb, i32* %in.addr.ldb
+ %in.addr.C = getelementptr %struct.arg* %in.addr, i32 0, i32 6
+ store float* %C, float** %in.addr.C
+ %in.addr.bytes_C = getelementptr %struct.arg* %in.addr, i32 0, i32 7
+ store i64 %bytesC, i64* %in.addr.bytes_C
+ %in.addr.ldc = getelementptr %struct.arg* %in.addr, i32 0, i32 8
+ store i32 %ldc, i32* %in.addr.ldc
+ %in.addr.k = getelementptr %struct.arg* %in.addr, i32 0, i32 9
+ store i32 %k, i32* %in.addr.k
+ %in.addr.alpha = getelementptr %struct.arg* %in.addr, i32 0, i32 10
+ %in.addr.alpha.cast = fptrunc double %conv33 to float
+ store float %in.addr.alpha.cast, float* %in.addr.alpha
+ %in.addr.beta = getelementptr %struct.arg* %in.addr, i32 0, i32 11
+ %in.addr.beta.cast = fptrunc double %conv34 to float
+ store float %in.addr.beta.cast, float* %in.addr.beta
+ %in.addr.dimX0 = getelementptr %struct.arg* %in.addr, i32 0, i32 12
+ store i32 2, i32* %in.addr.dimX0
+ %in.addr.dimY0 = getelementptr %struct.arg* %in.addr, i32 0, i32 13
+ store i32 16, i32* %in.addr.dimY0
+ %in.addr.dimX1 = getelementptr %struct.arg* %in.addr, i32 0, i32 14
+ store i32 %mul, i32* %in.addr.dimX1
+ %in.addr.dimY1 = getelementptr %struct.arg* %in.addr, i32 0, i32 15
+ store i32 %mul24, i32* %in.addr.dimY1
+ %args = bitcast %struct.arg* %in.addr to i8*
+ call void @llvm_visc_switchToTimer(i8** @viscTimerSet_GenVISC, i32 0)
+ %graph_Z9mysgemmNTPfiS_iS_iiffInternal_level2 = call i8* @llvm.visc.launch(i8* bitcast (%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float, i32, i32, i32, i32)* @_Z9mysgemmNTPfiS_iS_iiffInternal_level2 to i8*), i8* %args)
+ call void @llvm.visc.wait(i8* %graph_Z9mysgemmNTPfiS_iS_iiffInternal_level2)
+ br label %return
+
+return: ; preds = %if.end21, %_ZNKSt5ctypeIcE5widenEc.exit84, %_ZNKSt5ctypeIcE5widenEc.exit
+ ret void
+}
+
+declare %"class.std::basic_ostream"* @_ZNSolsEi(%"class.std::basic_ostream"*, i32) #0
+
+; Function Attrs: nounwind uwtable
+define i32 @main(i32 %argc, i8** %argv) #2 {
+entry:
+ %argc.addr = alloca i32, align 4
+ %timers = alloca %struct.pb_TimerSet, align 8
+ %matArow = alloca i32, align 4
+ %matAcol = alloca i32, align 4
+ %matBrow = alloca i32, align 4
+ %matBcol = alloca i32, align 4
+ %matA = alloca %"class.std::vector", align 8
+ %matBT = alloca %"class.std::vector", align 8
+ %matC = alloca %"class.std::vector", align 8
+ store i32 %argc, i32* %argc.addr, align 4, !tbaa !8
+ %0 = bitcast %struct.pb_TimerSet* %timers to i8*
+ call void @llvm.lifetime.start(i64 800, i8* %0) #1
+ %1 = bitcast %"class.std::vector"* %matA to i8*
+ call void @llvm.memset.p0i8.i64(i8* %1, i8 0, i64 24, i32 8, i1 false) #1
+ %2 = bitcast %"class.std::vector"* %matBT to i8*
+ call void @llvm.memset.p0i8.i64(i8* %2, i8 0, i64 24, i32 8, i1 false) #1
+ %call = call %struct.pb_Parameters* @pb_ReadParameters(i32* %argc.addr, i8** %argv) #1
+ %inpFiles = getelementptr inbounds %struct.pb_Parameters* %call, i64 0, i32 1
+ %3 = load i8*** %inpFiles, align 8, !tbaa !7
+ %4 = load i8** %3, align 8, !tbaa !7
+ %cmp = icmp eq i8* %4, null
+ br i1 %cmp, label %if.then, label %lor.lhs.false
+
+lor.lhs.false: ; preds = %entry
+ %arrayidx2 = getelementptr inbounds i8** %3, i64 1
+ %5 = load i8** %arrayidx2, align 8, !tbaa !7
+ %cmp3 = icmp eq i8* %5, null
+ br i1 %cmp3, label %if.then, label %lor.lhs.false4
+
+lor.lhs.false4: ; preds = %lor.lhs.false
+ %arrayidx6 = getelementptr inbounds i8** %3, i64 2
+ %6 = load i8** %arrayidx6, align 8, !tbaa !7
+ %cmp7 = icmp eq i8* %6, null
+ br i1 %cmp7, label %if.then, label %lor.lhs.false8
+
+lor.lhs.false8: ; preds = %lor.lhs.false4
+ %arrayidx10 = getelementptr inbounds i8** %3, i64 3
+ %7 = load i8** %arrayidx10, align 8, !tbaa !7
+ %cmp11 = icmp eq i8* %7, null
+ br i1 %cmp11, label %if.end, label %if.then
+
+if.then: ; preds = %lor.lhs.false8, %lor.lhs.false4, %lor.lhs.false, %entry
+ %8 = load %struct._IO_FILE** @stderr, align 8, !tbaa !7
+ %9 = call i64 @fwrite(i8* getelementptr inbounds ([33 x i8]* @.str4, i64 0, i64 0), i64 32, i64 1, %struct._IO_FILE* %8)
+ call void @exit(i32 -1) #7
+ unreachable
+
+if.end: ; preds = %lor.lhs.false8
+ %call15 = call zeroext i1 @_Z22readColMajorMatrixFilePKcRiS1_RSt6vectorIfSaIfEE(i8* %4, i32* %matArow, i32* %matAcol, %"class.std::vector"* %matA) #1
+ %10 = load i8*** %inpFiles, align 8, !tbaa !7
+ %arrayidx17 = getelementptr inbounds i8** %10, i64 2
+ %11 = load i8** %arrayidx17, align 8, !tbaa !7
+ %call18 = call zeroext i1 @_Z22readColMajorMatrixFilePKcRiS1_RSt6vectorIfSaIfEE(i8* %11, i32* %matBcol, i32* %matBrow, %"class.std::vector"* %matBT) #1
+ call void @pb_InitializeTimerSet(%struct.pb_TimerSet* %timers) #1
+ %12 = call i8* @llvm_visc_initializeTimerSet()
+ store i8* %12, i8** @viscTimerSet_GenVISC
+ call void @llvm_visc_switchToTimer(i8** @viscTimerSet_GenVISC, i32 0)
+ call void @llvm.visc.init()
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 6) #1
+ %13 = load i32* %matArow, align 4, !tbaa !8
+ %14 = load i32* %matAcol, align 4, !tbaa !8
+ %mul = mul nsw i32 %14, %13
+ %conv = sext i32 %mul to i64
+ %mul19 = shl nsw i64 %conv, 2
+ %15 = load i32* %matBrow, align 4, !tbaa !8
+ %16 = load i32* %matBcol, align 4, !tbaa !8
+ %mul20 = mul nsw i32 %16, %15
+ %conv21 = sext i32 %mul20 to i64
+ %mul22 = shl nsw i64 %conv21, 2
+ %mul23 = mul nsw i32 %16, %13
+ %conv24 = sext i32 %mul23 to i64
+ %mul25 = shl nsw i64 %conv24, 2
+ %17 = bitcast %"class.std::vector"* %matC to i8*
+ call void @llvm.memset.p0i8.i64(i8* %17, i8 0, i64 24, i32 8, i1 false) #1
+ %cmp.i.i.i.i = icmp eq i32 %mul23, 0
+ br i1 %cmp.i.i.i.i, label %_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i, label %cond.true.i.i.i.i
+
+cond.true.i.i.i.i: ; preds = %if.end
+ %cmp.i.i.i.i.i = icmp slt i32 %mul23, 0
+ br i1 %cmp.i.i.i.i.i, label %if.then.i.i.i.i.i, label %_ZN9__gnu_cxx13new_allocatorIfE8allocateEmPKv.exit.i.i.i.i, !prof !9
+
+if.then.i.i.i.i.i: ; preds = %cond.true.i.i.i.i
+ call void @_ZSt17__throw_bad_allocv() #7
+ unreachable
+
+_ZN9__gnu_cxx13new_allocatorIfE8allocateEmPKv.exit.i.i.i.i: ; preds = %cond.true.i.i.i.i
+ %call2.i.i.i.i.i = call noalias i8* @_Znwm(i64 %mul25) #1
+ %18 = bitcast i8* %call2.i.i.i.i.i to float*
+ br label %_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i
+
+_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i: ; preds = %_ZN9__gnu_cxx13new_allocatorIfE8allocateEmPKv.exit.i.i.i.i, %if.end
+ %cond.i.i.i.i = phi float* [ %18, %_ZN9__gnu_cxx13new_allocatorIfE8allocateEmPKv.exit.i.i.i.i ], [ null, %if.end ]
+ %_M_start.i.i.i81 = getelementptr inbounds %"class.std::vector"* %matC, i64 0, i32 0, i32 0, i32 0
+ store float* %cond.i.i.i.i, float** %_M_start.i.i.i81, align 8, !tbaa !7
+ %_M_finish.i.i.i = getelementptr inbounds %"class.std::vector"* %matC, i64 0, i32 0, i32 0, i32 1
+ store float* %cond.i.i.i.i, float** %_M_finish.i.i.i, align 8, !tbaa !7
+ %add.ptr.i.i.i = getelementptr inbounds float* %cond.i.i.i.i, i64 %conv24
+ %_M_end_of_storage.i.i.i = getelementptr inbounds %"class.std::vector"* %matC, i64 0, i32 0, i32 0, i32 2
+ store float* %add.ptr.i.i.i, float** %_M_end_of_storage.i.i.i, align 8, !tbaa !7
+ br i1 %cmp.i.i.i.i, label %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit, label %for.body.lr.ph.i.i.i.i.i.i.i.i
+
+for.body.lr.ph.i.i.i.i.i.i.i.i: ; preds = %_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i
+ %n.mod.vf.i.i.i.i.i.i.i.i = and i64 %conv24, 7
+ %n.vec.i.i.i.i.i.i.i.i = sub i64 %conv24, %n.mod.vf.i.i.i.i.i.i.i.i
+ %cmp.zero.i.i.i.i.i.i.i.i = icmp eq i64 %n.mod.vf.i.i.i.i.i.i.i.i, %conv24
+ %ptr.ind.end.i.i.i.i.i.i.i.i = getelementptr float* %cond.i.i.i.i, i64 %n.vec.i.i.i.i.i.i.i.i
+ br i1 %cmp.zero.i.i.i.i.i.i.i.i, label %middle.block.i.i.i.i.i.i.i.i, label %vector.body.i.i.i.i.i.i.i.i
+
+vector.body.i.i.i.i.i.i.i.i: ; preds = %vector.body.i.i.i.i.i.i.i.i, %for.body.lr.ph.i.i.i.i.i.i.i.i
+ %index.i.i.i.i.i.i.i.i = phi i64 [ %index.next.i.i.i.i.i.i.i.i, %vector.body.i.i.i.i.i.i.i.i ], [ 0, %for.body.lr.ph.i.i.i.i.i.i.i.i ]
+ %next.gep.i.i.i.i.i.i.i.i = getelementptr float* %cond.i.i.i.i, i64 %index.i.i.i.i.i.i.i.i
+ %19 = bitcast float* %next.gep.i.i.i.i.i.i.i.i to <4 x float>*
+ store <4 x float> zeroinitializer, <4 x float>* %19, align 4
+ %next.gep.sum41.i.i.i.i.i.i.i.i = or i64 %index.i.i.i.i.i.i.i.i, 4
+ %20 = getelementptr float* %cond.i.i.i.i, i64 %next.gep.sum41.i.i.i.i.i.i.i.i
+ %21 = bitcast float* %20 to <4 x float>*
+ store <4 x float> zeroinitializer, <4 x float>* %21, align 4
+ %index.next.i.i.i.i.i.i.i.i = add i64 %index.i.i.i.i.i.i.i.i, 8
+ %22 = icmp eq i64 %index.next.i.i.i.i.i.i.i.i, %n.vec.i.i.i.i.i.i.i.i
+ br i1 %22, label %middle.block.i.i.i.i.i.i.i.i, label %vector.body.i.i.i.i.i.i.i.i
+
+middle.block.i.i.i.i.i.i.i.i: ; preds = %vector.body.i.i.i.i.i.i.i.i, %for.body.lr.ph.i.i.i.i.i.i.i.i
+ %resume.val.i.i.i.i.i.i.i.i = phi float* [ %cond.i.i.i.i, %for.body.lr.ph.i.i.i.i.i.i.i.i ], [ %ptr.ind.end.i.i.i.i.i.i.i.i, %vector.body.i.i.i.i.i.i.i.i ]
+ %resume.val7.i.i.i.i.i.i.i.i = phi i64 [ %conv24, %for.body.lr.ph.i.i.i.i.i.i.i.i ], [ %n.mod.vf.i.i.i.i.i.i.i.i, %vector.body.i.i.i.i.i.i.i.i ]
+ %new.indc.resume.val.i.i.i.i.i.i.i.i = phi i64 [ 0, %for.body.lr.ph.i.i.i.i.i.i.i.i ], [ %n.vec.i.i.i.i.i.i.i.i, %vector.body.i.i.i.i.i.i.i.i ]
+ %cmp.n.i.i.i.i.i.i.i.i = icmp eq i64 %new.indc.resume.val.i.i.i.i.i.i.i.i, %conv24
+ br i1 %cmp.n.i.i.i.i.i.i.i.i, label %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit, label %for.body.i.i.i.i.i.i.i.i.preheader
+
+for.body.i.i.i.i.i.i.i.i.preheader: ; preds = %middle.block.i.i.i.i.i.i.i.i
+ %resume.val.i.i.i.i.i.i.i.i101 = bitcast float* %resume.val.i.i.i.i.i.i.i.i to i8*
+ %23 = shl nsw i64 %resume.val7.i.i.i.i.i.i.i.i, 2
+ call void @llvm.memset.p0i8.i64(i8* %resume.val.i.i.i.i.i.i.i.i101, i8 0, i64 %23, i32 4, i1 false)
+ br label %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit
+
+_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit: ; preds = %for.body.i.i.i.i.i.i.i.i.preheader, %middle.block.i.i.i.i.i.i.i.i, %_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i
+ store float* %add.ptr.i.i.i, float** %_M_finish.i.i.i, align 8, !tbaa !7
+ %_M_start.i.i = getelementptr inbounds %"class.std::vector"* %matA, i64 0, i32 0, i32 0, i32 0
+ %24 = load float** %_M_start.i.i, align 8, !tbaa !7
+ %25 = bitcast float* %24 to i8*
+ call void @llvm_visc_track_mem(i8* %25, i64 %mul19) #1
+ %_M_start.i.i82 = getelementptr inbounds %"class.std::vector"* %matBT, i64 0, i32 0, i32 0, i32 0
+ %26 = load float** %_M_start.i.i82, align 8, !tbaa !7
+ %27 = bitcast float* %26 to i8*
+ call void @llvm_visc_track_mem(i8* %27, i64 %mul22) #1
+ %28 = load float** %_M_start.i.i.i81, align 8, !tbaa !7
+ %29 = bitcast float* %28 to i8*
+ call void @llvm_visc_track_mem(i8* %29, i64 %mul25) #1
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 6) #1
+ %30 = load float** %_M_finish.i.i.i, align 8, !tbaa !7
+ %31 = load float** %_M_start.i.i.i81, align 8, !tbaa !7
+ %cmp3399 = icmp eq float* %30, %31
+ br i1 %cmp3399, label %for.end, label %for.body.lr.ph
+
+for.body.lr.ph: ; preds = %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit
+ %sub.ptr.lhs.cast.i = ptrtoint float* %30 to i64
+ %sub.ptr.rhs.cast.i = ptrtoint float* %31 to i64
+ %sub.ptr.sub.i = sub i64 %sub.ptr.lhs.cast.i, %sub.ptr.rhs.cast.i
+ %sub.ptr.div.i = ashr exact i64 %sub.ptr.sub.i, 2
+ br label %for.body
+
+for.body: ; preds = %for.body, %for.body.lr.ph
+ %i.0100 = phi i64 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
+ %add.ptr.i = getelementptr inbounds float* %31, i64 %i.0100
+ store float 0.000000e+00, float* %add.ptr.i, align 4, !tbaa !3
+ %inc = add i64 %i.0100, 1
+ %cmp33 = icmp ult i64 %inc, %sub.ptr.div.i
+ br i1 %cmp33, label %for.body, label %for.end
+
+for.end: ; preds = %for.body, %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 0) #1
+ %32 = load i32* %matArow, align 4, !tbaa !8
+ %33 = load i32* %matBcol, align 4, !tbaa !8
+ %34 = load i32* %matAcol, align 4, !tbaa !8
+ %35 = load float** %_M_start.i.i, align 8, !tbaa !7
+ %36 = load float** %_M_start.i.i82, align 8, !tbaa !7
+ %37 = load float** %_M_start.i.i.i81, align 8, !tbaa !7
+ call void @_Z10basicSgemmcciiifPfmiS_mifS_mi(i8 signext 78, i8 signext 84, i32 %32, i32 %33, i32 %34, float 1.000000e+00, float* %35, i64 %mul19, i32 %32, float* %36, i64 %mul22, i32 %33, float 0.000000e+00, float* %37, i64 %mul25, i32 %32)
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 3) #1
+ %38 = load float** %_M_start.i.i.i81, align 8, !tbaa !7
+ %39 = bitcast float* %38 to i8*
+ call void @llvm_visc_request_mem(i8* %39, i64 %mul25) #1
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 16) #1
+ %40 = load float** %_M_start.i.i, align 8, !tbaa !7
+ %41 = bitcast float* %40 to i8*
+ call void @llvm_visc_untrack_mem(i8* %41) #1
+ %42 = load float** %_M_start.i.i82, align 8, !tbaa !7
+ %43 = bitcast float* %42 to i8*
+ call void @llvm_visc_untrack_mem(i8* %43) #1
+ %44 = load float** %_M_start.i.i.i81, align 8, !tbaa !7
+ %45 = bitcast float* %44 to i8*
+ call void @llvm_visc_untrack_mem(i8* %45) #1
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 0) #1
+ call void @pb_PrintTimerSet(%struct.pb_TimerSet* %timers) #1
+ %Ptr = getelementptr [14 x i8]* @0, i64 0, i64 0
+ call void @llvm_visc_printTimerSet(i8** @viscTimerSet_GenVISC, i8* %Ptr)
+ call void @llvm.visc.cleanup()
+ %outFile = getelementptr inbounds %struct.pb_Parameters* %call, i64 0, i32 0
+ %46 = load i8** %outFile, align 8, !tbaa !7
+ %tobool = icmp eq i8* %46, null
+ br i1 %tobool, label %if.end45, label %if.then42
+
+if.then42: ; preds = %for.end
+ %47 = load i32* %matArow, align 4, !tbaa !8
+ %48 = load i32* %matBcol, align 4, !tbaa !8
+ %call44 = call zeroext i1 @_Z23writeColMajorMatrixFilePKciiRSt6vectorIfSaIfEE(i8* %46, i32 %47, i32 %48, %"class.std::vector"* %matC) #1
+ br label %if.end45
+
+if.end45: ; preds = %if.then42, %for.end
+ %arrayidx47 = getelementptr inbounds %struct.pb_TimerSet* %timers, i64 0, i32 4, i64 2
+ %call48 = call double @pb_GetElapsedTime(%struct.pb_Timer* %arrayidx47) #1
+ %call1.i88 = call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* @_ZSt4cout, i8* getelementptr inbounds ([10 x i8]* @.str5, i64 0, i64 0), i64 9) #1
+ %49 = load i32* %matArow, align 4, !tbaa !8
+ %conv50 = sitofp i32 %49 to double
+ %mul51 = fmul fast double %conv50, 2.000000e+00
+ %50 = load i32* %matBcol, align 4, !tbaa !8
+ %conv52 = sitofp i32 %50 to double
+ %mul53 = fmul fast double %mul51, %conv52
+ %51 = load i32* %matAcol, align 4, !tbaa !8
+ %conv54 = sitofp i32 %51 to double
+ %mul55 = fmul fast double %mul53, %conv54
+ %div = fdiv fast double %mul55, %call48
+ %div56 = fmul double %div, 1.000000e-09
+ %call.i = call %"class.std::basic_ostream"* @_ZNSo9_M_insertIdEERSoT_(%"class.std::basic_ostream"* @_ZSt4cout, double %div56) #1
+ %52 = bitcast %"class.std::basic_ostream"* %call.i to i8**
+ %vtable.i = load i8** %52, align 8, !tbaa !6
+ %vbase.offset.ptr.i = getelementptr i8* %vtable.i, i64 -24
+ %53 = bitcast i8* %vbase.offset.ptr.i to i64*
+ %vbase.offset.i = load i64* %53, align 8
+ %54 = bitcast %"class.std::basic_ostream"* %call.i to i8*
+ %add.ptr.sum.i = add i64 %vbase.offset.i, 240
+ %_M_ctype.i.i = getelementptr inbounds i8* %54, i64 %add.ptr.sum.i
+ %55 = bitcast i8* %_M_ctype.i.i to %"class.std::ctype"**
+ %56 = load %"class.std::ctype"** %55, align 8, !tbaa !7
+ %tobool.i.i.i = icmp eq %"class.std::ctype"* %56, null
+ br i1 %tobool.i.i.i, label %if.then.i.i.i, label %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit.i.i
+
+if.then.i.i.i: ; preds = %if.end45
+ call void @_ZSt16__throw_bad_castv() #7
+ unreachable
+
+_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit.i.i: ; preds = %if.end45
+ %_M_widen_ok.i.i.i = getelementptr inbounds %"class.std::ctype"* %56, i64 0, i32 6
+ %57 = load i8* %_M_widen_ok.i.i.i, align 1, !tbaa !4
+ %tobool.i3.i.i = icmp eq i8 %57, 0
+ br i1 %tobool.i3.i.i, label %if.end.i.i.i, label %if.then.i4.i.i
+
+if.then.i4.i.i: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit.i.i
+ %arrayidx.i.i.i = getelementptr inbounds %"class.std::ctype"* %56, i64 0, i32 7, i64 10
+ %58 = load i8* %arrayidx.i.i.i, align 1, !tbaa !4
+ br label %_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit
+
+if.end.i.i.i: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit.i.i
+ call void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"* %56) #1
+ %59 = bitcast %"class.std::ctype"* %56 to i8 (%"class.std::ctype"*, i8)***
+ %vtable.i.i.i = load i8 (%"class.std::ctype"*, i8)*** %59, align 8, !tbaa !6
+ %vfn.i.i.i = getelementptr inbounds i8 (%"class.std::ctype"*, i8)** %vtable.i.i.i, i64 6
+ %60 = load i8 (%"class.std::ctype"*, i8)** %vfn.i.i.i, align 8
+ %call.i.i.i = call signext i8 %60(%"class.std::ctype"* %56, i8 signext 10) #1
+ br label %_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit
+
+_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit: ; preds = %if.end.i.i.i, %if.then.i4.i.i
+ %retval.0.i.i.i = phi i8 [ %58, %if.then.i4.i.i ], [ %call.i.i.i, %if.end.i.i.i ]
+ %call1.i = call %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"* %call.i, i8 signext %retval.0.i.i.i) #1
+ %call.i.i = call %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"* %call1.i) #1
+ call void @pb_FreeParameters(%struct.pb_Parameters* %call) #1
+ %61 = load float** %_M_start.i.i.i81, align 8, !tbaa !7
+ %tobool.i.i.i.i78 = icmp eq float* %61, null
+ br i1 %tobool.i.i.i.i78, label %_ZNSt6vectorIfSaIfEED1Ev.exit80, label %if.then.i.i.i.i79
+
+if.then.i.i.i.i79: ; preds = %_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit
+ %62 = bitcast float* %61 to i8*
+ call void @_ZdlPv(i8* %62) #1
+ br label %_ZNSt6vectorIfSaIfEED1Ev.exit80
+
+_ZNSt6vectorIfSaIfEED1Ev.exit80: ; preds = %if.then.i.i.i.i79, %_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit
+ %63 = load float** %_M_start.i.i82, align 8, !tbaa !7
+ %tobool.i.i.i.i74 = icmp eq float* %63, null
+ br i1 %tobool.i.i.i.i74, label %_ZNSt6vectorIfSaIfEED1Ev.exit76, label %if.then.i.i.i.i75
+
+if.then.i.i.i.i75: ; preds = %_ZNSt6vectorIfSaIfEED1Ev.exit80
+ %64 = bitcast float* %63 to i8*
+ call void @_ZdlPv(i8* %64) #1
+ br label %_ZNSt6vectorIfSaIfEED1Ev.exit76
+
+_ZNSt6vectorIfSaIfEED1Ev.exit76: ; preds = %if.then.i.i.i.i75, %_ZNSt6vectorIfSaIfEED1Ev.exit80
+ %65 = load float** %_M_start.i.i, align 8, !tbaa !7
+ %tobool.i.i.i.i = icmp eq float* %65, null
+ br i1 %tobool.i.i.i.i, label %_ZNSt6vectorIfSaIfEED1Ev.exit, label %if.then.i.i.i.i
+
+if.then.i.i.i.i: ; preds = %_ZNSt6vectorIfSaIfEED1Ev.exit76
+ %66 = bitcast float* %65 to i8*
+ call void @_ZdlPv(i8* %66) #1
+ br label %_ZNSt6vectorIfSaIfEED1Ev.exit
+
+_ZNSt6vectorIfSaIfEED1Ev.exit: ; preds = %if.then.i.i.i.i, %_ZNSt6vectorIfSaIfEED1Ev.exit76
+ call void @llvm.lifetime.end(i64 800, i8* %0) #1
+ ret i32 0
+}
+
+; Function Attrs: nounwind
+declare void @llvm.lifetime.start(i64, i8* nocapture) #1
+
+declare %struct.pb_Parameters* @pb_ReadParameters(i32*, i8**) #0
+
+; Function Attrs: noreturn nounwind
+declare void @exit(i32) #4
+
+declare zeroext i1 @_Z22readColMajorMatrixFilePKcRiS1_RSt6vectorIfSaIfEE(i8*, i32*, i32*, %"class.std::vector"*) #0
+
+declare void @pb_InitializeTimerSet(%struct.pb_TimerSet*) #0
+
+declare void @pb_SwitchToTimer(%struct.pb_TimerSet*, i32) #0
+
+declare void @llvm_visc_track_mem(i8*, i64) #0
+
+declare void @llvm_visc_request_mem(i8*, i64) #0
+
+declare void @llvm_visc_untrack_mem(i8*) #0
+
+declare void @pb_PrintTimerSet(%struct.pb_TimerSet*) #0
+
+declare zeroext i1 @_Z23writeColMajorMatrixFilePKciiRSt6vectorIfSaIfEE(i8*, i32, i32, %"class.std::vector"*) #0
+
+declare double @pb_GetElapsedTime(%struct.pb_Timer*) #0
+
+declare void @pb_FreeParameters(%struct.pb_Parameters*) #0
+
+; Function Attrs: nounwind
+declare void @llvm.lifetime.end(i64, i8* nocapture) #1
+
+declare %"class.std::basic_ostream"* @_ZNSo9_M_insertIdEERSoT_(%"class.std::basic_ostream"*, double) #0
+
+; Function Attrs: noreturn
+declare void @_ZSt17__throw_bad_allocv() #5
+
+declare noalias i8* @_Znwm(i64) #0
+
+; Function Attrs: nounwind
+declare void @_ZdlPv(i8*) #6
+
+declare %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"*, i8 signext) #0
+
+declare void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"*) #0
+
+; Function Attrs: noreturn
+declare void @_ZSt16__throw_bad_castv() #5
+
+declare %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"*) #0
+
+declare %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"*, i8*, i64) #0
+
+; Function Attrs: nounwind
+define internal void @_GLOBAL__I_a() #1 section ".text.startup" {
+entry:
+ tail call void @_ZNSt8ios_base4InitC1Ev(%"class.std::ios_base::Init"* @_ZStL8__ioinit) #1
+ %0 = tail call i32 @__cxa_atexit(void (i8*)* bitcast (void (%"class.std::ios_base::Init"*)* @_ZNSt8ios_base4InitD1Ev to void (i8*)*), i8* getelementptr inbounds (%"class.std::ios_base::Init"* @_ZStL8__ioinit, i64 0, i32 0), i8* @__dso_handle) #1
+ ret void
+}
+
+; Function Attrs: nounwind
+declare i64 @fwrite(i8* nocapture, i64, i64, %struct._IO_FILE* nocapture) #1
+
+; Function Attrs: nounwind
+declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i32, i1) #1
+
+declare i8* @llvm_visc_initializeTimerSet()
+
+declare void @llvm_visc_switchToTimer(i8**, i32)
+
+declare void @llvm_visc_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.getNode() #1
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.getParentNode(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNodeInstanceID.x(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNumNodeInstances.x(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNodeInstanceID.y(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNumNodeInstances.y(i8*) #1
+
+; Function Attrs: nounwind uwtable
+define %rtype @_Z9mysgemmNTPfiS_iS_iiffInternal_level1(float* in %A, i64 %bytes_A, i32 %lda, float* in %B, i64 %bytes_B, i32 %ldb, float* in out %C, i64 %bytes_C, i32 %ldc, i32 %k, float %alpha, float %beta, i32 %dimX, i32 %dimY) #2 {
+entry:
+ %_Z9mysgemmNTPfiS_iS_iiff.node = call i8* @llvm.visc.createNode2D(i8* bitcast (%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float)* @_Z9mysgemmNTPfiS_iS_iiff to i8*), i32 %dimX, i32 %dimY)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 1, i32 1)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 2, i32 2)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 3, i32 3)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 4, i32 4)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 5, i32 5)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 6, i32 6)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 7, i32 7)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 8, i32 8)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 9, i32 9)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 10, i32 10)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 11, i32 11)
+ ret %rtype undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.createNode2D(i8*, i32, i32) #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.bind.input(i8*, i32, i32) #1
+
+; Function Attrs: nounwind uwtable
+define %rtype @_Z9mysgemmNTPfiS_iS_iiffInternal_level2(float* in %A, i64 %bytes_A, i32 %lda, float* in %B, i64 %bytes_B, i32 %ldb, float* in out %C, i64 %bytes_C, i32 %ldc, i32 %k, float %alpha, float %beta, i32 %dimX, i32 %dimY, i32 %dimX1, i32 %dimY2) #2 {
+entry:
+ %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node = call i8* @llvm.visc.createNode2D(i8* bitcast (%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float, i32, i32)* @_Z9mysgemmNTPfiS_iS_iiffInternal_level1 to i8*), i32 %dimX1, i32 %dimY2)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 1, i32 1)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 2, i32 2)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 3, i32 3)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 4, i32 4)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 5, i32 5)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 6, i32 6)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 7, i32 7)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 8, i32 8)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 9, i32 9)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 10, i32 10)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 11, i32 11)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 12, i32 12)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 13, i32 13)
+ ret %rtype undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.launch(i8*, i8*) #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.wait(i8*) #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.init() #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.cleanup() #1
+
+attributes #0 = { "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { nounwind }
+attributes #2 = { nounwind uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { noinline nounwind uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #4 = { noreturn nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #5 = { noreturn "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #6 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #7 = { noreturn nounwind }
+
+!visc_hint_gpu = !{!0, !1}
+!visc_hint_cpu = !{!2}
+
+!0 = metadata !{%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float)* @_Z9mysgemmNTPfiS_iS_iiff}
+!1 = metadata !{%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float, i32, i32)* @_Z9mysgemmNTPfiS_iS_iiffInternal_level1}
+!2 = metadata !{%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float, i32, i32, i32, i32)* @_Z9mysgemmNTPfiS_iS_iiffInternal_level2}
+!3 = metadata !{metadata !"float", metadata !4}
+!4 = metadata !{metadata !"omnipotent char", metadata !5}
+!5 = metadata !{metadata !"Simple C/C++ TBAA"}
+!6 = metadata !{metadata !"vtable pointer", metadata !5}
+!7 = metadata !{metadata !"any pointer", metadata !4}
+!8 = metadata !{metadata !"int", metadata !4}
+!9 = metadata !{metadata !"branch_weights", i32 4, i32 64}
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/Makefile b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/Makefile
new file mode 100644
index 0000000000..f74ee8921a
--- /dev/null
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/Makefile
@@ -0,0 +1,8 @@
+# (c) 2010 The Board of Trustees of the University of Illinois.
+
+LANGUAGE=visc
+SRCDIR_OBJS=io.ll #compute_gold.o
+VISC_OBJS=main.visc.ll
+APP_CUDALDFLAGS=-lm -lstdc++
+APP_CFLAGS=-ffast-math -O3
+APP_CXXFLAGS=-ffast-math -O3
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec_opt/io.cc b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/io.cc
similarity index 100%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec_opt/io.cc
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/io.cc
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec_opt/kernel.cl b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/kernel.cl
similarity index 100%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec_opt/kernel.cl
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/kernel.cl
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec_opt/main.cc b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/main.cc
similarity index 91%
rename from hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec_opt/main.cc
rename to hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/main.cc
index 8fbc45e08a..a4c252d8f1 100644
--- a/hpvm/test/parboil/benchmarks/sgemm/src/hpvm_vec_opt/main.cc
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/main.cc
@@ -10,7 +10,6 @@
* Main entry of dense matrix-matrix multiplication kernel
*/
-#include <hpvm.h>
#include <iostream>
#include <malloc.h>
#include <math.h>
@@ -20,6 +19,7 @@
#include <string.h>
#include <sys/time.h>
#include <vector>
+#include <visc.h>
// I/O routines
extern bool readColMajorMatrixFile(const char *fn, int &nr_row, int &nr_col,
@@ -42,8 +42,8 @@ extern char *readFile(const char *);
void mysgemmNT(float *A, int lda, float *B, int ldb, float *C, int ldc, int k,
float alpha, float beta) {
- __hpvm__hint(hpvm::SPIR_TARGET);
- __hpvm__attributes(3, A, B, C, 1, C);
+ __visc__hint(visc::SPIR_TARGET);
+ __visc__attributes(3, A, B, C, 1, C);
float c[TILE_N];
for (int i = 0; i < TILE_N; i++)
@@ -135,10 +135,10 @@ __attribute__((noinline)) void basicSgemm(char transa, char transb, int m,
unsigned db[2] = {TILE_N, TILE_TB_HEIGHT};
unsigned dg[2] = {m * TILE_N / TILE_M, n * TILE_TB_HEIGHT / TILE_N};
- void *sgemmDFG = __hpvm__node(mysgemmNT, 2, 2, db[0], db[1], dg[0] / db[0],
+ void *sgemmDFG = __visc__node(mysgemmNT, 2, 2, db[0], db[1], dg[0] / db[0],
dg[1] / db[1], 12, A, bytesA, lda, B, bytesB,
ldb, C, bytesC, ldc, k, alpha, beta, 0);
- __hpvm__wait(sgemmDFG);
+ __visc__wait(sgemmDFG);
}
int main(int argc, char *argv[]) {
@@ -168,7 +168,7 @@ int main(int argc, char *argv[]) {
readColMajorMatrixFile(params->inpFiles[2], matBcol, matBrow, matBT);
pb_InitializeTimerSet(&timers);
- __hpvm__init();
+ __visc__init();
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
// copy A to device memory
@@ -181,10 +181,10 @@ int main(int argc, char *argv[]) {
// OpenCL memory allocation
std::vector<float> matC(matArow * matBcol);
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_TRACK);
- llvm_hpvm_track_mem(&matA.front(), A_sz);
- llvm_hpvm_track_mem(&matBT.front(), B_sz);
- llvm_hpvm_track_mem(&matC.front(), C_sz);
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_TRACK);
+ llvm_visc_track_mem(&matA.front(), A_sz);
+ llvm_visc_track_mem(&matBT.front(), B_sz);
+ llvm_visc_track_mem(&matC.front(), C_sz);
// Copy A and B^T into device memory
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
@@ -200,16 +200,16 @@ int main(int argc, char *argv[]) {
matArow);
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
- llvm_hpvm_request_mem(&matC.front(), C_sz);
+ llvm_visc_request_mem(&matC.front(), C_sz);
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_UNTRACK);
- llvm_hpvm_untrack_mem(&matA.front());
- llvm_hpvm_untrack_mem(&matBT.front());
- llvm_hpvm_untrack_mem(&matC.front());
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_UNTRACK);
+ llvm_visc_untrack_mem(&matA.front());
+ llvm_visc_untrack_mem(&matBT.front());
+ llvm_visc_untrack_mem(&matC.front());
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
pb_PrintTimerSet(&timers);
- __hpvm__cleanup();
+ __visc__cleanup();
if (params->outFile) {
diff --git a/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/main.visc.ll b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/main.visc.ll
new file mode 100644
index 0000000000..b997cf7ebc
--- /dev/null
+++ b/hpvm/test/parboil/benchmarks/sgemm/src/visc_vec_opt/main.visc.ll
@@ -0,0 +1,889 @@
+; ModuleID = 'build/visc_vec_opt_default/main.ll'
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%"class.std::ios_base::Init" = type { i8 }
+%"class.std::basic_ostream" = type { i32 (...)**, %"class.std::basic_ios" }
+%"class.std::basic_ios" = type { %"class.std::ios_base", %"class.std::basic_ostream"*, i8, i8, %"class.std::basic_streambuf"*, %"class.std::ctype"*, %"class.std::num_put"*, %"class.std::num_get"* }
+%"class.std::ios_base" = type { i32 (...)**, i64, i64, i32, i32, i32, %"struct.std::ios_base::_Callback_list"*, %"struct.std::ios_base::_Words", [8 x %"struct.std::ios_base::_Words"], i32, %"struct.std::ios_base::_Words"*, %"class.std::locale" }
+%"struct.std::ios_base::_Callback_list" = type { %"struct.std::ios_base::_Callback_list"*, void (i32, %"class.std::ios_base"*, i32)*, i32, i32 }
+%"struct.std::ios_base::_Words" = type { i8*, i64 }
+%"class.std::locale" = type { %"class.std::locale::_Impl"* }
+%"class.std::locale::_Impl" = type { i32, %"class.std::locale::facet"**, i64, %"class.std::locale::facet"**, i8** }
+%"class.std::locale::facet" = type { i32 (...)**, i32 }
+%"class.std::basic_streambuf" = type { i32 (...)**, i8*, i8*, i8*, i8*, i8*, i8*, %"class.std::locale" }
+%"class.std::ctype" = type { %"class.std::locale::facet", %struct.__locale_struct*, i8, i32*, i32*, i16*, i8, [256 x i8], [256 x i8], i8 }
+%struct.__locale_struct = type { [13 x %struct.__locale_data*], i16*, i32*, i32*, [13 x i8*] }
+%struct.__locale_data = type opaque
+%"class.std::num_put" = type { %"class.std::locale::facet" }
+%"class.std::num_get" = type { %"class.std::locale::facet" }
+%struct._IO_FILE = type { i32, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, %struct._IO_marker*, %struct._IO_FILE*, i32, i32, i64, i16, i8, [1 x i8], i8*, i64, i8*, i8*, i8*, i8*, i64, i32, [20 x i8] }
+%struct._IO_marker = type { %struct._IO_marker*, %struct._IO_FILE*, i32 }
+%rtype = type {}
+%struct.arg = type <{ float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float, i32, i32, i32, i32, %rtype }>
+%struct.pb_TimerSet = type { i32, %struct.pb_async_time_marker_list*, i64, i64, [24 x %struct.pb_Timer], [24 x %struct.pb_SubTimerList*] }
+%struct.pb_async_time_marker_list = type { i8*, i32, i8*, %struct.pb_async_time_marker_list* }
+%struct.pb_Timer = type { i32, i64, i64 }
+%struct.pb_SubTimerList = type { %struct.pb_SubTimer*, %struct.pb_SubTimer* }
+%struct.pb_SubTimer = type { i8*, %struct.pb_Timer, %struct.pb_SubTimer* }
+%"class.std::vector" = type { %"struct.std::_Vector_base" }
+%"struct.std::_Vector_base" = type { %"struct.std::_Vector_base<float, std::allocator<float> >::_Vector_impl" }
+%"struct.std::_Vector_base<float, std::allocator<float> >::_Vector_impl" = type { float*, float*, float* }
+%struct.pb_Parameters = type { i8*, i8** }
+
+@_ZStL8__ioinit = internal global %"class.std::ios_base::Init" zeroinitializer, align 1
+@__dso_handle = external global i8
+@_ZSt4cerr = external global %"class.std::basic_ostream"
+@.str = private unnamed_addr constant [48 x i8] c"unsupported value of 'transa' in regtileSgemm()\00", align 1
+@.str1 = private unnamed_addr constant [48 x i8] c"unsupported value of 'transb' in regtileSgemm()\00", align 1
+@.str2 = private unnamed_addr constant [53 x i8] c"unsupported size of matrix. m should be multiple of \00", align 1
+@.str3 = private unnamed_addr constant [27 x i8] c"; n should be multiple of \00", align 1
+@stderr = external global %struct._IO_FILE*
+@.str4 = private unnamed_addr constant [33 x i8] c"Expecting three input filenames\0A\00", align 1
+@_ZSt4cout = external global %"class.std::basic_ostream"
+@.str5 = private unnamed_addr constant [10 x i8] c"GFLOPs = \00", align 1
+@llvm.global_ctors = appending global [1 x { i32, void ()* }] [{ i32, void ()* } { i32 65535, void ()* @_GLOBAL__I_a }]
+@viscTimerSet_GenVISC = common global i8* null
+@0 = internal constant [14 x i8] c"GenVISC_Timer\00"
+
+declare void @_ZNSt8ios_base4InitC1Ev(%"class.std::ios_base::Init"*) #0
+
+declare void @_ZNSt8ios_base4InitD1Ev(%"class.std::ios_base::Init"*) #0
+
+; Function Attrs: nounwind
+declare i32 @__cxa_atexit(void (i8*)*, i8*, i8*) #1
+
+; Function Attrs: nounwind uwtable
+define %rtype @_Z9mysgemmNTPfiS_iS_iiff(float* in %A, i64 %bytes_A, i32 %lda, float* in %B, i64 %bytes_B, i32 %ldb, float* in out %C, i64 %bytes_C, i32 %ldc, i32 %k, float %alpha, float %beta) #2 {
+entry:
+ %_Z9mysgemmNTPfiS_iS_iiff.node = call i8* @llvm.visc.getNode()
+ %_Z9mysgemmNTPfiS_iS_iiff.parentNode = call i8* @llvm.visc.getParentNode(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+
+; %call = call i32 @get_local_id(i32 1) #2
+ %call = call i32 @llvm.visc.getNodeInstanceID.y(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+
+; %call1 = call i32 @get_local_size(i32 0) #2
+ %call1 = call i32 @llvm.visc.getNumNodeInstances.x(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+
+ %mul = mul i32 %call1, %call
+
+; %call2 = call i32 @get_local_id(i32 0) #2
+ %call2 = call i32 @llvm.visc.getNodeInstanceID.x(i8* %_Z9mysgemmNTPfiS_iS_iiff.node)
+
+ %add = add i32 %mul, %call2
+
+; %call3 = call i32 @get_group_id(i32 0) #2
+ %call3 = call i32 @llvm.visc.getNodeInstanceID.x(i8* %_Z9mysgemmNTPfiS_iS_iiff.parentNode)
+
+ %mul4 = shl i32 %call3, 6
+ %add5 = add i32 %add, %mul4
+ %cmp89 = icmp sgt i32 %k, 0
+
+ %call6 = call i32 @llvm.visc.getNodeInstanceID.y(i8* %_Z9mysgemmNTPfiS_iS_iiff.parentNode)
+ %mul7 = shl i32 %call6, 3
+
+ br i1 %cmp89, label %for.body, label %for.end23
+
+for.body: ; preds = %entry, %for.inc21
+ %cp.091 = phi <8 x float> [ %add20, %for.inc21 ], [ zeroinitializer, %entry ]
+ %i.090 = phi i32 [ %add22, %for.inc21 ], [ 0, %entry ]
+; %call6 = call i32 @get_group_id(i32 1) #2
+; %call6 = call i32 @llvm.visc.getNodeInstanceID.y(i8* %_Z9mysgemmNTPfiS_iS_iiff.parentNode)
+; %mul7 = shl i32 %call6, 3
+ br label %for.body12
+
+for.body12: ; preds = %for.body12, %for.body
+ %cp.188 = phi <8 x float> [ %cp.091, %for.body ], [ %add20, %for.body12 ]
+ %j.087 = phi i32 [ 0, %for.body ], [ %inc, %for.body12 ]
+ %add13 = add i32 %j.087, %i.090
+ %mul14 = mul nsw i32 %add13, %lda
+ %add15 = add nsw i32 %mul14, %add5
+ %arrayidx = getelementptr inbounds float* %A, i32 %add15
+ %0 = load float* %arrayidx, align 4, !tbaa !3
+ %splat.splatinsert = insertelement <8 x float> undef, float %0, i32 0
+ %splat.splat = shufflevector <8 x float> %splat.splatinsert, <8 x float> undef, <8 x i32> zeroinitializer
+ %tmp83 = mul i32 %add13, %ldb
+ %add.ptr.sum = add i32 %tmp83, %mul7
+ %add.ptr17 = getelementptr inbounds float* %B, i32 %add.ptr.sum
+
+; %call18 = call <8 x float> @_Z6vload8jPKU3AS1f(i32 0, float* %add.ptr17) #2
+ %add.ptr17.cast = bitcast float* %add.ptr17 to <8 x float>*
+ %call18 = load <8 x float>* %add.ptr17.cast, align 8
+
+ %mul19 = fmul fast <8 x float> %call18, %splat.splat
+ %add20 = fadd fast <8 x float> %cp.188, %mul19
+ %inc = add nsw i32 %j.087, 1
+ %exitcond92 = icmp eq i32 %inc, 8
+ br i1 %exitcond92, label %for.inc21, label %for.body12
+
+for.inc21: ; preds = %for.body12
+ %add22 = add nsw i32 %i.090, 8
+ %cmp = icmp slt i32 %add22, %k
+ br i1 %cmp, label %for.body, label %for.end23
+
+for.end23: ; preds = %for.inc21, %entry
+ %cp.0.lcssa = phi <8 x float> [ zeroinitializer, %entry ], [ %add20, %for.inc21 ]
+ %splat.splatinsert24 = insertelement <8 x float> undef, float %alpha, i32 0
+ %splat.splat25 = shufflevector <8 x float> %splat.splatinsert24, <8 x float> undef, <8 x i32> zeroinitializer
+ %mul26 = fmul fast <8 x float> %splat.splat25, %cp.0.lcssa
+ %1 = extractelement <8 x float> %mul26, i32 0
+ %2 = extractelement <8 x float> %mul26, i32 1
+ %3 = extractelement <8 x float> %mul26, i32 2
+ %4 = extractelement <8 x float> %mul26, i32 3
+ %5 = extractelement <8 x float> %mul26, i32 4
+ %6 = extractelement <8 x float> %mul26, i32 5
+ %7 = extractelement <8 x float> %mul26, i32 6
+ %8 = extractelement <8 x float> %mul26, i32 7
+; %call35 = call i32 @get_group_id(i32 1) #2
+; %call35 = call i32 @llvm.visc.getNodeInstanceID.y(i8* %_Z9mysgemmNTPfiS_iS_iiff.parentNode)
+ %mul37 = shl i32 %call6, 3
+ %tmp85 = mul i32 %mul37, %ldc
+ %add44 = add i32 %tmp85, %add5
+ %arrayidx45 = getelementptr inbounds float* %C, i32 %add44
+ %9 = load float* %arrayidx45, align 4, !tbaa !3
+ %mul46 = fmul fast float %9, %beta
+ %add48 = fadd fast float %mul46, %1
+ store float %add48, float* %arrayidx45, align 4, !tbaa !3
+ %tmp84.193 = or i32 %mul37, 1
+ %tmp85.1 = mul i32 %tmp84.193, %ldc
+ %add44.1 = add i32 %tmp85.1, %add5
+ %arrayidx45.1 = getelementptr inbounds float* %C, i32 %add44.1
+ %10 = load float* %arrayidx45.1, align 4, !tbaa !3
+ %mul46.1 = fmul fast float %10, %beta
+ %add48.1 = fadd fast float %mul46.1, %2
+ store float %add48.1, float* %arrayidx45.1, align 4, !tbaa !3
+ %tmp84.294 = or i32 %mul37, 2
+ %tmp85.2 = mul i32 %tmp84.294, %ldc
+ %add44.2 = add i32 %tmp85.2, %add5
+ %arrayidx45.2 = getelementptr inbounds float* %C, i32 %add44.2
+ %11 = load float* %arrayidx45.2, align 4, !tbaa !3
+ %mul46.2 = fmul fast float %11, %beta
+ %add48.2 = fadd fast float %mul46.2, %3
+ store float %add48.2, float* %arrayidx45.2, align 4, !tbaa !3
+ %tmp84.395 = or i32 %mul37, 3
+ %tmp85.3 = mul i32 %tmp84.395, %ldc
+ %add44.3 = add i32 %tmp85.3, %add5
+ %arrayidx45.3 = getelementptr inbounds float* %C, i32 %add44.3
+ %12 = load float* %arrayidx45.3, align 4, !tbaa !3
+ %mul46.3 = fmul fast float %12, %beta
+ %add48.3 = fadd fast float %mul46.3, %4
+ store float %add48.3, float* %arrayidx45.3, align 4, !tbaa !3
+ %tmp84.496 = or i32 %mul37, 4
+ %tmp85.4 = mul i32 %tmp84.496, %ldc
+ %add44.4 = add i32 %tmp85.4, %add5
+ %arrayidx45.4 = getelementptr inbounds float* %C, i32 %add44.4
+ %13 = load float* %arrayidx45.4, align 4, !tbaa !3
+ %mul46.4 = fmul fast float %13, %beta
+ %add48.4 = fadd fast float %mul46.4, %5
+ store float %add48.4, float* %arrayidx45.4, align 4, !tbaa !3
+ %tmp84.597 = or i32 %mul37, 5
+ %tmp85.5 = mul i32 %tmp84.597, %ldc
+ %add44.5 = add i32 %tmp85.5, %add5
+ %arrayidx45.5 = getelementptr inbounds float* %C, i32 %add44.5
+ %14 = load float* %arrayidx45.5, align 4, !tbaa !3
+ %mul46.5 = fmul fast float %14, %beta
+ %add48.5 = fadd fast float %mul46.5, %6
+ store float %add48.5, float* %arrayidx45.5, align 4, !tbaa !3
+ %tmp84.698 = or i32 %mul37, 6
+ %tmp85.6 = mul i32 %tmp84.698, %ldc
+ %add44.6 = add i32 %tmp85.6, %add5
+ %arrayidx45.6 = getelementptr inbounds float* %C, i32 %add44.6
+ %15 = load float* %arrayidx45.6, align 4, !tbaa !3
+ %mul46.6 = fmul fast float %15, %beta
+ %add48.6 = fadd fast float %mul46.6, %7
+ store float %add48.6, float* %arrayidx45.6, align 4, !tbaa !3
+ %tmp84.799 = or i32 %mul37, 7
+ %tmp85.7 = mul i32 %tmp84.799, %ldc
+ %add44.7 = add i32 %tmp85.7, %add5
+ %arrayidx45.7 = getelementptr inbounds float* %C, i32 %add44.7
+ %16 = load float* %arrayidx45.7, align 4, !tbaa !3
+ %mul46.7 = fmul fast float %16, %beta
+ %add48.7 = fadd fast float %mul46.7, %8
+ store float %add48.7, float* %arrayidx45.7, align 4, !tbaa !3
+
+
+ ret %rtype undef
+}
+
+; Function Attrs: noinline nounwind uwtable
+define void @_Z10basicSgemmcciiifPfmiS_mifS_mi(i8 signext %transa, i8 signext %transb, i32 %m, i32 %n, i32 %k, float %alpha, float* %A, i64 %bytesA, i32 %lda, float* %B, i64 %bytesB, i32 %ldb, float %beta, float* %C, i64 %bytesC, i32 %ldc) #3 {
+entry:
+ switch i8 %transa, label %if.then [
+ i8 78, label %if.end
+ i8 110, label %if.end
+ ]
+
+if.then: ; preds = %entry
+ %call1.i = tail call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* @_ZSt4cerr, i8* getelementptr inbounds ([48 x i8]* @.str, i64 0, i64 0), i64 47) #1
+ %vtable.i = load i8** bitcast (%"class.std::basic_ostream"* @_ZSt4cerr to i8**), align 8, !tbaa !6
+ %vbase.offset.ptr.i = getelementptr i8* %vtable.i, i64 -24
+ %0 = bitcast i8* %vbase.offset.ptr.i to i64*
+ %vbase.offset.i = load i64* %0, align 8
+ %add.ptr.i.sum = add i64 %vbase.offset.i, 240
+ %_M_ctype.i = getelementptr inbounds i8* bitcast (%"class.std::basic_ostream"* @_ZSt4cerr to i8*), i64 %add.ptr.i.sum
+ %1 = bitcast i8* %_M_ctype.i to %"class.std::ctype"**
+ %2 = load %"class.std::ctype"** %1, align 8, !tbaa !7
+ %tobool.i93 = icmp eq %"class.std::ctype"* %2, null
+ br i1 %tobool.i93, label %if.then.i94, label %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit
+
+if.then.i94: ; preds = %if.then
+ tail call void @_ZSt16__throw_bad_castv() #7
+ unreachable
+
+_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit: ; preds = %if.then
+ %_M_widen_ok.i = getelementptr inbounds %"class.std::ctype"* %2, i64 0, i32 6
+ %3 = load i8* %_M_widen_ok.i, align 1, !tbaa !4
+ %tobool.i = icmp eq i8 %3, 0
+ br i1 %tobool.i, label %if.end.i, label %if.then.i
+
+if.then.i: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit
+ %arrayidx.i = getelementptr inbounds %"class.std::ctype"* %2, i64 0, i32 7, i64 10
+ %4 = load i8* %arrayidx.i, align 1, !tbaa !4
+ br label %_ZNKSt5ctypeIcE5widenEc.exit
+
+if.end.i: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit
+ tail call void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"* %2) #1
+ %5 = bitcast %"class.std::ctype"* %2 to i8 (%"class.std::ctype"*, i8)***
+ %vtable.i67 = load i8 (%"class.std::ctype"*, i8)*** %5, align 8, !tbaa !6
+ %vfn.i = getelementptr inbounds i8 (%"class.std::ctype"*, i8)** %vtable.i67, i64 6
+ %6 = load i8 (%"class.std::ctype"*, i8)** %vfn.i, align 8
+ %call.i68 = tail call signext i8 %6(%"class.std::ctype"* %2, i8 signext 10) #1
+ br label %_ZNKSt5ctypeIcE5widenEc.exit
+
+_ZNKSt5ctypeIcE5widenEc.exit: ; preds = %if.end.i, %if.then.i
+ %retval.0.i = phi i8 [ %4, %if.then.i ], [ %call.i68, %if.end.i ]
+ %call1.i43 = tail call %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"* @_ZSt4cerr, i8 signext %retval.0.i) #1
+ %call.i = tail call %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"* %call1.i43) #1
+ br label %return
+
+if.end: ; preds = %entry, %entry
+ switch i8 %transb, label %if.then9 [
+ i8 84, label %if.end12
+ i8 116, label %if.end12
+ ]
+
+if.then9: ; preds = %if.end
+ %call1.i45 = tail call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* @_ZSt4cerr, i8* getelementptr inbounds ([48 x i8]* @.str1, i64 0, i64 0), i64 47) #1
+ %vtable.i47 = load i8** bitcast (%"class.std::basic_ostream"* @_ZSt4cerr to i8**), align 8, !tbaa !6
+ %vbase.offset.ptr.i48 = getelementptr i8* %vtable.i47, i64 -24
+ %7 = bitcast i8* %vbase.offset.ptr.i48 to i64*
+ %vbase.offset.i49 = load i64* %7, align 8
+ %add.ptr.i50.sum = add i64 %vbase.offset.i49, 240
+ %_M_ctype.i69 = getelementptr inbounds i8* bitcast (%"class.std::basic_ostream"* @_ZSt4cerr to i8*), i64 %add.ptr.i50.sum
+ %8 = bitcast i8* %_M_ctype.i69 to %"class.std::ctype"**
+ %9 = load %"class.std::ctype"** %8, align 8, !tbaa !7
+ %tobool.i96 = icmp eq %"class.std::ctype"* %9, null
+ br i1 %tobool.i96, label %if.then.i97, label %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit99
+
+if.then.i97: ; preds = %if.then9
+ tail call void @_ZSt16__throw_bad_castv() #7
+ unreachable
+
+_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit99: ; preds = %if.then9
+ %_M_widen_ok.i71 = getelementptr inbounds %"class.std::ctype"* %9, i64 0, i32 6
+ %10 = load i8* %_M_widen_ok.i71, align 1, !tbaa !4
+ %tobool.i72 = icmp eq i8 %10, 0
+ br i1 %tobool.i72, label %if.end.i78, label %if.then.i74
+
+if.then.i74: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit99
+ %arrayidx.i73 = getelementptr inbounds %"class.std::ctype"* %9, i64 0, i32 7, i64 10
+ %11 = load i8* %arrayidx.i73, align 1, !tbaa !4
+ br label %_ZNKSt5ctypeIcE5widenEc.exit80
+
+if.end.i78: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit99
+ tail call void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"* %9) #1
+ %12 = bitcast %"class.std::ctype"* %9 to i8 (%"class.std::ctype"*, i8)***
+ %vtable.i75 = load i8 (%"class.std::ctype"*, i8)*** %12, align 8, !tbaa !6
+ %vfn.i76 = getelementptr inbounds i8 (%"class.std::ctype"*, i8)** %vtable.i75, i64 6
+ %13 = load i8 (%"class.std::ctype"*, i8)** %vfn.i76, align 8
+ %call.i77 = tail call signext i8 %13(%"class.std::ctype"* %9, i8 signext 10) #1
+ br label %_ZNKSt5ctypeIcE5widenEc.exit80
+
+_ZNKSt5ctypeIcE5widenEc.exit80: ; preds = %if.end.i78, %if.then.i74
+ %retval.0.i79 = phi i8 [ %11, %if.then.i74 ], [ %call.i77, %if.end.i78 ]
+ %call1.i52 = tail call %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"* @_ZSt4cerr, i8 signext %retval.0.i79) #1
+ %call.i53 = tail call %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"* %call1.i52) #1
+ br label %return
+
+if.end12: ; preds = %if.end, %if.end
+ %rem40 = and i32 %m, 63
+ %tobool = icmp eq i32 %rem40, 0
+ br i1 %tobool, label %lor.lhs.false, label %if.then15
+
+lor.lhs.false: ; preds = %if.end12
+ %rem1341 = and i32 %n, 7
+ %tobool14 = icmp eq i32 %rem1341, 0
+ br i1 %tobool14, label %if.end21, label %if.then15
+
+if.then15: ; preds = %lor.lhs.false, %if.end12
+ %call1.i55 = tail call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* @_ZSt4cerr, i8* getelementptr inbounds ([53 x i8]* @.str2, i64 0, i64 0), i64 52) #1
+ %call17 = tail call %"class.std::basic_ostream"* @_ZNSolsEi(%"class.std::basic_ostream"* @_ZSt4cerr, i32 64) #1
+ %call1.i57 = tail call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* %call17, i8* getelementptr inbounds ([27 x i8]* @.str3, i64 0, i64 0), i64 26) #1
+ %call19 = tail call %"class.std::basic_ostream"* @_ZNSolsEi(%"class.std::basic_ostream"* %call17, i32 8) #1
+ %14 = bitcast %"class.std::basic_ostream"* %call19 to i8**
+ %vtable.i59 = load i8** %14, align 8, !tbaa !6
+ %vbase.offset.ptr.i60 = getelementptr i8* %vtable.i59, i64 -24
+ %15 = bitcast i8* %vbase.offset.ptr.i60 to i64*
+ %vbase.offset.i61 = load i64* %15, align 8
+ %16 = bitcast %"class.std::basic_ostream"* %call19 to i8*
+ %add.ptr.i62.sum = add i64 %vbase.offset.i61, 240
+ %_M_ctype.i81 = getelementptr inbounds i8* %16, i64 %add.ptr.i62.sum
+ %17 = bitcast i8* %_M_ctype.i81 to %"class.std::ctype"**
+ %18 = load %"class.std::ctype"** %17, align 8, !tbaa !7
+ %tobool.i100 = icmp eq %"class.std::ctype"* %18, null
+ br i1 %tobool.i100, label %if.then.i101, label %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit103
+
+if.then.i101: ; preds = %if.then15
+ tail call void @_ZSt16__throw_bad_castv() #7
+ unreachable
+
+_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit103: ; preds = %if.then15
+ %_M_widen_ok.i83 = getelementptr inbounds %"class.std::ctype"* %18, i64 0, i32 6
+ %19 = load i8* %_M_widen_ok.i83, align 1, !tbaa !4
+ %tobool.i84 = icmp eq i8 %19, 0
+ br i1 %tobool.i84, label %if.end.i90, label %if.then.i86
+
+if.then.i86: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit103
+ %arrayidx.i85 = getelementptr inbounds %"class.std::ctype"* %18, i64 0, i32 7, i64 10
+ %20 = load i8* %arrayidx.i85, align 1, !tbaa !4
+ br label %_ZNKSt5ctypeIcE5widenEc.exit92
+
+if.end.i90: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit103
+ tail call void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"* %18) #1
+ %21 = bitcast %"class.std::ctype"* %18 to i8 (%"class.std::ctype"*, i8)***
+ %vtable.i87 = load i8 (%"class.std::ctype"*, i8)*** %21, align 8, !tbaa !6
+ %vfn.i88 = getelementptr inbounds i8 (%"class.std::ctype"*, i8)** %vtable.i87, i64 6
+ %22 = load i8 (%"class.std::ctype"*, i8)** %vfn.i88, align 8
+ %call.i89 = tail call signext i8 %22(%"class.std::ctype"* %18, i8 signext 10) #1
+ br label %_ZNKSt5ctypeIcE5widenEc.exit92
+
+_ZNKSt5ctypeIcE5widenEc.exit92: ; preds = %if.end.i90, %if.then.i86
+ %retval.0.i91 = phi i8 [ %20, %if.then.i86 ], [ %call.i89, %if.end.i90 ]
+ %call1.i64 = tail call %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"* %call19, i8 signext %retval.0.i91) #1
+ %call.i65 = tail call %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"* %call1.i64) #1
+ br label %return
+
+if.end21: ; preds = %lor.lhs.false
+ %mul = shl nsw i32 %m, 3
+ %div = sdiv i32 %mul, 64
+ %div27 = lshr i32 %div, 3
+ %div30 = lshr i32 %n, 3
+ %conv31 = fpext float %alpha to double
+ %conv32 = fpext float %beta to double
+ call void @llvm_visc_switchToTimer(i8** @viscTimerSet_GenVISC, i32 19)
+ %in.addr = alloca %struct.arg
+ %in.addr.A = getelementptr %struct.arg* %in.addr, i32 0, i32 0
+ store float* %A, float** %in.addr.A
+ %in.addr.bytes_A = getelementptr %struct.arg* %in.addr, i32 0, i32 1
+ store i64 %bytesA, i64* %in.addr.bytes_A
+ %in.addr.lda = getelementptr %struct.arg* %in.addr, i32 0, i32 2
+ store i32 %lda, i32* %in.addr.lda
+ %in.addr.B = getelementptr %struct.arg* %in.addr, i32 0, i32 3
+ store float* %B, float** %in.addr.B
+ %in.addr.bytes_B = getelementptr %struct.arg* %in.addr, i32 0, i32 4
+ store i64 %bytesB, i64* %in.addr.bytes_B
+ %in.addr.ldb = getelementptr %struct.arg* %in.addr, i32 0, i32 5
+ store i32 %ldb, i32* %in.addr.ldb
+ %in.addr.C = getelementptr %struct.arg* %in.addr, i32 0, i32 6
+ store float* %C, float** %in.addr.C
+ %in.addr.bytes_C = getelementptr %struct.arg* %in.addr, i32 0, i32 7
+ store i64 %bytesC, i64* %in.addr.bytes_C
+ %in.addr.ldc = getelementptr %struct.arg* %in.addr, i32 0, i32 8
+ store i32 %ldc, i32* %in.addr.ldc
+ %in.addr.k = getelementptr %struct.arg* %in.addr, i32 0, i32 9
+ store i32 %k, i32* %in.addr.k
+ %in.addr.alpha = getelementptr %struct.arg* %in.addr, i32 0, i32 10
+ %in.addr.alpha.cast = fptrunc double %conv31 to float
+ store float %in.addr.alpha.cast, float* %in.addr.alpha
+ %in.addr.beta = getelementptr %struct.arg* %in.addr, i32 0, i32 11
+ %in.addr.beta.cast = fptrunc double %conv32 to float
+ store float %in.addr.beta.cast, float* %in.addr.beta
+ %in.addr.dimX0 = getelementptr %struct.arg* %in.addr, i32 0, i32 12
+ store i32 8, i32* %in.addr.dimX0
+ %in.addr.dimY0 = getelementptr %struct.arg* %in.addr, i32 0, i32 13
+ store i32 8, i32* %in.addr.dimY0
+ %in.addr.dimX1 = getelementptr %struct.arg* %in.addr, i32 0, i32 14
+ store i32 %div27, i32* %in.addr.dimX1
+ %in.addr.dimY1 = getelementptr %struct.arg* %in.addr, i32 0, i32 15
+ store i32 %div30, i32* %in.addr.dimY1
+ %args = bitcast %struct.arg* %in.addr to i8*
+ call void @llvm_visc_switchToTimer(i8** @viscTimerSet_GenVISC, i32 0)
+ %graph_Z9mysgemmNTPfiS_iS_iiffInternal_level2 = call i8* @llvm.visc.launch(i8* bitcast (%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float, i32, i32, i32, i32)* @_Z9mysgemmNTPfiS_iS_iiffInternal_level2 to i8*), i8* %args)
+ call void @llvm.visc.wait(i8* %graph_Z9mysgemmNTPfiS_iS_iiffInternal_level2)
+ br label %return
+
+return: ; preds = %if.end21, %_ZNKSt5ctypeIcE5widenEc.exit92, %_ZNKSt5ctypeIcE5widenEc.exit80, %_ZNKSt5ctypeIcE5widenEc.exit
+ ret void
+}
+
+declare %"class.std::basic_ostream"* @_ZNSolsEi(%"class.std::basic_ostream"*, i32) #0
+
+; Function Attrs: nounwind uwtable
+define i32 @main(i32 %argc, i8** %argv) #2 {
+entry:
+ %argc.addr = alloca i32, align 4
+ %timers = alloca %struct.pb_TimerSet, align 8
+ %matArow = alloca i32, align 4
+ %matAcol = alloca i32, align 4
+ %matBrow = alloca i32, align 4
+ %matBcol = alloca i32, align 4
+ %matA = alloca %"class.std::vector", align 8
+ %matBT = alloca %"class.std::vector", align 8
+ %matC = alloca %"class.std::vector", align 8
+ store i32 %argc, i32* %argc.addr, align 4, !tbaa !8
+ %0 = bitcast %struct.pb_TimerSet* %timers to i8*
+ call void @llvm.lifetime.start(i64 800, i8* %0) #1
+ %1 = bitcast %"class.std::vector"* %matA to i8*
+ call void @llvm.memset.p0i8.i64(i8* %1, i8 0, i64 24, i32 8, i1 false) #1
+ %2 = bitcast %"class.std::vector"* %matBT to i8*
+ call void @llvm.memset.p0i8.i64(i8* %2, i8 0, i64 24, i32 8, i1 false) #1
+ %call = call %struct.pb_Parameters* @pb_ReadParameters(i32* %argc.addr, i8** %argv) #1
+ %inpFiles = getelementptr inbounds %struct.pb_Parameters* %call, i64 0, i32 1
+ %3 = load i8*** %inpFiles, align 8, !tbaa !7
+ %4 = load i8** %3, align 8, !tbaa !7
+ %cmp = icmp eq i8* %4, null
+ br i1 %cmp, label %if.then, label %lor.lhs.false
+
+lor.lhs.false: ; preds = %entry
+ %arrayidx2 = getelementptr inbounds i8** %3, i64 1
+ %5 = load i8** %arrayidx2, align 8, !tbaa !7
+ %cmp3 = icmp eq i8* %5, null
+ br i1 %cmp3, label %if.then, label %lor.lhs.false4
+
+lor.lhs.false4: ; preds = %lor.lhs.false
+ %arrayidx6 = getelementptr inbounds i8** %3, i64 2
+ %6 = load i8** %arrayidx6, align 8, !tbaa !7
+ %cmp7 = icmp eq i8* %6, null
+ br i1 %cmp7, label %if.then, label %lor.lhs.false8
+
+lor.lhs.false8: ; preds = %lor.lhs.false4
+ %arrayidx10 = getelementptr inbounds i8** %3, i64 3
+ %7 = load i8** %arrayidx10, align 8, !tbaa !7
+ %cmp11 = icmp eq i8* %7, null
+ br i1 %cmp11, label %if.end, label %if.then
+
+if.then: ; preds = %lor.lhs.false8, %lor.lhs.false4, %lor.lhs.false, %entry
+ %8 = load %struct._IO_FILE** @stderr, align 8, !tbaa !7
+ %9 = call i64 @fwrite(i8* getelementptr inbounds ([33 x i8]* @.str4, i64 0, i64 0), i64 32, i64 1, %struct._IO_FILE* %8)
+ call void @exit(i32 -1) #7
+ unreachable
+
+if.end: ; preds = %lor.lhs.false8
+ %call15 = call zeroext i1 @_Z22readColMajorMatrixFilePKcRiS1_RSt6vectorIfSaIfEE(i8* %4, i32* %matArow, i32* %matAcol, %"class.std::vector"* %matA) #1
+ %10 = load i8*** %inpFiles, align 8, !tbaa !7
+ %arrayidx17 = getelementptr inbounds i8** %10, i64 2
+ %11 = load i8** %arrayidx17, align 8, !tbaa !7
+ %call18 = call zeroext i1 @_Z22readColMajorMatrixFilePKcRiS1_RSt6vectorIfSaIfEE(i8* %11, i32* %matBcol, i32* %matBrow, %"class.std::vector"* %matBT) #1
+ call void @pb_InitializeTimerSet(%struct.pb_TimerSet* %timers) #1
+ %12 = call i8* @llvm_visc_initializeTimerSet()
+ store i8* %12, i8** @viscTimerSet_GenVISC
+ call void @llvm_visc_switchToTimer(i8** @viscTimerSet_GenVISC, i32 0)
+ call void @llvm.visc.init()
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 6) #1
+ %13 = load i32* %matArow, align 4, !tbaa !8
+ %14 = load i32* %matAcol, align 4, !tbaa !8
+ %mul = mul nsw i32 %14, %13
+ %conv = sext i32 %mul to i64
+ %mul19 = shl nsw i64 %conv, 2
+ %15 = load i32* %matBrow, align 4, !tbaa !8
+ %16 = load i32* %matBcol, align 4, !tbaa !8
+ %mul20 = mul nsw i32 %16, %15
+ %conv21 = sext i32 %mul20 to i64
+ %mul22 = shl nsw i64 %conv21, 2
+ %mul23 = mul nsw i32 %16, %13
+ %conv24 = sext i32 %mul23 to i64
+ %mul25 = shl nsw i64 %conv24, 2
+ %17 = bitcast %"class.std::vector"* %matC to i8*
+ call void @llvm.memset.p0i8.i64(i8* %17, i8 0, i64 24, i32 8, i1 false) #1
+ %cmp.i.i.i.i = icmp eq i32 %mul23, 0
+ br i1 %cmp.i.i.i.i, label %_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i, label %cond.true.i.i.i.i
+
+cond.true.i.i.i.i: ; preds = %if.end
+ %cmp.i.i.i.i.i = icmp slt i32 %mul23, 0
+ br i1 %cmp.i.i.i.i.i, label %if.then.i.i.i.i.i, label %_ZN9__gnu_cxx13new_allocatorIfE8allocateEmPKv.exit.i.i.i.i, !prof !9
+
+if.then.i.i.i.i.i: ; preds = %cond.true.i.i.i.i
+ call void @_ZSt17__throw_bad_allocv() #7
+ unreachable
+
+_ZN9__gnu_cxx13new_allocatorIfE8allocateEmPKv.exit.i.i.i.i: ; preds = %cond.true.i.i.i.i
+ %call2.i.i.i.i.i = call noalias i8* @_Znwm(i64 %mul25) #1
+ %18 = bitcast i8* %call2.i.i.i.i.i to float*
+ br label %_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i
+
+_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i: ; preds = %_ZN9__gnu_cxx13new_allocatorIfE8allocateEmPKv.exit.i.i.i.i, %if.end
+ %cond.i.i.i.i = phi float* [ %18, %_ZN9__gnu_cxx13new_allocatorIfE8allocateEmPKv.exit.i.i.i.i ], [ null, %if.end ]
+ %_M_start.i.i.i81 = getelementptr inbounds %"class.std::vector"* %matC, i64 0, i32 0, i32 0, i32 0
+ store float* %cond.i.i.i.i, float** %_M_start.i.i.i81, align 8, !tbaa !7
+ %_M_finish.i.i.i = getelementptr inbounds %"class.std::vector"* %matC, i64 0, i32 0, i32 0, i32 1
+ store float* %cond.i.i.i.i, float** %_M_finish.i.i.i, align 8, !tbaa !7
+ %add.ptr.i.i.i = getelementptr inbounds float* %cond.i.i.i.i, i64 %conv24
+ %_M_end_of_storage.i.i.i = getelementptr inbounds %"class.std::vector"* %matC, i64 0, i32 0, i32 0, i32 2
+ store float* %add.ptr.i.i.i, float** %_M_end_of_storage.i.i.i, align 8, !tbaa !7
+ br i1 %cmp.i.i.i.i, label %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit, label %for.body.lr.ph.i.i.i.i.i.i.i.i
+
+for.body.lr.ph.i.i.i.i.i.i.i.i: ; preds = %_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i
+ %n.mod.vf.i.i.i.i.i.i.i.i = and i64 %conv24, 7
+ %n.vec.i.i.i.i.i.i.i.i = sub i64 %conv24, %n.mod.vf.i.i.i.i.i.i.i.i
+ %cmp.zero.i.i.i.i.i.i.i.i = icmp eq i64 %n.mod.vf.i.i.i.i.i.i.i.i, %conv24
+ %ptr.ind.end.i.i.i.i.i.i.i.i = getelementptr float* %cond.i.i.i.i, i64 %n.vec.i.i.i.i.i.i.i.i
+ br i1 %cmp.zero.i.i.i.i.i.i.i.i, label %middle.block.i.i.i.i.i.i.i.i, label %vector.body.i.i.i.i.i.i.i.i
+
+vector.body.i.i.i.i.i.i.i.i: ; preds = %vector.body.i.i.i.i.i.i.i.i, %for.body.lr.ph.i.i.i.i.i.i.i.i
+ %index.i.i.i.i.i.i.i.i = phi i64 [ %index.next.i.i.i.i.i.i.i.i, %vector.body.i.i.i.i.i.i.i.i ], [ 0, %for.body.lr.ph.i.i.i.i.i.i.i.i ]
+ %next.gep.i.i.i.i.i.i.i.i = getelementptr float* %cond.i.i.i.i, i64 %index.i.i.i.i.i.i.i.i
+ %19 = bitcast float* %next.gep.i.i.i.i.i.i.i.i to <4 x float>*
+ store <4 x float> zeroinitializer, <4 x float>* %19, align 4
+ %next.gep.sum41.i.i.i.i.i.i.i.i = or i64 %index.i.i.i.i.i.i.i.i, 4
+ %20 = getelementptr float* %cond.i.i.i.i, i64 %next.gep.sum41.i.i.i.i.i.i.i.i
+ %21 = bitcast float* %20 to <4 x float>*
+ store <4 x float> zeroinitializer, <4 x float>* %21, align 4
+ %index.next.i.i.i.i.i.i.i.i = add i64 %index.i.i.i.i.i.i.i.i, 8
+ %22 = icmp eq i64 %index.next.i.i.i.i.i.i.i.i, %n.vec.i.i.i.i.i.i.i.i
+ br i1 %22, label %middle.block.i.i.i.i.i.i.i.i, label %vector.body.i.i.i.i.i.i.i.i
+
+middle.block.i.i.i.i.i.i.i.i: ; preds = %vector.body.i.i.i.i.i.i.i.i, %for.body.lr.ph.i.i.i.i.i.i.i.i
+ %resume.val.i.i.i.i.i.i.i.i = phi float* [ %cond.i.i.i.i, %for.body.lr.ph.i.i.i.i.i.i.i.i ], [ %ptr.ind.end.i.i.i.i.i.i.i.i, %vector.body.i.i.i.i.i.i.i.i ]
+ %resume.val7.i.i.i.i.i.i.i.i = phi i64 [ %conv24, %for.body.lr.ph.i.i.i.i.i.i.i.i ], [ %n.mod.vf.i.i.i.i.i.i.i.i, %vector.body.i.i.i.i.i.i.i.i ]
+ %new.indc.resume.val.i.i.i.i.i.i.i.i = phi i64 [ 0, %for.body.lr.ph.i.i.i.i.i.i.i.i ], [ %n.vec.i.i.i.i.i.i.i.i, %vector.body.i.i.i.i.i.i.i.i ]
+ %cmp.n.i.i.i.i.i.i.i.i = icmp eq i64 %new.indc.resume.val.i.i.i.i.i.i.i.i, %conv24
+ br i1 %cmp.n.i.i.i.i.i.i.i.i, label %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit, label %for.body.i.i.i.i.i.i.i.i.preheader
+
+for.body.i.i.i.i.i.i.i.i.preheader: ; preds = %middle.block.i.i.i.i.i.i.i.i
+ %resume.val.i.i.i.i.i.i.i.i101 = bitcast float* %resume.val.i.i.i.i.i.i.i.i to i8*
+ %23 = shl nsw i64 %resume.val7.i.i.i.i.i.i.i.i, 2
+ call void @llvm.memset.p0i8.i64(i8* %resume.val.i.i.i.i.i.i.i.i101, i8 0, i64 %23, i32 4, i1 false)
+ br label %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit
+
+_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit: ; preds = %for.body.i.i.i.i.i.i.i.i.preheader, %middle.block.i.i.i.i.i.i.i.i, %_ZNSt12_Vector_baseIfSaIfEEC2EmRKS0_.exit.i.i
+ store float* %add.ptr.i.i.i, float** %_M_finish.i.i.i, align 8, !tbaa !7
+ %_M_start.i.i = getelementptr inbounds %"class.std::vector"* %matA, i64 0, i32 0, i32 0, i32 0
+ %24 = load float** %_M_start.i.i, align 8, !tbaa !7
+ %25 = bitcast float* %24 to i8*
+ call void @llvm_visc_track_mem(i8* %25, i64 %mul19) #1
+ %_M_start.i.i82 = getelementptr inbounds %"class.std::vector"* %matBT, i64 0, i32 0, i32 0, i32 0
+ %26 = load float** %_M_start.i.i82, align 8, !tbaa !7
+ %27 = bitcast float* %26 to i8*
+ call void @llvm_visc_track_mem(i8* %27, i64 %mul22) #1
+ %28 = load float** %_M_start.i.i.i81, align 8, !tbaa !7
+ %29 = bitcast float* %28 to i8*
+ call void @llvm_visc_track_mem(i8* %29, i64 %mul25) #1
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 6) #1
+ %30 = load float** %_M_finish.i.i.i, align 8, !tbaa !7
+ %31 = load float** %_M_start.i.i.i81, align 8, !tbaa !7
+ %cmp3399 = icmp eq float* %30, %31
+ br i1 %cmp3399, label %for.end, label %for.body.lr.ph
+
+for.body.lr.ph: ; preds = %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit
+ %sub.ptr.lhs.cast.i = ptrtoint float* %30 to i64
+ %sub.ptr.rhs.cast.i = ptrtoint float* %31 to i64
+ %sub.ptr.sub.i = sub i64 %sub.ptr.lhs.cast.i, %sub.ptr.rhs.cast.i
+ %sub.ptr.div.i = ashr exact i64 %sub.ptr.sub.i, 2
+ br label %for.body
+
+for.body: ; preds = %for.body, %for.body.lr.ph
+ %i.0100 = phi i64 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
+ %add.ptr.i = getelementptr inbounds float* %31, i64 %i.0100
+ store float 0.000000e+00, float* %add.ptr.i, align 4, !tbaa !3
+ %inc = add i64 %i.0100, 1
+ %cmp33 = icmp ult i64 %inc, %sub.ptr.div.i
+ br i1 %cmp33, label %for.body, label %for.end
+
+for.end: ; preds = %for.body, %_ZNSt6vectorIfSaIfEEC1EmRKfRKS0_.exit
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 0) #1
+ %32 = load i32* %matArow, align 4, !tbaa !8
+ %33 = load i32* %matBcol, align 4, !tbaa !8
+ %34 = load i32* %matAcol, align 4, !tbaa !8
+ %35 = load float** %_M_start.i.i, align 8, !tbaa !7
+ %36 = load float** %_M_start.i.i82, align 8, !tbaa !7
+ %37 = load float** %_M_start.i.i.i81, align 8, !tbaa !7
+ call void @_Z10basicSgemmcciiifPfmiS_mifS_mi(i8 signext 78, i8 signext 84, i32 %32, i32 %33, i32 %34, float 1.000000e+00, float* %35, i64 %mul19, i32 %32, float* %36, i64 %mul22, i32 %33, float 0.000000e+00, float* %37, i64 %mul25, i32 %32)
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 3) #1
+ %38 = load float** %_M_start.i.i.i81, align 8, !tbaa !7
+ %39 = bitcast float* %38 to i8*
+ call void @llvm_visc_request_mem(i8* %39, i64 %mul25) #1
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 16) #1
+ %40 = load float** %_M_start.i.i, align 8, !tbaa !7
+ %41 = bitcast float* %40 to i8*
+ call void @llvm_visc_untrack_mem(i8* %41) #1
+ %42 = load float** %_M_start.i.i82, align 8, !tbaa !7
+ %43 = bitcast float* %42 to i8*
+ call void @llvm_visc_untrack_mem(i8* %43) #1
+ %44 = load float** %_M_start.i.i.i81, align 8, !tbaa !7
+ %45 = bitcast float* %44 to i8*
+ call void @llvm_visc_untrack_mem(i8* %45) #1
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 0) #1
+ call void @pb_PrintTimerSet(%struct.pb_TimerSet* %timers) #1
+ %Ptr = getelementptr [14 x i8]* @0, i64 0, i64 0
+ call void @llvm_visc_printTimerSet(i8** @viscTimerSet_GenVISC, i8* %Ptr)
+ call void @llvm.visc.cleanup()
+ %outFile = getelementptr inbounds %struct.pb_Parameters* %call, i64 0, i32 0
+ %46 = load i8** %outFile, align 8, !tbaa !7
+ %tobool = icmp eq i8* %46, null
+ br i1 %tobool, label %if.end45, label %if.then42
+
+if.then42: ; preds = %for.end
+ %47 = load i32* %matArow, align 4, !tbaa !8
+ %48 = load i32* %matBcol, align 4, !tbaa !8
+ %call44 = call zeroext i1 @_Z23writeColMajorMatrixFilePKciiRSt6vectorIfSaIfEE(i8* %46, i32 %47, i32 %48, %"class.std::vector"* %matC) #1
+ br label %if.end45
+
+if.end45: ; preds = %if.then42, %for.end
+ %arrayidx47 = getelementptr inbounds %struct.pb_TimerSet* %timers, i64 0, i32 4, i64 2
+ %call48 = call double @pb_GetElapsedTime(%struct.pb_Timer* %arrayidx47) #1
+ %call1.i88 = call %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"* @_ZSt4cout, i8* getelementptr inbounds ([10 x i8]* @.str5, i64 0, i64 0), i64 9) #1
+ %49 = load i32* %matArow, align 4, !tbaa !8
+ %conv50 = sitofp i32 %49 to double
+ %mul51 = fmul fast double %conv50, 2.000000e+00
+ %50 = load i32* %matBcol, align 4, !tbaa !8
+ %conv52 = sitofp i32 %50 to double
+ %mul53 = fmul fast double %mul51, %conv52
+ %51 = load i32* %matAcol, align 4, !tbaa !8
+ %conv54 = sitofp i32 %51 to double
+ %mul55 = fmul fast double %mul53, %conv54
+ %div = fdiv fast double %mul55, %call48
+ %div56 = fmul double %div, 1.000000e-09
+ %call.i = call %"class.std::basic_ostream"* @_ZNSo9_M_insertIdEERSoT_(%"class.std::basic_ostream"* @_ZSt4cout, double %div56) #1
+ %52 = bitcast %"class.std::basic_ostream"* %call.i to i8**
+ %vtable.i = load i8** %52, align 8, !tbaa !6
+ %vbase.offset.ptr.i = getelementptr i8* %vtable.i, i64 -24
+ %53 = bitcast i8* %vbase.offset.ptr.i to i64*
+ %vbase.offset.i = load i64* %53, align 8
+ %54 = bitcast %"class.std::basic_ostream"* %call.i to i8*
+ %add.ptr.sum.i = add i64 %vbase.offset.i, 240
+ %_M_ctype.i.i = getelementptr inbounds i8* %54, i64 %add.ptr.sum.i
+ %55 = bitcast i8* %_M_ctype.i.i to %"class.std::ctype"**
+ %56 = load %"class.std::ctype"** %55, align 8, !tbaa !7
+ %tobool.i.i.i = icmp eq %"class.std::ctype"* %56, null
+ br i1 %tobool.i.i.i, label %if.then.i.i.i, label %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit.i.i
+
+if.then.i.i.i: ; preds = %if.end45
+ call void @_ZSt16__throw_bad_castv() #7
+ unreachable
+
+_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit.i.i: ; preds = %if.end45
+ %_M_widen_ok.i.i.i = getelementptr inbounds %"class.std::ctype"* %56, i64 0, i32 6
+ %57 = load i8* %_M_widen_ok.i.i.i, align 1, !tbaa !4
+ %tobool.i3.i.i = icmp eq i8 %57, 0
+ br i1 %tobool.i3.i.i, label %if.end.i.i.i, label %if.then.i4.i.i
+
+if.then.i4.i.i: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit.i.i
+ %arrayidx.i.i.i = getelementptr inbounds %"class.std::ctype"* %56, i64 0, i32 7, i64 10
+ %58 = load i8* %arrayidx.i.i.i, align 1, !tbaa !4
+ br label %_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit
+
+if.end.i.i.i: ; preds = %_ZSt13__check_facetISt5ctypeIcEERKT_PS3_.exit.i.i
+ call void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"* %56) #1
+ %59 = bitcast %"class.std::ctype"* %56 to i8 (%"class.std::ctype"*, i8)***
+ %vtable.i.i.i = load i8 (%"class.std::ctype"*, i8)*** %59, align 8, !tbaa !6
+ %vfn.i.i.i = getelementptr inbounds i8 (%"class.std::ctype"*, i8)** %vtable.i.i.i, i64 6
+ %60 = load i8 (%"class.std::ctype"*, i8)** %vfn.i.i.i, align 8
+ %call.i.i.i = call signext i8 %60(%"class.std::ctype"* %56, i8 signext 10) #1
+ br label %_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit
+
+_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit: ; preds = %if.end.i.i.i, %if.then.i4.i.i
+ %retval.0.i.i.i = phi i8 [ %58, %if.then.i4.i.i ], [ %call.i.i.i, %if.end.i.i.i ]
+ %call1.i = call %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"* %call.i, i8 signext %retval.0.i.i.i) #1
+ %call.i.i = call %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"* %call1.i) #1
+ call void @pb_FreeParameters(%struct.pb_Parameters* %call) #1
+ %61 = load float** %_M_start.i.i.i81, align 8, !tbaa !7
+ %tobool.i.i.i.i78 = icmp eq float* %61, null
+ br i1 %tobool.i.i.i.i78, label %_ZNSt6vectorIfSaIfEED1Ev.exit80, label %if.then.i.i.i.i79
+
+if.then.i.i.i.i79: ; preds = %_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit
+ %62 = bitcast float* %61 to i8*
+ call void @_ZdlPv(i8* %62) #1
+ br label %_ZNSt6vectorIfSaIfEED1Ev.exit80
+
+_ZNSt6vectorIfSaIfEED1Ev.exit80: ; preds = %if.then.i.i.i.i79, %_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_.exit
+ %63 = load float** %_M_start.i.i82, align 8, !tbaa !7
+ %tobool.i.i.i.i74 = icmp eq float* %63, null
+ br i1 %tobool.i.i.i.i74, label %_ZNSt6vectorIfSaIfEED1Ev.exit76, label %if.then.i.i.i.i75
+
+if.then.i.i.i.i75: ; preds = %_ZNSt6vectorIfSaIfEED1Ev.exit80
+ %64 = bitcast float* %63 to i8*
+ call void @_ZdlPv(i8* %64) #1
+ br label %_ZNSt6vectorIfSaIfEED1Ev.exit76
+
+_ZNSt6vectorIfSaIfEED1Ev.exit76: ; preds = %if.then.i.i.i.i75, %_ZNSt6vectorIfSaIfEED1Ev.exit80
+ %65 = load float** %_M_start.i.i, align 8, !tbaa !7
+ %tobool.i.i.i.i = icmp eq float* %65, null
+ br i1 %tobool.i.i.i.i, label %_ZNSt6vectorIfSaIfEED1Ev.exit, label %if.then.i.i.i.i
+
+if.then.i.i.i.i: ; preds = %_ZNSt6vectorIfSaIfEED1Ev.exit76
+ %66 = bitcast float* %65 to i8*
+ call void @_ZdlPv(i8* %66) #1
+ br label %_ZNSt6vectorIfSaIfEED1Ev.exit
+
+_ZNSt6vectorIfSaIfEED1Ev.exit: ; preds = %if.then.i.i.i.i, %_ZNSt6vectorIfSaIfEED1Ev.exit76
+ call void @llvm.lifetime.end(i64 800, i8* %0) #1
+ ret i32 0
+}
+
+; Function Attrs: nounwind
+declare void @llvm.lifetime.start(i64, i8* nocapture) #1
+
+declare %struct.pb_Parameters* @pb_ReadParameters(i32*, i8**) #0
+
+; Function Attrs: noreturn nounwind
+declare void @exit(i32) #4
+
+declare zeroext i1 @_Z22readColMajorMatrixFilePKcRiS1_RSt6vectorIfSaIfEE(i8*, i32*, i32*, %"class.std::vector"*) #0
+
+declare void @pb_InitializeTimerSet(%struct.pb_TimerSet*) #0
+
+declare void @pb_SwitchToTimer(%struct.pb_TimerSet*, i32) #0
+
+declare void @llvm_visc_track_mem(i8*, i64) #0
+
+declare void @llvm_visc_request_mem(i8*, i64) #0
+
+declare void @llvm_visc_untrack_mem(i8*) #0
+
+declare void @pb_PrintTimerSet(%struct.pb_TimerSet*) #0
+
+declare zeroext i1 @_Z23writeColMajorMatrixFilePKciiRSt6vectorIfSaIfEE(i8*, i32, i32, %"class.std::vector"*) #0
+
+declare double @pb_GetElapsedTime(%struct.pb_Timer*) #0
+
+declare void @pb_FreeParameters(%struct.pb_Parameters*) #0
+
+; Function Attrs: nounwind
+declare void @llvm.lifetime.end(i64, i8* nocapture) #1
+
+declare %"class.std::basic_ostream"* @_ZNSo9_M_insertIdEERSoT_(%"class.std::basic_ostream"*, double) #0
+
+; Function Attrs: noreturn
+declare void @_ZSt17__throw_bad_allocv() #5
+
+declare noalias i8* @_Znwm(i64) #0
+
+; Function Attrs: nounwind
+declare void @_ZdlPv(i8*) #6
+
+declare %"class.std::basic_ostream"* @_ZNSo3putEc(%"class.std::basic_ostream"*, i8 signext) #0
+
+declare void @_ZNKSt5ctypeIcE13_M_widen_initEv(%"class.std::ctype"*) #0
+
+; Function Attrs: noreturn
+declare void @_ZSt16__throw_bad_castv() #5
+
+declare %"class.std::basic_ostream"* @_ZNSo5flushEv(%"class.std::basic_ostream"*) #0
+
+declare %"class.std::basic_ostream"* @_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l(%"class.std::basic_ostream"*, i8*, i64) #0
+
+; Function Attrs: nounwind
+define internal void @_GLOBAL__I_a() #1 section ".text.startup" {
+entry:
+ tail call void @_ZNSt8ios_base4InitC1Ev(%"class.std::ios_base::Init"* @_ZStL8__ioinit) #1
+ %0 = tail call i32 @__cxa_atexit(void (i8*)* bitcast (void (%"class.std::ios_base::Init"*)* @_ZNSt8ios_base4InitD1Ev to void (i8*)*), i8* getelementptr inbounds (%"class.std::ios_base::Init"* @_ZStL8__ioinit, i64 0, i32 0), i8* @__dso_handle) #1
+ ret void
+}
+
+; Function Attrs: nounwind
+declare i64 @fwrite(i8* nocapture, i64, i64, %struct._IO_FILE* nocapture) #1
+
+; Function Attrs: nounwind
+declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i32, i1) #1
+
+declare i8* @llvm_visc_initializeTimerSet()
+
+declare void @llvm_visc_switchToTimer(i8**, i32)
+
+declare void @llvm_visc_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.getNode() #1
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.getParentNode(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNodeInstanceID.y(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNumNodeInstances.x(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNodeInstanceID.x(i8*) #1
+
+; Function Attrs: nounwind uwtable
+define %rtype @_Z9mysgemmNTPfiS_iS_iiffInternal_level1(float* in %A, i64 %bytes_A, i32 %lda, float* in %B, i64 %bytes_B, i32 %ldb, float* in out %C, i64 %bytes_C, i32 %ldc, i32 %k, float %alpha, float %beta, i32 %dimX, i32 %dimY) #2 {
+entry:
+ %_Z9mysgemmNTPfiS_iS_iiff.node = call i8* @llvm.visc.createNode2D(i8* bitcast (%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float)* @_Z9mysgemmNTPfiS_iS_iiff to i8*), i32 %dimX, i32 %dimY)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 1, i32 1)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 2, i32 2)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 3, i32 3)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 4, i32 4)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 5, i32 5)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 6, i32 6)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 7, i32 7)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 8, i32 8)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 9, i32 9)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 10, i32 10)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiff.node, i32 11, i32 11)
+ ret %rtype undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.createNode2D(i8*, i32, i32) #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.bind.input(i8*, i32, i32) #1
+
+; Function Attrs: nounwind uwtable
+define %rtype @_Z9mysgemmNTPfiS_iS_iiffInternal_level2(float* in %A, i64 %bytes_A, i32 %lda, float* in %B, i64 %bytes_B, i32 %ldb, float* in out %C, i64 %bytes_C, i32 %ldc, i32 %k, float %alpha, float %beta, i32 %dimX, i32 %dimY, i32 %dimX1, i32 %dimY2) #2 {
+entry:
+ %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node = call i8* @llvm.visc.createNode2D(i8* bitcast (%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float, i32, i32)* @_Z9mysgemmNTPfiS_iS_iiffInternal_level1 to i8*), i32 %dimX1, i32 %dimY2)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 1, i32 1)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 2, i32 2)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 3, i32 3)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 4, i32 4)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 5, i32 5)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 6, i32 6)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 7, i32 7)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 8, i32 8)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 9, i32 9)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 10, i32 10)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 11, i32 11)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 12, i32 12)
+ call void @llvm.visc.bind.input(i8* %_Z9mysgemmNTPfiS_iS_iiffInternal_level1.node, i32 13, i32 13)
+ ret %rtype undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.launch(i8*, i8*) #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.wait(i8*) #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.init() #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.cleanup() #1
+
+attributes #0 = { "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { nounwind }
+attributes #2 = { nounwind uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { noinline nounwind uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #4 = { noreturn nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #5 = { noreturn "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #6 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #7 = { noreturn nounwind }
+
+!visc_hint_gpu = !{!0, !1}
+!visc_hint_cpu = !{!2}
+
+!0 = metadata !{%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float)* @_Z9mysgemmNTPfiS_iS_iiff}
+!1 = metadata !{%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float, i32, i32)* @_Z9mysgemmNTPfiS_iS_iiffInternal_level1}
+!2 = metadata !{%rtype (float*, i64, i32, float*, i64, i32, float*, i64, i32, i32, float, float, i32, i32, i32, i32)* @_Z9mysgemmNTPfiS_iS_iiffInternal_level2}
+!3 = metadata !{metadata !"float", metadata !4}
+!4 = metadata !{metadata !"omnipotent char", metadata !5}
+!5 = metadata !{metadata !"Simple C/C++ TBAA"}
+!6 = metadata !{metadata !"vtable pointer", metadata !5}
+!7 = metadata !{metadata !"any pointer", metadata !4}
+!8 = metadata !{metadata !"int", metadata !4}
+!9 = metadata !{metadata !"branch_weights", i32 4, i32 64}
diff --git a/hpvm/test/parboil/benchmarks/spmv/Makefile b/hpvm/test/parboil/benchmarks/spmv/Makefile
index aff3e54712..23e1d49900 100644
--- a/hpvm/test/parboil/benchmarks/spmv/Makefile
+++ b/hpvm/test/parboil/benchmarks/spmv/Makefile
@@ -1,9 +1,9 @@
PARBOIL_ROOT = $(LLVM_SRC_ROOT)/tools/hpvm/test/parboil
APP = spmv
-# Default compile hpvm
+# Default compile visc
ifeq ($(VERSION),)
- VERSION = hpvm
+ VERSION = visc
endif
# Default use small test case
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu/main.c b/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu/main.c
index 8bff8a1d0a..a19184a965 100644
--- a/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu/main.c
+++ b/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu/main.c
@@ -81,7 +81,7 @@ int main(int argc, char **argv) {
input_vec(parameters->inpFiles[1], h_x_vector, dim);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
// parameters declaration
cl_int clStatus;
@@ -127,7 +127,7 @@ int main(int argc, char **argv) {
cl_mem jds_ptr_int;
cl_mem sh_zcnt_int;
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
/*const char* clSource[] = {readFile("src/opencl_base/kernel.cl")};*/
/*cl_program clProgram =
* clCreateProgramWithSource(clContext,1,clSource,NULL,&clStatus);*/
@@ -217,7 +217,7 @@ int main(int argc, char **argv) {
// printf("!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!grid is %d and block is
// %d=\n",grid,block); printf("!!! dim is %d\n",dim);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 0, sizeof(cl_mem), &d_Ax_vector);
CHECK_ERROR("clSetKernelArg")
clStatus = clSetKernelArg(clKernel, 1, sizeof(cl_mem), &d_data);
@@ -240,7 +240,7 @@ int main(int argc, char **argv) {
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
int i;
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (int j = 0; j < 20; j++) {
for (i = 0; i < 50; i++) {
/*pb_SwitchToTimer(&timers, pb_TimerID_KERNEL);*/
@@ -260,7 +260,7 @@ int main(int argc, char **argv) {
dim * sizeof(float), h_Ax_vector, 0, NULL, NULL);
CHECK_ERROR("clEnqueueReadBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseKernel(clKernel);
clStatus = clReleaseProgram(clProgram);
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu_baseline/main.c b/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu_baseline/main.c
index f704f96ed2..d4fc026b73 100644
--- a/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu_baseline/main.c
+++ b/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu_baseline/main.c
@@ -236,7 +236,7 @@ int main(int argc, char **argv) {
// main execution
int i;
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
/*for(int j=0; j<20; j++) {*/
for (i = 0; i < 50; i++) {
clStatus = clEnqueueNDRangeKernel(clCommandQueue, clKernel, 1, NULL, &grid,
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu_huge/main.c b/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu_huge/main.c
index a6fe5012f9..42ffab597d 100644
--- a/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu_huge/main.c
+++ b/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu_huge/main.c
@@ -81,7 +81,7 @@ int main(int argc, char **argv) {
input_vec(parameters->inpFiles[1], h_x_vector, dim);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
// parameters declaration
cl_int clStatus;
@@ -127,7 +127,7 @@ int main(int argc, char **argv) {
cl_mem jds_ptr_int;
cl_mem sh_zcnt_int;
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
/*const char* clSource[] = {readFile("src/opencl_base/kernel.cl")};*/
/*cl_program clProgram =
* clCreateProgramWithSource(clContext,1,clSource,NULL,&clStatus);*/
@@ -217,7 +217,7 @@ int main(int argc, char **argv) {
// printf("!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!grid is %d and block is
// %d=\n",grid,block); printf("!!! dim is %d\n",dim);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 0, sizeof(cl_mem), &d_Ax_vector);
CHECK_ERROR("clSetKernelArg")
clStatus = clSetKernelArg(clKernel, 1, sizeof(cl_mem), &d_data);
@@ -240,7 +240,7 @@ int main(int argc, char **argv) {
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
int i;
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (int j = 0; j < 1; j++) {
for (i = 0; i < 50; i++) {
/*pb_SwitchToTimer(&timers, pb_TimerID_KERNEL);*/
@@ -260,7 +260,7 @@ int main(int argc, char **argv) {
dim * sizeof(float), h_Ax_vector, 0, NULL, NULL);
CHECK_ERROR("clEnqueueReadBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseKernel(clKernel);
clStatus = clReleaseProgram(clProgram);
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu_large/main.c b/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu_large/main.c
index bc3655c4ab..fbd272b32f 100644
--- a/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu_large/main.c
+++ b/hpvm/test/parboil/benchmarks/spmv/src/opencl_cpu_large/main.c
@@ -81,7 +81,7 @@ int main(int argc, char **argv) {
input_vec(parameters->inpFiles[1], h_x_vector, dim);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
// parameters declaration
cl_int clStatus;
@@ -127,7 +127,7 @@ int main(int argc, char **argv) {
cl_mem jds_ptr_int;
cl_mem sh_zcnt_int;
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
/*const char* clSource[] = {readFile("src/opencl_base/kernel.cl")};*/
/*cl_program clProgram =
* clCreateProgramWithSource(clContext,1,clSource,NULL,&clStatus);*/
@@ -217,7 +217,7 @@ int main(int argc, char **argv) {
// printf("!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!grid is %d and block is
// %d=\n",grid,block); printf("!!! dim is %d\n",dim);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 0, sizeof(cl_mem), &d_Ax_vector);
CHECK_ERROR("clSetKernelArg")
clStatus = clSetKernelArg(clKernel, 1, sizeof(cl_mem), &d_data);
@@ -240,7 +240,7 @@ int main(int argc, char **argv) {
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
int i;
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (int j = 0; j < 20; j++) {
for (i = 0; i < 50; i++) {
/*pb_SwitchToTimer(&timers, pb_TimerID_KERNEL);*/
@@ -260,7 +260,7 @@ int main(int argc, char **argv) {
dim * sizeof(float), h_Ax_vector, 0, NULL, NULL);
CHECK_ERROR("clEnqueueReadBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseKernel(clKernel);
clStatus = clReleaseProgram(clProgram);
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/opencl_nvidia/main.c b/hpvm/test/parboil/benchmarks/spmv/src/opencl_nvidia/main.c
index 88fd0c878b..343814149a 100644
--- a/hpvm/test/parboil/benchmarks/spmv/src/opencl_nvidia/main.c
+++ b/hpvm/test/parboil/benchmarks/spmv/src/opencl_nvidia/main.c
@@ -224,7 +224,7 @@ int main(int argc, char **argv) {
// main execution
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
int i;
for (i = 0; i < 50; i++) {
clStatus = clEnqueueNDRangeKernel(clCommandQueue, clKernel, 1, NULL, &grid,
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/opencl_nvidia_huge/main.c b/hpvm/test/parboil/benchmarks/spmv/src/opencl_nvidia_huge/main.c
index ca538e3a95..4600a3e6b8 100644
--- a/hpvm/test/parboil/benchmarks/spmv/src/opencl_nvidia_huge/main.c
+++ b/hpvm/test/parboil/benchmarks/spmv/src/opencl_nvidia_huge/main.c
@@ -83,7 +83,7 @@ int main(int argc, char **argv) {
printf("Col count = %d, dim = %d\n", col_count, dim);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_platform_id clPlatform;
@@ -137,7 +137,7 @@ int main(int argc, char **argv) {
cl_mem jds_ptr_int;
cl_mem sh_zcnt_int;
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCLDeviceProp clDeviceProp;
clStatus = clGetDeviceInfo(clDevice, CL_DEVICE_COMPUTE_CAPABILITY_MAJOR_NV,
@@ -215,7 +215,7 @@ int main(int argc, char **argv) {
compute_active_thread(&block, &grid, nzcnt_len, pad, clDeviceProp.major,
clDeviceProp.minor, clDeviceProp.multiProcessorCount);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 0, sizeof(cl_mem), &d_Ax_vector);
CHECK_ERROR("clSetKernelArg")
clStatus = clSetKernelArg(clKernel, 1, sizeof(cl_mem), &d_data);
@@ -237,7 +237,7 @@ int main(int argc, char **argv) {
// main execution
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
int i;
for (int j = 0; j < 5; j++) {
for (i = 0; i < 50; i++) {
@@ -260,7 +260,7 @@ int main(int argc, char **argv) {
dim * sizeof(float), h_Ax_vector, 0, NULL, NULL);
CHECK_ERROR("clEnqueueReadBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseKernel(clKernel);
clStatus = clReleaseProgram(clProgram);
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/opencl_nvidia_large/main.c b/hpvm/test/parboil/benchmarks/spmv/src/opencl_nvidia_large/main.c
index 21973c2fa7..d2375af91d 100644
--- a/hpvm/test/parboil/benchmarks/spmv/src/opencl_nvidia_large/main.c
+++ b/hpvm/test/parboil/benchmarks/spmv/src/opencl_nvidia_large/main.c
@@ -83,7 +83,7 @@ int main(int argc, char **argv) {
printf("Col count = %d, dim = %d\n", col_count, dim);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_platform_id clPlatform;
@@ -137,7 +137,7 @@ int main(int argc, char **argv) {
cl_mem jds_ptr_int;
cl_mem sh_zcnt_int;
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
OpenCLDeviceProp clDeviceProp;
clStatus = clGetDeviceInfo(clDevice, CL_DEVICE_COMPUTE_CAPABILITY_MAJOR_NV,
@@ -215,7 +215,7 @@ int main(int argc, char **argv) {
compute_active_thread(&block, &grid, nzcnt_len, pad, clDeviceProp.major,
clDeviceProp.minor, clDeviceProp.multiProcessorCount);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 0, sizeof(cl_mem), &d_Ax_vector);
CHECK_ERROR("clSetKernelArg")
clStatus = clSetKernelArg(clKernel, 1, sizeof(cl_mem), &d_data);
@@ -237,7 +237,7 @@ int main(int argc, char **argv) {
// main execution
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
int i;
for (int j = 0; j < 100; j++) {
for (i = 0; i < 50; i++) {
@@ -260,7 +260,7 @@ int main(int argc, char **argv) {
dim * sizeof(float), h_Ax_vector, 0, NULL, NULL);
CHECK_ERROR("clEnqueueReadBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseKernel(clKernel);
clStatus = clReleaseProgram(clProgram);
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/hpvm/Makefile b/hpvm/test/parboil/benchmarks/spmv/src/visc/Makefile
similarity index 88%
rename from hpvm/test/parboil/benchmarks/spmv/src/hpvm/Makefile
rename to hpvm/test/parboil/benchmarks/spmv/src/visc/Makefile
index 06af6bebea..a289d68f34 100644
--- a/hpvm/test/parboil/benchmarks/spmv/src/hpvm/Makefile
+++ b/hpvm/test/parboil/benchmarks/spmv/src/visc/Makefile
@@ -1,9 +1,9 @@
# (c) 2010 The Board of Trustees of the University of Illinois.
-LANGUAGE=hpvm
+LANGUAGE=visc
TOOLS_SRC=common_src/convert-dataset
SRCDIR_OBJS=gpu_info.ll file.ll
-HPVM_OBJS=main.hpvm.ll
+VISC_OBJS=main.visc.ll
APP_CUDALDFLAGS=-lm
APP_CFLAGS=-ffast-math -O1 -I$(TOOLS_SRC)
APP_CXXFLAGS=-ffast-math -O1 -I$(TOOLS_SRC)
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/hpvm/file.cpp b/hpvm/test/parboil/benchmarks/spmv/src/visc/file.cpp
similarity index 100%
rename from hpvm/test/parboil/benchmarks/spmv/src/hpvm/file.cpp
rename to hpvm/test/parboil/benchmarks/spmv/src/visc/file.cpp
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/hpvm/file.h b/hpvm/test/parboil/benchmarks/spmv/src/visc/file.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/spmv/src/hpvm/file.h
rename to hpvm/test/parboil/benchmarks/spmv/src/visc/file.h
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/hpvm/gpu_info.cpp b/hpvm/test/parboil/benchmarks/spmv/src/visc/gpu_info.cpp
similarity index 100%
rename from hpvm/test/parboil/benchmarks/spmv/src/hpvm/gpu_info.cpp
rename to hpvm/test/parboil/benchmarks/spmv/src/visc/gpu_info.cpp
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/hpvm/gpu_info.h b/hpvm/test/parboil/benchmarks/spmv/src/visc/gpu_info.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/spmv/src/hpvm/gpu_info.h
rename to hpvm/test/parboil/benchmarks/spmv/src/visc/gpu_info.h
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/hpvm/kernel.cl b/hpvm/test/parboil/benchmarks/spmv/src/visc/kernel.cl
similarity index 100%
rename from hpvm/test/parboil/benchmarks/spmv/src/hpvm/kernel.cl
rename to hpvm/test/parboil/benchmarks/spmv/src/visc/kernel.cl
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/hpvm/main.cpp b/hpvm/test/parboil/benchmarks/spmv/src/visc/main.cpp
similarity index 68%
rename from hpvm/test/parboil/benchmarks/spmv/src/hpvm/main.cpp
rename to hpvm/test/parboil/benchmarks/spmv/src/visc/main.cpp
index 4414744b49..f6ce5ccfb2 100644
--- a/hpvm/test/parboil/benchmarks/spmv/src/hpvm/main.cpp
+++ b/hpvm/test/parboil/benchmarks/spmv/src/visc/main.cpp
@@ -8,11 +8,11 @@
//#include <CL/cl.h>
//#include <CL/cl_ext.h>
-#include <hpvm.h>
#include <parboil.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
+#include <visc.h>
#include "convert_dataset.h"
#include "file.h"
@@ -54,15 +54,15 @@ void spmv_jds(float *dst_vector, size_t bytes_dst_vector, float *d_data,
size_t bytes_x_vec, int dim, int *jds_ptr_int,
size_t bytes_jds_ptr_int, int *sh_zcnt_int,
size_t bytes_sh_zcnt_int) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(7, dst_vector, d_data, d_index, d_perm, x_vec, jds_ptr_int,
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(7, dst_vector, d_data, d_index, d_perm, x_vec, jds_ptr_int,
sh_zcnt_int, 1, dst_vector);
- void *thisNode = __hpvm__getNode();
- void *parentNode = __hpvm__getParentNode(thisNode);
- int lx = __hpvm__getNodeInstanceID_x(thisNode);
- int gx = __hpvm__getNodeInstanceID_x(parentNode);
- int gridx = __hpvm__getNumNodeInstances_x(thisNode);
+ void *thisNode = __visc__getNode();
+ void *parentNode = __visc__getParentNode(thisNode);
+ int lx = __visc__getNodeInstanceID_x(thisNode);
+ int gx = __visc__getNodeInstanceID_x(parentNode);
+ int gridx = __visc__getNumNodeInstances_x(thisNode);
int ix = gx * gridx + lx;
int warp_id = ix >> WARP_BITS;
@@ -126,25 +126,25 @@ void spmvLvl1(float *dst_vector, size_t bytes_dst_vector, float *d_data,
size_t bytes_x_vec, int dim, int *jds_ptr_int,
size_t bytes_jds_ptr_int, int *sh_zcnt_int,
size_t bytes_sh_zcnt_int, size_t dim_X1) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(7, dst_vector, d_data, d_index, d_perm, x_vec, jds_ptr_int,
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(7, dst_vector, d_data, d_index, d_perm, x_vec, jds_ptr_int,
sh_zcnt_int, 1, dst_vector);
- void *spmv_node = __hpvm__createNodeND(1, spmv_jds, dim_X1);
- __hpvm__bindIn(spmv_node, 0, 0, 0);
- __hpvm__bindIn(spmv_node, 1, 1, 0);
- __hpvm__bindIn(spmv_node, 2, 2, 0);
- __hpvm__bindIn(spmv_node, 3, 3, 0);
- __hpvm__bindIn(spmv_node, 4, 4, 0);
- __hpvm__bindIn(spmv_node, 5, 5, 0);
- __hpvm__bindIn(spmv_node, 6, 6, 0);
- __hpvm__bindIn(spmv_node, 7, 7, 0);
- __hpvm__bindIn(spmv_node, 8, 8, 0);
- __hpvm__bindIn(spmv_node, 9, 9, 0);
- __hpvm__bindIn(spmv_node, 10, 10, 0);
- __hpvm__bindIn(spmv_node, 11, 11, 0);
- __hpvm__bindIn(spmv_node, 12, 12, 0);
- __hpvm__bindIn(spmv_node, 13, 13, 0);
- __hpvm__bindIn(spmv_node, 14, 14, 0);
+ void *spmv_node = __visc__createNodeND(1, spmv_jds, dim_X1);
+ __visc__bindIn(spmv_node, 0, 0, 0);
+ __visc__bindIn(spmv_node, 1, 1, 0);
+ __visc__bindIn(spmv_node, 2, 2, 0);
+ __visc__bindIn(spmv_node, 3, 3, 0);
+ __visc__bindIn(spmv_node, 4, 4, 0);
+ __visc__bindIn(spmv_node, 5, 5, 0);
+ __visc__bindIn(spmv_node, 6, 6, 0);
+ __visc__bindIn(spmv_node, 7, 7, 0);
+ __visc__bindIn(spmv_node, 8, 8, 0);
+ __visc__bindIn(spmv_node, 9, 9, 0);
+ __visc__bindIn(spmv_node, 10, 10, 0);
+ __visc__bindIn(spmv_node, 11, 11, 0);
+ __visc__bindIn(spmv_node, 12, 12, 0);
+ __visc__bindIn(spmv_node, 13, 13, 0);
+ __visc__bindIn(spmv_node, 14, 14, 0);
}
void spmvLvl2(float *dst_vector, size_t bytes_dst_vector, float *d_data,
@@ -153,26 +153,26 @@ void spmvLvl2(float *dst_vector, size_t bytes_dst_vector, float *d_data,
size_t bytes_x_vec, int dim, int *jds_ptr_int,
size_t bytes_jds_ptr_int, int *sh_zcnt_int,
size_t bytes_sh_zcnt_int, size_t dim_X1, size_t dim_X2) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(7, dst_vector, d_data, d_index, d_perm, x_vec, jds_ptr_int,
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(7, dst_vector, d_data, d_index, d_perm, x_vec, jds_ptr_int,
sh_zcnt_int, 1, dst_vector);
- void *spmv_node = __hpvm__createNodeND(1, spmvLvl1, dim_X2);
- __hpvm__bindIn(spmv_node, 0, 0, 0);
- __hpvm__bindIn(spmv_node, 1, 1, 0);
- __hpvm__bindIn(spmv_node, 2, 2, 0);
- __hpvm__bindIn(spmv_node, 3, 3, 0);
- __hpvm__bindIn(spmv_node, 4, 4, 0);
- __hpvm__bindIn(spmv_node, 5, 5, 0);
- __hpvm__bindIn(spmv_node, 6, 6, 0);
- __hpvm__bindIn(spmv_node, 7, 7, 0);
- __hpvm__bindIn(spmv_node, 8, 8, 0);
- __hpvm__bindIn(spmv_node, 9, 9, 0);
- __hpvm__bindIn(spmv_node, 10, 10, 0);
- __hpvm__bindIn(spmv_node, 11, 11, 0);
- __hpvm__bindIn(spmv_node, 12, 12, 0);
- __hpvm__bindIn(spmv_node, 13, 13, 0);
- __hpvm__bindIn(spmv_node, 14, 14, 0);
- __hpvm__bindIn(spmv_node, 15, 15, 0);
+ void *spmv_node = __visc__createNodeND(1, spmvLvl1, dim_X2);
+ __visc__bindIn(spmv_node, 0, 0, 0);
+ __visc__bindIn(spmv_node, 1, 1, 0);
+ __visc__bindIn(spmv_node, 2, 2, 0);
+ __visc__bindIn(spmv_node, 3, 3, 0);
+ __visc__bindIn(spmv_node, 4, 4, 0);
+ __visc__bindIn(spmv_node, 5, 5, 0);
+ __visc__bindIn(spmv_node, 6, 6, 0);
+ __visc__bindIn(spmv_node, 7, 7, 0);
+ __visc__bindIn(spmv_node, 8, 8, 0);
+ __visc__bindIn(spmv_node, 9, 9, 0);
+ __visc__bindIn(spmv_node, 10, 10, 0);
+ __visc__bindIn(spmv_node, 11, 11, 0);
+ __visc__bindIn(spmv_node, 12, 12, 0);
+ __visc__bindIn(spmv_node, 13, 13, 0);
+ __visc__bindIn(spmv_node, 14, 14, 0);
+ __visc__bindIn(spmv_node, 15, 15, 0);
}
void spmvLvl3(float *dst_vector, size_t bytes_dst_vector, float *d_data,
@@ -181,27 +181,27 @@ void spmvLvl3(float *dst_vector, size_t bytes_dst_vector, float *d_data,
size_t bytes_x_vec, int dim, int *jds_ptr_int,
size_t bytes_jds_ptr_int, int *sh_zcnt_int,
size_t bytes_sh_zcnt_int, size_t dim_X1, size_t dim_X2) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(7, dst_vector, d_data, d_index, d_perm, x_vec, jds_ptr_int,
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(7, dst_vector, d_data, d_index, d_perm, x_vec, jds_ptr_int,
sh_zcnt_int, 1, dst_vector);
- void *spmv_node = __hpvm__createNodeND(1, spmvLvl2, dim_X2);
- __hpvm__bindIn(spmv_node, 0, 0, 0);
- __hpvm__bindIn(spmv_node, 1, 1, 0);
- __hpvm__bindIn(spmv_node, 2, 2, 0);
- __hpvm__bindIn(spmv_node, 3, 3, 0);
- __hpvm__bindIn(spmv_node, 4, 4, 0);
- __hpvm__bindIn(spmv_node, 5, 5, 0);
- __hpvm__bindIn(spmv_node, 6, 6, 0);
- __hpvm__bindIn(spmv_node, 7, 7, 0);
- __hpvm__bindIn(spmv_node, 8, 8, 0);
- __hpvm__bindIn(spmv_node, 9, 9, 0);
- __hpvm__bindIn(spmv_node, 10, 10, 0);
- __hpvm__bindIn(spmv_node, 11, 11, 0);
- __hpvm__bindIn(spmv_node, 12, 12, 0);
- __hpvm__bindIn(spmv_node, 13, 13, 0);
- __hpvm__bindIn(spmv_node, 14, 14, 0);
- __hpvm__bindIn(spmv_node, 15, 15, 0);
- __hpvm__bindIn(spmv_node, 16, 16, 0);
+ void *spmv_node = __visc__createNodeND(1, spmvLvl2, dim_X2);
+ __visc__bindIn(spmv_node, 0, 0, 0);
+ __visc__bindIn(spmv_node, 1, 1, 0);
+ __visc__bindIn(spmv_node, 2, 2, 0);
+ __visc__bindIn(spmv_node, 3, 3, 0);
+ __visc__bindIn(spmv_node, 4, 4, 0);
+ __visc__bindIn(spmv_node, 5, 5, 0);
+ __visc__bindIn(spmv_node, 6, 6, 0);
+ __visc__bindIn(spmv_node, 7, 7, 0);
+ __visc__bindIn(spmv_node, 8, 8, 0);
+ __visc__bindIn(spmv_node, 9, 9, 0);
+ __visc__bindIn(spmv_node, 10, 10, 0);
+ __visc__bindIn(spmv_node, 11, 11, 0);
+ __visc__bindIn(spmv_node, 12, 12, 0);
+ __visc__bindIn(spmv_node, 13, 13, 0);
+ __visc__bindIn(spmv_node, 14, 14, 0);
+ __visc__bindIn(spmv_node, 15, 15, 0);
+ __visc__bindIn(spmv_node, 16, 16, 0);
}
int main(int argc, char **argv) {
@@ -261,7 +261,7 @@ int main(int argc, char **argv) {
input_vec(parameters->inpFiles[1], h_x_vector, dim);
pb_InitializeTimerSet(&timers);
- __hpvm__init();
+ __visc__init();
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
memset(h_Ax_vector, 0, dim * sizeof(float));
@@ -271,14 +271,14 @@ int main(int argc, char **argv) {
compute_active_thread(&block, &grid, nzcnt_len, pad, 3, 0, 8);
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_TRACK);
- llvm_hpvm_track_mem(h_Ax_vector, dim * sizeof(float));
- llvm_hpvm_track_mem(h_data, len * sizeof(float));
- llvm_hpvm_track_mem(h_indices, len * sizeof(int));
- llvm_hpvm_track_mem(h_perm, dim * sizeof(int));
- llvm_hpvm_track_mem(h_x_vector, dim * sizeof(float));
- llvm_hpvm_track_mem(h_ptr, depth * sizeof(int));
- llvm_hpvm_track_mem(h_nzcnt, nzcnt_len * sizeof(int));
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_TRACK);
+ llvm_visc_track_mem(h_Ax_vector, dim * sizeof(float));
+ llvm_visc_track_mem(h_data, len * sizeof(float));
+ llvm_visc_track_mem(h_indices, len * sizeof(int));
+ llvm_visc_track_mem(h_perm, dim * sizeof(int));
+ llvm_visc_track_mem(h_x_vector, dim * sizeof(float));
+ llvm_visc_track_mem(h_ptr, depth * sizeof(int));
+ llvm_visc_track_mem(h_nzcnt, nzcnt_len * sizeof(int));
// main execution
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
@@ -306,9 +306,9 @@ int main(int argc, char **argv) {
block,
(grid / block)};
*(RootIn *)root_in = root_in_local;
- void *spmvDFG = __hpvm__launch(0, spmvLvl3, root_in);
+ void *spmvDFG = __visc__launch(0, spmvLvl3, root_in);
- __hpvm__wait(spmvDFG);
+ __visc__wait(spmvDFG);
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
/******************************* Issues *******************************
@@ -326,21 +326,21 @@ int main(int argc, char **argv) {
// HtoD memory copy
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
- llvm_hpvm_request_mem(h_Ax_vector, dim * sizeof(float));
+ llvm_visc_request_mem(h_Ax_vector, dim * sizeof(float));
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_UNTRACK);
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_UNTRACK);
- llvm_hpvm_untrack_mem(h_Ax_vector);
- llvm_hpvm_untrack_mem(h_data);
- llvm_hpvm_untrack_mem(h_indices);
- llvm_hpvm_untrack_mem(h_perm);
- llvm_hpvm_untrack_mem(h_x_vector);
- llvm_hpvm_untrack_mem(h_ptr);
- llvm_hpvm_untrack_mem(h_nzcnt);
+ llvm_visc_untrack_mem(h_Ax_vector);
+ llvm_visc_untrack_mem(h_data);
+ llvm_visc_untrack_mem(h_indices);
+ llvm_visc_untrack_mem(h_perm);
+ llvm_visc_untrack_mem(h_x_vector);
+ llvm_visc_untrack_mem(h_ptr);
+ llvm_visc_untrack_mem(h_nzcnt);
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
pb_PrintTimerSet(&timers);
- __hpvm__cleanup();
+ __visc__cleanup();
if (parameters->outFile) {
/*pb_SwitchToTimer(&timers, pb_TimerID_IO);*/
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/visc/main.visc.ll.kernels.bc b/hpvm/test/parboil/benchmarks/spmv/src/visc/main.visc.ll.kernels.bc
new file mode 100644
index 0000000000000000000000000000000000000000..b804d14d16cff805c0c1850d1f5079ab6e973ecf
GIT binary patch
literal 2464
zcmZuxe^3+o75{D$vH>>9MuH}iFdG9jwK^^V5d<W^&`|M+=&6WPH~f-zL<497oM$&b
zND$)@m9uf~I+L{YhCi+yYfoxtv=X8mMYu!2pXXQ*j?=dG@E6somVP(!tlo6zyKmq7
z?t7p2dGGCab^6{r3Jid1769=M0>B4Y0B_W;JgObw6mfBN&N0M=rN;22$(+L#w78I^
zI~K3&!qwtzo}?&FZAdZ3`4q~Fr8Fj0G^C&dMo~p~pj*>Fi$I)>5&-n@nKZlP**v@u
zKhLW2H2XbSG}^54)EhleW|3%9t;a&3y|O;Ld?qLZeeyd&{*X4gFDM@j@+T(cqapdd
zplmwGAGFIygL2TwL#=*K1K`=3nDAzv2f}929-9%h1w0KtPnHU`8a>upwyg=Z)w1DR
z3;8{Kh&g0VzSS$62=Z?!WcLs`Xz_VkR48rnzz`b`wLzW?H>Z_{wrEf~c{-Rph!B4Z
z$!8R@p`h&l#db$3@_TF~+EUAIq&yI(1q>-<H_R{=wgMEvo-j&+1As#W08*%{ml$J-
zs#ebNC0#Bkrd&avctLTyH{R7%j2o381BhpM8uJGLYZLR@6l<l`ZQ4PVvslDcWa5^>
zOta8VFIb*_QT1yq-6*ZbD+;$)v~64~t?;P_3D<3zmW%-jpElE;eLgfd;LPeqZQD@T
zyZ0=fH+BB+HVW#1u!TbR1yHL7Wt0)_eH?9}nN{Z#GzCy*MK-Mgw4Op+aMTJ-5^V^e
zjRe{P6==SVF!T1?nYZ80L4-G;<_ISHYpA)mW9Kvsf!EyTz6a%tKoN{jJ(MF{2;Kn6
zzai@=VZD!G+hAgx842A8bU&=+4GG=$!>9oIKN&%T-zlV556}h;1Iw6XRSp0=I7UDX
z9E;2rfL>oJ!ki`mXvKOKC!zuU2~NZffYTa)!aa4eAD4=Xecsr#b^RRuVkH=_0r!R@
z-p#Mhj^5QTS(5iTeiw9_c@mYYRON=NA+n(2P<blgTGn(89*~e69*~0y+4Q9Bpgs8!
zl00a(n+5-v6-;V`KRZPknc>G6Ru9=_-64lUKBE!1CGEdb>C;`Ng{@^|mos-HjO%_X
z-W7T*crYQj4}{-pgkMhxo&fF?Etn*Pdd}Bn_^@vEXY9OF`W!l>Iu}xX8d^*v%iDA+
zS0AoF)LE-?<LzB!eB7&#UvlM!e)Ui|`vq%t1b@ZH?7Z=?m@-rf?~}r*+4y@t!9O*^
z3#4GyB;3RyB$MaZd7aK&FFW^aXmzJE@2o!OR5-6V7B-RT$s*XD3?41JI4P?UBl&Im
zY**N;ga4Bzw9#LW%>dy8T=0YzWMI1FzL5Njpsa?BiMTnFhS>Q7`g|`tf2hg3&w{?k
zLz|5riv?}5u&qrVI7^6KPelKL3yGJF?<ZY+l5{u$T<*XQA;f5S^Uu}BEb?XJ5xBqN
z1F-CZtIANpxl(YwM_W99m)HA9sjpx~%?qNO@yAL%3z#7u1`m7!XWcR!;>W<-&`z(7
z6(dJQbg2r7Q^Fkp3UN~*QY|rkj<gb{B9@ZN0bY9pVxM%=I%k=?m-l1pkwY;RyIB(=
zc5k!L6O2)Xo5=oz0V+W4Agl7s7%Q&d5<Ye^w9IMZOvP4CY$TWYHI|U?A!ij&B?<FO
zrT$P;p9~j!%iMMecd2_f6=M!8Q9126cWK*R%4KgS7H`%{x_Ku;3EZWuDihnBrxkdA
z<o5pHwpJ55)#vL_AP3q{vICXdXfAP<8ffZMam3z-u1yk^#Dh!r@WugFBW+VajN2O0
zZI9~{aB84&Pdnnaw}V6u3G=eH<I_}j1q@!w5o5}@=Ql*XyG{h4n~EiKKsSYj0l-+P
ztR&8JKi>Pzf0#_mVWxOUh1Qc2k%UdIr>JbEr(N*w>E+xo80Tk&@Y&(W)W;Z(b!=WM
zih-MOYV#R|1*63^okaMlK~VYExs}+#x?<;ird4fRSG7G>y`yBB%-UPRy*svMRPv>Q
zEcxZ*o}BanOa9UD_#fvpw^BrFNv5+kalxf$XYBmrJIjS^gNRmy-~IMQ-PXf~^feDw
zI8u;)3sx|SU967lXK48ULXaJ)8YD|JT3{U8De9XxoN$1R9BRNK9kgTD6LP#aRT?oO
zQSaDamo18Lc>18=NJnPw-W<Vq=J1`QG6&;4bfmnK^(v|LM5K_$zi;V^0$x3z%E!*!
z7(XrEG+jQRA+91~Lq?(h3=t{ecA7Kd|A!=p`W2IFBhhv#hc!FW_8G-fW9RlH9^Uuj
z<VZ1W)RyXJhlwJ(I@MY7Z${Db+e*GV8qHokRr4Vct#^@yXf2Q%8d<TW&&ITwE8pj=
zn>4Jsf_<_XgsE?OeiP+(^SoAkeW%D_PCXt)`%_A^>a$mI*5fB>=@_q)jKvFq8~eqc
zjHt3`udtZj;CdpQ*BiVm^Is2Q&FII7Lkv@_+MBTcyRojK#qUg`U(h!<Oco6P`J+NH
zm7sVz`f!&n<E1e}dKqBO4}TcVXxsCX7!tG987@tX<?X42`oHROWA2U}s9P@_#S-zN
zHKot_^&76eOO7{;rEP)ot52M;SS7!t++=^2`Q=B>^cO-=`oiBwy7zRei>i%_Fu#62
zDN(m$ASHX4(qEb$J0L*P0pP>$vfQ0!?7`mg?Ur_dTWXW`boE(Q?PY?Cd{$lSggdvj
XpiXK~&^_31@sx7zHXZ?kUS0nM9J*WM
literal 0
HcmV?d00001
diff --git a/hpvm/test/parboil/benchmarks/spmv/src/visc/main.visc.ll.kernels.ll b/hpvm/test/parboil/benchmarks/spmv/src/visc/main.visc.ll.kernels.ll
new file mode 100644
index 0000000000..5604d70e8a
--- /dev/null
+++ b/hpvm/test/parboil/benchmarks/spmv/src/visc/main.visc.ll.kernels.ll
@@ -0,0 +1,138 @@
+; ModuleID = 'build/visc_default/main.visc.ll.kernels.bc'
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024"
+target triple = "spir64-unknown-unknown"
+
+%rtype = type {}
+
+; Function Attrs: optsize zeroext
+define void @spmv_jds(float* %dst_vector, i64 %bytes_dst_vector, float* %d_data, i64 %bytes_d_data, i32* %d_index, i64 %bytes_d_index, i32* %d_perm, i64 %bytes_d_perm, float* %x_vec, i64 %bytes_x_vec, i32 %dim, i32* %jds_ptr_int, i64 %bytes_jds_ptr_int, i32* %sh_zcnt_int, i64 %bytes_sh_zcnt_int) #0 {
+entry:
+ ;%0 = call i64 @_Z12get_group_idj(i32 0)
+ ;%1 = trunc i64 %0 to i32
+ ;%2 = call i64 @_Z14get_local_sizej(i32 0)
+ ;%3 = trunc i64 %2 to i32
+ ;%4 = mul i32 %1, %3
+ ;%5 = call i64 @_Z12get_local_idj(i32 0)
+ ;%6 = trunc i64 %5 to i32
+ ;%7 = add i32 %4, %6
+ %0 = add i32 0, 0
+ %1 = add i32 0, 0
+ %2 = add i32 0, 0
+ %3 = add i32 0, 0
+ %4 = add i32 0, 0
+ %5 = add i32 0, 0
+ %6 = call i64 @_Z13get_global_idj(i32 0)
+ %7 = trunc i64 %6 to i32
+ %cmp = icmp slt i32 %7, %dim
+ br i1 %cmp, label %if.then, label %if.end38
+
+if.then: ; preds = %entry
+ %shr = ashr i32 %7, 5
+ %idxprom = sext i32 %shr to i64
+ %arrayidx = getelementptr inbounds i32* %sh_zcnt_int, i64 %idxprom
+ %8 = load i32* %arrayidx, align 4, !tbaa !4
+ %9 = load i32* %jds_ptr_int, align 4, !tbaa !4
+ %add = add nsw i32 %9, %7
+ %idxprom3 = sext i32 %add to i64
+ %arrayidx4 = getelementptr inbounds float* %d_data, i64 %idxprom3
+ %10 = load float* %arrayidx4, align 4, !tbaa !8
+ %arrayidx6 = getelementptr inbounds i32* %d_index, i64 %idxprom3
+ %11 = load i32* %arrayidx6, align 4, !tbaa !4
+ %idxprom7 = sext i32 %11 to i64
+ %arrayidx8 = getelementptr inbounds float* %x_vec, i64 %idxprom7
+ %12 = load float* %arrayidx8, align 4, !tbaa !8
+ %cmp9 = icmp sgt i32 %8, 1
+ br i1 %cmp9, label %if.then10, label %if.end
+
+if.then10: ; preds = %if.then
+ %arrayidx11 = getelementptr inbounds i32* %jds_ptr_int, i64 1
+ %.pn77 = load i32* %arrayidx11, align 4
+ %idxprom13.pn.in78 = add nsw i32 %.pn77, %7
+ %idxprom13.pn79 = sext i32 %idxprom13.pn.in78 to i64
+ %i.0.in80 = getelementptr inbounds i32* %d_index, i64 %idxprom13.pn79
+ %i.081 = load i32* %i.0.in80, align 4
+ %cmp1582 = icmp sgt i32 %8, 2
+ %arrayidx1783 = getelementptr inbounds float* %d_data, i64 %idxprom13.pn79
+ %13 = load float* %arrayidx1783, align 4, !tbaa !8
+ br i1 %cmp1582, label %for.body, label %for.end
+
+for.body: ; preds = %for.body, %if.then10
+ %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 2, %if.then10 ]
+ %14 = phi float [ %16, %for.body ], [ %13, %if.then10 ]
+ %i.088 = phi i32 [ %i.0, %for.body ], [ %i.081, %if.then10 ]
+ %sum.086 = phi float [ %add25, %for.body ], [ 0.000000e+00, %if.then10 ]
+ %t.085 = phi float [ %15, %for.body ], [ %12, %if.then10 ]
+ %d.084 = phi float [ %14, %for.body ], [ %10, %if.then10 ]
+ %arrayidx19 = getelementptr inbounds i32* %jds_ptr_int, i64 %indvars.iv
+ %idxprom23 = sext i32 %i.088 to i64
+ %arrayidx24 = getelementptr inbounds float* %x_vec, i64 %idxprom23
+ %15 = load float* %arrayidx24, align 4, !tbaa !8
+ %mul = fmul fast float %d.084, %t.085
+ %add25 = fadd fast float %sum.086, %mul
+ %indvars.iv.next = add i64 %indvars.iv, 1
+ %.pn = load i32* %arrayidx19, align 4
+ %idxprom13.pn.in = add nsw i32 %.pn, %7
+ %idxprom13.pn = sext i32 %idxprom13.pn.in to i64
+ %i.0.in = getelementptr inbounds i32* %d_index, i64 %idxprom13.pn
+ %i.0 = load i32* %i.0.in, align 4
+ %arrayidx17 = getelementptr inbounds float* %d_data, i64 %idxprom13.pn
+ %16 = load float* %arrayidx17, align 4, !tbaa !8
+ %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+ %exitcond = icmp eq i32 %lftr.wideiv, %8
+ br i1 %exitcond, label %for.end, label %for.body
+
+for.end: ; preds = %for.body, %if.then10
+ %.lcssa = phi float [ %13, %if.then10 ], [ %16, %for.body ]
+ %i.0.lcssa = phi i32 [ %i.081, %if.then10 ], [ %i.0, %for.body ]
+ %sum.0.lcssa = phi float [ 0.000000e+00, %if.then10 ], [ %add25, %for.body ]
+ %t.0.lcssa = phi float [ %12, %if.then10 ], [ %15, %for.body ]
+ %d.0.lcssa = phi float [ %10, %if.then10 ], [ %14, %for.body ]
+ %idxprom28 = sext i32 %i.0.lcssa to i64
+ %arrayidx29 = getelementptr inbounds float* %x_vec, i64 %idxprom28
+ %17 = load float* %arrayidx29, align 4, !tbaa !8
+ %mul30 = fmul fast float %d.0.lcssa, %t.0.lcssa
+ %add31 = fadd fast float %sum.0.lcssa, %mul30
+ br label %if.end
+
+if.end: ; preds = %for.end, %if.then
+ %d.1 = phi float [ %.lcssa, %for.end ], [ %10, %if.then ]
+ %t.1 = phi float [ %17, %for.end ], [ %12, %if.then ]
+ %sum.1 = phi float [ %add31, %for.end ], [ 0.000000e+00, %if.then ]
+ %mul32 = fmul fast float %d.1, %t.1
+ %add33 = fadd fast float %sum.1, %mul32
+ %idxprom34 = sext i32 %7 to i64
+ %arrayidx35 = getelementptr inbounds i32* %d_perm, i64 %idxprom34
+ %18 = load i32* %arrayidx35, align 4, !tbaa !4
+ %idxprom36 = sext i32 %18 to i64
+ %arrayidx37 = getelementptr inbounds float* %dst_vector, i64 %idxprom36
+ store float %add33, float* %arrayidx37, align 4, !tbaa !8
+ br label %if.end38
+
+if.end38: ; preds = %if.end, %entry
+ ret void
+}
+
+declare i64 @_Z13get_global_idj(i32)
+
+declare i64 @_Z12get_group_idj(i32)
+
+declare i64 @_Z14get_local_sizej(i32)
+
+declare i64 @_Z12get_local_idj(i32)
+
+attributes #0 = { optsize zeroext "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+
+!visc_hint_gpu = !{}
+!visc_hint_cpu = !{!0, !1}
+!opencl.kernels = !{!2}
+
+!0 = metadata !{%rtype (float*, i64, float*, i64, i32*, i64, i32*, i64, float*, i64, i32, i32*, i64, i32*, i64, i32)* undef}
+!1 = metadata !{%rtype (float*, i64, float*, i64, i32*, i64, i32*, i64, float*, i64, i32, i32*, i64, i32*, i64, i32, i32)* undef}
+!2 = metadata !{void (float*, i64, float*, i64, i32*, i64, i32*, i64, float*, i64, i32, i32*, i64, i32*, i64)* @spmv_jds, metadata !3}
+!3 = metadata !{metadata !"kernel_arg_type", metadata !"float*", metadata !"i64", metadata !"float*", metadata !"i64", metadata !"i32*", metadata !"i64", metadata !"i32*", metadata !"i64", metadata !"float*", metadata !"i64", metadata !"i32", metadata !"i32*", metadata !"i64", metadata !"i32*", metadata !"i64"}
+!4 = metadata !{metadata !5, metadata !5, i64 0}
+!5 = metadata !{metadata !"int", metadata !6}
+!6 = metadata !{metadata !"omnipotent char", metadata !7}
+!7 = metadata !{metadata !"Simple C/C++ TBAA"}
+!8 = metadata !{metadata !9, metadata !9, i64 0}
+!9 = metadata !{metadata !"float", metadata !6}
diff --git a/hpvm/test/parboil/benchmarks/stencil/Makefile b/hpvm/test/parboil/benchmarks/stencil/Makefile
index e761d7b4f5..a44dd0dbf0 100644
--- a/hpvm/test/parboil/benchmarks/stencil/Makefile
+++ b/hpvm/test/parboil/benchmarks/stencil/Makefile
@@ -1,9 +1,9 @@
PARBOIL_ROOT = $(LLVM_SRC_ROOT)/tools/hpvm/test/parboil
APP = stencil
-# Default compile hpvm
+# Default compile visc
ifeq ($(VERSION),)
- VERSION = hpvm
+ VERSION = visc
endif
# Default use small test case
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/opencl_base/main.c b/hpvm/test/parboil/benchmarks/stencil/src/opencl_base/main.c
index 1157b61988..ec47c22227 100644
--- a/hpvm/test/parboil/benchmarks/stencil/src/opencl_base/main.c
+++ b/hpvm/test/parboil/benchmarks/stencil/src/opencl_base/main.c
@@ -174,7 +174,7 @@ int main(int argc, char **argv) {
CHECK_ERROR("clSetKernelArg")
// main execution
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
int t;
for (t = 0; t < iteration; t++) {
clStatus = clEnqueueNDRangeKernel(clCommandQueue, clKernel, 3, NULL, grid,
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_default/main.c b/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_default/main.c
index 70a86245b7..61382182d1 100644
--- a/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_default/main.c
+++ b/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_default/main.c
@@ -89,7 +89,7 @@ int main(int argc, char **argv) {
fclose(fp);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_platform_id clPlatform;
clStatus = clGetPlatformIDs(1, &clPlatform, NULL);
@@ -140,7 +140,7 @@ int main(int argc, char **argv) {
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
memcpy(h_Anext, h_A0, sizeof(float) * size);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
// memory allocation
d_A0 = clCreateBuffer(clContext, CL_MEM_READ_WRITE, size * sizeof(float),
@@ -170,7 +170,7 @@ int main(int argc, char **argv) {
// printf("block x is %d and y is %d z \n",block[0],block[1]);
// printf("grid x is %d and y is %d\n",grid[0],grid[1]);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 0, sizeof(float), (void *)&c0);
clStatus = clSetKernelArg(clKernel, 1, sizeof(float), (void *)&c1);
clStatus = clSetKernelArg(clKernel, 2, sizeof(cl_mem), (void *)&d_A0);
@@ -182,7 +182,7 @@ int main(int argc, char **argv) {
// main execution
/*pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);*/
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (int i = 0; i < 10; i++) {
int t;
for (t = 0; t < iteration; t++) {
@@ -219,7 +219,7 @@ int main(int argc, char **argv) {
size * sizeof(float), h_Anext, 0, NULL, NULL);
CHECK_ERROR("clEnqueueReadBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseMemObject(d_A0);
clStatus = clReleaseMemObject(d_Anext);
clStatus = clReleaseKernel(clKernel);
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_large/main.c b/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_large/main.c
index 3a5dfa3b3a..217352e036 100644
--- a/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_large/main.c
+++ b/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_large/main.c
@@ -89,7 +89,7 @@ int main(int argc, char **argv) {
fclose(fp);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_platform_id clPlatform;
clStatus = clGetPlatformIDs(1, &clPlatform, NULL);
@@ -140,7 +140,7 @@ int main(int argc, char **argv) {
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
memcpy(h_Anext, h_A0, sizeof(float) * size);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
// memory allocation
d_A0 = clCreateBuffer(clContext, CL_MEM_READ_WRITE, size * sizeof(float),
@@ -170,7 +170,7 @@ int main(int argc, char **argv) {
// printf("block x is %d and y is %d z \n",block[0],block[1]);
// printf("grid x is %d and y is %d\n",grid[0],grid[1]);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 0, sizeof(float), (void *)&c0);
clStatus = clSetKernelArg(clKernel, 1, sizeof(float), (void *)&c1);
clStatus = clSetKernelArg(clKernel, 2, sizeof(cl_mem), (void *)&d_A0);
@@ -182,7 +182,7 @@ int main(int argc, char **argv) {
// main execution
/*pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);*/
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (int i = 0; i < 1; i++) {
int t;
for (t = 0; t < iteration; t++) {
@@ -219,7 +219,7 @@ int main(int argc, char **argv) {
size * sizeof(float), h_Anext, 0, NULL, NULL);
CHECK_ERROR("clEnqueueReadBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseMemObject(d_A0);
clStatus = clReleaseMemObject(d_Anext);
clStatus = clReleaseKernel(clKernel);
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_strided/main.c b/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_strided/main.c
index 264cec20a9..28c0e5fd7b 100644
--- a/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_strided/main.c
+++ b/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_strided/main.c
@@ -89,7 +89,7 @@ int main(int argc, char **argv) {
fclose(fp);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_platform_id clPlatform;
clStatus = clGetPlatformIDs(1, &clPlatform, NULL);
@@ -140,7 +140,7 @@ int main(int argc, char **argv) {
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
memcpy(h_Anext, h_A0, sizeof(float) * size);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
// memory allocation
d_A0 = clCreateBuffer(clContext, CL_MEM_READ_WRITE, size * sizeof(float),
@@ -170,7 +170,7 @@ int main(int argc, char **argv) {
// printf("block x is %d and y is %d z \n",block[0],block[1]);
// printf("grid x is %d and y is %d\n",grid[0],grid[1]);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 0, sizeof(float), (void *)&c0);
clStatus = clSetKernelArg(clKernel, 1, sizeof(float), (void *)&c1);
clStatus = clSetKernelArg(clKernel, 2, sizeof(cl_mem), (void *)&d_A0);
@@ -195,7 +195,7 @@ int main(int argc, char **argv) {
d_A0 = d_Anext;
d_Anext = d_temp;
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 2, sizeof(cl_mem), (void *)&d_A0);
clStatus = clSetKernelArg(clKernel, 3, sizeof(cl_mem), (void *)&d_Anext);
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
@@ -213,7 +213,7 @@ int main(int argc, char **argv) {
size * sizeof(float), h_Anext, 0, NULL, NULL);
CHECK_ERROR("clEnqueueReadBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseMemObject(d_A0);
clStatus = clReleaseMemObject(d_Anext);
clStatus = clReleaseKernel(clKernel);
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_vec/main.c b/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_vec/main.c
index 7b5db72237..f767f6a9d2 100644
--- a/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_vec/main.c
+++ b/hpvm/test/parboil/benchmarks/stencil/src/opencl_base_vec/main.c
@@ -89,7 +89,7 @@ int main(int argc, char **argv) {
fclose(fp);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_platform_id clPlatform;
clStatus = clGetPlatformIDs(1, &clPlatform, NULL);
@@ -140,7 +140,7 @@ int main(int argc, char **argv) {
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
memcpy(h_Anext, h_A0, sizeof(float) * size);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
// memory allocation
d_A0 = clCreateBuffer(clContext, CL_MEM_READ_WRITE, size * sizeof(float),
@@ -170,7 +170,7 @@ int main(int argc, char **argv) {
// printf("block x is %d and y is %d z \n",block[0],block[1]);
// printf("grid x is %d and y is %d\n",grid[0],grid[1]);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 0, sizeof(float), (void *)&c0);
clStatus = clSetKernelArg(clKernel, 1, sizeof(float), (void *)&c1);
clStatus = clSetKernelArg(clKernel, 2, sizeof(cl_mem), (void *)&d_A0);
@@ -195,7 +195,7 @@ int main(int argc, char **argv) {
d_A0 = d_Anext;
d_Anext = d_temp;
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 2, sizeof(cl_mem), (void *)&d_A0);
clStatus = clSetKernelArg(clKernel, 3, sizeof(cl_mem), (void *)&d_Anext);
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
@@ -213,7 +213,7 @@ int main(int argc, char **argv) {
size * sizeof(float), h_Anext, 0, NULL, NULL);
CHECK_ERROR("clEnqueueReadBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseMemObject(d_A0);
clStatus = clReleaseMemObject(d_Anext);
clStatus = clReleaseKernel(clKernel);
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu/main.c b/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu/main.c
index 51c263f0ef..10626bed59 100644
--- a/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu/main.c
+++ b/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu/main.c
@@ -89,7 +89,7 @@ int main(int argc, char **argv) {
fclose(fp);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_uint numPlatforms;
@@ -145,7 +145,7 @@ int main(int argc, char **argv) {
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
memcpy(h_Anext, h_A0, sizeof(float) * size);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
// memory allocation
d_A0 = clCreateBuffer(clContext, CL_MEM_READ_WRITE, size * sizeof(float),
@@ -175,7 +175,7 @@ int main(int argc, char **argv) {
// printf("block x is %d and y is %d z \n",block[0],block[1]);
// printf("grid x is %d and y is %d\n",grid[0],grid[1]);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 0, sizeof(float), (void *)&c0);
clStatus = clSetKernelArg(clKernel, 1, sizeof(float), (void *)&c1);
clStatus = clSetKernelArg(clKernel, 2, sizeof(cl_mem), (void *)&d_A0);
@@ -200,7 +200,7 @@ int main(int argc, char **argv) {
d_A0 = d_Anext;
d_Anext = d_temp;
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 2, sizeof(cl_mem), (void *)&d_A0);
clStatus = clSetKernelArg(clKernel, 3, sizeof(cl_mem), (void *)&d_Anext);
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
@@ -218,7 +218,7 @@ int main(int argc, char **argv) {
size * sizeof(float), h_Anext, 0, NULL, NULL);
CHECK_ERROR("clEnqueueReadBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseMemObject(d_A0);
clStatus = clReleaseMemObject(d_Anext);
clStatus = clReleaseKernel(clKernel);
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_baseline/main.c b/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_baseline/main.c
index a2a98e9233..1d03111f20 100644
--- a/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_baseline/main.c
+++ b/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_baseline/main.c
@@ -89,7 +89,7 @@ int main(int argc, char **argv) {
fclose(fp);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
cl_int clStatus;
cl_uint numPlatforms;
@@ -184,7 +184,7 @@ int main(int argc, char **argv) {
// main execution
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
int t;
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
/*for(int i=0; i<1; i++) {*/
for (t = 0; t < iteration; t++) {
clStatus = clEnqueueNDRangeKernel(clCommandQueue, clKernel, 3, NULL, grid,
@@ -216,7 +216,7 @@ int main(int argc, char **argv) {
size * sizeof(float), h_Anext, 0, NULL, NULL);
CHECK_ERROR("clEnqueueReadBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
clStatus = clReleaseMemObject(d_A0);
clStatus = clReleaseMemObject(d_Anext);
clStatus = clReleaseKernel(clKernel);
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_default/kernel-spir64.ll b/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_default/kernel-spir64.ll
index a288b7649a..9ea545c184 100644
--- a/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_default/kernel-spir64.ll
+++ b/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_default/kernel-spir64.ll
@@ -1,4 +1,4 @@
-; ModuleID = '/home/psrivas2.hpvm.llvm/test/HPVM/parboil/benchmarks/stencil/src/opencl_cpu/kernel_offline.cl'
+; ModuleID = '/home/psrivas2/visc/llvm/test/VISC/parboil/benchmarks/stencil/src/opencl_cpu/kernel_offline.cl'
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024"
target triple = "spir64-unknown-unknown"
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_default/main.c b/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_default/main.c
index 9fc78af4b9..cf86734a86 100644
--- a/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_default/main.c
+++ b/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_default/main.c
@@ -89,7 +89,7 @@ int main(int argc, char **argv) {
fclose(fp);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_uint numPlatforms;
@@ -145,7 +145,7 @@ int main(int argc, char **argv) {
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
memcpy(h_Anext, h_A0, sizeof(float) * size);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
// memory allocation
d_A0 = clCreateBuffer(clContext, CL_MEM_READ_WRITE, size * sizeof(float),
@@ -177,7 +177,7 @@ int main(int argc, char **argv) {
printf("grid(%lu, %lu, %lu), block(%lu, %lu, %lu)\n", grid[0], grid[1],
grid[2], block[0], block[1], block[2]);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 0, sizeof(float), (void *)&c0);
clStatus = clSetKernelArg(clKernel, 1, sizeof(float), (void *)&c1);
clStatus = clSetKernelArg(clKernel, 2, sizeof(cl_mem), (void *)&d_A0);
@@ -190,7 +190,7 @@ int main(int argc, char **argv) {
// main execution
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
int t;
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (int i = 0; i < 2; i++) {
for (t = 0; t < iteration; t++) {
/*pb_SwitchToTimer(&timers, pb_TimerID_KERNEL);*/
@@ -204,7 +204,7 @@ int main(int argc, char **argv) {
d_A0 = d_Anext;
d_Anext = d_temp;
- /*pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);*/
+ /*pb_SwitchToTimer(&timers, visc_TimerID_SETUP);*/
clStatus = clSetKernelArg(clKernel, 2, sizeof(cl_mem), (void *)&d_A0);
clStatus = clSetKernelArg(clKernel, 3, sizeof(cl_mem), (void *)&d_Anext);
/*pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);*/
@@ -226,7 +226,7 @@ int main(int argc, char **argv) {
size * sizeof(float), h_Anext, 0, NULL, NULL);
CHECK_ERROR("clEnqueueReadBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseMemObject(d_A0);
clStatus = clReleaseMemObject(d_Anext);
clStatus = clReleaseKernel(clKernel);
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_large/kernel-spir64.ll b/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_large/kernel-spir64.ll
index a288b7649a..9ea545c184 100644
--- a/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_large/kernel-spir64.ll
+++ b/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_large/kernel-spir64.ll
@@ -1,4 +1,4 @@
-; ModuleID = '/home/psrivas2.hpvm.llvm/test/HPVM/parboil/benchmarks/stencil/src/opencl_cpu/kernel_offline.cl'
+; ModuleID = '/home/psrivas2/visc/llvm/test/VISC/parboil/benchmarks/stencil/src/opencl_cpu/kernel_offline.cl'
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024"
target triple = "spir64-unknown-unknown"
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_large/main.c b/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_large/main.c
index a1e1c4e74e..3b009e370e 100644
--- a/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_large/main.c
+++ b/hpvm/test/parboil/benchmarks/stencil/src/opencl_cpu_large/main.c
@@ -89,7 +89,7 @@ int main(int argc, char **argv) {
fclose(fp);
pb_InitializeTimerSet(&timers);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
cl_int clStatus;
cl_uint numPlatforms;
@@ -145,7 +145,7 @@ int main(int argc, char **argv) {
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
memcpy(h_Anext, h_A0, sizeof(float) * size);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
// memory allocation
d_A0 = clCreateBuffer(clContext, CL_MEM_READ_WRITE, size * sizeof(float),
@@ -177,7 +177,7 @@ int main(int argc, char **argv) {
printf("grid(%lu, %lu, %lu), block(%lu, %lu, %lu)\n", grid[0], grid[1],
grid[2], block[0], block[1], block[2]);
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clSetKernelArg(clKernel, 0, sizeof(float), (void *)&c0);
clStatus = clSetKernelArg(clKernel, 1, sizeof(float), (void *)&c1);
clStatus = clSetKernelArg(clKernel, 2, sizeof(cl_mem), (void *)&d_A0);
@@ -190,7 +190,7 @@ int main(int argc, char **argv) {
// main execution
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
int t;
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
for (int i = 0; i < 1; i++) {
for (t = 0; t < iteration; t++) {
/*pb_SwitchToTimer(&timers, pb_TimerID_KERNEL);*/
@@ -204,7 +204,7 @@ int main(int argc, char **argv) {
d_A0 = d_Anext;
d_Anext = d_temp;
- /*pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);*/
+ /*pb_SwitchToTimer(&timers, visc_TimerID_SETUP);*/
clStatus = clSetKernelArg(clKernel, 2, sizeof(cl_mem), (void *)&d_A0);
clStatus = clSetKernelArg(clKernel, 3, sizeof(cl_mem), (void *)&d_Anext);
/*pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);*/
@@ -226,7 +226,7 @@ int main(int argc, char **argv) {
size * sizeof(float), h_Anext, 0, NULL, NULL);
CHECK_ERROR("clEnqueueReadBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_SETUP);
+ pb_SwitchToTimer(&timers, visc_TimerID_SETUP);
clStatus = clReleaseMemObject(d_A0);
clStatus = clReleaseMemObject(d_Anext);
clStatus = clReleaseKernel(clKernel);
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/hpvm/Makefile b/hpvm/test/parboil/benchmarks/stencil/src/visc/Makefile
similarity index 80%
rename from hpvm/test/parboil/benchmarks/stencil/src/hpvm/Makefile
rename to hpvm/test/parboil/benchmarks/stencil/src/visc/Makefile
index 35b36dcf3c..cf61fb3a6c 100644
--- a/hpvm/test/parboil/benchmarks/stencil/src/hpvm/Makefile
+++ b/hpvm/test/parboil/benchmarks/stencil/src/visc/Makefile
@@ -1,8 +1,8 @@
# (c) 2010 The Board of Trustees of the University of Illinois.
-LANGUAGE=hpvm
+LANGUAGE=visc
SRCDIR_OBJS=file.ll
-HPVM_OBJS=stencil.hpvm.ll
+VISC_OBJS=stencil.visc.ll
APP_CUDALDFLAGS=-lm
APP_CFLAGS=-ffast-math -O3
APP_CXXFLAGS=-ffast-math -O3
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/hpvm/common.h b/hpvm/test/parboil/benchmarks/stencil/src/visc/common.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/stencil/src/hpvm/common.h
rename to hpvm/test/parboil/benchmarks/stencil/src/visc/common.h
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/hpvm/file.cc b/hpvm/test/parboil/benchmarks/stencil/src/visc/file.cc
similarity index 100%
rename from hpvm/test/parboil/benchmarks/stencil/src/hpvm/file.cc
rename to hpvm/test/parboil/benchmarks/stencil/src/visc/file.cc
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/hpvm/file.h b/hpvm/test/parboil/benchmarks/stencil/src/visc/file.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/stencil/src/hpvm/file.h
rename to hpvm/test/parboil/benchmarks/stencil/src/visc/file.h
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/hpvm/kernel.cl b/hpvm/test/parboil/benchmarks/stencil/src/visc/kernel.cl
similarity index 100%
rename from hpvm/test/parboil/benchmarks/stencil/src/hpvm/kernel.cl
rename to hpvm/test/parboil/benchmarks/stencil/src/visc/kernel.cl
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/hpvm/stencil.cpp b/hpvm/test/parboil/benchmarks/stencil/src/visc/stencil.cpp
similarity index 66%
rename from hpvm/test/parboil/benchmarks/stencil/src/hpvm/stencil.cpp
rename to hpvm/test/parboil/benchmarks/stencil/src/visc/stencil.cpp
index e5810fc810..5672a3ee49 100644
--- a/hpvm/test/parboil/benchmarks/stencil/src/hpvm/stencil.cpp
+++ b/hpvm/test/parboil/benchmarks/stencil/src/visc/stencil.cpp
@@ -9,11 +9,11 @@
#include "common.h"
#include "file.h"
-#include <hpvm.h>
#include <parboil.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
+#include <visc.h>
static int read_data(float *A0, int nx, int ny, int nz, FILE *fp) {
int s = 0;
@@ -42,23 +42,23 @@ typedef struct __attribute__((__packed__)) {
void naive_kernel(float c0, float c1, float *A0, size_t bytes_A0, float *Anext,
size_t bytes_Anext, int nx, int ny, int nz) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(2, A0, Anext, 1, Anext);
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(2, A0, Anext, 1, Anext);
- void *thisNode = __hpvm__getNode();
- void *parentNode = __hpvm__getParentNode(thisNode);
+ void *thisNode = __visc__getNode();
+ void *parentNode = __visc__getParentNode(thisNode);
- int lx = __hpvm__getNodeInstanceID_x(thisNode);
- int ly = __hpvm__getNodeInstanceID_y(thisNode);
- int lz = __hpvm__getNodeInstanceID_z(thisNode);
+ int lx = __visc__getNodeInstanceID_x(thisNode);
+ int ly = __visc__getNodeInstanceID_y(thisNode);
+ int lz = __visc__getNodeInstanceID_z(thisNode);
- int gx = __hpvm__getNodeInstanceID_x(parentNode);
- int gy = __hpvm__getNodeInstanceID_y(parentNode);
- int gz = __hpvm__getNodeInstanceID_z(parentNode);
+ int gx = __visc__getNodeInstanceID_x(parentNode);
+ int gy = __visc__getNodeInstanceID_y(parentNode);
+ int gz = __visc__getNodeInstanceID_z(parentNode);
- int gridx = __hpvm__getNumNodeInstances_x(thisNode);
- int gridy = __hpvm__getNumNodeInstances_y(thisNode);
- int gridz = __hpvm__getNumNodeInstances_z(thisNode);
+ int gridx = __visc__getNumNodeInstances_x(thisNode);
+ int gridy = __visc__getNumNodeInstances_y(thisNode);
+ int gridz = __visc__getNumNodeInstances_z(thisNode);
int i = gx * gridx + lx + 1;
int j = gy * gridy + ly + 1;
@@ -78,65 +78,65 @@ void naive_kernel(float c0, float c1, float *A0, size_t bytes_A0, float *Anext,
void stencilLvl1(float c0, float c1, float *A0, size_t bytes_A0, float *Anext,
size_t bytes_Anext, int nx, int ny, int nz, size_t dim_X1,
size_t dim_Y1, size_t dim_Z1) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(2, A0, Anext, 1, Anext);
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(2, A0, Anext, 1, Anext);
void *stencil_node =
- __hpvm__createNodeND(3, naive_kernel, dim_X1, dim_Y1, dim_Z1);
- __hpvm__bindIn(stencil_node, 0, 0, 0);
- __hpvm__bindIn(stencil_node, 1, 1, 0);
- __hpvm__bindIn(stencil_node, 2, 2, 0);
- __hpvm__bindIn(stencil_node, 3, 3, 0);
- __hpvm__bindIn(stencil_node, 4, 4, 0);
- __hpvm__bindIn(stencil_node, 5, 5, 0);
- __hpvm__bindIn(stencil_node, 6, 6, 0);
- __hpvm__bindIn(stencil_node, 7, 7, 0);
- __hpvm__bindIn(stencil_node, 8, 8, 0);
+ __visc__createNodeND(3, naive_kernel, dim_X1, dim_Y1, dim_Z1);
+ __visc__bindIn(stencil_node, 0, 0, 0);
+ __visc__bindIn(stencil_node, 1, 1, 0);
+ __visc__bindIn(stencil_node, 2, 2, 0);
+ __visc__bindIn(stencil_node, 3, 3, 0);
+ __visc__bindIn(stencil_node, 4, 4, 0);
+ __visc__bindIn(stencil_node, 5, 5, 0);
+ __visc__bindIn(stencil_node, 6, 6, 0);
+ __visc__bindIn(stencil_node, 7, 7, 0);
+ __visc__bindIn(stencil_node, 8, 8, 0);
}
void stencilLvl2(float c0, float c1, float *A0, size_t bytes_A0, float *Anext,
size_t bytes_Anext, int nx, int ny, int nz, size_t dim_X1,
size_t dim_Y1, size_t dim_Z1, size_t dim_X2, size_t dim_Y2,
size_t dim_Z2) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(2, A0, Anext, 1, Anext);
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(2, A0, Anext, 1, Anext);
void *stencil_node =
- __hpvm__createNodeND(3, stencilLvl1, dim_X2, dim_Y2, dim_Z2);
- __hpvm__bindIn(stencil_node, 0, 0, 0);
- __hpvm__bindIn(stencil_node, 1, 1, 0);
- __hpvm__bindIn(stencil_node, 2, 2, 0);
- __hpvm__bindIn(stencil_node, 3, 3, 0);
- __hpvm__bindIn(stencil_node, 4, 4, 0);
- __hpvm__bindIn(stencil_node, 5, 5, 0);
- __hpvm__bindIn(stencil_node, 6, 6, 0);
- __hpvm__bindIn(stencil_node, 7, 7, 0);
- __hpvm__bindIn(stencil_node, 8, 8, 0);
- __hpvm__bindIn(stencil_node, 9, 9, 0);
- __hpvm__bindIn(stencil_node, 10, 10, 0);
- __hpvm__bindIn(stencil_node, 11, 11, 0);
+ __visc__createNodeND(3, stencilLvl1, dim_X2, dim_Y2, dim_Z2);
+ __visc__bindIn(stencil_node, 0, 0, 0);
+ __visc__bindIn(stencil_node, 1, 1, 0);
+ __visc__bindIn(stencil_node, 2, 2, 0);
+ __visc__bindIn(stencil_node, 3, 3, 0);
+ __visc__bindIn(stencil_node, 4, 4, 0);
+ __visc__bindIn(stencil_node, 5, 5, 0);
+ __visc__bindIn(stencil_node, 6, 6, 0);
+ __visc__bindIn(stencil_node, 7, 7, 0);
+ __visc__bindIn(stencil_node, 8, 8, 0);
+ __visc__bindIn(stencil_node, 9, 9, 0);
+ __visc__bindIn(stencil_node, 10, 10, 0);
+ __visc__bindIn(stencil_node, 11, 11, 0);
}
void stencilLvl3(float c0, float c1, float *A0, size_t bytes_A0, float *Anext,
size_t bytes_Anext, int nx, int ny, int nz, size_t dim_X1,
size_t dim_Y1, size_t dim_Z1, size_t dim_X2, size_t dim_Y2,
size_t dim_Z2) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(2, A0, Anext, 1, Anext);
- void *stencil_node = __hpvm__createNodeND(0, stencilLvl2);
- __hpvm__bindIn(stencil_node, 0, 0, 0);
- __hpvm__bindIn(stencil_node, 1, 1, 0);
- __hpvm__bindIn(stencil_node, 2, 2, 0);
- __hpvm__bindIn(stencil_node, 3, 3, 0);
- __hpvm__bindIn(stencil_node, 4, 4, 0);
- __hpvm__bindIn(stencil_node, 5, 5, 0);
- __hpvm__bindIn(stencil_node, 6, 6, 0);
- __hpvm__bindIn(stencil_node, 7, 7, 0);
- __hpvm__bindIn(stencil_node, 8, 8, 0);
- __hpvm__bindIn(stencil_node, 9, 9, 0);
- __hpvm__bindIn(stencil_node, 10, 10, 0);
- __hpvm__bindIn(stencil_node, 11, 11, 0);
- __hpvm__bindIn(stencil_node, 12, 12, 0);
- __hpvm__bindIn(stencil_node, 13, 13, 0);
- __hpvm__bindIn(stencil_node, 14, 14, 0);
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(2, A0, Anext, 1, Anext);
+ void *stencil_node = __visc__createNodeND(0, stencilLvl2);
+ __visc__bindIn(stencil_node, 0, 0, 0);
+ __visc__bindIn(stencil_node, 1, 1, 0);
+ __visc__bindIn(stencil_node, 2, 2, 0);
+ __visc__bindIn(stencil_node, 3, 3, 0);
+ __visc__bindIn(stencil_node, 4, 4, 0);
+ __visc__bindIn(stencil_node, 5, 5, 0);
+ __visc__bindIn(stencil_node, 6, 6, 0);
+ __visc__bindIn(stencil_node, 7, 7, 0);
+ __visc__bindIn(stencil_node, 8, 8, 0);
+ __visc__bindIn(stencil_node, 9, 9, 0);
+ __visc__bindIn(stencil_node, 10, 10, 0);
+ __visc__bindIn(stencil_node, 11, 11, 0);
+ __visc__bindIn(stencil_node, 12, 12, 0);
+ __visc__bindIn(stencil_node, 13, 13, 0);
+ __visc__bindIn(stencil_node, 14, 14, 0);
}
int main(int argc, char **argv) {
@@ -195,11 +195,11 @@ int main(int argc, char **argv) {
fclose(fp);
pb_InitializeTimerSet(&timers);
- __hpvm__init();
+ __visc__init();
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_TRACK);
- llvm_hpvm_track_mem(h_A0, sizeof(float) * size);
- llvm_hpvm_track_mem(h_Anext, sizeof(float) * size);
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_TRACK);
+ llvm_visc_track_mem(h_A0, sizeof(float) * size);
+ llvm_visc_track_mem(h_Anext, sizeof(float) * size);
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
@@ -241,9 +241,9 @@ int main(int argc, char **argv) {
grid[1] / block[1],
grid[2] / block[2]};
*(RootIn *)root_in = root_in_local;
- void *stencilDFG = __hpvm__launch(0, stencilLvl3, root_in);
+ void *stencilDFG = __visc__launch(0, stencilLvl3, root_in);
- __hpvm__wait(stencilDFG);
+ __visc__wait(stencilDFG);
// printf("iteration %d\n",t);
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
float *h_temp = h_A0;
@@ -255,19 +255,19 @@ int main(int argc, char **argv) {
h_A0 = h_Anext;
h_Anext = h_temp;
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
- llvm_hpvm_request_mem(h_Anext, bytes);
+ llvm_visc_request_mem(h_Anext, bytes);
printf("A[126,1,1] = %f\n", h_Anext[Index3D(nx, ny, 126, 1, 1)]);
printf("A[125,1,1] = %f\n", h_Anext[Index3D(nx, ny, 125, 1, 1)]);
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_UNTRACK);
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_UNTRACK);
- llvm_hpvm_untrack_mem(h_A0);
- llvm_hpvm_untrack_mem(h_Anext);
+ llvm_visc_untrack_mem(h_A0);
+ llvm_visc_untrack_mem(h_Anext);
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
pb_PrintTimerSet(&timers);
- __hpvm__cleanup();
+ __visc__cleanup();
if (parameters->outFile) {
/*pb_SwitchToTimer(&timers, pb_TimerID_IO);*/
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/hpvm_vec/common.h b/hpvm/test/parboil/benchmarks/stencil/src/visc_vec/common.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/stencil/src/hpvm_vec/common.h
rename to hpvm/test/parboil/benchmarks/stencil/src/visc_vec/common.h
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/hpvm_vec/stencil.c b/hpvm/test/parboil/benchmarks/stencil/src/visc_vec/stencil.c
similarity index 90%
rename from hpvm/test/parboil/benchmarks/stencil/src/hpvm_vec/stencil.c
rename to hpvm/test/parboil/benchmarks/stencil/src/visc_vec/stencil.c
index 35c5ed960c..bb6e45c932 100644
--- a/hpvm/test/parboil/benchmarks/stencil/src/hpvm_vec/stencil.c
+++ b/hpvm/test/parboil/benchmarks/stencil/src/visc_vec/stencil.c
@@ -9,11 +9,11 @@
#include "common.h"
#include "file.h"
-#include <hpvm.h>
#include <parboil.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
+#include <visc.h>
static int read_data(float *A0, int nx, int ny, int nz, FILE *fp) {
int s = 0;
@@ -31,7 +31,7 @@ static int read_data(float *A0, int nx, int ny, int nz, FILE *fp) {
void naive_kernel(float c0, float c1, float *A0, float *Anext, int nx, int ny,
int nz) {
- __hpvm__attributes(2, A0, Anext, 1, Anext);
+ __visc__attributes(2, A0, Anext, 1, Anext);
int i = get_global_id(0) + 1;
int j = get_global_id(1) + 1;
int k = get_global_id(2) + 1;
@@ -106,11 +106,11 @@ int main(int argc, char **argv) {
fclose(fp);
pb_InitializeTimerSet(&timers);
- __hpvm__init();
+ __visc__init();
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_TRACK);
- llvm_hpvm_track_mem(h_A0, sizeof(float) * size);
- llvm_hpvm_track_mem(h_Anext, sizeof(float) * size);
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_TRACK);
+ llvm_visc_track_mem(h_A0, sizeof(float) * size);
+ llvm_visc_track_mem(h_Anext, sizeof(float) * size);
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
@@ -133,11 +133,11 @@ int main(int argc, char **argv) {
printf("A[125,1,1] = %f\n", h_A0[Index3D(nx, ny, 125, 1, 1)]);
for (t = 0; t < iteration; t++) {
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
- unsigned stencilDFG = __hpvm__node(
+ unsigned stencilDFG = __visc__node(
naive_kernel, 2, 3, block[0], block[1], block[2], grid[0] / block[0],
grid[1] / block[1], grid[2] / block[2], 9, (float)c0, (float)c1, h_A0,
bytes, h_Anext, bytes, nx, ny, nz, 0);
- __hpvm__wait(stencilDFG);
+ __visc__wait(stencilDFG);
// printf("iteration %d\n",t);
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
float *h_temp = h_A0;
@@ -149,19 +149,19 @@ int main(int argc, char **argv) {
h_A0 = h_Anext;
h_Anext = h_temp;
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
- llvm_hpvm_request_mem(h_Anext, bytes);
+ llvm_visc_request_mem(h_Anext, bytes);
printf("A[126,1,1] = %f\n", h_Anext[Index3D(nx, ny, 126, 1, 1)]);
printf("A[125,1,1] = %f\n", h_Anext[Index3D(nx, ny, 125, 1, 1)]);
- pb_SwitchToTimer(&timers, hpvm_TimerID_MEM_UNTRACK);
+ pb_SwitchToTimer(&timers, visc_TimerID_MEM_UNTRACK);
- llvm_hpvm_untrack_mem(h_A0);
- llvm_hpvm_untrack_mem(h_Anext);
+ llvm_visc_untrack_mem(h_A0);
+ llvm_visc_untrack_mem(h_Anext);
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
pb_PrintTimerSet(&timers);
- __hpvm__cleanup();
+ __visc__cleanup();
if (parameters->outFile) {
/*pb_SwitchToTimer(&timers, pb_TimerID_IO);*/
diff --git a/hpvm/test/parboil/benchmarks/stencil/src/visc_vec/stencil.visc.ll b/hpvm/test/parboil/benchmarks/stencil/src/visc_vec/stencil.visc.ll
new file mode 100644
index 0000000000..7dc32f3760
--- /dev/null
+++ b/hpvm/test/parboil/benchmarks/stencil/src/visc_vec/stencil.visc.ll
@@ -0,0 +1,673 @@
+; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG -o %t.ll -S %s
+; RUN: llvm-link %llvm_src/../libclc/built_libs/nvptx--nvidiacl.bc %s.kernels.ll -o %t.ll.kernels.linked.bc
+; RUN: clang -O3 -target nvptx %t.ll.kernels.linked.bc -S -o %s.nvptx.s
+; RUN: llvm-link %t.ll %llvm_src/projects/visc-rt/visc-rt.ll parboil.ll -S -o %t.linked.ll
+; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -o %t.bin
+; ModuleID = 'build/visc_vec_default/stencil.ll'
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%rtype = type {}
+%struct.pb_TimerSet = type { i32, %struct.pb_async_time_marker_list*, i64, i64, [24 x %struct.pb_Timer], [24 x %struct.pb_SubTimerList*] }
+%struct.pb_async_time_marker_list = type { i8*, i32, i8*, %struct.pb_async_time_marker_list* }
+%struct.pb_Timer = type { i32, i64, i64 }
+%struct.pb_SubTimerList = type { %struct.pb_SubTimer*, %struct.pb_SubTimer* }
+%struct.pb_SubTimer = type { i8*, %struct.pb_Timer, %struct.pb_SubTimer* }
+%struct.pb_Parameters = type { i8*, i8** }
+%struct._IO_FILE = type { i32, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, i8*, %struct._IO_marker*, %struct._IO_FILE*, i32, i32, i64, i16, i8, [1 x i8], i8*, i64, i8*, i8*, i8*, i8*, i64, i32, [20 x i8] }
+%struct._IO_marker = type { %struct._IO_marker*, %struct._IO_FILE*, i32 }
+%struct.arg = type <{ float, float, float*, i64, float*, i64, i32, i32, i32, i32, i32, i32, i32, i32, i32, %rtype }>
+
+@.str3 = private unnamed_addr constant [3 x i8] c"rb\00", align 1
+@.str4 = private unnamed_addr constant [37 x i8] c"grid(%d, %d, %d), block(%d, %d, %d)\0A\00", align 1
+@.str5 = private unnamed_addr constant [17 x i8] c"A[126,1,1] = %f\0A\00", align 1
+@.str6 = private unnamed_addr constant [17 x i8] c"A[125,1,1] = %f\0A\00", align 1
+@str = private unnamed_addr constant [46 x i8] c"OpenCL accelerated 7 points stencil codes****\00"
+@str7 = private unnamed_addr constant [45 x i8] c"Author: Li-Wen Chang <lchang20@illinois.edu>\00"
+@str8 = private unnamed_addr constant [106 x i8] c"Usage: probe nx ny nz t\0Anx: the grid size x\0Any: the grid size y\0Anz: the grid size z\0At: the iteration time\00"
+@viscTimerSet_GenVISC = common global i8* null
+@0 = internal constant [14 x i8] c"GenVISC_Timer\00"
+
+; Function Attrs: nounwind uwtable
+define %rtype @naive_kernel(float %c0, float %c1, float* in %A0, i64 %bytes_A0, float* in out %Anext, i64 %bytes_Anext, i32 %nx, i32 %ny, i32 %nz) #0 {
+entry:
+ %naive_kernel.node = call i8* @llvm.visc.getNode()
+ %naive_kernel.parentNode = call i8* @llvm.visc.getParentNode(i8* %naive_kernel.node)
+ %a0 = call i32 @llvm.visc.getNodeInstanceID.x(i8* %naive_kernel.parentNode)
+ %a1 = call i32 @llvm.visc.getNumNodeInstances.x(i8* %naive_kernel.node)
+ %a2 = mul i32 %a0, %a1
+ %a3 = call i32 @llvm.visc.getNodeInstanceID.x(i8* %naive_kernel.node)
+ %a4 = add i32 %a2, %a3
+ ;%add = add nsw i32 %4, 1
+ %a5 = call i32 @llvm.visc.getNodeInstanceID.y(i8* %naive_kernel.parentNode)
+ %a6 = call i32 @llvm.visc.getNumNodeInstances.y(i8* %naive_kernel.node)
+ %a7 = mul i32 %a5, %a6
+ %a8 = call i32 @llvm.visc.getNodeInstanceID.y(i8* %naive_kernel.node)
+ %a9 = add i32 %a7, %a8
+ ;%add3 = add nsw i32 %9, 1
+ %a10 = call i32 @llvm.visc.getNodeInstanceID.z(i8* %naive_kernel.parentNode)
+ %a11 = call i32 @llvm.visc.getNumNodeInstances.z(i8* %naive_kernel.node)
+ %a12 = mul i32 %a10, %a11
+ %a13 = call i32 @llvm.visc.getNodeInstanceID.z(i8* %naive_kernel.node)
+ %a14 = add i32 %a12, %a13
+ ;%sub = add nsw i32 %nx, -1
+ ;%cmp = icmp slt i32 %add, %sub
+ ;br i1 %cmp, label %if.then, label %if.end
+
+
+ ;%call = tail call i32 @get_global_id(i32 0) #2
+ ;%mul = shl i32 %call, 2
+ %mul = shl i32 %a4, 2
+ %add258 = or i32 %mul, 1
+ ;%call1 = tail call i32 @get_global_id(i32 1) #2
+ ;%add2 = add i32 %call1, 1
+ %add2 = add i32 %a9, 1
+ ;%call3 = tail call i32 @get_global_id(i32 2) #2
+ ;%add4 = add i32 %call3, 1
+ %add4 = add i32 %a14, 1
+ %sub = add i32 %add258, 3
+ %sub6 = add i32 %nx, -1
+ %cmp = icmp slt i32 %sub, %sub6
+ br i1 %cmp, label %if.then, label %if.else
+
+if.then: ; preds = %entry
+ %mul7 = mul nsw i32 %add4, %ny
+ %add8 = add nsw i32 %mul7, %add2
+ %mul9 = mul nsw i32 %add8, %nx
+ %add11 = add i32 %sub, %mul9
+ %add.ptr = getelementptr inbounds float* %A0, i32 %add11
+ ;%call12 = tail call <4 x float> @_Z6vload4jPKU3AS1f(i32 0, float* %add.ptr) #2
+ %vadd.ptr = bitcast float* %add.ptr to <4 x float>*
+ %call12 = load <4 x float>* %vadd.ptr
+
+ %add13 = add i32 %a14, 2
+ %mul14 = mul nsw i32 %add13, %ny
+ %add15 = add nsw i32 %mul14, %add2
+ %mul16 = mul nsw i32 %add15, %nx
+ %add18 = add i32 %sub, %mul16
+ %add.ptr19 = getelementptr inbounds float* %A0, i32 %add18
+ ;%call20 = tail call <4 x float> @_Z6vload4jPKU3AS1f(i32 0, float* %add.ptr19) #2
+ %vadd.ptr19 = bitcast float* %add.ptr19 to <4 x float>*
+ %call20 = load <4 x float>* %vadd.ptr19
+
+ %mul22 = mul nsw i32 %a14, %ny
+ %add23 = add nsw i32 %mul22, %add2
+ %mul24 = mul nsw i32 %add23, %nx
+ %add26 = add i32 %sub, %mul24
+ %add.ptr27 = getelementptr inbounds float* %A0, i32 %add26
+ ;%call28 = tail call <4 x float> @_Z6vload4jPKU3AS1f(i32 0, float* %add.ptr27) #2
+ %vadd.ptr27 = bitcast float* %add.ptr27 to <4 x float>*
+ %call28 = load <4 x float>* %vadd.ptr27
+
+ %add29 = add i32 %a9, 2
+ %add31 = add nsw i32 %add29, %mul7
+ %mul32 = mul nsw i32 %add31, %nx
+ %add34 = add i32 %sub, %mul32
+ %add.ptr35 = getelementptr inbounds float* %A0, i32 %add34
+ ;%call36 = tail call <4 x float> @_Z6vload4jPKU3AS1f(i32 0, float* %add.ptr35) #2
+ %vadd.ptr35 = bitcast float* %add.ptr35 to <4 x float>*
+ %call36 = load <4 x float>* %vadd.ptr35
+
+ %add39 = add nsw i32 %mul7, %a9
+ %mul40 = mul nsw i32 %add39, %nx
+ %add42 = add i32 %sub, %mul40
+ %add.ptr43 = getelementptr inbounds float* %A0, i32 %add42
+ ;%call44 = tail call <4 x float> @_Z6vload4jPKU3AS1f(i32 0, float* %add.ptr43) #2
+ %vadd.ptr43 = bitcast float* %add.ptr43 to <4 x float>*
+ %call44 = load <4 x float>* %vadd.ptr43
+
+ %add49 = add i32 %add258, 4
+ %add50 = add i32 %add49, %mul9
+ %arrayidx = getelementptr inbounds float* %A0, i32 %add50
+ %0 = load float* %arrayidx, align 4, !tbaa !2
+ %add55261 = or i32 %mul, 3
+ %add56 = add i32 %add55261, %mul9
+ %arrayidx57 = getelementptr inbounds float* %A0, i32 %add56
+ %1 = load float* %arrayidx57, align 4, !tbaa !2
+ %2 = shufflevector <4 x float> %call12, <4 x float> undef, <3 x i32> <i32 1, i32 2, i32 3>
+ %vext = shufflevector <3 x float> %2, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef>
+ %vecinit58 = insertelement <4 x float> %vext, float %0, i32 3
+ %vecinit60 = insertelement <4 x float> undef, float %1, i32 0
+ %vecinit62 = shufflevector <4 x float> %vecinit60, <4 x float> %call12, <4 x i32> <i32 0, i32 4, i32 5, i32 6>
+ %splat.splatinsert = insertelement <4 x float> undef, float %c1, i32 0
+ %splat.splat = shufflevector <4 x float> %splat.splatinsert, <4 x float> undef, <4 x i32> zeroinitializer
+ %add63 = fadd fast <4 x float> %call20, %call28
+ %add64 = fadd fast <4 x float> %add63, %call36
+ %add65 = fadd fast <4 x float> %add64, %call44
+ %add66 = fadd fast <4 x float> %add65, %vecinit58
+ %add67 = fadd fast <4 x float> %add66, %vecinit62
+ %mul68 = fmul fast <4 x float> %splat.splat, %add67
+ %splat.splatinsert69 = insertelement <4 x float> undef, float %c0, i32 0
+ %splat.splat70 = shufflevector <4 x float> %splat.splatinsert69, <4 x float> undef, <4 x i32> zeroinitializer
+ %mul71 = fmul fast <4 x float> %splat.splat70, %call12
+ %sub72 = fsub fast <4 x float> %mul68, %mul71
+ %add.ptr78 = getelementptr inbounds float* %Anext, i32 %add11
+ ;tail call void @_Z7vstore4Dv4_fjPU3AS1f(<4 x float> %sub72, i32 0, float* %add.ptr78) #2
+ %vadd.ptr78 = bitcast float* %add.ptr78 to <4 x float>*
+ store <4 x float> %sub72, <4 x float>* %vadd.ptr78
+
+ br label %if.end146
+
+if.else: ; preds = %entry
+ %cmp80 = icmp slt i32 %add258, %sub6
+ br i1 %cmp80, label %for.body.lr.ph, label %if.end146
+
+for.body.lr.ph: ; preds = %if.else
+ %add84 = add i32 %a14, 2
+ %mul85 = mul nsw i32 %add84, %ny
+ %add86 = add nsw i32 %mul85, %add2
+ %mul87 = mul nsw i32 %add86, %nx
+ %add88 = add i32 %mul87, 3
+ %mul92 = mul nsw i32 %a14, %ny
+ %add93 = add nsw i32 %mul92, %add2
+ %mul94 = mul nsw i32 %add93, %nx
+ %add95 = add i32 %mul94, 3
+ %add99 = add i32 %a9, 2
+ %mul100 = mul nsw i32 %add4, %ny
+ %add101 = add nsw i32 %add99, %mul100
+ %mul102 = mul nsw i32 %add101, %nx
+ %add103 = add i32 %mul102, 3
+ %add109 = add nsw i32 %mul100, %a9
+ %mul110 = mul nsw i32 %add109, %nx
+ %add111 = add i32 %mul110, 3
+ %add117 = add nsw i32 %mul100, %add2
+ %mul118 = mul nsw i32 %add117, %nx
+ %add119 = add i32 %mul118, 3
+ %add127 = add i32 %mul118, 2
+ br label %for.body
+
+for.body: ; preds = %for.body, %for.body.lr.ph
+ %vid.0260 = phi i32 [ %add258, %for.body.lr.ph ], [ %add115, %for.body ]
+ %add89 = add i32 %add88, %vid.0260
+ %arrayidx90 = getelementptr inbounds float* %A0, i32 %add89
+ %3 = load float* %arrayidx90, align 4, !tbaa !2
+ %add96 = add i32 %add95, %vid.0260
+ %arrayidx97 = getelementptr inbounds float* %A0, i32 %add96
+ %4 = load float* %arrayidx97, align 4, !tbaa !2
+ %add98 = fadd fast float %3, %4
+ %add104 = add i32 %add103, %vid.0260
+ %arrayidx105 = getelementptr inbounds float* %A0, i32 %add104
+ %5 = load float* %arrayidx105, align 4, !tbaa !2
+ %add106 = fadd fast float %add98, %5
+ %add112 = add i32 %add111, %vid.0260
+ %arrayidx113 = getelementptr inbounds float* %A0, i32 %add112
+ %6 = load float* %arrayidx113, align 4, !tbaa !2
+ %add114 = fadd fast float %add106, %6
+ %add115 = add nsw i32 %vid.0260, 1
+ %add120 = add i32 %add119, %add115
+ %arrayidx121 = getelementptr inbounds float* %A0, i32 %add120
+ %7 = load float* %arrayidx121, align 4, !tbaa !2
+ %add122 = fadd fast float %add114, %7
+ %add128 = add i32 %add127, %vid.0260
+ %arrayidx129 = getelementptr inbounds float* %A0, i32 %add128
+ %8 = load float* %arrayidx129, align 4, !tbaa !2
+ %add130 = fadd fast float %add122, %8
+ %mul131 = fmul fast float %add130, %c1
+ %add136 = add i32 %add119, %vid.0260
+ %arrayidx137 = getelementptr inbounds float* %A0, i32 %add136
+ %9 = load float* %arrayidx137, align 4, !tbaa !2
+ %mul138 = fmul fast float %9, %c0
+ %sub139 = fsub fast float %mul131, %mul138
+ %arrayidx145 = getelementptr inbounds float* %Anext, i32 %add136
+ store float %sub139, float* %arrayidx145, align 4, !tbaa !2
+ %exitcond = icmp eq i32 %add115, %sub6
+ br i1 %exitcond, label %if.end146, label %for.body
+
+if.end146: ; preds = %for.body, %if.else, %if.then
+ ;ret void
+
+
+
+
+
+;if.then: ; preds = %entry
+ ;%add5 = add nsw i32 %14, 1
+ ;%add6 = add nsw i32 %14, 2
+ ;%mul = mul nsw i32 %add6, %ny
+ ;%add7 = add nsw i32 %mul, %add3
+ ;%mul8 = mul nsw i32 %add7, %nx
+ ;%add9 = add i32 %4, 4
+ ;%add10 = add i32 %add9, %mul8
+ ;%idxprom = sext i32 %add10 to i64
+ ;%arrayidx = getelementptr inbounds float* %A0, i64 %idxprom
+ ;%15 = load float* %arrayidx, align 4, !tbaa !2
+ ;%mul12 = mul nsw i32 %14, %ny
+ ;%add13 = add nsw i32 %mul12, %add3
+ ;%mul14 = mul nsw i32 %add13, %nx
+ ;%add16 = add i32 %add9, %mul14
+ ;%idxprom17 = sext i32 %add16 to i64
+ ;%arrayidx18 = getelementptr inbounds float* %A0, i64 %idxprom17
+ ;%16 = load float* %arrayidx18, align 4, !tbaa !2
+ ;%add19 = fadd fast float %15, %16
+ ;%add20 = add nsw i32 %9, 2
+ ;%mul21 = mul nsw i32 %add5, %ny
+ ;%add22 = add nsw i32 %add20, %mul21
+ ;%mul23 = mul nsw i32 %add22, %nx
+ ;%add25 = add i32 %add9, %mul23
+ ;%idxprom26 = sext i32 %add25 to i64
+ ;%arrayidx27 = getelementptr inbounds float* %A0, i64 %idxprom26
+ ;%17 = load float* %arrayidx27, align 4, !tbaa !2
+ ;%add28 = fadd fast float %add19, %17
+ ;%add31 = add nsw i32 %mul21, %9
+ ;%mul32 = mul nsw i32 %add31, %nx
+ ;%add34 = add i32 %add9, %mul32
+ ;%idxprom35 = sext i32 %add34 to i64
+ ;%arrayidx36 = getelementptr inbounds float* %A0, i64 %idxprom35
+ ;%18 = load float* %arrayidx36, align 4, !tbaa !2
+ ;%add37 = fadd fast float %add28, %18
+ ;%add40 = add nsw i32 %mul21, %add3
+ ;%mul41 = mul nsw i32 %add40, %nx
+ ;%add42 = add i32 %4, 5
+ ;%add43 = add i32 %add42, %mul41
+ ;%idxprom44 = sext i32 %add43 to i64
+ ;%arrayidx45 = getelementptr inbounds float* %A0, i64 %idxprom44
+ ;%19 = load float* %arrayidx45, align 4, !tbaa !2
+ ;%add46 = fadd fast float %add37, %19
+ ;%add51 = add i32 %4, 3
+ ;%add52 = add i32 %add51, %mul41
+ ;%idxprom53 = sext i32 %add52 to i64
+ ;%arrayidx54 = getelementptr inbounds float* %A0, i64 %idxprom53
+ ;%20 = load float* %arrayidx54, align 4, !tbaa !2
+ ;%add55 = fadd fast float %add46, %20
+ ;%mul56 = fmul fast float %add55, %c1
+ ;%add61 = add i32 %add9, %mul41
+ ;%idxprom62 = sext i32 %add61 to i64
+ ;%arrayidx63 = getelementptr inbounds float* %A0, i64 %idxprom62
+ ;%21 = load float* %arrayidx63, align 4, !tbaa !2
+ ;%mul64 = fmul fast float %21, %c0
+ ;%sub65 = fsub fast float %mul56, %mul64
+ ;%arrayidx72 = getelementptr inbounds float* %Anext, i64 %idxprom62
+ ;store float %sub65, float* %arrayidx72, align 4, !tbaa !2
+ ;br label %if.end
+
+;if.end: ; preds = %if.then, %entry
+ ret %rtype undef
+}
+
+; Function Attrs: nounwind uwtable
+define i32 @main(i32 %argc, i8** %argv) #0 {
+entry:
+ %argc.addr = alloca i32, align 4
+ %timers = alloca %struct.pb_TimerSet, align 8
+ store i32 %argc, i32* %argc.addr, align 4, !tbaa !5
+ %0 = bitcast %struct.pb_TimerSet* %timers to i8*
+ call void @llvm.lifetime.start(i64 800, i8* %0) #1
+ %puts = call i32 @puts(i8* getelementptr inbounds ([46 x i8]* @str, i64 0, i64 0))
+ %puts186 = call i32 @puts(i8* getelementptr inbounds ([45 x i8]* @str7, i64 0, i64 0))
+ %call2 = call %struct.pb_Parameters* @pb_ReadParameters(i32* %argc.addr, i8** %argv) #1
+ %1 = load i32* %argc.addr, align 4, !tbaa !5
+ %cmp = icmp slt i32 %1, 5
+ br i1 %cmp, label %if.then, label %if.end
+
+if.then: ; preds = %entry
+ %puts187 = call i32 @puts(i8* getelementptr inbounds ([106 x i8]* @str8, i64 0, i64 0))
+ br label %cleanup
+
+if.end: ; preds = %entry
+ %arrayidx = getelementptr inbounds i8** %argv, i64 1
+ %2 = load i8** %arrayidx, align 8, !tbaa !6
+ %call.i = call i64 @strtol(i8* nocapture %2, i8** null, i32 10) #1
+ %conv.i = trunc i64 %call.i to i32
+ %cmp5 = icmp slt i32 %conv.i, 1
+ br i1 %cmp5, label %cleanup, label %if.end7
+
+if.end7: ; preds = %if.end
+ %arrayidx8 = getelementptr inbounds i8** %argv, i64 2
+ %3 = load i8** %arrayidx8, align 8, !tbaa !6
+ %call.i188 = call i64 @strtol(i8* nocapture %3, i8** null, i32 10) #1
+ %conv.i189 = trunc i64 %call.i188 to i32
+ %cmp10 = icmp slt i32 %conv.i189, 1
+ br i1 %cmp10, label %cleanup, label %if.end12
+
+if.end12: ; preds = %if.end7
+ %arrayidx13 = getelementptr inbounds i8** %argv, i64 3
+ %4 = load i8** %arrayidx13, align 8, !tbaa !6
+ %call.i190 = call i64 @strtol(i8* nocapture %4, i8** null, i32 10) #1
+ %conv.i191 = trunc i64 %call.i190 to i32
+ %cmp15 = icmp slt i32 %conv.i191, 1
+ br i1 %cmp15, label %cleanup, label %if.end17
+
+if.end17: ; preds = %if.end12
+ %arrayidx18 = getelementptr inbounds i8** %argv, i64 4
+ %5 = load i8** %arrayidx18, align 8, !tbaa !6
+ %call.i192 = call i64 @strtol(i8* nocapture %5, i8** null, i32 10) #1
+ %conv.i193 = trunc i64 %call.i192 to i32
+ %cmp20 = icmp slt i32 %conv.i193, 1
+ br i1 %cmp20, label %cleanup, label %for.cond1.preheader.lr.ph.i
+
+for.cond1.preheader.lr.ph.i: ; preds = %if.end17
+ %mul = shl i64 %call.i, 32
+ %mul23 = mul i64 %mul, %call.i188
+ %sext = mul i64 %mul23, %call.i190
+ %add = ashr exact i64 %sext, 30
+ %mul24 = add i64 %add, 12
+ %call25 = call noalias i8* @malloc(i64 %mul24) #1
+ %6 = bitcast i8* %call25 to float*
+ %call27 = call noalias i8* @malloc(i64 %mul24) #1
+ %7 = bitcast i8* %call27 to float*
+ %inpFiles = getelementptr inbounds %struct.pb_Parameters* %call2, i64 0, i32 1
+ %8 = load i8*** %inpFiles, align 8, !tbaa !6
+ %9 = load i8** %8, align 8, !tbaa !6
+ %call29 = call %struct._IO_FILE* @fopen(i8* %9, i8* getelementptr inbounds ([3 x i8]* @.str3, i64 0, i64 0)) #1
+ %add.ptr = getelementptr inbounds i8* %call25, i64 12
+ %10 = bitcast i8* %add.ptr to float*
+ %cmp24.i = icmp sgt i32 %conv.i189, 0
+ %cmp51.i = icmp sgt i32 %conv.i, 0
+ %or.cond = and i1 %cmp24.i, %cmp51.i
+ br i1 %or.cond, label %for.cond4.preheader.lr.ph.us.i.preheader.split.us, label %read_data.exit
+
+for.cond4.preheader.lr.ph.us.i.preheader.split.us: ; preds = %for.cond1.preheader.lr.ph.i
+ %11 = mul i32 %conv.i, %conv.i189
+ br label %for.body6.lr.ph.us.us.i.preheader.us
+
+for.body6.lr.ph.us.us.i.us: ; preds = %for.body6.lr.ph.us.us.i.preheader.us, %for.inc8.us.us.i.us
+ %j.06.us.us.i.us = phi i32 [ %inc9.us.us.i.us, %for.inc8.us.us.i.us ], [ 0, %for.body6.lr.ph.us.us.i.preheader.us ]
+ %s.15.us.us.i.us = phi i32 [ %14, %for.inc8.us.us.i.us ], [ %s.09.us.i.us, %for.body6.lr.ph.us.us.i.preheader.us ]
+ %12 = sext i32 %s.15.us.us.i.us to i64
+ br label %for.body6.us.us.i.us
+
+for.body6.us.us.i.us: ; preds = %for.body6.us.us.i.us, %for.body6.lr.ph.us.us.i.us
+ %indvars.iv.i.us = phi i64 [ %indvars.iv.next.i.us, %for.body6.us.us.i.us ], [ %12, %for.body6.lr.ph.us.us.i.us ]
+ %k.03.us.us.i.us = phi i32 [ %inc7.us.us.i.us, %for.body6.us.us.i.us ], [ 0, %for.body6.lr.ph.us.us.i.us ]
+ %add.ptr.us.us.i.us = getelementptr inbounds float* %10, i64 %indvars.iv.i.us
+ %13 = bitcast float* %add.ptr.us.us.i.us to i8*
+ %call.us.us.i.us = call i64 @fread(i8* %13, i64 4, i64 1, %struct._IO_FILE* %call29) #1
+ %indvars.iv.next.i.us = add i64 %indvars.iv.i.us, 1
+ %inc7.us.us.i.us = add nsw i32 %k.03.us.us.i.us, 1
+ %exitcond.i.us = icmp eq i32 %inc7.us.us.i.us, %conv.i
+ br i1 %exitcond.i.us, label %for.inc8.us.us.i.us, label %for.body6.us.us.i.us
+
+for.inc8.us.us.i.us: ; preds = %for.body6.us.us.i.us
+ %14 = add i32 %s.15.us.us.i.us, %conv.i
+ %inc9.us.us.i.us = add nsw i32 %j.06.us.us.i.us, 1
+ %exitcond33.i.us = icmp eq i32 %inc9.us.us.i.us, %conv.i189
+ br i1 %exitcond33.i.us, label %for.inc11.us.i.us, label %for.body6.lr.ph.us.us.i.us
+
+for.inc11.us.i.us: ; preds = %for.inc8.us.us.i.us
+ %15 = add i32 %11, %s.09.us.i.us
+ %inc12.us.i.us = add nsw i32 %i.010.us.i.us, 1
+ %exitcond34.i.us = icmp eq i32 %inc12.us.i.us, %conv.i191
+ br i1 %exitcond34.i.us, label %read_data.exit, label %for.body6.lr.ph.us.us.i.preheader.us
+
+for.body6.lr.ph.us.us.i.preheader.us: ; preds = %for.inc11.us.i.us, %for.cond4.preheader.lr.ph.us.i.preheader.split.us
+ %i.010.us.i.us = phi i32 [ %inc12.us.i.us, %for.inc11.us.i.us ], [ 0, %for.cond4.preheader.lr.ph.us.i.preheader.split.us ]
+ %s.09.us.i.us = phi i32 [ %15, %for.inc11.us.i.us ], [ 0, %for.cond4.preheader.lr.ph.us.i.preheader.split.us ]
+ br label %for.body6.lr.ph.us.us.i.us
+
+read_data.exit: ; preds = %for.inc11.us.i.us, %for.cond1.preheader.lr.ph.i
+ %call31 = call i32 @fclose(%struct._IO_FILE* %call29) #1
+ call void @pb_InitializeTimerSet(%struct.pb_TimerSet* %timers) #1
+ %16 = call i8* @llvm_visc_initializeTimerSet()
+ store i8* %16, i8** @viscTimerSet_GenVISC
+ call void @llvm_visc_switchToTimer(i8** @viscTimerSet_GenVISC, i32 0)
+ call void @llvm.visc.init()
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 15) #1
+ call void @llvm_visc_track_mem(i8* %call25, i64 %mul24) #1
+ call void @llvm_visc_track_mem(i8* %call27, i64 %mul24) #1
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 6) #1
+ call void @llvm.memcpy.p0i8.p0i8.i64(i8* %call27, i8* %call25, i64 %mul24, i32 4, i1 false)
+ %sub40 = add nsw i32 %conv.i, 253
+ %div = sdiv i32 %sub40, 256
+ %mul42 = shl nsw i32 %div, 6
+ %sub44 = add nsw i32 %conv.i189, -2
+ %sub46 = add nsw i32 %conv.i191, -2
+ %call53 = call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([37 x i8]* @.str4, i64 0, i64 0), i32 %mul42, i32 %sub44, i32 %sub46, i32 64, i32 1, i32 1) #1
+ %add56 = add nsw i32 %conv.i189, 1
+ %mul57 = mul nsw i32 %add56, %conv.i
+ %add59 = add nsw i32 %mul57, 129
+ %idxprom = sext i32 %add59 to i64
+ %arrayidx60 = getelementptr inbounds float* %6, i64 %idxprom
+ %17 = load float* %arrayidx60, align 4, !tbaa !2
+ %conv61 = fpext float %17 to double
+ %call62 = call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([17 x i8]* @.str5, i64 0, i64 0), double %conv61) #1
+ %add67 = add nsw i32 %mul57, 128
+ %idxprom68 = sext i32 %add67 to i64
+ %arrayidx69 = getelementptr inbounds float* %6, i64 %idxprom68
+ %18 = load float* %arrayidx69, align 4, !tbaa !2
+ %conv70 = fpext float %18 to double
+ %call71 = call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([17 x i8]* @.str6, i64 0, i64 0), double %conv70) #1
+ %cmp72194 = icmp sgt i32 %conv.i193, 0
+ br i1 %cmp72194, label %for.body, label %for.end
+
+for.body: ; preds = %for.body, %read_data.exit
+ %h_A0.0197 = phi float* [ %h_Anext.0196, %for.body ], [ %6, %read_data.exit ]
+ %h_Anext.0196 = phi float* [ %h_A0.0197, %for.body ], [ %7, %read_data.exit ]
+ %t.0195 = phi i32 [ %inc, %for.body ], [ 0, %read_data.exit ]
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 0) #1
+ call void @llvm_visc_switchToTimer(i8** @viscTimerSet_GenVISC, i32 19)
+ %in.addr = alloca %struct.arg
+ %in.addr.c0 = getelementptr %struct.arg* %in.addr, i32 0, i32 0
+ %in.addr.c0.cast = fptrunc double 0x3FC5555560000000 to float
+ store float %in.addr.c0.cast, float* %in.addr.c0
+ %in.addr.c1 = getelementptr %struct.arg* %in.addr, i32 0, i32 1
+ %in.addr.c1.cast = fptrunc double 0x3F9C71C720000000 to float
+ store float %in.addr.c1.cast, float* %in.addr.c1
+ %in.addr.A0 = getelementptr %struct.arg* %in.addr, i32 0, i32 2
+ store float* %h_A0.0197, float** %in.addr.A0
+ %in.addr.bytes_A0 = getelementptr %struct.arg* %in.addr, i32 0, i32 3
+ store i64 %mul24, i64* %in.addr.bytes_A0
+ %in.addr.Anext = getelementptr %struct.arg* %in.addr, i32 0, i32 4
+ store float* %h_Anext.0196, float** %in.addr.Anext
+ %in.addr.bytes_Anext = getelementptr %struct.arg* %in.addr, i32 0, i32 5
+ store i64 %mul24, i64* %in.addr.bytes_Anext
+ %in.addr.nx = getelementptr %struct.arg* %in.addr, i32 0, i32 6
+ store i32 %conv.i, i32* %in.addr.nx
+ %in.addr.ny = getelementptr %struct.arg* %in.addr, i32 0, i32 7
+ store i32 %conv.i189, i32* %in.addr.ny
+ %in.addr.nz = getelementptr %struct.arg* %in.addr, i32 0, i32 8
+ store i32 %conv.i191, i32* %in.addr.nz
+ %in.addr.dimX0 = getelementptr %struct.arg* %in.addr, i32 0, i32 9
+ store i32 64, i32* %in.addr.dimX0
+ %in.addr.dimY0 = getelementptr %struct.arg* %in.addr, i32 0, i32 10
+ store i32 1, i32* %in.addr.dimY0
+ %in.addr.dimZ1 = getelementptr %struct.arg* %in.addr, i32 0, i32 11
+ store i32 1, i32* %in.addr.dimZ1
+ %in.addr.dimX1 = getelementptr %struct.arg* %in.addr, i32 0, i32 12
+ store i32 %div, i32* %in.addr.dimX1
+ %in.addr.dimY2 = getelementptr %struct.arg* %in.addr, i32 0, i32 13
+ store i32 %sub44, i32* %in.addr.dimY2
+ %in.addr.dimZ2 = getelementptr %struct.arg* %in.addr, i32 0, i32 14
+ store i32 %sub46, i32* %in.addr.dimZ2
+ %args = bitcast %struct.arg* %in.addr to i8*
+ call void @llvm_visc_switchToTimer(i8** @viscTimerSet_GenVISC, i32 0)
+ %graphnaive_kernelInternal_level2 = call i8* @llvm.visc.launch(i8* bitcast (%rtype (float, float, float*, i64, float*, i64, i32, i32, i32, i32, i32, i32, i32, i32, i32)* @naive_kernelInternal_level2 to i8*), i8* %args)
+ call void @llvm.visc.wait(i8* %graphnaive_kernelInternal_level2)
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 6) #1
+ %inc = add nsw i32 %t.0195, 1
+ %exitcond = icmp eq i32 %inc, %conv.i193
+ br i1 %exitcond, label %for.end, label %for.body
+
+for.end: ; preds = %for.body, %read_data.exit
+ %h_A0.0.lcssa = phi float* [ %6, %read_data.exit ], [ %h_Anext.0196, %for.body ]
+ %h_Anext.0.lcssa = phi float* [ %7, %read_data.exit ], [ %h_A0.0197, %for.body ]
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 3) #1
+ %19 = bitcast float* %h_A0.0.lcssa to i8*
+ call void @llvm_visc_request_mem(i8* %19, i64 %mul24) #1
+ %arrayidx97 = getelementptr inbounds float* %h_A0.0.lcssa, i64 %idxprom
+ %20 = load float* %arrayidx97, align 4, !tbaa !2
+ %conv98 = fpext float %20 to double
+ %call99 = call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([17 x i8]* @.str5, i64 0, i64 0), double %conv98) #1
+ %arrayidx106 = getelementptr inbounds float* %h_A0.0.lcssa, i64 %idxprom68
+ %21 = load float* %arrayidx106, align 4, !tbaa !2
+ %conv107 = fpext float %21 to double
+ %call108 = call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([17 x i8]* @.str6, i64 0, i64 0), double %conv107) #1
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 16) #1
+ %22 = bitcast float* %h_Anext.0.lcssa to i8*
+ call void @llvm_visc_untrack_mem(i8* %22) #1
+ call void @llvm_visc_untrack_mem(i8* %19) #1
+ call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 0) #1
+ call void @pb_PrintTimerSet(%struct.pb_TimerSet* %timers) #1
+ %Ptr = getelementptr [14 x i8]* @0, i64 0, i64 0
+ call void @llvm_visc_printTimerSet(i8** @viscTimerSet_GenVISC, i8* %Ptr)
+ call void @llvm.visc.cleanup()
+ %outFile = getelementptr inbounds %struct.pb_Parameters* %call2, i64 0, i32 0
+ %23 = load i8** %outFile, align 8, !tbaa !6
+ %tobool = icmp eq i8* %23, null
+ br i1 %tobool, label %if.end113, label %if.then110
+
+if.then110: ; preds = %for.end
+ %add.ptr112 = getelementptr inbounds float* %h_A0.0.lcssa, i64 3
+ call void @outputData(i8* %23, float* %add.ptr112, i32 %conv.i, i32 %conv.i189, i32 %conv.i191) #1
+ br label %if.end113
+
+if.end113: ; preds = %if.then110, %for.end
+ call void @free(i8* %22) #1
+ call void @free(i8* %19) #1
+ call void @pb_FreeParameters(%struct.pb_Parameters* %call2) #1
+ br label %cleanup
+
+cleanup: ; preds = %if.end113, %if.end17, %if.end12, %if.end7, %if.end, %if.then
+ %retval.0 = phi i32 [ -1, %if.then ], [ 0, %if.end113 ], [ -1, %if.end ], [ -1, %if.end7 ], [ -1, %if.end12 ], [ -1, %if.end17 ]
+ call void @llvm.lifetime.end(i64 800, i8* %0) #1
+ ret i32 %retval.0
+}
+
+; Function Attrs: nounwind
+declare void @llvm.lifetime.start(i64, i8* nocapture) #1
+
+; Function Attrs: nounwind
+declare i32 @printf(i8* nocapture, ...) #2
+
+declare %struct.pb_Parameters* @pb_ReadParameters(i32*, i8**) #3
+
+; Function Attrs: nounwind
+declare noalias i8* @malloc(i64) #2
+
+; Function Attrs: nounwind
+declare noalias %struct._IO_FILE* @fopen(i8* nocapture, i8* nocapture) #2
+
+; Function Attrs: nounwind
+declare i32 @fclose(%struct._IO_FILE* nocapture) #2
+
+declare void @pb_InitializeTimerSet(%struct.pb_TimerSet*) #3
+
+declare void @pb_SwitchToTimer(%struct.pb_TimerSet*, i32) #3
+
+declare void @llvm_visc_track_mem(i8*, i64) #3
+
+; Function Attrs: nounwind
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) #1
+
+declare void @llvm_visc_request_mem(i8*, i64) #3
+
+declare void @llvm_visc_untrack_mem(i8*) #3
+
+declare void @pb_PrintTimerSet(%struct.pb_TimerSet*) #3
+
+declare void @outputData(i8*, float*, i32, i32, i32) #3
+
+; Function Attrs: nounwind
+declare void @free(i8* nocapture) #2
+
+declare void @pb_FreeParameters(%struct.pb_Parameters*) #3
+
+; Function Attrs: nounwind
+declare void @llvm.lifetime.end(i64, i8* nocapture) #1
+
+; Function Attrs: nounwind
+declare i64 @fread(i8* nocapture, i64, i64, %struct._IO_FILE* nocapture) #2
+
+; Function Attrs: nounwind
+declare i64 @strtol(i8*, i8** nocapture, i32) #2
+
+; Function Attrs: nounwind
+declare i32 @puts(i8* nocapture) #1
+
+declare i8* @llvm_visc_initializeTimerSet()
+
+declare void @llvm_visc_switchToTimer(i8**, i32)
+
+declare void @llvm_visc_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind
+declare void @llvm.visc.init() #1
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.getNode() #1
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.getParentNode(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNodeInstanceID.x(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNumNodeInstances.x(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNodeInstanceID.y(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNumNodeInstances.y(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNodeInstanceID.z(i8*) #1
+
+; Function Attrs: nounwind
+declare i32 @llvm.visc.getNumNodeInstances.z(i8*) #1
+
+; Function Attrs: nounwind uwtable
+define %rtype @naive_kernelInternal_level1(float %c0, float %c1, float* in %A0, i64 %bytes_A0, float* in out %Anext, i64 %bytes_Anext, i32 %nx, i32 %ny, i32 %nz, i32 %dimX, i32 %dimY, i32 %dimZ) #0 {
+entry:
+ %naive_kernel.node = call i8* @llvm.visc.createNode3D(i8* bitcast (%rtype (float, float, float*, i64, float*, i64, i32, i32, i32)* @naive_kernel to i8*), i32 %dimX, i32 %dimY, i32 %dimZ)
+ call void @llvm.visc.bind.input(i8* %naive_kernel.node, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %naive_kernel.node, i32 1, i32 1)
+ call void @llvm.visc.bind.input(i8* %naive_kernel.node, i32 2, i32 2)
+ call void @llvm.visc.bind.input(i8* %naive_kernel.node, i32 3, i32 3)
+ call void @llvm.visc.bind.input(i8* %naive_kernel.node, i32 4, i32 4)
+ call void @llvm.visc.bind.input(i8* %naive_kernel.node, i32 5, i32 5)
+ call void @llvm.visc.bind.input(i8* %naive_kernel.node, i32 6, i32 6)
+ call void @llvm.visc.bind.input(i8* %naive_kernel.node, i32 7, i32 7)
+ call void @llvm.visc.bind.input(i8* %naive_kernel.node, i32 8, i32 8)
+ ret %rtype undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.createNode3D(i8*, i32, i32, i32) #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.bind.input(i8*, i32, i32) #1
+
+; Function Attrs: nounwind uwtable
+define %rtype @naive_kernelInternal_level2(float %c0, float %c1, float* in %A0, i64 %bytes_A0, float* in out %Anext, i64 %bytes_Anext, i32 %nx, i32 %ny, i32 %nz, i32 %dimX, i32 %dimY, i32 %dimZ, i32 %dimX1, i32 %dimY2, i32 %dimZ3) #0 {
+entry:
+ %naive_kernelInternal_level1.node = call i8* @llvm.visc.createNode3D(i8* bitcast (%rtype (float, float, float*, i64, float*, i64, i32, i32, i32, i32, i32, i32)* @naive_kernelInternal_level1 to i8*), i32 %dimX1, i32 %dimY2, i32 %dimZ3)
+ call void @llvm.visc.bind.input(i8* %naive_kernelInternal_level1.node, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %naive_kernelInternal_level1.node, i32 1, i32 1)
+ call void @llvm.visc.bind.input(i8* %naive_kernelInternal_level1.node, i32 2, i32 2)
+ call void @llvm.visc.bind.input(i8* %naive_kernelInternal_level1.node, i32 3, i32 3)
+ call void @llvm.visc.bind.input(i8* %naive_kernelInternal_level1.node, i32 4, i32 4)
+ call void @llvm.visc.bind.input(i8* %naive_kernelInternal_level1.node, i32 5, i32 5)
+ call void @llvm.visc.bind.input(i8* %naive_kernelInternal_level1.node, i32 6, i32 6)
+ call void @llvm.visc.bind.input(i8* %naive_kernelInternal_level1.node, i32 7, i32 7)
+ call void @llvm.visc.bind.input(i8* %naive_kernelInternal_level1.node, i32 8, i32 8)
+ call void @llvm.visc.bind.input(i8* %naive_kernelInternal_level1.node, i32 9, i32 9)
+ call void @llvm.visc.bind.input(i8* %naive_kernelInternal_level1.node, i32 10, i32 10)
+ call void @llvm.visc.bind.input(i8* %naive_kernelInternal_level1.node, i32 11, i32 11)
+ ret %rtype undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.visc.launch(i8*, i8*) #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.wait(i8*) #1
+
+; Function Attrs: nounwind
+declare void @llvm.visc.cleanup() #1
+
+attributes #0 = { nounwind uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { nounwind }
+attributes #2 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "use-soft-float"="false" }
+
+!visc_hint_gpu = !{}
+!visc_hint_cpu = !{!0, !1}
+
+!0 = metadata !{%rtype (float, float, float*, i64, float*, i64, i32, i32, i32, i32, i32, i32)* @naive_kernelInternal_level1}
+!1 = metadata !{%rtype (float, float, float*, i64, float*, i64, i32, i32, i32, i32, i32, i32, i32, i32, i32)* @naive_kernelInternal_level2}
+!2 = metadata !{metadata !"float", metadata !3}
+!3 = metadata !{metadata !"omnipotent char", metadata !4}
+!4 = metadata !{metadata !"Simple C/C++ TBAA"}
+!5 = metadata !{metadata !"int", metadata !3}
+!6 = metadata !{metadata !"any pointer", metadata !3}
diff --git a/hpvm/test/parboil/benchmarks/tpacf/Makefile b/hpvm/test/parboil/benchmarks/tpacf/Makefile
index e76139ba38..6140acd5ac 100644
--- a/hpvm/test/parboil/benchmarks/tpacf/Makefile
+++ b/hpvm/test/parboil/benchmarks/tpacf/Makefile
@@ -1,9 +1,9 @@
PARBOIL_ROOT = $(LLVM_SRC_ROOT)/tools/hpvm/test/parboil
APP = tpacf
-# Default compile hpvm
+# Default compile visc
ifeq ($(VERSION),)
- VERSION = hpvm
+ VERSION = visc
endif
# Default use small test case
diff --git a/hpvm/test/parboil/benchmarks/tpacf/src/opencl_base/main.cc b/hpvm/test/parboil/benchmarks/tpacf/src/opencl_base/main.cc
index d89d556a10..d945bccf4e 100644
--- a/hpvm/test/parboil/benchmarks/tpacf/src/opencl_base/main.cc
+++ b/hpvm/test/parboil/benchmarks/tpacf/src/opencl_base/main.cc
@@ -199,7 +199,7 @@ int main(int argc, char **argv) {
3 * f_mem_size, h_x_data, 0, NULL, NULL);
CHECK_ERROR("clEnqueueWriteBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
TPACF(d_hists, d_x_data, dev_binb, clCommandQueue, clKernel);
diff --git a/hpvm/test/parboil/benchmarks/tpacf/src/opencl_cpu_base/main.cc b/hpvm/test/parboil/benchmarks/tpacf/src/opencl_cpu_base/main.cc
index ef2a21daed..791b5fbdd6 100644
--- a/hpvm/test/parboil/benchmarks/tpacf/src/opencl_cpu_base/main.cc
+++ b/hpvm/test/parboil/benchmarks/tpacf/src/opencl_cpu_base/main.cc
@@ -203,7 +203,7 @@ int main(int argc, char **argv) {
3 * f_mem_size, h_x_data, 0, NULL, NULL);
CHECK_ERROR("clEnqueueWriteBuffer")
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
TPACF(d_hists, d_x_data, dev_binb, clCommandQueue, clKernel);
diff --git a/hpvm/test/parboil/benchmarks/tpacf/src/hpvm/Makefile b/hpvm/test/parboil/benchmarks/tpacf/src/visc/Makefile
similarity index 82%
rename from hpvm/test/parboil/benchmarks/tpacf/src/hpvm/Makefile
rename to hpvm/test/parboil/benchmarks/tpacf/src/visc/Makefile
index 040e2c7994..ba6459d78a 100644
--- a/hpvm/test/parboil/benchmarks/tpacf/src/hpvm/Makefile
+++ b/hpvm/test/parboil/benchmarks/tpacf/src/visc/Makefile
@@ -1,8 +1,8 @@
# (c) 2010 The Board of Trustees of the University of Illinois.
-LANGUAGE=hpvm
+LANGUAGE=visc
SRCDIR_OBJS=args.ll model.ll
-HPVM_OBJS=main.hpvm.ll
+VISC_OBJS=main.visc.ll
APP_CUDALDFLAGS=-lm
APP_CFLAGS=-ffast-math -O3
APP_CXXFLAGS=-ffast-math -O3
diff --git a/hpvm/test/parboil/benchmarks/tpacf/src/hpvm/args.cc b/hpvm/test/parboil/benchmarks/tpacf/src/visc/args.cc
similarity index 100%
rename from hpvm/test/parboil/benchmarks/tpacf/src/hpvm/args.cc
rename to hpvm/test/parboil/benchmarks/tpacf/src/visc/args.cc
diff --git a/hpvm/test/parboil/benchmarks/tpacf/src/hpvm/args.h b/hpvm/test/parboil/benchmarks/tpacf/src/visc/args.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/tpacf/src/hpvm/args.h
rename to hpvm/test/parboil/benchmarks/tpacf/src/visc/args.h
diff --git a/hpvm/test/parboil/benchmarks/tpacf/src/hpvm/main.cc b/hpvm/test/parboil/benchmarks/tpacf/src/visc/main.cc
similarity index 76%
rename from hpvm/test/parboil/benchmarks/tpacf/src/hpvm/main.cc
rename to hpvm/test/parboil/benchmarks/tpacf/src/visc/main.cc
index 49208f579c..3239be6c92 100644
--- a/hpvm/test/parboil/benchmarks/tpacf/src/hpvm/main.cc
+++ b/hpvm/test/parboil/benchmarks/tpacf/src/visc/main.cc
@@ -14,7 +14,7 @@
#include "args.h"
#include "model.h"
-#include <hpvm.h>
+#include <visc.h>
extern unsigned int NUM_SETS;
extern unsigned int NUM_ELEMENTS;
@@ -62,13 +62,13 @@ void packData(RootIn *args, hist_t *histograms, size_t bytes_histograms,
void Allocation(long block) {
// Memory shared between threadblocks
- // void* data_s = __hpvm__malloc(sizeof(struct cartesian)*BLOCK_SIZE);
+ // void* data_s = __visc__malloc(sizeof(struct cartesian)*BLOCK_SIZE);
void *warp_hists =
- __hpvm__malloc(sizeof(unsigned int) * NUM_BINS * NUM_HISTOGRAMS);
+ __visc__malloc(sizeof(unsigned int) * NUM_BINS * NUM_HISTOGRAMS);
- //__hpvm__return(data_s, sizeof(struct cartesian)*BLOCK_SIZE,
+ //__visc__return(data_s, sizeof(struct cartesian)*BLOCK_SIZE,
// warp_hists, sizeof(unsigned int)*NUM_BINS*NUM_HISTOGRAMS);
- __hpvm__return(2, warp_hists,
+ __visc__return(2, warp_hists,
sizeof(unsigned int) * NUM_BINS * NUM_HISTOGRAMS);
}
@@ -80,14 +80,14 @@ void TPACFLeaf(hist_t *histograms, size_t bytes_histograms,
// struct cartesian* data_s, size_t bytes_data_s,
unsigned int *warp_hists, size_t bytes_warp_hists) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(2, all_x_data, binb, 1, histograms);
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(2, all_x_data, binb, 1, histograms);
- void *thisNode = __hpvm__getNode();
- void *parentNode = __hpvm__getParentNode(thisNode);
- int lx = __hpvm__getNodeInstanceID_x(thisNode);
- int gx = __hpvm__getNodeInstanceID_x(parentNode);
- int dimx = __hpvm__getNumNodeInstances_x(thisNode);
+ void *thisNode = __visc__getNode();
+ void *parentNode = __visc__getParentNode(thisNode);
+ int lx = __visc__getNodeInstanceID_x(thisNode);
+ int gx = __visc__getNodeInstanceID_x(parentNode);
+ int dimx = __visc__getNumNodeInstances_x(thisNode);
float *all_y_data = all_x_data + NUM_ELEMENTS * (NUM_SETS + 1);
float *all_z_data = all_y_data + NUM_ELEMENTS * (NUM_SETS + 1);
@@ -170,7 +170,7 @@ void TPACFLeaf(hist_t *histograms, size_t bytes_histograms,
unsigned int warpnum = tid / (WARP_SIZE / HISTS_PER_WARP);
if ((distance < binb[min]) && (distance >= binb[max]) &&
(!do_self || (tid + j > k)) && ((tid + j) < NUM_ELEMENTS)) {
- __hpvm__atomic_add((int *)&(warp_hists(bin_index, warpnum)), 1);
+ __visc__atomic_add((int *)&(warp_hists(bin_index, warpnum)), 1);
}
}
}
@@ -181,7 +181,7 @@ void TPACFLeaf(hist_t *histograms, size_t bytes_histograms,
for (unsigned int offset = NUM_HISTOGRAMS >> 1; offset > 0; offset >>= 1) {
for (unsigned int bin_base = 0; bin_base < NUM_BINS;
bin_base += BLOCK_SIZE / (NUM_HISTOGRAMS >> 1)) {
- __hpvm__barrier();
+ __visc__barrier();
if (warp_index < offset && bin_base + bin_index < NUM_BINS) {
unsigned long sum =
warp_hists(bin_base + bin_index, warp_index) +
@@ -191,7 +191,7 @@ void TPACFLeaf(hist_t *histograms, size_t bytes_histograms,
}
}
- __hpvm__barrier();
+ __visc__barrier();
// Put the results back in the real histogram
// warp_hists(x, 0) holds sum of all locations of bin x
@@ -207,26 +207,26 @@ void BlockingTPACF(hist_t *histograms, size_t bytes_histograms,
float *binb, size_t bytes_binb, int NUM_SETS,
int NUM_ELEMENTS, long block) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(2, all_x_data, binb, 1, histograms);
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(2, all_x_data, binb, 1, histograms);
- void *AllocationNode = __hpvm__createNodeND(0, Allocation);
- void *TPACFLeafNode = __hpvm__createNodeND(1, TPACFLeaf, block);
+ void *AllocationNode = __visc__createNodeND(0, Allocation);
+ void *TPACFLeafNode = __visc__createNodeND(1, TPACFLeaf, block);
// Bind Inputs
- __hpvm__bindIn(AllocationNode, 8, 0, 0); // Bind block
- __hpvm__bindIn(TPACFLeafNode, 0, 0, 0); // Bind histograms
- __hpvm__bindIn(TPACFLeafNode, 1, 1, 0); // Bind bytes_histograms
- __hpvm__bindIn(TPACFLeafNode, 2, 2, 0); // Bind all_x_data
- __hpvm__bindIn(TPACFLeafNode, 3, 3, 0); // Bind bytes_all_data
- __hpvm__bindIn(TPACFLeafNode, 4, 4, 0); // Bind binb
- __hpvm__bindIn(TPACFLeafNode, 5, 5, 0); // Bind bytes_binb
- __hpvm__bindIn(TPACFLeafNode, 6, 6, 0); // Bind NUM_SETS
- __hpvm__bindIn(TPACFLeafNode, 7, 7, 0); // Bind NUM_ELEMENTS
+ __visc__bindIn(AllocationNode, 8, 0, 0); // Bind block
+ __visc__bindIn(TPACFLeafNode, 0, 0, 0); // Bind histograms
+ __visc__bindIn(TPACFLeafNode, 1, 1, 0); // Bind bytes_histograms
+ __visc__bindIn(TPACFLeafNode, 2, 2, 0); // Bind all_x_data
+ __visc__bindIn(TPACFLeafNode, 3, 3, 0); // Bind bytes_all_data
+ __visc__bindIn(TPACFLeafNode, 4, 4, 0); // Bind binb
+ __visc__bindIn(TPACFLeafNode, 5, 5, 0); // Bind bytes_binb
+ __visc__bindIn(TPACFLeafNode, 6, 6, 0); // Bind NUM_SETS
+ __visc__bindIn(TPACFLeafNode, 7, 7, 0); // Bind NUM_ELEMENTS
// Create Edges
- __hpvm__edge(AllocationNode, TPACFLeafNode, 1, 0, 8, 0); // Edge warp_hists
- __hpvm__edge(AllocationNode, TPACFLeafNode, 1, 1, 9,
+ __visc__edge(AllocationNode, TPACFLeafNode, 1, 0, 8, 0); // Edge warp_hists
+ __visc__edge(AllocationNode, TPACFLeafNode, 1, 1, 9,
0); // Edge bytes_warp_hists
}
@@ -236,21 +236,21 @@ void TPACFRoot(hist_t *histograms, size_t bytes_histograms, float *all_x_data,
float *binb, size_t bytes_binb, int NUM_SETS, int NUM_ELEMENTS,
long block, long grid) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(2, all_x_data, binb, 1, histograms);
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(2, all_x_data, binb, 1, histograms);
- void *BlockingTPACFNode = __hpvm__createNodeND(1, BlockingTPACF, grid);
+ void *BlockingTPACFNode = __visc__createNodeND(1, BlockingTPACF, grid);
// Bind Inputs
- __hpvm__bindIn(BlockingTPACFNode, 0, 0, 0); // Bind histograms
- __hpvm__bindIn(BlockingTPACFNode, 1, 1, 0); // Bind bytes_histograms
- __hpvm__bindIn(BlockingTPACFNode, 2, 2, 0); // Bind all_x_data
- __hpvm__bindIn(BlockingTPACFNode, 3, 3, 0); // Bind bytes_all_data
- __hpvm__bindIn(BlockingTPACFNode, 4, 4, 0); // Bind binb
- __hpvm__bindIn(BlockingTPACFNode, 5, 5, 0); // Bind bytes_binb
- __hpvm__bindIn(BlockingTPACFNode, 6, 6, 0); // Bind NUM_SETS
- __hpvm__bindIn(BlockingTPACFNode, 7, 7, 0); // Bind NUM_ELEMENTS
- __hpvm__bindIn(BlockingTPACFNode, 8, 8, 0); // Bind block
+ __visc__bindIn(BlockingTPACFNode, 0, 0, 0); // Bind histograms
+ __visc__bindIn(BlockingTPACFNode, 1, 1, 0); // Bind bytes_histograms
+ __visc__bindIn(BlockingTPACFNode, 2, 2, 0); // Bind all_x_data
+ __visc__bindIn(BlockingTPACFNode, 3, 3, 0); // Bind bytes_all_data
+ __visc__bindIn(BlockingTPACFNode, 4, 4, 0); // Bind binb
+ __visc__bindIn(BlockingTPACFNode, 5, 5, 0); // Bind bytes_binb
+ __visc__bindIn(BlockingTPACFNode, 6, 6, 0); // Bind NUM_SETS
+ __visc__bindIn(BlockingTPACFNode, 7, 7, 0); // Bind NUM_ELEMENTS
+ __visc__bindIn(BlockingTPACFNode, 8, 8, 0); // Bind block
}
void TPACFWrapper(hist_t *histograms, size_t bytes_histograms,
@@ -258,22 +258,22 @@ void TPACFWrapper(hist_t *histograms, size_t bytes_histograms,
// next arg is read-only constant
float *binb, size_t bytes_binb, int NUM_SETS,
int NUM_ELEMENTS, long block, long grid) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(2, all_x_data, binb, 1, histograms);
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(2, all_x_data, binb, 1, histograms);
- void *BlockingTPACFNode = __hpvm__createNodeND(0, TPACFRoot);
+ void *BlockingTPACFNode = __visc__createNodeND(0, TPACFRoot);
// Bind Inputs
- __hpvm__bindIn(BlockingTPACFNode, 0, 0, 0); // Bind histograms
- __hpvm__bindIn(BlockingTPACFNode, 1, 1, 0); // Bind bytes_histograms
- __hpvm__bindIn(BlockingTPACFNode, 2, 2, 0); // Bind all_x_data
- __hpvm__bindIn(BlockingTPACFNode, 3, 3, 0); // Bind bytes_all_data
- __hpvm__bindIn(BlockingTPACFNode, 4, 4, 0); // Bind binb
- __hpvm__bindIn(BlockingTPACFNode, 5, 5, 0); // Bind bytes_binb
- __hpvm__bindIn(BlockingTPACFNode, 6, 6, 0); // Bind NUM_SETS
- __hpvm__bindIn(BlockingTPACFNode, 7, 7, 0); // Bind NUM_ELEMENTS
- __hpvm__bindIn(BlockingTPACFNode, 8, 8, 0); // Bind block
- __hpvm__bindIn(BlockingTPACFNode, 9, 9, 0); // Bind grid
+ __visc__bindIn(BlockingTPACFNode, 0, 0, 0); // Bind histograms
+ __visc__bindIn(BlockingTPACFNode, 1, 1, 0); // Bind bytes_histograms
+ __visc__bindIn(BlockingTPACFNode, 2, 2, 0); // Bind all_x_data
+ __visc__bindIn(BlockingTPACFNode, 3, 3, 0); // Bind bytes_all_data
+ __visc__bindIn(BlockingTPACFNode, 4, 4, 0); // Bind binb
+ __visc__bindIn(BlockingTPACFNode, 5, 5, 0); // Bind bytes_binb
+ __visc__bindIn(BlockingTPACFNode, 6, 6, 0); // Bind NUM_SETS
+ __visc__bindIn(BlockingTPACFNode, 7, 7, 0); // Bind NUM_ELEMENTS
+ __visc__bindIn(BlockingTPACFNode, 8, 8, 0); // Bind block
+ __visc__bindIn(BlockingTPACFNode, 9, 9, 0); // Bind grid
}
// **===-----------------------------------------------------------===**
@@ -324,14 +324,14 @@ int main(int argc, char **argv) {
}
pb_InitializeTimerSet(&timers);
- __hpvm__init();
+ __visc__init();
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
// split into x, y, and z arrays
// AOS to SOA transformation
size_t bytes_h_x_data = 3 * f_mem_size;
float *h_x_data = (float *)malloc(bytes_h_x_data);
- llvm_hpvm_track_mem(h_x_data, bytes_h_x_data);
+ llvm_visc_track_mem(h_x_data, bytes_h_x_data);
float *h_y_data = h_x_data + NUM_ELEMENTS * (NUM_SETS + 1);
float *h_z_data = h_y_data + NUM_ELEMENTS * (NUM_SETS + 1);
@@ -349,12 +349,12 @@ int main(int argc, char **argv) {
// allocate system memory for final histograms
size_t bytes_hists = NUM_BINS * (NUM_SETS * 2 + 1) * sizeof(hist_t);
hist_t *hists = (hist_t *)malloc(bytes_hists);
- llvm_hpvm_track_mem(hists, bytes_hists);
+ llvm_visc_track_mem(hists, bytes_hists);
// Initialize the boundary constants for bin search
size_t bytes_binb = (NUM_BINS + 1) * sizeof(float);
float *binb = (float *)malloc(bytes_binb);
- llvm_hpvm_track_mem(binb, bytes_binb);
+ llvm_visc_track_mem(binb, bytes_binb);
for (int k = 0; k < NUM_BINS + 1; k++) {
binb[k] = cos(pow(10.0, (log10(min_arcmin) + k * 1.0 / bins_per_dec)) /
@@ -369,17 +369,17 @@ int main(int argc, char **argv) {
RootIn *graph_args = (RootIn *)malloc(sizeof(RootIn));
packData(graph_args, hists, bytes_hists, h_x_data, bytes_h_x_data, binb,
bytes_binb, NUM_SETS, NUM_ELEMENTS, block, grid);
- pb_SwitchToTimer(&timers, hpvm_TimerID_COMPUTATION);
+ pb_SwitchToTimer(&timers, visc_TimerID_COMPUTATION);
- void *TPACF_DFG = __hpvm__launch(0, TPACFRoot, (void *)graph_args);
- __hpvm__wait(TPACF_DFG);
+ void *TPACF_DFG = __visc__launch(0, TPACFRoot, (void *)graph_args);
+ __visc__wait(TPACF_DFG);
pb_SwitchToTimer(&timers, pb_TimerID_COPY);
pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
// **===-------------------------------------------------------------===**
- llvm_hpvm_request_mem(hists, bytes_hists);
+ llvm_visc_request_mem(hists, bytes_hists);
// references into output histograms
hist_t *dd_hist = hists;
hist_t *rr_hist = dd_hist + NUM_BINS;
@@ -407,7 +407,7 @@ int main(int argc, char **argv) {
pb_SwitchToTimer(&timers, pb_TimerID_NONE);
pb_PrintTimerSet(&timers);
- __hpvm__cleanup();
+ __visc__cleanup();
FILE *outfile;
if ((outfile = fopen(params->outFile, "w")) == NULL) {
diff --git a/hpvm/test/parboil/benchmarks/tpacf/src/hpvm/model.cc b/hpvm/test/parboil/benchmarks/tpacf/src/visc/model.cc
similarity index 100%
rename from hpvm/test/parboil/benchmarks/tpacf/src/hpvm/model.cc
rename to hpvm/test/parboil/benchmarks/tpacf/src/visc/model.cc
diff --git a/hpvm/test/parboil/benchmarks/tpacf/src/hpvm/model.h b/hpvm/test/parboil/benchmarks/tpacf/src/visc/model.h
similarity index 100%
rename from hpvm/test/parboil/benchmarks/tpacf/src/hpvm/model.h
rename to hpvm/test/parboil/benchmarks/tpacf/src/visc/model.h
diff --git a/hpvm/test/parboil/common/include/parboil.h b/hpvm/test/parboil/common/include/parboil.h
index ba25726c02..30ad6721c3 100644
--- a/hpvm/test/parboil/common/include/parboil.h
+++ b/hpvm/test/parboil/common/include/parboil.h
@@ -102,23 +102,23 @@ enum pb_TimerID {
* host activity: automatically filled in,
* not intended for direct usage */
// GPU FUNCTION
- hpvm_TimerID_INIT_CTX,
- hpvm_TimerID_CLEAR_CTX,
- hpvm_TimerID_COPY_SCALAR,
- hpvm_TimerID_COPY_PTR,
- hpvm_TimerID_MEM_FREE,
- hpvm_TimerID_READ_OUTPUT,
- hpvm_TimerID_SETUP,
- hpvm_TimerID_MEM_TRACK,
- hpvm_TimerID_MEM_UNTRACK,
- hpvm_TimerID_MISC,
+ visc_TimerID_INIT_CTX,
+ visc_TimerID_CLEAR_CTX,
+ visc_TimerID_COPY_SCALAR,
+ visc_TimerID_COPY_PTR,
+ visc_TimerID_MEM_FREE,
+ visc_TimerID_READ_OUTPUT,
+ visc_TimerID_SETUP,
+ visc_TimerID_MEM_TRACK,
+ visc_TimerID_MEM_UNTRACK,
+ visc_TimerID_MISC,
// LAUNCH FUNCTION
- hpvm_TimerID_PTHREAD_CREATE,
- hpvm_TimerID_ARG_PACK,
- hpvm_TimerID_ARG_UNPACK,
- hpvm_TimerID_COMPUTATION,
- hpvm_TimerID_OUTPUT_PACK,
- hpvm_TimerID_OUTPUT_UNPACK,
+ visc_TimerID_PTHREAD_CREATE,
+ visc_TimerID_ARG_PACK,
+ visc_TimerID_ARG_UNPACK,
+ visc_TimerID_COMPUTATION,
+ visc_TimerID_OUTPUT_PACK,
+ visc_TimerID_OUTPUT_UNPACK,
pb_TimerID_LAST /* Number of timer IDs */
};
diff --git a/hpvm/test/parboil/common/mk/hpvm.mk b/hpvm/test/parboil/common/mk/visc.mk
similarity index 81%
rename from hpvm/test/parboil/common/mk/hpvm.mk
rename to hpvm/test/parboil/common/mk/visc.mk
index 1c59d4d8fd..eb11371ccd 100755
--- a/hpvm/test/parboil/common/mk/hpvm.mk
+++ b/hpvm/test/parboil/common/mk/visc.mk
@@ -9,38 +9,38 @@ CFLAGS=$(LANG_CFLAGS) $(PLATFORM_CFLAGS) $(APP_CFLAGS)
CXXFLAGS=$(LANG_CXXFLAGS) $(PLATFORM_CXXFLAGS) $(APP_CXXFLAGS)
LDFLAGS=$(LANG_LDFLAGS) $(PLATFORM_LDFLAGS) $(APP_LDFLAGS)
-# HPVM
+# VISC
LIBCLC_LIB_PATH = $(LLVM_SRC_ROOT)/../libclc/built_libs
-#HPVM_RT_PATH = $(LLVM_SRC_ROOT)/../build/projects/hpvm-rt
-HPVM_RT_PATH = $(LLVM_SRC_ROOT)/tools/hpvm/projects/hpvm-rt
+#VISC_RT_PATH = $(LLVM_SRC_ROOT)/../build/projects/visc-rt
+VISC_RT_PATH = $(LLVM_SRC_ROOT)/tools/hpvm/projects/visc-rt
-HPVM_RT_LIB = $(HPVM_RT_PATH)/hpvm-rt.ll
+VISC_RT_LIB = $(VISC_RT_PATH)/visc-rt.ll
#LIBCLC_NVPTX_LIB = $(LIBCLC_LIB_PATH)/nvptx--nvidiacl.bc
LIBCLC_NVPTX_LIB = $(LIBCLC_LIB_PATH)/nvptx64--nvidiacl.bc
#LIBCLC_NVPTX_LIB = nvptx64--nvidiacl.bc
LLVM_34_AS = /opt/llvm/bin/llvm-as
-TESTGEN_OPTFLAGS = -load LLVMGenHPVM.so -genhpvm -globaldce
+TESTGEN_OPTFLAGS = -load LLVMGenVISC.so -genvisc -globaldce
KERNEL_GEN_FLAGS = -O3 -target nvptx64-nvidia-nvcl
ifeq ($(TARGET),x86)
DEVICE = SPIR_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_SPIR.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-spir -dfg2llvm-x86 -clearDFG
+ VISC_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_SPIR.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-spir -dfg2llvm-x86 -clearDFG
CFLAGS += -DOPENCL_CPU
else ifeq ($(TARGET),seq)
DEVICE = CPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG
+ VISC_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG
else ifeq ($(TARGET),seqx86)
DEVICE = CPU_OR_SPIR_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_SPIR.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-spir -dfg2llvm-x86 -clearDFG
+ VISC_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_SPIR.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-spir -dfg2llvm-x86 -clearDFG
CFLAGS += -DOPENCL_CPU
else ifeq ($(TARGET),seqgpu)
DEVICE = CPU_OR_GPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG
+ VISC_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG
else
DEVICE = GPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG
+ VISC_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG
endif
CFLAGS += -DDEVICE=$(DEVICE)
@@ -49,31 +49,31 @@ CXXFLAGS += -DDEVICE=$(DEVICE)
HOST_LINKFLAGS =
ifeq ($(TIMER),x86)
- HPVM_OPTFLAGS += -hpvm-timers-x86
+ VISC_OPTFLAGS += -visc-timers-x86
else ifeq ($(TIMER),ptx)
- HPVM_OPTFLAGS += -hpvm-timers-ptx
+ VISC_OPTFLAGS += -visc-timers-ptx
else ifeq ($(TIMER),gen)
- TESTGEN_OPTFLAGS += -hpvm-timers-gen
+ TESTGEN_OPTFLAGS += -visc-timers-gen
else ifeq ($(TIMER),spir)
- TESTGEN_OPTFLAGS += -hpvm-timers-spir
+ TESTGEN_OPTFLAGS += -visc-timers-spir
else ifeq ($(TIMER),no)
else
ifeq ($(TARGET),x86)
- HPVM_OPTFLAGS += -hpvm-timers-x86 -hpvm-timers-spir
+ VISC_OPTFLAGS += -visc-timers-x86 -visc-timers-spir
else ifeq ($(TARGET),seq)
- HPVM_OPTFLAGS += -hpvm-timers-x86
+ VISC_OPTFLAGS += -visc-timers-x86
else ifeq ($(TARGET),seqx86)
- HPVM_OPTFLAGS += -hpvm-timers-x86 -hpvm-timers-spir
+ VISC_OPTFLAGS += -visc-timers-x86 -visc-timers-spir
else ifeq ($(TARGET),seqgpu)
- HPVM_OPTFLAGS += -hpvm-timers-x86 -hpvm-timers-ptx
+ VISC_OPTFLAGS += -visc-timers-x86 -visc-timers-ptx
else
- HPVM_OPTFLAGS += -hpvm-timers-x86 -hpvm-timers-ptx
+ VISC_OPTFLAGS += -visc-timers-x86 -visc-timers-ptx
endif
- TESTGEN_OPTFLAGS += -hpvm-timers-gen
+ TESTGEN_OPTFLAGS += -visc-timers-gen
endif
ifeq ($(DABSTRACTION),true)
- HPVM_OPTFLAGS += -hpvm-eda
+ VISC_OPTFLAGS += -visc-eda
endif
# Rules common to all makefiles
@@ -121,7 +121,7 @@ endif
########################################
OBJS = $(call INBUILDDIR,$(SRCDIR_OBJS))
-TEST_OBJS = $(call INBUILDDIR,$(HPVM_OBJS))
+TEST_OBJS = $(call INBUILDDIR,$(VISC_OBJS))
PARBOIL_OBJS = $(call INBUILDDIR,parboil.ll)
KERNEL = $(TEST_OBJS).kernels.ll
KERNEL_OPT = $(BUILDDIR)/$(APP).kernels.opt.ll
@@ -190,14 +190,14 @@ $(KERNEL_OPT) : $(KERNEL)
$(BIN) : $(HOST_LINKED)
$(CXX) -O3 $(LDFLAGS) $< -o $@
-$(HOST_LINKED) : $(HOST) $(OBJS) $(BUILDDIR)/parboil.ll $(HPVM_RT_LIB)
+$(HOST_LINKED) : $(HOST) $(OBJS) $(BUILDDIR)/parboil.ll $(VISC_RT_LIB)
$(LLVM_LINK) $^ -S -o $@
-$(HPVM_RT_LIB) : $(HPVM_RT_PATH)/hpvm-rt.cpp
+$(VISC_RT_LIB) : $(VISC_RT_PATH)/visc-rt.cpp
make -C $(LLVM_LIB_PATH)
-$(HOST) $(KERNEL): $(BUILDDIR)/$(HPVM_OBJS)
- $(OPT) --debug $(HPVM_OPTFLAGS) -S $< -o $(HOST)
+$(HOST) $(KERNEL): $(BUILDDIR)/$(VISC_OBJS)
+ $(OPT) --debug $(VISC_OPTFLAGS) -S $< -o $(HOST)
$(RUNDIR) :
mkdir -p $(RUNDIR)
@@ -214,11 +214,11 @@ $(BUILDDIR)/%.ll : $(SRCDIR)/%.cc
$(BUILDDIR)/%.ll : $(SRCDIR)/%.cpp
$(CXX) $(CXXFLAGS) -S -emit-llvm $< -o $@
-$(BUILDDIR)/%.hpvm.ll: $(BUILDDIR)/%.ll
+$(BUILDDIR)/%.visc.ll: $(BUILDDIR)/%.ll
$(OPT) $(TESTGEN_OPTFLAGS) $< -S -o $@
cat $(LLVM_SRC_ROOT)/tools/hpvm/test/parboil/RUN.parboil.script $@ > $@.tmp
- mv $@.tmp $(BUILDDIR)/$(APP).hpvm.ll
- #@cp $(HPVM_OBJS) $(BUILDDIR)/$(HPVM_OBJS)
+ mv $@.tmp $(BUILDDIR)/$(APP).visc.ll
+ #@cp $(VISC_OBJS) $(BUILDDIR)/$(VISC_OBJS)
$(BUILDDIR)/%.o : $(SRCDIR)/%.c
$(CC) $(CFLAGS) -c $< -o $@
diff --git a/hpvm/test/parboil/common/platform/hpvm.default.mk b/hpvm/test/parboil/common/platform/visc.default.mk
similarity index 61%
rename from hpvm/test/parboil/common/platform/hpvm.default.mk
rename to hpvm/test/parboil/common/platform/visc.default.mk
index ca90d453a3..03a9b0874a 100644
--- a/hpvm/test/parboil/common/platform/hpvm.default.mk
+++ b/hpvm/test/parboil/common/platform/visc.default.mk
@@ -12,20 +12,20 @@
#OPENCL_LIB_PATH=$(OPENCL_PATH)/lib/x86_64
#build
-HPVM_BUILD_DIR = $(LLVM_SRC_ROOT)/../build
+VISC_BUILD_DIR = $(LLVM_SRC_ROOT)/../build
# gcc (default)
-CC = $(HPVM_BUILD_DIR)/bin/clang
-OCLBE = $(HPVM_BUILD_DIR)/bin/llvm-cbe
-PLATFORM_CFLAGS = -I$(LLVM_SRC_ROOT)/include -I$(HPVM_BUILD_DIR)/include -I../../../include
+CC = $(VISC_BUILD_DIR)/bin/clang
+OCLBE = $(VISC_BUILD_DIR)/bin/llvm-cbe
+PLATFORM_CFLAGS = -I$(LLVM_SRC_ROOT)/include -I$(VISC_BUILD_DIR)/include -I../../../include
-CXX = $(HPVM_BUILD_DIR)/bin/clang++
-PLATFORM_CXXFLAGS = -I$(LLVM_SRC_ROOT)/include -I$(HPVM_BUILD_DIR)/include -I../../../include
+CXX = $(VISC_BUILD_DIR)/bin/clang++
+PLATFORM_CXXFLAGS = -I$(LLVM_SRC_ROOT)/include -I$(VISC_BUILD_DIR)/include -I../../../include
-LINKER = $(HPVM_BUILD_DIR)/bin/clang++
+LINKER = $(VISC_BUILD_DIR)/bin/clang++
PLATFORM_LDFLAGS = -lm -lpthread -lOpenCL
-LLVM_LIB_PATH = $(HPVM_BUILD_DIR)/lib
-LLVM_BIN_PATH = $(HPVM_BUILD_DIR)/bin
+LLVM_LIB_PATH = $(VISC_BUILD_DIR)/lib
+LLVM_BIN_PATH = $(VISC_BUILD_DIR)/bin
OPT = $(LLVM_BIN_PATH)/opt
LLVM_LINK = $(LLVM_BIN_PATH)/llvm-link
diff --git a/hpvm/test/parboil/driver/options.py b/hpvm/test/parboil/driver/options.py
index e15883b753..b80fc16168 100644
--- a/hpvm/test/parboil/driver/options.py
+++ b/hpvm/test/parboil/driver/options.py
@@ -264,7 +264,7 @@ def time_options(progname, cmd, args):
label_ptx = 'NVPTX_Timer'
#label_ptx = 'SPIR_Timer'
label_x86 = 'X86_Timer'
- label_gen = 'GenHPVM_Timer'
+ label_gen = 'GenVISC_Timer'
timings[label_f] = {}
timings[label_f]['IO'] = addTime([(label_pb, 'IO')], timings)
timings[label_f]['Memory Track'] = addTime([(label_pb, 'Mem_Track')], timings)
@@ -297,11 +297,11 @@ def time_options(progname, cmd, args):
timerName = 'Parboil'
timings[timerName] = {}
continue
- if line.startswith('Printing HPVM Timer'):
- regex = re.search('Printing HPVM Timer: *(?P<name>[a-zA-Z0-9 _]+)', line)
+ if line.startswith('Printing VISC Timer'):
+ regex = re.search('Printing VISC Timer: *(?P<name>[a-zA-Z0-9 _]+)', line)
timerName = regex.group('name').strip()
timings[timerName] = {}
- if timerName != 'NVPTX_Timer' and timerName != 'X86_Timer' and timerName != 'GenHPVM_Timer' and timerName != 'KernelTimer' and timerName != 'SPIR_Timer':
+ if timerName != 'NVPTX_Timer' and timerName != 'X86_Timer' and timerName != 'GenVISC_Timer' and timerName != 'KernelTimer' and timerName != 'SPIR_Timer':
print "Warning: Found unknown timer " + timerName
continue
m = re.search('(?P<timerID>[a-zA-Z _/]+) *: *(?P<value>[0-9]*\.[0-9]*) *$', line)
@@ -352,67 +352,67 @@ def time_options(progname, cmd, args):
globals.verbose = opts.verbose
configs = [
- ('spmv', { 'VERSION' : ["opencl_nvidia", "hpvm"],
+ ('spmv', { 'VERSION' : ["opencl_nvidia", "visc"],
'TEST' : [("large", 10)]
}
)
- ,('sgemm', { 'VERSION' : ["opencl_nvidia", "hpvm_sh"],
+ ,('sgemm', { 'VERSION' : ["opencl_nvidia", "visc_sh"],
'TEST' : [("4K", 10)]
}
)
- ,('lbm', { 'VERSION' : ["opencl_nvidia", "hpvm"],
+ ,('lbm', { 'VERSION' : ["opencl_nvidia", "visc"],
'TEST' : [("long", 10)]
}
)
- ,('stencil', { 'VERSION' : ["opencl_base", "hpvm"],
+ ,('stencil', { 'VERSION' : ["opencl_base", "visc"],
'TEST' : [("large", 10)]
}
)
- ,('bfs', { 'VERSION' : ["opencl_nvidia", "hpvm"],
+ ,('bfs', { 'VERSION' : ["opencl_nvidia", "visc"],
'TEST' : [("1M", 10), ("SF", 10)]
}
)
- ,('tpacf', { 'VERSION' : ["opencl_base", "hpvm"],
+ ,('tpacf', { 'VERSION' : ["opencl_base", "visc"],
'TEST' : [("large", 10)]
}
)
- ,('cutcp', { 'VERSION' : ["opencl_nvidia", "hpvm"],
+ ,('cutcp', { 'VERSION' : ["opencl_nvidia", "visc"],
'TEST' : [("large", 10)]
}
)
- #('histo', { 'VERSION' : ["opencl_nvidia", "hpvm"],
+ #('histo', { 'VERSION' : ["opencl_nvidia", "visc"],
#'TEST' : [("default", 10), ("large", 10)]
#}
#)
- #('spmv', { 'VERSION' : ["opencl_cpu_baseline", "hpvm"],
+ #('spmv', { 'VERSION' : ["opencl_cpu_baseline", "visc"],
#'TEST' : [("large", 10), ("huge", 10)]
#}
#)
- #('sgemm', { 'VERSION' : ["opencl_cpu_sm", "hpvm_sh"],
+ #('sgemm', { 'VERSION' : ["opencl_cpu_sm", "visc_sh"],
#'TEST' : [("medium", 1), ("4K", 1)]
#}
#)
- #('lbm', { 'VERSION' : ["opencl_cpu_baseline", "hpvm"],
+ #('lbm', { 'VERSION' : ["opencl_cpu_baseline", "visc"],
#'TEST' : [("short", 10), ("long", 10)]
#}
#)
- #,('stencil', { 'VERSION' : ["opencl_cpu_baseline", "hpvm"],
+ #,('stencil', { 'VERSION' : ["opencl_cpu_baseline", "visc"],
#'TEST' : [("default", 10), ("large", 10)]
#}
#)
- #('bfs', { 'VERSION' : ["opencl_cpu_baseline", "hpvm_base"],
+ #('bfs', { 'VERSION' : ["opencl_cpu_baseline", "visc_base"],
#'TEST' : [("1M", 5), ("SF", 5)]
#}
#)
- #,('tpacf', { 'VERSION' : ["opencl_cpu_base", "hpvm"],
+ #,('tpacf', { 'VERSION' : ["opencl_cpu_base", "visc"],
#'TEST' : [("medium", 1), ("large", 1)]
#}
#)
- #,('cutcp', { 'VERSION' : ["opencl_cpu_baseline", "hpvm"],
+ #,('cutcp', { 'VERSION' : ["opencl_cpu_baseline", "visc"],
#'TEST' : [("small", 1), ("large", 1)]
#}
#)
- #,('histo', { 'VERSION' : ["opencl_cpu_baseline", "hpvm"],
+ #,('histo', { 'VERSION' : ["opencl_cpu_baseline", "visc"],
#'TEST' : [("default", 1), ("large", 1)]
#}
#)
diff --git a/hpvm/test/parboil/parboilParser.py b/hpvm/test/parboil/parboilParser.py
index 5ea1346349..0d1f10b686 100755
--- a/hpvm/test/parboil/parboilParser.py
+++ b/hpvm/test/parboil/parboilParser.py
@@ -77,7 +77,7 @@ def parseCSVFile(filename):
file.close()
#print csvDict['a']['b']['c']['d']['e']
- #print csvDict['sgemm']['hpvm']['c']['d']['e']
+ #print csvDict['sgemm']['visc']['c']['d']['e']
#print csvDict['sgemm']['opencl_base']['c']['d']['e']
#print csvDict['sgemm']['opencl_base']['small']['d']['e']
#print csvDict['sgemm']['opencl_base']['small']['Final']['e']
@@ -96,14 +96,14 @@ def parseCSVFile(filename):
# returns a list of available tests for the given application
-# the tests are found based on the hpvm version, because it exists
+# the tests are found based on the visc version, because it exists
# for all apps in the dict
def getTests(app, csvDict):
- return csvDict[app]["hpvm"].keys()
+ return csvDict[app]["visc"].keys()
-def isHPVMVersion(version):
- return version.startswith("hpvm")
+def isViscVersion(version):
+ return version.startswith("visc")
def getAllVersions(csvDict):
@@ -142,7 +142,7 @@ def printTimerDecomposition(csvDict, version):
# get apps
apps = csvDict.keys()
- isHPVM = isHPVMVersion(version)
+ isVisc = isViscVersion(version)
# get tests for each app
tests = dict()
@@ -150,7 +150,7 @@ def printTimerDecomposition(csvDict, version):
tests[app] = csvDict[app][version].keys()
# list of timer-category pairs
- if isHPVM:
+ if isVisc:
timers =[('Final', 'Kernel'),
('Final', 'Load Program Binary'),
('Final', 'Argument Unpack'),
@@ -170,7 +170,7 @@ def printTimerDecomposition(csvDict, version):
('Parboil', 'Clear_Ctx'),
('Final', 'Timer Wall - IO'),
('Final', 'IO'),
- ('GenHPVM_Timer', 'Timer Wall Time')]
+ ('GenVISC_Timer', 'Timer Wall Time')]
else:
timers =[('Final', 'Init_Ctx'),
('Final', 'Arg_Unpack'),
diff --git a/hpvm/test/pipeline/Makefile b/hpvm/test/pipeline/Makefile
index 3fc794393c..421c9a8532 100644
--- a/hpvm/test/pipeline/Makefile
+++ b/hpvm/test/pipeline/Makefile
@@ -23,12 +23,12 @@ CURRENT_DIR := $(dir $(abspath $(lastword $(MAKEFILE_LIST))))
EXE = pipeline-$(TARGET)
INCLUDES += -I$(SRC_DIR) -I$(CAM_PIPE_SRC_DIR)
-INCLUDES += -I$(LLVM_SRC_ROOT)/include -I../include -I$(HPVM_BUILD_DIR)/include
+INCLUDES += -I$(LLVM_SRC_ROOT)/include -I../include -I$(VISC_BUILD_DIR)/include
## BEGIN HPVM MAKEFILE
SRCDIR_OBJS= io.ll
OBJS_SRC=src/io.cc
-HPVM_OBJS=main.hpvm.ll
+VISC_OBJS=main.visc.ll
APP = $(EXE)
APP_CFLAGS += $(INCLUDES) -ffast-math -O3 -fno-lax-vector-conversions -fno-vectorize -fno-slp-vectorize
APP_CXXFLAGS += $(INCLUDES) -ffast-math -O3 -fno-lax-vector-conversions -fno-vectorize -fno-slp-vectorize
@@ -39,21 +39,21 @@ OBJS_CFLAGS = $(APP_CFLAGS) $(PLATFORM_CFLAGS)
CXXFLAGS = $(APP_CXXFLAGS) $(PLATFORM_CXXFLAGS)
LDFLAGS= $(APP_LDFLAGS) $(PLATFORM_LDFLAGS)
-HPVM_RT_PATH = $(LLVM_SRC_ROOT)/tools/hpvm/projects/hpvm-rt
-HPVM_RT_LIB = $(HPVM_RT_PATH)/hpvm-rt.ll
+VISC_RT_PATH = $(LLVM_SRC_ROOT)/tools/hpvm/projects/visc-rt
+VISC_RT_LIB = $(VISC_RT_PATH)/visc-rt.ll
-TESTGEN_OPTFLAGS = -load LLVMGenHPVM.so -genhpvm -globaldce
+TESTGEN_OPTFLAGS = -load LLVMGenVISC.so -genvisc -globaldce
ifeq ($(TARGET),seq)
DEVICE = CPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG
- HPVM_OPTFLAGS += -hpvm-timers-x86
+ VISC_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG
+ VISC_OPTFLAGS += -visc-timers-x86
else
DEVICE = GPU_TARGET
- HPVM_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG
- HPVM_OPTFLAGS += -hpvm-timers-x86 -hpvm-timers-ptx
+ VISC_OPTFLAGS = -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -localmem -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG
+ VISC_OPTFLAGS += -visc-timers-x86 -visc-timers-ptx
endif
- TESTGEN_OPTFLAGS += -hpvm-timers-gen
+ TESTGEN_OPTFLAGS += -visc-timers-gen
CFLAGS += -DDEVICE=$(DEVICE)
CXXFLAGS += -DDEVICE=$(DEVICE)
@@ -64,7 +64,7 @@ INBUILDDIR=$(addprefix $(BUILD_DIR)/,$(1))
.PRECIOUS: $(BUILD_DIR)/%.ll
OBJS = $(call INBUILDDIR,$(SRCDIR_OBJS))
-TEST_OBJS = $(call INBUILDDIR,$(HPVM_OBJS))
+TEST_OBJS = $(call INBUILDDIR,$(VISC_OBJS))
KERNEL = $(TEST_OBJS).kernels.ll
ifeq ($(TARGET),seq)
@@ -91,14 +91,14 @@ $(KERNEL_OCL) : $(KERNEL)
$(EXE) : $(HOST_LINKED)
$(CXX) -O3 $(LDFLAGS) $< -o $@
-$(HOST_LINKED) : $(HOST) $(OBJS) $(HPVM_RT_LIB)
+$(HOST_LINKED) : $(HOST) $(OBJS) $(VISC_RT_LIB)
$(LLVM_LINK) $^ -S -o $@
-$(HPVM_RT_LIB) : $(HPVM_RT_PATH)/hpvm-rt.cpp
+$(VISC_RT_LIB) : $(VISC_RT_PATH)/visc-rt.cpp
make -C $(LLVM_LIB_PATH)
-$(HOST) $(KERNEL): $(BUILD_DIR)/$(HPVM_OBJS)
- $(OPT) -debug $(HPVM_OPTFLAGS) -S $< -o $(HOST)
+$(HOST) $(KERNEL): $(BUILD_DIR)/$(VISC_OBJS)
+ $(OPT) -debug $(VISC_OPTFLAGS) -S $< -o $(HOST)
$(BUILD_DIR):
mkdir -p $(BUILD_DIR)
@@ -109,7 +109,7 @@ $(BUILD_DIR)/%.ll : $(SRC_DIR)/%.cc
$(BUILD_DIR)/main.ll : $(SRC_DIR)/main.cc
$(CC) $(CXXFLAGS) -emit-llvm -S -o $@ $<
-$(BUILD_DIR)/main.hpvm.ll : $(BUILD_DIR)/main.ll
- $(OPT) -debug-only=genhpvm $(TESTGEN_OPTFLAGS) $< -S -o $@
+$(BUILD_DIR)/main.visc.ll : $(BUILD_DIR)/main.ll
+ $(OPT) -debug-only=genvisc $(TESTGEN_OPTFLAGS) $< -S -o $@
## END HPVM MAKEFILE
diff --git a/hpvm/test/pipeline/copyToVersions.sh b/hpvm/test/pipeline/copyToVersions.sh
index 67551aff2f..3b9c19bad6 100755
--- a/hpvm/test/pipeline/copyToVersions.sh
+++ b/hpvm/test/pipeline/copyToVersions.sh
@@ -1,12 +1,12 @@
-declare -a versionList=("hpvmGPU" "hpvmVector" "hpvmScalar" "hpvmGPU-Scalar-MaxG" "hpvmVector-Scalar-MaxG" "hpvmGPU-Scalar-ZC" "hpvmVector-Scalar-ZC")
+declare -a versionList=("viscGPU" "viscVector" "viscScalar" "viscGPU-Scalar-MaxG" "viscVector-Scalar-MaxG" "viscGPU-Scalar-ZC" "viscVector-Scalar-ZC")
declare -a fileList=("Makefile" "io.cc" "main.cc")
for version in "${versionList[@]}"; do
echo $version
for filename in "${fileList[@]}"; do
- echo cp ./src/hpvm_parallel/$filename ./src/$version/
- cp ./src/hpvm_parallel/$filename ./src/$version/
+ echo cp ./src/visc_parallel/$filename ./src/$version/
+ cp ./src/visc_parallel/$filename ./src/$version/
done
echo
done
diff --git a/hpvm/test/pipeline/gradient.hpvm.merged.experiments.notimer.ll b/hpvm/test/pipeline/gradient.visc.merged.experiments.notimer.ll
similarity index 95%
rename from hpvm/test/pipeline/gradient.hpvm.merged.experiments.notimer.ll
rename to hpvm/test/pipeline/gradient.visc.merged.experiments.notimer.ll
index 8056cc12ee..06ec055bb7 100644
--- a/hpvm/test/pipeline/gradient.hpvm.merged.experiments.notimer.ll
+++ b/hpvm/test/pipeline/gradient.visc.merged.experiments.notimer.ll
@@ -1,4 +1,4 @@
-; ModuleID = 'build/Gradient_default/main.hpvm.ll'
+; ModuleID = 'build/Gradient_default/main.visc.ll'
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"
@@ -167,9 +167,9 @@ entry:
; Function Attrs: nounwind uwtable
define %emptyStruct @squareRoot(float* nocapture in %Gx, i64 %bytesGx, float* nocapture in %Gy, i64 %bytesGy, float* nocapture out %G, i64 %bytesG, i32 %m, i32 %n, i32 %dummyH, i32 %dummyV) #2 {
entry:
- %call3 = tail call i8* @llvm.hpvm.getNode()
- %call14 = tail call i32 @llvm.hpvm.getNodeInstanceID.x(i8* %call3)
- %call25 = tail call i32 @llvm.hpvm.getNodeInstanceID.y(i8* %call3)
+ %call3 = tail call i8* @llvm.visc.getNode()
+ %call14 = tail call i32 @llvm.visc.getNodeInstanceID.x(i8* %call3)
+ %call25 = tail call i32 @llvm.visc.getNodeInstanceID.y(i8* %call3)
%cmp = icmp slt i32 %call14, %n
%cmp3 = icmp slt i32 %call25, %m
%or.cond = and i1 %cmp, %cmp3
@@ -198,51 +198,51 @@ if.end: ; preds = %if.then, %entry
; Function Attrs: nounwind uwtable
define %emptyStruct.23 @WrapperSquareRoot(float* nocapture in %Gx, i64 %bytesGx, float* nocapture in %Gy, i64 %bytesGy, float* nocapture out %G, i64 %bytesG, i32 %m, i32 %n, i32 %dummyH, i32 %dummyV) #2 {
entry:
- %squareRoot.node = tail call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct (float*, i64, float*, i64, float*, i64, i32, i32, i32, i32)* @squareRoot to i8*), i32 %m, i32 %n)
- tail call void @llvm.hpvm.bind.input(i8* %squareRoot.node, i32 0, i32 0, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %squareRoot.node, i32 1, i32 1, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %squareRoot.node, i32 2, i32 2, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %squareRoot.node, i32 3, i32 3, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %squareRoot.node, i32 4, i32 4, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %squareRoot.node, i32 5, i32 5, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %squareRoot.node, i32 6, i32 6, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %squareRoot.node, i32 7, i32 7, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %squareRoot.node, i32 8, i32 8, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %squareRoot.node, i32 9, i32 9, i1 false)
+ %squareRoot.node = tail call i8* @llvm.visc.createNode2D(i8* bitcast (%emptyStruct (float*, i64, float*, i64, float*, i64, i32, i32, i32, i32)* @squareRoot to i8*), i32 %m, i32 %n)
+ tail call void @llvm.visc.bind.input(i8* %squareRoot.node, i32 0, i32 0, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %squareRoot.node, i32 1, i32 1, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %squareRoot.node, i32 2, i32 2, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %squareRoot.node, i32 3, i32 3, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %squareRoot.node, i32 4, i32 4, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %squareRoot.node, i32 5, i32 5, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %squareRoot.node, i32 6, i32 6, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %squareRoot.node, i32 7, i32 7, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %squareRoot.node, i32 8, i32 8, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %squareRoot.node, i32 9, i32 9, i1 false)
ret %emptyStruct.23 undef
}
; Function Attrs: nounwind uwtable
define %emptyStruct.24 @Gradient(float* nocapture in %Is, i64 %bytesIs, float* nocapture in %Sx, i64 %bytesSx, float* nocapture in %Sy, i64 %bytesSy, float* nocapture out %Gx, i64 %bytesGx, float* nocapture out %Gy, i64 %bytesGy, float* nocapture out %G, i64 %bytesG, i32 %m, i32 %n) #2 {
entry:
- %WrapperHorizontal_WrapperVertical.node = tail call i8* @llvm.hpvm.createNode(i8* bitcast (%WrapperHorizontal.WrapperVertical.ty (float*, i64, float*, i64, float*, i64, i32, i32, float*, i64, float*, i64, float*, i64, i32, i32)* @WrapperHorizontal_WrapperVertical to i8*))
- %WrapperSquareRoot.node = tail call i8* @llvm.hpvm.createNode(i8* bitcast (%emptyStruct.23 (float*, i64, float*, i64, float*, i64, i32, i32, i32, i32)* @WrapperSquareRoot to i8*))
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 0, i32 0, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 1, i32 1, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 2, i32 2, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 3, i32 3, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 6, i32 4, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 7, i32 5, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 12, i32 6, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 13, i32 7, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 0, i32 8, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 1, i32 9, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 4, i32 10, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 5, i32 11, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 8, i32 12, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 9, i32 13, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 12, i32 14, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 13, i32 15, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperSquareRoot.node, i32 6, i32 0, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperSquareRoot.node, i32 7, i32 1, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperSquareRoot.node, i32 8, i32 2, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperSquareRoot.node, i32 9, i32 3, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperSquareRoot.node, i32 10, i32 4, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperSquareRoot.node, i32 11, i32 5, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperSquareRoot.node, i32 12, i32 6, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperSquareRoot.node, i32 13, i32 7, i1 false)
- %output.repl = tail call i8* @llvm.hpvm.createEdge(i8* %WrapperHorizontal_WrapperVertical.node, i8* %WrapperSquareRoot.node, i1 false, i32 0, i32 8, i1 false)
- %output1.repl = tail call i8* @llvm.hpvm.createEdge(i8* %WrapperHorizontal_WrapperVertical.node, i8* %WrapperSquareRoot.node, i1 false, i32 1, i32 9, i1 false)
+ %WrapperHorizontal_WrapperVertical.node = tail call i8* @llvm.visc.createNode(i8* bitcast (%WrapperHorizontal.WrapperVertical.ty (float*, i64, float*, i64, float*, i64, i32, i32, float*, i64, float*, i64, float*, i64, i32, i32)* @WrapperHorizontal_WrapperVertical to i8*))
+ %WrapperSquareRoot.node = tail call i8* @llvm.visc.createNode(i8* bitcast (%emptyStruct.23 (float*, i64, float*, i64, float*, i64, i32, i32, i32, i32)* @WrapperSquareRoot to i8*))
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 0, i32 0, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 1, i32 1, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 2, i32 2, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 3, i32 3, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 6, i32 4, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 7, i32 5, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 12, i32 6, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 13, i32 7, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 0, i32 8, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 1, i32 9, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 4, i32 10, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 5, i32 11, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 8, i32 12, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 9, i32 13, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 12, i32 14, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperHorizontal_WrapperVertical.node, i32 13, i32 15, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperSquareRoot.node, i32 6, i32 0, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperSquareRoot.node, i32 7, i32 1, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperSquareRoot.node, i32 8, i32 2, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperSquareRoot.node, i32 9, i32 3, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperSquareRoot.node, i32 10, i32 4, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperSquareRoot.node, i32 11, i32 5, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperSquareRoot.node, i32 12, i32 6, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperSquareRoot.node, i32 13, i32 7, i1 false)
+ %output.repl = tail call i8* @llvm.visc.createEdge(i8* %WrapperHorizontal_WrapperVertical.node, i8* %WrapperSquareRoot.node, i1 false, i32 0, i32 8, i1 false)
+ %output1.repl = tail call i8* @llvm.visc.createEdge(i8* %WrapperHorizontal_WrapperVertical.node, i8* %WrapperSquareRoot.node, i1 false, i32 1, i32 9, i1 false)
ret %emptyStruct.24 undef
}
@@ -866,7 +866,7 @@ cond.false: ; preds = %land.lhs.true58, %l
cond.end: ; preds = %land.lhs.true58
call void @pb_InitializeTimerSet(%struct.pb_TimerSet* %timers) #1
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%103 = load i32** %p.i.i.i.i, align 8, !tbaa !5
%104 = load i32* %103, align 4, !tbaa !9
%arrayidx.i296 = getelementptr inbounds i32* %103, i64 1
@@ -1137,15 +1137,15 @@ cond.false87: ; preds = %_Z12getNextFrameRN2
unreachable
cond.end88: ; preds = %_Z12getNextFrameRN2cv12VideoCaptureERNS_3MatE.exit335
- call void @llvm_hpvm_track_mem(i8* %150, i64 %mul65) #1
- call void @llvm_hpvm_track_mem(i8* %106, i64 36) #1
- call void @llvm_hpvm_track_mem(i8* %113, i64 36) #1
+ call void @llvm_visc_track_mem(i8* %150, i64 %mul65) #1
+ call void @llvm_visc_track_mem(i8* %106, i64 36) #1
+ call void @llvm_visc_track_mem(i8* %113, i64 36) #1
%176 = load i8** %data73, align 8, !tbaa !5
- call void @llvm_hpvm_track_mem(i8* %176, i64 %mul65) #1
+ call void @llvm_visc_track_mem(i8* %176, i64 %mul65) #1
%177 = load i8** %data74, align 8, !tbaa !5
- call void @llvm_hpvm_track_mem(i8* %177, i64 %mul65) #1
+ call void @llvm_visc_track_mem(i8* %177, i64 %mul65) #1
%178 = load i8** %data75, align 8, !tbaa !5
- call void @llvm_hpvm_track_mem(i8* %178, i64 %mul65) #1
+ call void @llvm_visc_track_mem(i8* %178, i64 %mul65) #1
%179 = load i8** %data, align 8, !tbaa !5
%180 = bitcast i8* %179 to float*
store float* %180, float** %I1.i, align 1, !tbaa !5
@@ -1154,8 +1154,8 @@ cond.end88: ; preds = %_Z12getNextFrameRN2
for.body: ; preds = %for.body, %cond.end88
%j.0480 = phi i32 [ 0, %cond.end88 ], [ %inc, %for.body ]
- %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%emptyStruct.24 (float*, i64, float*, i64, float*, i64, float*, i64, float*, i64, float*, i64, i32, i32)* @Gradient to i8*), i8* %call66, i1 false)
- call void @llvm.hpvm.wait(i8* %graphID)
+ %graphID = call i8* @llvm.visc.launch(i8* bitcast (%emptyStruct.24 (float*, i64, float*, i64, float*, i64, float*, i64, float*, i64, float*, i64, i32, i32)* @Gradient to i8*), i8* %call66, i1 false)
+ call void @llvm.visc.wait(i8* %graphID)
%inc = add i32 %j.0480, 1
%exitcond = icmp eq i32 %inc, 2994
br i1 %exitcond, label %for.end, label %for.body
@@ -1163,19 +1163,19 @@ for.body: ; preds = %for.body, %cond.end
for.end: ; preds = %for.body
call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 0) #1
%181 = load i8** %data75, align 8, !tbaa !5
- call void @llvm_hpvm_request_mem(i8* %181, i64 %mul65) #1
+ call void @llvm_visc_request_mem(i8* %181, i64 %mul65) #1
%182 = load i8** %data, align 8, !tbaa !5
- call void @llvm_hpvm_untrack_mem(i8* %182) #1
- call void @llvm_hpvm_untrack_mem(i8* %106) #1
- call void @llvm_hpvm_untrack_mem(i8* %113) #1
+ call void @llvm_visc_untrack_mem(i8* %182) #1
+ call void @llvm_visc_untrack_mem(i8* %106) #1
+ call void @llvm_visc_untrack_mem(i8* %113) #1
%183 = load i8** %data73, align 8, !tbaa !5
- call void @llvm_hpvm_untrack_mem(i8* %183) #1
+ call void @llvm_visc_untrack_mem(i8* %183) #1
%184 = load i8** %data74, align 8, !tbaa !5
- call void @llvm_hpvm_untrack_mem(i8* %184) #1
+ call void @llvm_visc_untrack_mem(i8* %184) #1
%185 = load i8** %data75, align 8, !tbaa !5
- call void @llvm_hpvm_untrack_mem(i8* %185) #1
+ call void @llvm_visc_untrack_mem(i8* %185) #1
call void @pb_PrintTimerSet(%struct.pb_TimerSet* %timers) #1
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.cleanup()
call void @pb_FreeParameters(%struct.pb_Parameters* %call3) #1
%u.i.i.i342 = getelementptr inbounds %"class.cv::Mat"* %out, i64 0, i32 9
%186 = load %"struct.cv::UMatData"** %u.i.i.i342, align 8, !tbaa !5
@@ -1647,13 +1647,13 @@ declare noalias i8* @malloc(i64) #5
declare void @_ZN2cv12VideoCaptureD1Ev(%"class.cv::VideoCapture"*) #0
-declare void @llvm_hpvm_track_mem(i8*, i64) #0
+declare void @llvm_visc_track_mem(i8*, i64) #0
declare void @pb_SwitchToTimer(%struct.pb_TimerSet*, i32) #0
-declare void @llvm_hpvm_request_mem(i8*, i64) #0
+declare void @llvm_visc_request_mem(i8*, i64) #0
-declare void @llvm_hpvm_untrack_mem(i8*) #0
+declare void @llvm_visc_untrack_mem(i8*) #0
declare void @pb_PrintTimerSet(%struct.pb_TimerSet*) #0
@@ -1713,50 +1713,50 @@ entry:
declare i64 @fwrite(i8* nocapture, i64, i64, %struct._IO_FILE* nocapture) #1
; Function Attrs: nounwind readnone
-declare i8* @llvm.hpvm.getNode() #7
+declare i8* @llvm.visc.getNode() #7
; Function Attrs: nounwind readnone
-declare i32 @llvm.hpvm.getNodeInstanceID.x(i8*) #7
+declare i32 @llvm.visc.getNodeInstanceID.x(i8*) #7
; Function Attrs: nounwind readnone
-declare i32 @llvm.hpvm.getNodeInstanceID.y(i8*) #7
+declare i32 @llvm.visc.getNodeInstanceID.y(i8*) #7
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode2D(i8*, i32, i32) #1
+declare i8* @llvm.visc.createNode2D(i8*, i32, i32) #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1) #1
+declare void @llvm.visc.bind.input(i8*, i32, i32, i1) #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32, i1) #1
+declare void @llvm.visc.bind.output(i8*, i32, i32, i1) #1
; Function Attrs: nounwind readonly
declare float @llvm.sqrt.f32(float) #8
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #1
+declare i8* @llvm.visc.createNode(i8*) #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32, i1) #1
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32, i1) #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #1
+declare i8* @llvm.visc.launch(i8*, i8*, i1) #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #1
+declare void @llvm.visc.wait(i8*) #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
define %horizontal.vertical.ty @horizontal_vertical(float* nocapture in %n1_Is, i64 %n1_bytesIs, float* nocapture in %n1_Sx, i64 %n1_bytesSx, float* nocapture out %n1_Gx, i64 %n1_bytesGx, i32 %n1_m, i32 %n1_n, float* nocapture in %n2_Is, i64 %n2_bytesIs, float* nocapture in %n2_Sy, i64 %n2_bytesSy, float* nocapture out %n2_Gy, i64 %n2_bytesGy, i32 %n2_m, i32 %n2_n) #1 {
entry:
- %call3.i = tail call i8* @llvm.hpvm.getNode() #1
- %call14.i = tail call i32 @llvm.hpvm.getNodeInstanceID.x(i8* %call3.i) #1
- %call25.i = tail call i32 @llvm.hpvm.getNodeInstanceID.y(i8* %call3.i) #1
+ %call3.i = tail call i8* @llvm.visc.getNode() #1
+ %call14.i = tail call i32 @llvm.visc.getNodeInstanceID.x(i8* %call3.i) #1
+ %call25.i = tail call i32 @llvm.visc.getNodeInstanceID.y(i8* %call3.i) #1
%mul.i = mul nsw i32 %call25.i, %n1_n
%add.i = add nsw i32 %mul.i, %call14.i
%cmp.i = icmp slt i32 %call14.i, %n1_n
@@ -2139,25 +2139,25 @@ vertical.exit: ; preds = %if.end42.2.i67.us,
; Function Attrs: nounwind
define %WrapperHorizontal.WrapperVertical.ty @WrapperHorizontal_WrapperVertical(float* nocapture in %n1_Is, i64 %n1_bytesIs, float* nocapture in %n1_Sx, i64 %n1_bytesSx, float* nocapture out %n1_Gx, i64 %n1_bytesGx, i32 %n1_m, i32 %n1_n, float* nocapture in %n2_Is, i64 %n2_bytesIs, float* nocapture in %n2_Sy, i64 %n2_bytesSy, float* nocapture out %n2_Gy, i64 %n2_bytesGy, i32 %n2_m, i32 %n2_n) #1 {
entry:
- %horizontal_vertical.node = tail call i8* @llvm.hpvm.createNode2D(i8* bitcast (%horizontal.vertical.ty (float*, i64, float*, i64, float*, i64, i32, i32, float*, i64, float*, i64, float*, i64, i32, i32)* @horizontal_vertical to i8*), i32 %n1_m, i32 %n1_n)
- tail call void @llvm.hpvm.bind.output(i8* %horizontal_vertical.node, i32 0, i32 0, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 7, i32 7, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 6, i32 6, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 5, i32 5, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 4, i32 4, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 3, i32 3, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 2, i32 2, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 1, i32 1, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 0, i32 0, i1 false)
- tail call void @llvm.hpvm.bind.output(i8* %horizontal_vertical.node, i32 1, i32 1, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 15, i32 15, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 14, i32 14, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 13, i32 13, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 12, i32 12, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 11, i32 11, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 10, i32 10, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 9, i32 9, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %horizontal_vertical.node, i32 8, i32 8, i1 false)
+ %horizontal_vertical.node = tail call i8* @llvm.visc.createNode2D(i8* bitcast (%horizontal.vertical.ty (float*, i64, float*, i64, float*, i64, i32, i32, float*, i64, float*, i64, float*, i64, i32, i32)* @horizontal_vertical to i8*), i32 %n1_m, i32 %n1_n)
+ tail call void @llvm.visc.bind.output(i8* %horizontal_vertical.node, i32 0, i32 0, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 7, i32 7, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 6, i32 6, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 5, i32 5, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 4, i32 4, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 3, i32 3, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 2, i32 2, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 1, i32 1, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 0, i32 0, i1 false)
+ tail call void @llvm.visc.bind.output(i8* %horizontal_vertical.node, i32 1, i32 1, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 15, i32 15, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 14, i32 14, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 13, i32 13, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 12, i32 12, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 11, i32 11, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 10, i32 10, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 9, i32 9, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %horizontal_vertical.node, i32 8, i32 8, i1 false)
ret %WrapperHorizontal.WrapperVertical.ty undef
}
@@ -2172,9 +2172,9 @@ attributes #7 = { nounwind readnone }
attributes #8 = { nounwind readonly }
attributes #9 = { noreturn nounwind }
-!hpvm_hint_gpu = !{!0, !1}
-!hpvm_hint_cpu = !{!2, !3, !4}
-!hpvm_hint_spir = !{}
+!visc_hint_gpu = !{!0, !1}
+!visc_hint_cpu = !{!2, !3, !4}
+!visc_hint_spir = !{}
!0 = metadata !{%emptyStruct (float*, i64, float*, i64, float*, i64, i32, i32, i32, i32)* @squareRoot}
!1 = metadata !{%horizontal.vertical.ty (float*, i64, float*, i64, float*, i64, i32, i32, float*, i64, float*, i64, float*, i64, i32, i32)* @horizontal_vertical}
diff --git a/hpvm/test/pipeline/laplacian.hpvm.merged.experiments.notimer.ll b/hpvm/test/pipeline/laplacian.visc.merged.experiments.notimer.ll
similarity index 95%
rename from hpvm/test/pipeline/laplacian.hpvm.merged.experiments.notimer.ll
rename to hpvm/test/pipeline/laplacian.visc.merged.experiments.notimer.ll
index aa4a0d19a0..4b04586251 100644
--- a/hpvm/test/pipeline/laplacian.hpvm.merged.experiments.notimer.ll
+++ b/hpvm/test/pipeline/laplacian.visc.merged.experiments.notimer.ll
@@ -1,4 +1,4 @@
-; ModuleID = 'build/Laplacian_default/main.hpvm.ll'
+; ModuleID = 'build/Laplacian_default/main.visc.ll'
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"
@@ -170,9 +170,9 @@ declare void @llvm.lifetime.end(i64, i8* nocapture) #1
; Function Attrs: nounwind uwtable
define %emptyStruct @lincomb(float* nocapture in %Is, i64 %bytesIs, float* nocapture in %D, i64 %bytesD, float* nocapture in %E, i64 %bytesE, float* nocapture out %L, i64 %bytesL, i32 %m, i32 %n, i32 %dummyD, i32 %dummyE) #2 {
entry:
- %call3 = tail call i8* @llvm.hpvm.getNode()
- %call14 = tail call i32 @llvm.hpvm.getNodeInstanceID.x(i8* %call3)
- %call25 = tail call i32 @llvm.hpvm.getNodeInstanceID.y(i8* %call3)
+ %call3 = tail call i8* @llvm.visc.getNode()
+ %call14 = tail call i32 @llvm.visc.getNodeInstanceID.x(i8* %call3)
+ %call25 = tail call i32 @llvm.visc.getNodeInstanceID.y(i8* %call3)
%cmp = icmp slt i32 %call14, %n
%cmp3 = icmp slt i32 %call25, %m
%or.cond = and i1 %cmp, %cmp3
@@ -202,55 +202,55 @@ if.end: ; preds = %if.then, %entry
; Function Attrs: nounwind uwtable
define %emptyStruct.23 @WrapperLincomb(float* nocapture in %Is, i64 %bytesIs, float* nocapture in %D, i64 %bytesD, float* nocapture in %E, i64 %bytesE, float* nocapture out %L, i64 %bytesL, i32 %m, i32 %n, i32 %dummyD, i32 %dummyE) #2 {
entry:
- %lincomb.node = tail call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct (float*, i64, float*, i64, float*, i64, float*, i64, i32, i32, i32, i32)* @lincomb to i8*), i32 %m, i32 %n)
- tail call void @llvm.hpvm.bind.input(i8* %lincomb.node, i32 0, i32 0, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %lincomb.node, i32 1, i32 1, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %lincomb.node, i32 2, i32 2, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %lincomb.node, i32 3, i32 3, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %lincomb.node, i32 4, i32 4, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %lincomb.node, i32 5, i32 5, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %lincomb.node, i32 6, i32 6, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %lincomb.node, i32 7, i32 7, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %lincomb.node, i32 8, i32 8, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %lincomb.node, i32 9, i32 9, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %lincomb.node, i32 10, i32 10, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %lincomb.node, i32 11, i32 11, i1 false)
+ %lincomb.node = tail call i8* @llvm.visc.createNode2D(i8* bitcast (%emptyStruct (float*, i64, float*, i64, float*, i64, float*, i64, i32, i32, i32, i32)* @lincomb to i8*), i32 %m, i32 %n)
+ tail call void @llvm.visc.bind.input(i8* %lincomb.node, i32 0, i32 0, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %lincomb.node, i32 1, i32 1, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %lincomb.node, i32 2, i32 2, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %lincomb.node, i32 3, i32 3, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %lincomb.node, i32 4, i32 4, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %lincomb.node, i32 5, i32 5, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %lincomb.node, i32 6, i32 6, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %lincomb.node, i32 7, i32 7, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %lincomb.node, i32 8, i32 8, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %lincomb.node, i32 9, i32 9, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %lincomb.node, i32 10, i32 10, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %lincomb.node, i32 11, i32 11, i1 false)
ret %emptyStruct.23 undef
}
; Function Attrs: nounwind uwtable
define %emptyStruct.24 @LaplacianEstimate(float* nocapture in %Is, i64 %bytesIs, float* nocapture in %B, i64 %bytesB, float* nocapture out %D, i64 %bytesD, float* nocapture out %E, i64 %bytesE, float* nocapture out %L, i64 %bytesL, i32 %m, i32 %n) #2 {
entry:
- %WrapperDilate_WrapperErode.node = tail call i8* @llvm.hpvm.createNode(i8* bitcast (%WrapperDilate.WrapperErode.ty (float*, i64, float*, i64, float*, i64, i32, i32, float*, i64, float*, i64, float*, i64, i32, i32)* @WrapperDilate_WrapperErode to i8*))
- %WrapperLincomb.node = tail call i8* @llvm.hpvm.createNode(i8* bitcast (%emptyStruct.23 (float*, i64, float*, i64, float*, i64, float*, i64, i32, i32, i32, i32)* @WrapperLincomb to i8*))
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 0, i32 0, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 1, i32 1, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 2, i32 2, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 3, i32 3, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 4, i32 4, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 5, i32 5, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 10, i32 6, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 11, i32 7, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 0, i32 8, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 1, i32 9, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 2, i32 10, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 3, i32 11, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 6, i32 12, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 7, i32 13, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 10, i32 14, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 11, i32 15, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperLincomb.node, i32 0, i32 0, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperLincomb.node, i32 1, i32 1, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperLincomb.node, i32 4, i32 2, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperLincomb.node, i32 5, i32 3, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperLincomb.node, i32 6, i32 4, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperLincomb.node, i32 7, i32 5, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperLincomb.node, i32 8, i32 6, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperLincomb.node, i32 9, i32 7, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperLincomb.node, i32 10, i32 8, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %WrapperLincomb.node, i32 11, i32 9, i1 false)
- %output.repl = tail call i8* @llvm.hpvm.createEdge(i8* %WrapperDilate_WrapperErode.node, i8* %WrapperLincomb.node, i1 false, i32 0, i32 10, i1 false)
- %output1.repl = tail call i8* @llvm.hpvm.createEdge(i8* %WrapperDilate_WrapperErode.node, i8* %WrapperLincomb.node, i1 false, i32 1, i32 11, i1 false)
+ %WrapperDilate_WrapperErode.node = tail call i8* @llvm.visc.createNode(i8* bitcast (%WrapperDilate.WrapperErode.ty (float*, i64, float*, i64, float*, i64, i32, i32, float*, i64, float*, i64, float*, i64, i32, i32)* @WrapperDilate_WrapperErode to i8*))
+ %WrapperLincomb.node = tail call i8* @llvm.visc.createNode(i8* bitcast (%emptyStruct.23 (float*, i64, float*, i64, float*, i64, float*, i64, i32, i32, i32, i32)* @WrapperLincomb to i8*))
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 0, i32 0, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 1, i32 1, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 2, i32 2, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 3, i32 3, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 4, i32 4, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 5, i32 5, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 10, i32 6, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 11, i32 7, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 0, i32 8, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 1, i32 9, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 2, i32 10, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 3, i32 11, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 6, i32 12, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 7, i32 13, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 10, i32 14, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperDilate_WrapperErode.node, i32 11, i32 15, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperLincomb.node, i32 0, i32 0, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperLincomb.node, i32 1, i32 1, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperLincomb.node, i32 4, i32 2, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperLincomb.node, i32 5, i32 3, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperLincomb.node, i32 6, i32 4, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperLincomb.node, i32 7, i32 5, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperLincomb.node, i32 8, i32 6, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperLincomb.node, i32 9, i32 7, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperLincomb.node, i32 10, i32 8, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %WrapperLincomb.node, i32 11, i32 9, i1 false)
+ %output.repl = tail call i8* @llvm.visc.createEdge(i8* %WrapperDilate_WrapperErode.node, i8* %WrapperLincomb.node, i1 false, i32 0, i32 10, i1 false)
+ %output1.repl = tail call i8* @llvm.visc.createEdge(i8* %WrapperDilate_WrapperErode.node, i8* %WrapperLincomb.node, i1 false, i32 1, i32 11, i1 false)
ret %emptyStruct.24 undef
}
@@ -873,7 +873,7 @@ cond.false: ; preds = %land.lhs.true58, %l
cond.end: ; preds = %land.lhs.true58
call void @pb_InitializeTimerSet(%struct.pb_TimerSet* %timers) #1
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%103 = load i32** %p.i.i.i.i, align 8, !tbaa !5
%104 = load i32* %103, align 4, !tbaa !9
%arrayidx.i290 = getelementptr inbounds i32* %103, i64 1
@@ -1062,18 +1062,18 @@ _Z12getNextFrameRN2cv12VideoCaptureERNS_3MatE.exit332: ; preds = %if.then.i328,
call void @llvm.lifetime.end(i64 24, i8* %134) #1
%data = getelementptr inbounds %"class.cv::Mat"* %src, i64 0, i32 4
%139 = load i8** %data, align 8, !tbaa !5
- call void @llvm_hpvm_track_mem(i8* %139, i64 %mul65) #1
+ call void @llvm_visc_track_mem(i8* %139, i64 %mul65) #1
%arraydecay = getelementptr inbounds [9 x float]* %B, i64 0, i64 0
- call void @llvm_hpvm_track_mem(i8* %106, i64 36) #1
+ call void @llvm_visc_track_mem(i8* %106, i64 36) #1
%data81 = getelementptr inbounds %"class.cv::Mat"* %D, i64 0, i32 4
%140 = load i8** %data81, align 8, !tbaa !5
- call void @llvm_hpvm_track_mem(i8* %140, i64 %mul65) #1
+ call void @llvm_visc_track_mem(i8* %140, i64 %mul65) #1
%data82 = getelementptr inbounds %"class.cv::Mat"* %E, i64 0, i32 4
%141 = load i8** %data82, align 8, !tbaa !5
- call void @llvm_hpvm_track_mem(i8* %141, i64 %mul65) #1
+ call void @llvm_visc_track_mem(i8* %141, i64 %mul65) #1
%data83 = getelementptr inbounds %"class.cv::Mat"* %L, i64 0, i32 4
%142 = load i8** %data83, align 8, !tbaa !5
- call void @llvm_hpvm_track_mem(i8* %142, i64 %mul65) #1
+ call void @llvm_visc_track_mem(i8* %142, i64 %mul65) #1
%143 = load i8** %data, align 8, !tbaa !5
%144 = bitcast i8* %143 to float*
%145 = load i8** %data81, align 8, !tbaa !5
@@ -1126,8 +1126,8 @@ _Z12getNextFrameRN2cv12VideoCaptureERNS_3MatE.exit332: ; preds = %if.then.i328,
for.body: ; preds = %for.body, %_Z12getNextFrameRN2cv12VideoCaptureERNS_3MatE.exit332
%j.0474 = phi i32 [ 0, %_Z12getNextFrameRN2cv12VideoCaptureERNS_3MatE.exit332 ], [ %inc, %for.body ]
- %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%emptyStruct.24 (float*, i64, float*, i64, float*, i64, float*, i64, float*, i64, i32, i32)* @LaplacianEstimate to i8*), i8* %call66, i1 false)
- call void @llvm.hpvm.wait(i8* %graphID)
+ %graphID = call i8* @llvm.visc.launch(i8* bitcast (%emptyStruct.24 (float*, i64, float*, i64, float*, i64, float*, i64, float*, i64, i32, i32)* @LaplacianEstimate to i8*), i8* %call66, i1 false)
+ call void @llvm.visc.wait(i8* %graphID)
%inc = add nsw i32 %j.0474, 1
%exitcond = icmp eq i32 %inc, 2994
br i1 %exitcond, label %for.end, label %for.body
@@ -1135,18 +1135,18 @@ for.body: ; preds = %for.body, %_Z12getN
for.end: ; preds = %for.body
call void @pb_SwitchToTimer(%struct.pb_TimerSet* %timers, i32 0) #1
%165 = load i8** %data83, align 8, !tbaa !5
- call void @llvm_hpvm_request_mem(i8* %165, i64 %mul65) #1
+ call void @llvm_visc_request_mem(i8* %165, i64 %mul65) #1
%166 = load i8** %data, align 8, !tbaa !5
- call void @llvm_hpvm_untrack_mem(i8* %166) #1
- call void @llvm_hpvm_untrack_mem(i8* %106) #1
+ call void @llvm_visc_untrack_mem(i8* %166) #1
+ call void @llvm_visc_untrack_mem(i8* %106) #1
%167 = load i8** %data81, align 8, !tbaa !5
- call void @llvm_hpvm_untrack_mem(i8* %167) #1
+ call void @llvm_visc_untrack_mem(i8* %167) #1
%168 = load i8** %data82, align 8, !tbaa !5
- call void @llvm_hpvm_untrack_mem(i8* %168) #1
+ call void @llvm_visc_untrack_mem(i8* %168) #1
%169 = load i8** %data83, align 8, !tbaa !5
- call void @llvm_hpvm_untrack_mem(i8* %169) #1
+ call void @llvm_visc_untrack_mem(i8* %169) #1
call void @pb_PrintTimerSet(%struct.pb_TimerSet* %timers) #1
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.cleanup()
call void @pb_FreeParameters(%struct.pb_Parameters* %call3) #1
%u.i.i.i336 = getelementptr inbounds %"class.cv::Mat"* %out, i64 0, i32 9
%170 = load %"struct.cv::UMatData"** %u.i.i.i336, align 8, !tbaa !5
@@ -1614,13 +1614,13 @@ declare noalias i8* @malloc(i64) #5
declare void @_ZN2cv12VideoCaptureD1Ev(%"class.cv::VideoCapture"*) #0
-declare void @llvm_hpvm_track_mem(i8*, i64) #0
+declare void @llvm_visc_track_mem(i8*, i64) #0
declare void @pb_SwitchToTimer(%struct.pb_TimerSet*, i32) #0
-declare void @llvm_hpvm_request_mem(i8*, i64) #0
+declare void @llvm_visc_request_mem(i8*, i64) #0
-declare void @llvm_hpvm_untrack_mem(i8*) #0
+declare void @llvm_visc_untrack_mem(i8*) #0
declare void @pb_PrintTimerSet(%struct.pb_TimerSet*) #0
@@ -1677,47 +1677,47 @@ declare i64 @fwrite(i8* nocapture, i64, i64, %struct._IO_FILE* nocapture) #1
declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i32, i1) #1
; Function Attrs: nounwind readnone
-declare i8* @llvm.hpvm.getNode() #7
+declare i8* @llvm.visc.getNode() #7
; Function Attrs: nounwind readnone
-declare i32 @llvm.hpvm.getNodeInstanceID.x(i8*) #7
+declare i32 @llvm.visc.getNodeInstanceID.x(i8*) #7
; Function Attrs: nounwind readnone
-declare i32 @llvm.hpvm.getNodeInstanceID.y(i8*) #7
+declare i32 @llvm.visc.getNodeInstanceID.y(i8*) #7
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode2D(i8*, i32, i32) #1
+declare i8* @llvm.visc.createNode2D(i8*, i32, i32) #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1) #1
+declare void @llvm.visc.bind.input(i8*, i32, i32, i1) #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32, i1) #1
+declare void @llvm.visc.bind.output(i8*, i32, i32, i1) #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #1
+declare i8* @llvm.visc.createNode(i8*) #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32, i1) #1
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32, i1) #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #1
+declare i8* @llvm.visc.launch(i8*, i8*, i1) #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #1
+declare void @llvm.visc.wait(i8*) #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
define %dilate.erode.ty @dilate_erode(float* nocapture in %n1_Is, i64 %n1_bytesIs, float* nocapture in %n1_B, i64 %n1_bytesB, float* nocapture out %n1_D, i64 %n1_bytesD, i32 %n1_m, i32 %n1_n, float* nocapture in %n2_Is, i64 %n2_bytesIs, float* nocapture in %n2_B, i64 %n2_bytesB, float* nocapture out %n2_E, i64 %n2_bytesE, i32 %n2_m, i32 %n2_n) #1 {
entry:
- %call3.i = tail call i8* @llvm.hpvm.getNode() #1
- %call14.i = tail call i32 @llvm.hpvm.getNodeInstanceID.x(i8* %call3.i) #1
- %call25.i = tail call i32 @llvm.hpvm.getNodeInstanceID.y(i8* %call3.i) #1
+ %call3.i = tail call i8* @llvm.visc.getNode() #1
+ %call14.i = tail call i32 @llvm.visc.getNodeInstanceID.x(i8* %call3.i) #1
+ %call25.i = tail call i32 @llvm.visc.getNodeInstanceID.y(i8* %call3.i) #1
%cmp.i = icmp slt i32 %call14.i, %n1_n
%cmp3.i = icmp slt i32 %call25.i, %n1_m
%or.cond.i = and i1 %cmp.i, %cmp3.i
@@ -2070,25 +2070,25 @@ erode.exit: ; preds = %dilate.exit, %cond.
; Function Attrs: nounwind
define %WrapperDilate.WrapperErode.ty @WrapperDilate_WrapperErode(float* nocapture in %n1_Is, i64 %n1_bytesIs, float* nocapture in %n1_B, i64 %n1_bytesB, float* nocapture out %n1_D, i64 %n1_bytesD, i32 %n1_m, i32 %n1_n, float* nocapture in %n2_Is, i64 %n2_bytesIs, float* nocapture in %n2_B, i64 %n2_bytesB, float* nocapture out %n2_E, i64 %n2_bytesE, i32 %n2_m, i32 %n2_n) #1 {
entry:
- %dilate_erode.node = tail call i8* @llvm.hpvm.createNode2D(i8* bitcast (%dilate.erode.ty (float*, i64, float*, i64, float*, i64, i32, i32, float*, i64, float*, i64, float*, i64, i32, i32)* @dilate_erode to i8*), i32 %n1_m, i32 %n1_n)
- tail call void @llvm.hpvm.bind.output(i8* %dilate_erode.node, i32 0, i32 0, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 7, i32 7, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 6, i32 6, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 5, i32 5, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 4, i32 4, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 3, i32 3, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 2, i32 2, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 1, i32 1, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 0, i32 0, i1 false)
- tail call void @llvm.hpvm.bind.output(i8* %dilate_erode.node, i32 1, i32 1, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 15, i32 15, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 14, i32 14, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 13, i32 13, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 12, i32 12, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 11, i32 11, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 10, i32 10, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 9, i32 9, i1 false)
- tail call void @llvm.hpvm.bind.input(i8* %dilate_erode.node, i32 8, i32 8, i1 false)
+ %dilate_erode.node = tail call i8* @llvm.visc.createNode2D(i8* bitcast (%dilate.erode.ty (float*, i64, float*, i64, float*, i64, i32, i32, float*, i64, float*, i64, float*, i64, i32, i32)* @dilate_erode to i8*), i32 %n1_m, i32 %n1_n)
+ tail call void @llvm.visc.bind.output(i8* %dilate_erode.node, i32 0, i32 0, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 7, i32 7, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 6, i32 6, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 5, i32 5, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 4, i32 4, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 3, i32 3, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 2, i32 2, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 1, i32 1, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 0, i32 0, i1 false)
+ tail call void @llvm.visc.bind.output(i8* %dilate_erode.node, i32 1, i32 1, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 15, i32 15, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 14, i32 14, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 13, i32 13, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 12, i32 12, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 11, i32 11, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 10, i32 10, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 9, i32 9, i1 false)
+ tail call void @llvm.visc.bind.input(i8* %dilate_erode.node, i32 8, i32 8, i1 false)
ret %WrapperDilate.WrapperErode.ty undef
}
@@ -2103,9 +2103,9 @@ attributes #7 = { nounwind readnone }
attributes #8 = { noreturn nounwind }
attributes #9 = { nounwind readonly }
-!hpvm_hint_gpu = !{!0, !1}
-!hpvm_hint_cpu = !{!2, !3, !4}
-!hpvm_hint_spir = !{}
+!visc_hint_gpu = !{!0, !1}
+!visc_hint_cpu = !{!2, !3, !4}
+!visc_hint_spir = !{}
!0 = metadata !{%emptyStruct (float*, i64, float*, i64, float*, i64, float*, i64, i32, i32, i32, i32)* @lincomb}
!1 = metadata !{%dilate.erode.ty (float*, i64, float*, i64, float*, i64, i32, i32, float*, i64, float*, i64, float*, i64, i32, i32)* @dilate_erode}
diff --git a/hpvm/test/pipeline/run.sh b/hpvm/test/pipeline/run.sh
index 5ac734026b..0c8435764b 100755
--- a/hpvm/test/pipeline/run.sh
+++ b/hpvm/test/pipeline/run.sh
@@ -4,7 +4,7 @@ echo Pipeline Script $1 $2
version=$1
pos=$2
-if [[ ($version == *"GPU"*) || ($version == "hpvm_parallel") ]]
+if [[ ($version == *"GPU"*) || ($version == "visc_parallel") ]]
then
target=""
elif [[ $version == *"Vector"* ]]
diff --git a/hpvm/test/pipeline/runscript.sh b/hpvm/test/pipeline/runscript.sh
index c95af8f831..5a2933e788 100755
--- a/hpvm/test/pipeline/runscript.sh
+++ b/hpvm/test/pipeline/runscript.sh
@@ -2,21 +2,21 @@
echo Pipeline Script
# Compile all version
-make VERSION=hpvmGPU clean
-make VERSION=hpvmVector TARGET=x86 clean
-make VERSION=hpvmScalar TARGET=seq clean
+make VERSION=viscGPU clean
+make VERSION=viscVector TARGET=x86 clean
+make VERSION=viscScalar TARGET=seq clean
-make VERSION=hpvmGPU
-make VERSION=hpvmVector TARGET=x86
-make VERSION=hpvmScalar TARGET=seq
+make VERSION=viscGPU
+make VERSION=viscVector TARGET=x86
+make VERSION=viscScalar TARGET=seq
#Run all version
-make VERSION=hpvmGPU run &
+make VERSION=viscGPU run &
ID_GPU=$!
-make VERSION=hpvmVector TARGET=x86 run &
+make VERSION=viscVector TARGET=x86 run &
ID_Vector=$!
-make VERSION=hpvmScalar TARGET=seq run
+make VERSION=viscScalar TARGET=seq run
ID_Scalar=$!
#echo Wait 60 seconds
diff --git a/hpvm/test/pipeline/src/Makefile b/hpvm/test/pipeline/src/Makefile
index 55acb2e098..ec39b86f1c 100644
--- a/hpvm/test/pipeline/src/Makefile
+++ b/hpvm/test/pipeline/src/Makefile
@@ -1,8 +1,8 @@
# (c) 2010 The Board of Trustees of the University of Illinois.
-LANGUAGE=hpvm
+LANGUAGE=visc
SRCDIR_OBJS=io.ll #compute_gold.o
-HPVM_OBJS=main.hpvm.ll
+VISC_OBJS=main.visc.ll
APP_CUDALDFLAGS=-lm -lstdc++
APP_CFLAGS+=-ffast-math -O3 -I/opt/opencv/include
APP_CXXFLAGS+=-ffast-math -O3 -I/opt/opencv/include
diff --git a/hpvm/test/pipeline/src/main.cc b/hpvm/test/pipeline/src/main.cc
index ef9d8412c7..9314833d25 100644
--- a/hpvm/test/pipeline/src/main.cc
+++ b/hpvm/test/pipeline/src/main.cc
@@ -13,7 +13,6 @@
#include "opencv2/ocl/ocl.hpp"
#include "opencv2/opencv.hpp"
#include <cassert>
-#include <hpvm.h>
#include <iostream>
#include <malloc.h>
#include <math.h>
@@ -21,6 +20,7 @@
#include <stdlib.h>
#include <string.h>
#include <sys/time.h>
+#include <visc.h>
#define NUM_RUNS 100
#define DEPTH 3
@@ -147,12 +147,12 @@ void packData(struct InStruct *args, float *I, size_t bytesI, float *Is,
void gaussianSmoothing(float *I, size_t bytesI, float *Gs, size_t bytesGs,
float *Is, size_t bytesIs, long m, long n) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(2, I, Gs, 1, Is);
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(2, I, Gs, 1, Is);
- void *thisNode = __hpvm__getNode();
- long gx = __hpvm__getNodeInstanceID_x(thisNode);
- long gy = __hpvm__getNodeInstanceID_y(thisNode);
+ void *thisNode = __visc__getNode();
+ long gx = __visc__getNodeInstanceID_x(thisNode);
+ long gy = __visc__getNodeInstanceID_y(thisNode);
int gloc = gx + gy * n;
@@ -187,26 +187,26 @@ void gaussianSmoothing(float *I, size_t bytesI, float *Gs, size_t bytesGs,
Is[gloc] = smoothedVal;
}
- __hpvm__return(2, bytesIs, bytesIs);
+ __visc__return(2, bytesIs, bytesIs);
}
void WrapperGaussianSmoothing(float *I, size_t bytesI, float *Gs,
size_t bytesGs, float *Is, size_t bytesIs, long m,
long n) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(2, I, Gs, 1, Is);
- void *GSNode = __hpvm__createNodeND(2, gaussianSmoothing, m, n);
- __hpvm__bindIn(GSNode, 0, 0, 0); // Bind I
- __hpvm__bindIn(GSNode, 1, 1, 0); // Bind bytesI
- __hpvm__bindIn(GSNode, 2, 2, 0); // Bind Gs
- __hpvm__bindIn(GSNode, 3, 3, 0); // Bind bytesGs
- __hpvm__bindIn(GSNode, 4, 4, 0); // Bind Is
- __hpvm__bindIn(GSNode, 5, 5, 0); // Bind bytesIs
- __hpvm__bindIn(GSNode, 6, 6, 0); // Bind m
- __hpvm__bindIn(GSNode, 7, 7, 0); // Bind n
-
- __hpvm__bindOut(GSNode, 0, 0, 0); // bind output bytesIs
- __hpvm__bindOut(GSNode, 1, 1, 0); // bind output bytesIs
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(2, I, Gs, 1, Is);
+ void *GSNode = __visc__createNodeND(2, gaussianSmoothing, m, n);
+ __visc__bindIn(GSNode, 0, 0, 0); // Bind I
+ __visc__bindIn(GSNode, 1, 1, 0); // Bind bytesI
+ __visc__bindIn(GSNode, 2, 2, 0); // Bind Gs
+ __visc__bindIn(GSNode, 3, 3, 0); // Bind bytesGs
+ __visc__bindIn(GSNode, 4, 4, 0); // Bind Is
+ __visc__bindIn(GSNode, 5, 5, 0); // Bind bytesIs
+ __visc__bindIn(GSNode, 6, 6, 0); // Bind m
+ __visc__bindIn(GSNode, 7, 7, 0); // Bind n
+
+ __visc__bindOut(GSNode, 0, 0, 0); // bind output bytesIs
+ __visc__bindOut(GSNode, 1, 1, 0); // bind output bytesIs
}
/* Compute a non-linear laplacian estimate of input image I of size m x n */
@@ -220,14 +220,14 @@ void WrapperGaussianSmoothing(float *I, size_t bytesI, float *Gs,
void laplacianEstimate(float *Is, size_t bytesIs, float *B, size_t bytesB,
float *L, size_t bytesL, long m, long n) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(2, Is, B, 1, L);
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(2, Is, B, 1, L);
// 3x3 image area
float imageArea[SZB * SZB];
- void *thisNode = __hpvm__getNode();
- long gx = __hpvm__getNodeInstanceID_x(thisNode);
- long gy = __hpvm__getNodeInstanceID_y(thisNode);
+ void *thisNode = __visc__getNode();
+ long gx = __visc__getNodeInstanceID_x(thisNode);
+ long gy = __visc__getNodeInstanceID_y(thisNode);
int i, j;
if ((gx < n) && (gy < m)) {
@@ -300,25 +300,25 @@ void laplacianEstimate(float *Is, size_t bytesIs, float *B, size_t bytesB,
float laplacian = dilatedPixel + erodedPixel - 2 * imageArea[1 * SZB + 1];
L[gy * n + gx] = laplacian;
}
- __hpvm__return(1, bytesL);
+ __visc__return(1, bytesL);
}
void WrapperlaplacianEstimate(float *Is, size_t bytesIs, float *B,
size_t bytesB, float *L, size_t bytesL, long m,
long n) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(2, Is, B, 1, L);
- void *LNode = __hpvm__createNodeND(2, laplacianEstimate, m, n);
- __hpvm__bindIn(LNode, 0, 0, 0); // Bind Is
- __hpvm__bindIn(LNode, 1, 1, 0); // Bind bytesIs
- __hpvm__bindIn(LNode, 2, 2, 0); // Bind B
- __hpvm__bindIn(LNode, 3, 3, 0); // Bind bytesB
- __hpvm__bindIn(LNode, 4, 4, 0); // Bind L
- __hpvm__bindIn(LNode, 5, 5, 0); // Bind bytesL
- __hpvm__bindIn(LNode, 6, 6, 0); // Bind m
- __hpvm__bindIn(LNode, 7, 7, 0); // Bind n
-
- __hpvm__bindOut(LNode, 0, 0, 0); // bind output bytesL
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(2, Is, B, 1, L);
+ void *LNode = __visc__createNodeND(2, laplacianEstimate, m, n);
+ __visc__bindIn(LNode, 0, 0, 0); // Bind Is
+ __visc__bindIn(LNode, 1, 1, 0); // Bind bytesIs
+ __visc__bindIn(LNode, 2, 2, 0); // Bind B
+ __visc__bindIn(LNode, 3, 3, 0); // Bind bytesB
+ __visc__bindIn(LNode, 4, 4, 0); // Bind L
+ __visc__bindIn(LNode, 5, 5, 0); // Bind bytesL
+ __visc__bindIn(LNode, 6, 6, 0); // Bind m
+ __visc__bindIn(LNode, 7, 7, 0); // Bind n
+
+ __visc__bindOut(LNode, 0, 0, 0); // bind output bytesL
}
/* Compute the zero crossings of input image L of size m x n */
@@ -331,16 +331,16 @@ void WrapperlaplacianEstimate(float *Is, size_t bytesIs, float *B,
*/
void computeZeroCrossings(float *L, size_t bytesL, float *B, size_t bytesB,
float *S, size_t bytesS, long m, long n) {
- __hpvm__hint(hpvm::DEVICE);
- //__hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(2, L, B, 1, S);
+ __visc__hint(visc::DEVICE);
+ //__visc__hint(visc::CPU_TARGET);
+ __visc__attributes(2, L, B, 1, S);
// 3x3 image area
float imageArea[SZB][SZB];
- void *thisNode = __hpvm__getNode();
- long gx = __hpvm__getNodeInstanceID_x(thisNode);
- long gy = __hpvm__getNodeInstanceID_y(thisNode);
+ void *thisNode = __visc__getNode();
+ long gx = __visc__getNodeInstanceID_x(thisNode);
+ long gy = __visc__getNodeInstanceID_y(thisNode);
int i, j;
if ((gx < n) && (gy < m)) {
@@ -416,25 +416,25 @@ void computeZeroCrossings(float *L, size_t bytesL, float *B, size_t bytesB,
float pixelSign = dilatedPixel - erodedPixel;
S[gy * n + gx] = pixelSign;
}
- __hpvm__return(1, bytesS);
+ __visc__return(1, bytesS);
}
void WrapperComputeZeroCrossings(float *L, size_t bytesL, float *B,
size_t bytesB, float *S, size_t bytesS, long m,
long n) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(2, L, B, 1, S);
- void *ZCNode = __hpvm__createNodeND(2, computeZeroCrossings, m, n);
- __hpvm__bindIn(ZCNode, 0, 0, 0); // Bind L
- __hpvm__bindIn(ZCNode, 1, 1, 0); // Bind bytesL
- __hpvm__bindIn(ZCNode, 2, 2, 0); // Bind B
- __hpvm__bindIn(ZCNode, 3, 3, 0); // Bind bytesB
- __hpvm__bindIn(ZCNode, 4, 4, 0); // Bind S
- __hpvm__bindIn(ZCNode, 5, 5, 0); // Bind bytesS
- __hpvm__bindIn(ZCNode, 6, 6, 0); // Bind m
- __hpvm__bindIn(ZCNode, 7, 7, 0); // Bind n
-
- __hpvm__bindOut(ZCNode, 0, 0, 0); // bind output bytesS
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(2, L, B, 1, S);
+ void *ZCNode = __visc__createNodeND(2, computeZeroCrossings, m, n);
+ __visc__bindIn(ZCNode, 0, 0, 0); // Bind L
+ __visc__bindIn(ZCNode, 1, 1, 0); // Bind bytesL
+ __visc__bindIn(ZCNode, 2, 2, 0); // Bind B
+ __visc__bindIn(ZCNode, 3, 3, 0); // Bind bytesB
+ __visc__bindIn(ZCNode, 4, 4, 0); // Bind S
+ __visc__bindIn(ZCNode, 5, 5, 0); // Bind bytesS
+ __visc__bindIn(ZCNode, 6, 6, 0); // Bind m
+ __visc__bindIn(ZCNode, 7, 7, 0); // Bind n
+
+ __visc__bindOut(ZCNode, 0, 0, 0); // bind output bytesS
}
/*
@@ -458,12 +458,12 @@ void computeGradient(float *Is, size_t bytesIs, float *Sx, size_t bytesSx,
float *Sy, size_t bytesSy, float *G, size_t bytesG, long m,
long n) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(3, Is, Sx, Sy, 1, G);
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(3, Is, Sx, Sy, 1, G);
- void *thisNode = __hpvm__getNode();
- long gx = __hpvm__getNodeInstanceID_x(thisNode);
- long gy = __hpvm__getNodeInstanceID_y(thisNode);
+ void *thisNode = __visc__getNode();
+ long gx = __visc__getNodeInstanceID_x(thisNode);
+ long gy = __visc__getNodeInstanceID_y(thisNode);
int gloc = gx + gy * n;
@@ -498,27 +498,27 @@ void computeGradient(float *Is, size_t bytesIs, float *Sx, size_t bytesSx,
G[gloc] = sqrt(Gx * Gx + Gy * Gy);
}
- __hpvm__return(1, bytesG);
+ __visc__return(1, bytesG);
}
void WrapperComputeGradient(float *Is, size_t bytesIs, float *Sx,
size_t bytesSx, float *Sy, size_t bytesSy, float *G,
size_t bytesG, long m, long n) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(3, Is, Sx, Sy, 1, G);
- void *CGNode = __hpvm__createNodeND(2, computeGradient, m, n);
- __hpvm__bindIn(CGNode, 0, 0, 0); // Bind Is
- __hpvm__bindIn(CGNode, 1, 1, 0); // Bind bytesIs
- __hpvm__bindIn(CGNode, 2, 2, 0); // Bind Sx
- __hpvm__bindIn(CGNode, 3, 3, 0); // Bind bytesSx
- __hpvm__bindIn(CGNode, 4, 4, 0); // Bind Sy
- __hpvm__bindIn(CGNode, 5, 5, 0); // Bind bytesSy
- __hpvm__bindIn(CGNode, 6, 6, 0); // Bind G
- __hpvm__bindIn(CGNode, 7, 7, 0); // Bind bytesG
- __hpvm__bindIn(CGNode, 8, 8, 0); // Bind m
- __hpvm__bindIn(CGNode, 9, 9, 0); // Bind n
-
- __hpvm__bindOut(CGNode, 0, 0, 0); // bind output bytesG
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(3, Is, Sx, Sy, 1, G);
+ void *CGNode = __visc__createNodeND(2, computeGradient, m, n);
+ __visc__bindIn(CGNode, 0, 0, 0); // Bind Is
+ __visc__bindIn(CGNode, 1, 1, 0); // Bind bytesIs
+ __visc__bindIn(CGNode, 2, 2, 0); // Bind Sx
+ __visc__bindIn(CGNode, 3, 3, 0); // Bind bytesSx
+ __visc__bindIn(CGNode, 4, 4, 0); // Bind Sy
+ __visc__bindIn(CGNode, 5, 5, 0); // Bind bytesSy
+ __visc__bindIn(CGNode, 6, 6, 0); // Bind G
+ __visc__bindIn(CGNode, 7, 7, 0); // Bind bytesG
+ __visc__bindIn(CGNode, 8, 8, 0); // Bind m
+ __visc__bindIn(CGNode, 9, 9, 0); // Bind n
+
+ __visc__bindOut(CGNode, 0, 0, 0); // bind output bytesG
}
/*
@@ -531,13 +531,13 @@ void WrapperComputeGradient(float *Is, size_t bytesIs, float *Sx,
void computeMaxGradientLeaf(float *G, size_t bytesG, float *maxG,
size_t bytesMaxG, long m, long n) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(1, G, 1, maxG);
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(1, G, 1, maxG);
- void *thisNode = __hpvm__getNode();
+ void *thisNode = __visc__getNode();
- long lx = __hpvm__getNodeInstanceID_x(thisNode); // threadIdx.x
- long dimx = __hpvm__getNumNodeInstances_x(thisNode); // blockDim.x
+ long lx = __visc__getNodeInstanceID_x(thisNode); // threadIdx.x
+ long dimx = __visc__getNumNodeInstances_x(thisNode); // blockDim.x
// Assume a single thread block
// Thread block iterates over all elements
@@ -556,39 +556,39 @@ void computeMaxGradientLeaf(float *G, size_t bytesG, float *maxG,
*maxG = G[lx];
}
- __hpvm__return(1, bytesMaxG);
+ __visc__return(1, bytesMaxG);
}
void computeMaxGradientTB(float *G, size_t bytesG, float *maxG,
size_t bytesMaxG, long m, long n, long block_x) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(2, G, maxG, 1, maxG);
- void *CMGLeafNode = __hpvm__createNodeND(1, computeMaxGradientLeaf, block_x);
- __hpvm__bindIn(CMGLeafNode, 0, 0, 0); // Bind G
- __hpvm__bindIn(CMGLeafNode, 1, 1, 0); // Bind bytesG
- __hpvm__bindIn(CMGLeafNode, 2, 2, 0); // Bind maxG
- __hpvm__bindIn(CMGLeafNode, 3, 3, 0); // Bind bytesMaxG
- __hpvm__bindIn(CMGLeafNode, 4, 4, 0); // Bind m
- __hpvm__bindIn(CMGLeafNode, 5, 5, 0); // Bind n
-
- __hpvm__bindOut(CMGLeafNode, 0, 0, 0); // bind output bytesMaxG
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(2, G, maxG, 1, maxG);
+ void *CMGLeafNode = __visc__createNodeND(1, computeMaxGradientLeaf, block_x);
+ __visc__bindIn(CMGLeafNode, 0, 0, 0); // Bind G
+ __visc__bindIn(CMGLeafNode, 1, 1, 0); // Bind bytesG
+ __visc__bindIn(CMGLeafNode, 2, 2, 0); // Bind maxG
+ __visc__bindIn(CMGLeafNode, 3, 3, 0); // Bind bytesMaxG
+ __visc__bindIn(CMGLeafNode, 4, 4, 0); // Bind m
+ __visc__bindIn(CMGLeafNode, 5, 5, 0); // Bind n
+
+ __visc__bindOut(CMGLeafNode, 0, 0, 0); // bind output bytesMaxG
}
void WrapperComputeMaxGradient(float *G, size_t bytesG, float *maxG,
size_t bytesMaxG, long m, long n, long block_x,
long grid_x) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(2, G, maxG, 1, maxG);
- void *CMGTBNode = __hpvm__createNodeND(1, computeMaxGradientTB, grid_x);
- __hpvm__bindIn(CMGTBNode, 0, 0, 0); // Bind G
- __hpvm__bindIn(CMGTBNode, 1, 1, 0); // Bind bytesG
- __hpvm__bindIn(CMGTBNode, 2, 2, 0); // Bind maxG
- __hpvm__bindIn(CMGTBNode, 3, 3, 0); // Bind bytesMaxG
- __hpvm__bindIn(CMGTBNode, 4, 4, 0); // Bind m
- __hpvm__bindIn(CMGTBNode, 5, 5, 0); // Bind n
- __hpvm__bindIn(CMGTBNode, 6, 6, 0); // Bind block_x
-
- __hpvm__bindOut(CMGTBNode, 0, 0, 0); // bind output bytesMaxG
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(2, G, maxG, 1, maxG);
+ void *CMGTBNode = __visc__createNodeND(1, computeMaxGradientTB, grid_x);
+ __visc__bindIn(CMGTBNode, 0, 0, 0); // Bind G
+ __visc__bindIn(CMGTBNode, 1, 1, 0); // Bind bytesG
+ __visc__bindIn(CMGTBNode, 2, 2, 0); // Bind maxG
+ __visc__bindIn(CMGTBNode, 3, 3, 0); // Bind bytesMaxG
+ __visc__bindIn(CMGTBNode, 4, 4, 0); // Bind m
+ __visc__bindIn(CMGTBNode, 5, 5, 0); // Bind n
+ __visc__bindIn(CMGTBNode, 6, 6, 0); // Bind block_x
+
+ __visc__bindOut(CMGTBNode, 0, 0, 0); // bind output bytesMaxG
}
/* Reject the zero crossings where the gradient is below a threshold */
@@ -604,39 +604,39 @@ void WrapperComputeMaxGradient(float *G, size_t bytesG, float *maxG,
void rejectZeroCrossings(float *S, size_t bytesS, float *G, size_t bytesG,
float *maxG, size_t bytesMaxG, float *E, size_t bytesE,
long m, long n) {
- __hpvm__hint(hpvm::DEVICE);
- __hpvm__attributes(3, S, G, maxG, 1, E);
+ __visc__hint(visc::DEVICE);
+ __visc__attributes(3, S, G, maxG, 1, E);
- void *thisNode = __hpvm__getNode();
- int gx = __hpvm__getNodeInstanceID_x(thisNode);
- int gy = __hpvm__getNodeInstanceID_y(thisNode);
+ void *thisNode = __visc__getNode();
+ int gx = __visc__getNodeInstanceID_x(thisNode);
+ int gy = __visc__getNodeInstanceID_y(thisNode);
float mG = *maxG;
if ((gx < n) && (gy < m)) {
E[gy * n + gx] =
((S[gy * n + gx] > 0.0) && (G[gy * n + gx] > THETA * mG)) ? 1.0 : 0.0;
}
- __hpvm__return(1, bytesE);
+ __visc__return(1, bytesE);
}
void WrapperRejectZeroCrossings(float *S, size_t bytesS, float *G,
size_t bytesG, float *maxG, size_t bytesMaxG,
float *E, size_t bytesE, long m, long n) {
- __hpvm__hint(hpvm::CPU_TARGET);
- __hpvm__attributes(3, S, G, maxG, 1, E);
- void *RZCNode = __hpvm__createNodeND(2, rejectZeroCrossings, m, n);
- __hpvm__bindIn(RZCNode, 0, 0, 0); // Bind S
- __hpvm__bindIn(RZCNode, 1, 1, 0); // Bind bytesS
- __hpvm__bindIn(RZCNode, 2, 2, 0); // Bind G
- __hpvm__bindIn(RZCNode, 3, 3, 0); // Bind bytesG
- __hpvm__bindIn(RZCNode, 4, 4, 0); // Bind maxG
- __hpvm__bindIn(RZCNode, 5, 5, 0); // Bind bytesMaxG
- __hpvm__bindIn(RZCNode, 6, 6, 0); // Bind E
- __hpvm__bindIn(RZCNode, 7, 7, 0); // Bind bytesE
- __hpvm__bindIn(RZCNode, 8, 8, 0); // Bind m
- __hpvm__bindIn(RZCNode, 9, 9, 0); // Bind n
-
- __hpvm__bindOut(RZCNode, 0, 0, 0); // bind output bytesE
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(3, S, G, maxG, 1, E);
+ void *RZCNode = __visc__createNodeND(2, rejectZeroCrossings, m, n);
+ __visc__bindIn(RZCNode, 0, 0, 0); // Bind S
+ __visc__bindIn(RZCNode, 1, 1, 0); // Bind bytesS
+ __visc__bindIn(RZCNode, 2, 2, 0); // Bind G
+ __visc__bindIn(RZCNode, 3, 3, 0); // Bind bytesG
+ __visc__bindIn(RZCNode, 4, 4, 0); // Bind maxG
+ __visc__bindIn(RZCNode, 5, 5, 0); // Bind bytesMaxG
+ __visc__bindIn(RZCNode, 6, 6, 0); // Bind E
+ __visc__bindIn(RZCNode, 7, 7, 0); // Bind bytesE
+ __visc__bindIn(RZCNode, 8, 8, 0); // Bind m
+ __visc__bindIn(RZCNode, 9, 9, 0); // Bind n
+
+ __visc__bindOut(RZCNode, 0, 0, 0); // bind output bytesE
}
// Pipelined Root node
@@ -656,80 +656,80 @@ void edgeDetection(float *I, size_t bytesI, // 0
long block_x, // 24
long grid_x // 25
) {
- __hpvm__attributes(5, I, Gs, B, Sx, Sy, 6, Is, L, S, G, maxG, E);
- __hpvm__hint(hpvm::CPU_TARGET);
- void *GSNode = __hpvm__createNodeND(0, WrapperGaussianSmoothing);
- void *LNode = __hpvm__createNodeND(0, WrapperlaplacianEstimate);
- void *CZCNode = __hpvm__createNodeND(0, WrapperComputeZeroCrossings);
- void *CGNode = __hpvm__createNodeND(0, WrapperComputeGradient);
- void *CMGNode = __hpvm__createNodeND(0, WrapperComputeMaxGradient);
- void *RZCNode = __hpvm__createNodeND(0, WrapperRejectZeroCrossings);
+ __visc__attributes(5, I, Gs, B, Sx, Sy, 6, Is, L, S, G, maxG, E);
+ __visc__hint(visc::CPU_TARGET);
+ void *GSNode = __visc__createNodeND(0, WrapperGaussianSmoothing);
+ void *LNode = __visc__createNodeND(0, WrapperlaplacianEstimate);
+ void *CZCNode = __visc__createNodeND(0, WrapperComputeZeroCrossings);
+ void *CGNode = __visc__createNodeND(0, WrapperComputeGradient);
+ void *CMGNode = __visc__createNodeND(0, WrapperComputeMaxGradient);
+ void *RZCNode = __visc__createNodeND(0, WrapperRejectZeroCrossings);
// Gaussian Inputs
- __hpvm__bindIn(GSNode, 0, 0, 1); // Bind I
- __hpvm__bindIn(GSNode, 1, 1, 1); // Bind bytesI
- __hpvm__bindIn(GSNode, 14, 2, 1); // Bind Gs
- __hpvm__bindIn(GSNode, 15, 3, 1); // Bind bytesGs
- __hpvm__bindIn(GSNode, 2, 4, 1); // Bind Is
- __hpvm__bindIn(GSNode, 3, 5, 1); // Bind bytesIs
- __hpvm__bindIn(GSNode, 22, 6, 1); // Bind m
- __hpvm__bindIn(GSNode, 23, 7, 1); // Bind n
+ __visc__bindIn(GSNode, 0, 0, 1); // Bind I
+ __visc__bindIn(GSNode, 1, 1, 1); // Bind bytesI
+ __visc__bindIn(GSNode, 14, 2, 1); // Bind Gs
+ __visc__bindIn(GSNode, 15, 3, 1); // Bind bytesGs
+ __visc__bindIn(GSNode, 2, 4, 1); // Bind Is
+ __visc__bindIn(GSNode, 3, 5, 1); // Bind bytesIs
+ __visc__bindIn(GSNode, 22, 6, 1); // Bind m
+ __visc__bindIn(GSNode, 23, 7, 1); // Bind n
// Laplacian Inputs
- __hpvm__bindIn(LNode, 2, 0, 1); // Bind Is
- __hpvm__edge(GSNode, LNode, 1, 0, 1, 1); // Get bytesIs
- __hpvm__bindIn(LNode, 16, 2, 1); // Bind B
- __hpvm__bindIn(LNode, 17, 3, 1); // Bind bytesB
- __hpvm__bindIn(LNode, 4, 4, 1); // Bind L
- __hpvm__bindIn(LNode, 5, 5, 1); // Bind bytesL
- __hpvm__bindIn(LNode, 22, 6, 1); // Bind m
- __hpvm__bindIn(LNode, 23, 7, 1); // Bind n
+ __visc__bindIn(LNode, 2, 0, 1); // Bind Is
+ __visc__edge(GSNode, LNode, 1, 0, 1, 1); // Get bytesIs
+ __visc__bindIn(LNode, 16, 2, 1); // Bind B
+ __visc__bindIn(LNode, 17, 3, 1); // Bind bytesB
+ __visc__bindIn(LNode, 4, 4, 1); // Bind L
+ __visc__bindIn(LNode, 5, 5, 1); // Bind bytesL
+ __visc__bindIn(LNode, 22, 6, 1); // Bind m
+ __visc__bindIn(LNode, 23, 7, 1); // Bind n
// Compute ZC Inputs
- __hpvm__bindIn(CZCNode, 4, 0, 1); // Bind L
- __hpvm__edge(LNode, CZCNode, 1, 0, 1, 1); // Get bytesL
- __hpvm__bindIn(CZCNode, 16, 2, 1); // Bind B
- __hpvm__bindIn(CZCNode, 17, 3, 1); // Bind bytesB
- __hpvm__bindIn(CZCNode, 6, 4, 1); // Bind S
- __hpvm__bindIn(CZCNode, 7, 5, 1); // Bind bytesS
- __hpvm__bindIn(CZCNode, 22, 6, 1); // Bind m
- __hpvm__bindIn(CZCNode, 23, 7, 1); // Bind n
+ __visc__bindIn(CZCNode, 4, 0, 1); // Bind L
+ __visc__edge(LNode, CZCNode, 1, 0, 1, 1); // Get bytesL
+ __visc__bindIn(CZCNode, 16, 2, 1); // Bind B
+ __visc__bindIn(CZCNode, 17, 3, 1); // Bind bytesB
+ __visc__bindIn(CZCNode, 6, 4, 1); // Bind S
+ __visc__bindIn(CZCNode, 7, 5, 1); // Bind bytesS
+ __visc__bindIn(CZCNode, 22, 6, 1); // Bind m
+ __visc__bindIn(CZCNode, 23, 7, 1); // Bind n
// Gradient Inputs
- __hpvm__bindIn(CGNode, 2, 0, 1); // Bind Is
- __hpvm__edge(GSNode, CGNode, 1, 1, 1, 1); // Get bytesIs
- __hpvm__bindIn(CGNode, 18, 2, 1); // Bind Sx
- __hpvm__bindIn(CGNode, 19, 3, 1); // Bind bytesSx
- __hpvm__bindIn(CGNode, 20, 4, 1); // Bind Sy
- __hpvm__bindIn(CGNode, 21, 5, 1); // Bind bytesSy
- __hpvm__bindIn(CGNode, 8, 6, 1); // Bind G
- __hpvm__bindIn(CGNode, 9, 7, 1); // Bind bytesG
- __hpvm__bindIn(CGNode, 22, 8, 1); // Bind m
- __hpvm__bindIn(CGNode, 23, 9, 1); // Bind n
+ __visc__bindIn(CGNode, 2, 0, 1); // Bind Is
+ __visc__edge(GSNode, CGNode, 1, 1, 1, 1); // Get bytesIs
+ __visc__bindIn(CGNode, 18, 2, 1); // Bind Sx
+ __visc__bindIn(CGNode, 19, 3, 1); // Bind bytesSx
+ __visc__bindIn(CGNode, 20, 4, 1); // Bind Sy
+ __visc__bindIn(CGNode, 21, 5, 1); // Bind bytesSy
+ __visc__bindIn(CGNode, 8, 6, 1); // Bind G
+ __visc__bindIn(CGNode, 9, 7, 1); // Bind bytesG
+ __visc__bindIn(CGNode, 22, 8, 1); // Bind m
+ __visc__bindIn(CGNode, 23, 9, 1); // Bind n
// Max Gradient Inputs
- __hpvm__bindIn(CMGNode, 8, 0, 1); // Bind G
- __hpvm__edge(CGNode, CMGNode, 1, 0, 1, 1); // Get bytesG
- __hpvm__bindIn(CMGNode, 10, 2, 1); // Bind maxG
- __hpvm__bindIn(CMGNode, 11, 3, 1); // Bind bytesMaxG
- __hpvm__bindIn(CMGNode, 22, 4, 1); // Bind m
- __hpvm__bindIn(CMGNode, 23, 5, 1); // Bind n
- __hpvm__bindIn(CMGNode, 24, 6, 1); // Bind block_x
- __hpvm__bindIn(CMGNode, 25, 7, 1); // Bind grid_x
+ __visc__bindIn(CMGNode, 8, 0, 1); // Bind G
+ __visc__edge(CGNode, CMGNode, 1, 0, 1, 1); // Get bytesG
+ __visc__bindIn(CMGNode, 10, 2, 1); // Bind maxG
+ __visc__bindIn(CMGNode, 11, 3, 1); // Bind bytesMaxG
+ __visc__bindIn(CMGNode, 22, 4, 1); // Bind m
+ __visc__bindIn(CMGNode, 23, 5, 1); // Bind n
+ __visc__bindIn(CMGNode, 24, 6, 1); // Bind block_x
+ __visc__bindIn(CMGNode, 25, 7, 1); // Bind grid_x
// Reject ZC Inputs
- __hpvm__bindIn(RZCNode, 6, 0, 1); // Bind S
- __hpvm__edge(CZCNode, RZCNode, 1, 0, 1, 1); // Get bytesS
- __hpvm__bindIn(RZCNode, 8, 2, 1); // Bind G
- __hpvm__bindIn(RZCNode, 9, 3, 1); // Bind bytesG
- __hpvm__bindIn(RZCNode, 10, 4, 1); // Bind maxG
- __hpvm__edge(CMGNode, RZCNode, 1, 0, 5, 1); // Get bytesMaxG
- __hpvm__bindIn(RZCNode, 12, 6, 1); // Bind E
- __hpvm__bindIn(RZCNode, 13, 7, 1); // Bind bytesE
- __hpvm__bindIn(RZCNode, 22, 8, 1); // Bind m
- __hpvm__bindIn(RZCNode, 23, 9, 1); // Bind n
-
- __hpvm__bindOut(RZCNode, 0, 0, 1); // Bind output
+ __visc__bindIn(RZCNode, 6, 0, 1); // Bind S
+ __visc__edge(CZCNode, RZCNode, 1, 0, 1, 1); // Get bytesS
+ __visc__bindIn(RZCNode, 8, 2, 1); // Bind G
+ __visc__bindIn(RZCNode, 9, 3, 1); // Bind bytesG
+ __visc__bindIn(RZCNode, 10, 4, 1); // Bind maxG
+ __visc__edge(CMGNode, RZCNode, 1, 0, 5, 1); // Get bytesMaxG
+ __visc__bindIn(RZCNode, 12, 6, 1); // Bind E
+ __visc__bindIn(RZCNode, 13, 7, 1); // Bind bytesE
+ __visc__bindIn(RZCNode, 22, 8, 1); // Bind m
+ __visc__bindIn(RZCNode, 23, 9, 1); // Bind n
+
+ __visc__bindOut(RZCNode, 0, 0, 1); // Bind output
}
}
@@ -796,7 +796,7 @@ int main(int argc, char *argv[]) {
assert(src.isContinuous() && Is.isContinuous() && L.isContinuous() &&
S.isContinuous() && G.isContinuous() && E.isContinuous());
- __hpvm__init();
+ __visc__init();
// copy A to device memory
I_sz = src.size[0] * src.size[1] * sizeof(float);
@@ -843,7 +843,7 @@ int main(int argc, char *argv[]) {
for (unsigned j = 0; j < NUM_RUNS; j++) {
std::cout << "Run: " << j << "\n";
- void *DFG = __hpvm__launch(1, edgeDetection, (void *)args);
+ void *DFG = __visc__launch(1, edgeDetection, (void *)args);
cap = VideoCapture(inFile);
getNextFrame(cap, src);
@@ -855,25 +855,25 @@ int main(int argc, char *argv[]) {
*maxG = 0.0;
- llvm_hpvm_track_mem(src.data, I_sz);
- llvm_hpvm_track_mem(Is.data, I_sz);
- llvm_hpvm_track_mem(L.data, I_sz);
- llvm_hpvm_track_mem(S.data, I_sz);
- llvm_hpvm_track_mem(G.data, I_sz);
- llvm_hpvm_track_mem(maxG, bytesMaxG);
- llvm_hpvm_track_mem(E.data, I_sz);
- llvm_hpvm_track_mem(Gs, bytesGs);
- llvm_hpvm_track_mem(B, bytesB);
- llvm_hpvm_track_mem(Sx, bytesSx);
- llvm_hpvm_track_mem(Sy, bytesSy);
-
- __hpvm__push(DFG, args);
- void *ret = __hpvm__pop(DFG);
+ llvm_visc_track_mem(src.data, I_sz);
+ llvm_visc_track_mem(Is.data, I_sz);
+ llvm_visc_track_mem(L.data, I_sz);
+ llvm_visc_track_mem(S.data, I_sz);
+ llvm_visc_track_mem(G.data, I_sz);
+ llvm_visc_track_mem(maxG, bytesMaxG);
+ llvm_visc_track_mem(E.data, I_sz);
+ llvm_visc_track_mem(Gs, bytesGs);
+ llvm_visc_track_mem(B, bytesB);
+ llvm_visc_track_mem(Sx, bytesSx);
+ llvm_visc_track_mem(Sy, bytesSy);
+
+ __visc__push(DFG, args);
+ void *ret = __visc__pop(DFG);
std::cout << "Returned size: " << *(size_t *)ret << " expected " << I_sz
<< '\n';
- llvm_hpvm_request_mem(maxG, bytesMaxG);
- llvm_hpvm_request_mem(E.data, I_sz);
+ llvm_visc_request_mem(maxG, bytesMaxG);
+ llvm_visc_request_mem(E.data, I_sz);
Mat in, out;
resize(src, in, Size(HEIGHT, WIDTH));
@@ -882,26 +882,26 @@ int main(int argc, char *argv[]) {
imshow(input_window, in);
waitKey(1);
- llvm_hpvm_untrack_mem(src.data);
- llvm_hpvm_untrack_mem(Is.data);
- llvm_hpvm_untrack_mem(L.data);
- llvm_hpvm_untrack_mem(S.data);
- llvm_hpvm_untrack_mem(G.data);
- llvm_hpvm_untrack_mem(maxG);
- llvm_hpvm_untrack_mem(E.data);
- llvm_hpvm_untrack_mem(Gs);
- llvm_hpvm_untrack_mem(B);
- llvm_hpvm_untrack_mem(Sx);
- llvm_hpvm_untrack_mem(Sy);
+ llvm_visc_untrack_mem(src.data);
+ llvm_visc_untrack_mem(Is.data);
+ llvm_visc_untrack_mem(L.data);
+ llvm_visc_untrack_mem(S.data);
+ llvm_visc_untrack_mem(G.data);
+ llvm_visc_untrack_mem(maxG);
+ llvm_visc_untrack_mem(E.data);
+ llvm_visc_untrack_mem(Gs);
+ llvm_visc_untrack_mem(B);
+ llvm_visc_untrack_mem(Sx);
+ llvm_visc_untrack_mem(Sy);
getNextFrame(cap, src);
}
} else {
- __hpvm__push(DFG, args);
- __hpvm__pop(DFG);
+ __visc__push(DFG, args);
+ __visc__pop(DFG);
}
- __hpvm__wait(DFG);
+ __visc__wait(DFG);
}
- __hpvm__cleanup();
+ __visc__cleanup();
return 0;
}
diff --git a/hpvm/test/unitTests/CreateNodeAndEdge.c b/hpvm/test/unitTests/CreateNodeAndEdge.c
index c3f58c95d6..1b6b1cff21 100644
--- a/hpvm/test/unitTests/CreateNodeAndEdge.c
+++ b/hpvm/test/unitTests/CreateNodeAndEdge.c
@@ -1,4 +1,4 @@
-#include "hpvm.h"
+#include "visc.h"
#include <stdio.h>
struct Root {
@@ -7,33 +7,33 @@ struct Root {
};
void Func1(int *In, int *Out) {
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(1, In, 1, Out);
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(1, In, 1, Out);
- __hpvm__return(1, Out);
+ __visc__return(1, Out);
}
void Func2(int *BindIn, int *SrcIn, int *Out) {
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(2, BindIn, SrcIn, 1, Out);
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(2, BindIn, SrcIn, 1, Out);
- __hpvm__return(1, Out);
+ __visc__return(1, Out);
}
void PipeRoot(int *In, int *Out) {
- __hpvm__hint(CPU_TARGET);
+ __visc__hint(CPU_TARGET);
- __hpvm__attributes(1, In, 1, Out);
+ __visc__attributes(1, In, 1, Out);
- void *SrcNode = __hpvm__createNodeND(0, Func1);
- void *DestNode = __hpvm__createNodeND(0, Func2);
+ void *SrcNode = __visc__createNodeND(0, Func1);
+ void *DestNode = __visc__createNodeND(0, Func2);
- __hpvm__bindIn(SrcNode, 0, 0, 0);
+ __visc__bindIn(SrcNode, 0, 0, 0);
- __hpvm__bindIn(DestNode, 0, 0, 0);
- __hpvm__edge(SrcNode, DestNode, 1, 0, 1, 0);
+ __visc__bindIn(DestNode, 0, 0, 0);
+ __visc__edge(SrcNode, DestNode, 1, 0, 1, 0);
- __hpvm__bindOut(SrcNode, 0, 0, 0);
+ __visc__bindOut(SrcNode, 0, 0, 0);
}
int main(void) {
@@ -41,10 +41,10 @@ int main(void) {
int Out = 0;
struct Root RootArgs = {(int *)&In, (int *)&Out};
- __hpvm__init();
- void *PipeDFG = __hpvm__launch(0, PipeRoot, (void *)&RootArgs);
- __hpvm__wait(PipeDFG);
- __hpvm__cleanup();
+ __visc__init();
+ void *PipeDFG = __visc__launch(0, PipeRoot, (void *)&RootArgs);
+ __visc__wait(PipeDFG);
+ __visc__cleanup();
return 0;
}
diff --git a/hpvm/test/unitTests/Makefile b/hpvm/test/unitTests/Makefile
index 15580e9300..539ee5e8fb 100644
--- a/hpvm/test/unitTests/Makefile
+++ b/hpvm/test/unitTests/Makefile
@@ -2,8 +2,8 @@ PASSES :=
.PHONY: clean
-LLVM_INSTALL:=/home/psrivas2/Hetero/HPVM/Code/trunk/llvm-install
-LIBCLC:=/home/psrivas2/Hetero/HPVM/Code/trunk/libclc
+LLVM_INSTALL:=/home/psrivas2/Hetero/VISC/Code/trunk/llvm-install
+LIBCLC:=/home/psrivas2/Hetero/VISC/Code/trunk/libclc
HOST:=gemm_opencl
KERNELS:=matrixMul
LLVM_CC:=$(LLVM_INSTALL)/bin/clang
diff --git a/hpvm/test/unitTests/MallocIntrinsic.c b/hpvm/test/unitTests/MallocIntrinsic.c
index 173f6b3b16..cfd041a991 100644
--- a/hpvm/test/unitTests/MallocIntrinsic.c
+++ b/hpvm/test/unitTests/MallocIntrinsic.c
@@ -1,4 +1,4 @@
-#include "hpvm.h"
+#include "visc.h"
#include <stdlib.h>
struct Root {
@@ -7,12 +7,12 @@ struct Root {
};
void PipeRoot(int *In, int *Out) {
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(1, In, 1, Out);
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(1, In, 1, Out);
- Out = (int *)__hpvm__malloc(*In);
+ Out = (int *)__visc__malloc(*In);
- __hpvm__return(1, Out);
+ __visc__return(1, Out);
}
int main(void) {
@@ -26,12 +26,12 @@ int main(void) {
RootArgs->input = (int *)&In;
RootArgs->output = (int *)&Out;
- __hpvm__init();
+ __visc__init();
- void *PipeDFG = __hpvm__launch(0, PipeRoot, (void *)RootArgs);
- __hpvm__wait(PipeDFG);
+ void *PipeDFG = __visc__launch(0, PipeRoot, (void *)RootArgs);
+ __visc__wait(PipeDFG);
- __hpvm__cleanup();
+ __visc__cleanup();
return 0;
}
diff --git a/hpvm/test/unitTests/PipelineIntrinsics.c b/hpvm/test/unitTests/PipelineIntrinsics.c
index 43ba0ef56c..2a9bf83402 100644
--- a/hpvm/test/unitTests/PipelineIntrinsics.c
+++ b/hpvm/test/unitTests/PipelineIntrinsics.c
@@ -1,4 +1,4 @@
-#include "hpvm.h"
+#include "visc.h"
#include <stdlib.h>
struct Root {
@@ -7,9 +7,9 @@ struct Root {
};
void PipeRoot(int *In, int *Out) {
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(1, In, 1, Out);
- __hpvm__return(1, Out);
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(1, In, 1, Out);
+ __visc__return(1, Out);
}
int main(void) {
@@ -23,12 +23,12 @@ int main(void) {
RootArgs->input = (int *)&In;
RootArgs->output = (int *)&Out;
- __hpvm__init();
+ __visc__init();
- void *PipeDFG = __hpvm__launch(0, PipeRoot, (void *)RootArgs);
- __hpvm__wait(PipeDFG);
+ void *PipeDFG = __visc__launch(0, PipeRoot, (void *)RootArgs);
+ __visc__wait(PipeDFG);
- __hpvm__cleanup();
+ __visc__cleanup();
return 0;
}
diff --git a/hpvm/test/unitTests/PipelineIntrinsics.malloc.c b/hpvm/test/unitTests/PipelineIntrinsics.malloc.c
index c2deed9867..36fc02d22b 100644
--- a/hpvm/test/unitTests/PipelineIntrinsics.malloc.c
+++ b/hpvm/test/unitTests/PipelineIntrinsics.malloc.c
@@ -1,4 +1,4 @@
-#include "hpvm.h"
+#include "visc.h"
#include <stdlib.h>
struct Root {
@@ -7,24 +7,24 @@ struct Root {
};
void PipeRoot(int *In, int *Out) {
- __hpvm__hint(CPU_TARGET);
- __hpvm__attributes(1, In, 1, Out);
- __hpvm__return(1, Out);
+ __visc__hint(CPU_TARGET);
+ __visc__attributes(1, In, 1, Out);
+ __visc__return(1, Out);
}
int main(void) {
int In, Out;
- __hpvm__init();
+ __visc__init();
struct Root *RootArgs = (struct Root *)malloc(sizeof(struct Root));
RootArgs->input = (int *)&In;
RootArgs->output = (int *)&Out;
- void *PipeDFG = __hpvm__launch(0, PipeRoot, (void *)RootArgs);
- __hpvm__wait(PipeDFG);
+ void *PipeDFG = __visc__launch(0, PipeRoot, (void *)RootArgs);
+ __visc__wait(PipeDFG);
- __hpvm__cleanup();
+ __visc__cleanup();
return 0;
}
diff --git a/hpvm/test/unitTests/temp/3level.ll b/hpvm/test/unitTests/temp/3level.ll
index 2e3753f140..168e7b4232 100644
--- a/hpvm/test/unitTests/temp/3level.ll
+++ b/hpvm/test/unitTests/temp/3level.ll
@@ -1,5 +1,5 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG -o %t.ll -S < %s
-; RUN: llvm-link %t.ll ~/current-src/projects/hpvm-rt/hpvm-rt.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll ~/current-src/projects/visc-rt/visc-rt.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -lrt -o %t.bin
; RUN: %t.bin 5
; ModuleID = '/home/psrivas2/current-test/unitTests/3level.ll'
@@ -13,31 +13,31 @@ target triple = "x86_64-unknown-linux-gnu"
@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #0
+declare i8* @llvm.visc.createNode(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32) #0
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*) #0
+declare i8* @llvm.visc.launch(i8*, i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #0
+declare void @llvm.visc.wait(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.getNode() #0
+declare i8* @llvm.visc.getNode() #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32)
+declare void @llvm.visc.bind.input(i8*, i32, i32)
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32)
+declare void @llvm.visc.bind.output(i8*, i32, i32)
; Function Attrs: nounwind uwtable
define i32 @main(i32 %argc, i8** nocapture %argv) #1 {
@@ -47,18 +47,18 @@ entry:
%0 = load i8** %arrayidx, align 8, !tbaa !0
%call.i = tail call i64 @strtol(i8* nocapture %0, i8** null, i32 10) #0
%conv.i = trunc i64 %call.i to i32
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%1 = bitcast %struct.arg* %in.addr to i32*
store i32 %conv.i, i32* %1
%args = bitcast %struct.arg* %in.addr to i8*
- %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
+ %graphID = call i8* @llvm.visc.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
%call1 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %conv.i) #0
- call void @llvm.hpvm.wait(i8* %graphID)
+ call void @llvm.visc.wait(i8* %graphID)
%2 = getelementptr %struct.arg* %in.addr, i32 0, i32 1
%outputstruct = load %rtype* %2
%output1 = extractvalue %rtype %outputstruct, 0
%output2 = extractvalue %rtype %outputstruct, 1
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.cleanup()
%call2 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %output1) #0
%call3 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %output2) #0
ret i32 0
@@ -83,21 +83,21 @@ define %rtype_internal @foo(i32 %id) {
}
define %rtype_internal @subNode(i32 %id) {
- %foo_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype_internal (i32)* @foo to i8*))
- call void @llvm.hpvm.bind.input(i8* %foo_node, i32 0, i32 0)
- call void @llvm.hpvm.bind.output(i8* %foo_node, i32 0, i32 0)
+ %foo_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype_internal (i32)* @foo to i8*))
+ call void @llvm.visc.bind.input(i8* %foo_node, i32 0, i32 0)
+ call void @llvm.visc.bind.output(i8* %foo_node, i32 0, i32 0)
ret %rtype_internal zeroinitializer
}
define %rtype @Root(i32 %id) {
- %p_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype_internal (i32)* @producer to i8*))
- %c_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype_internal (i32)* @consumer to i8*))
- %sub_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype_internal (i32)* @subNode to i8*))
- %edge = call i8* @llvm.hpvm.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
- call void @llvm.hpvm.bind.input(i8* %p_node, i32 0, i32 0)
- call void @llvm.hpvm.bind.output(i8* %c_node, i32 0, i32 0)
- call void @llvm.hpvm.bind.input(i8* %sub_node, i32 0, i32 0)
- call void @llvm.hpvm.bind.output(i8* %sub_node, i32 0, i32 1)
+ %p_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype_internal (i32)* @producer to i8*))
+ %c_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype_internal (i32)* @consumer to i8*))
+ %sub_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype_internal (i32)* @subNode to i8*))
+ %edge = call i8* @llvm.visc.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %p_node, i32 0, i32 0)
+ call void @llvm.visc.bind.output(i8* %c_node, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %sub_node, i32 0, i32 0)
+ call void @llvm.visc.bind.output(i8* %sub_node, i32 0, i32 1)
ret %rtype zeroinitializer
}
diff --git a/hpvm/test/unitTests/temp/Makefile b/hpvm/test/unitTests/temp/Makefile
index 15580e9300..539ee5e8fb 100644
--- a/hpvm/test/unitTests/temp/Makefile
+++ b/hpvm/test/unitTests/temp/Makefile
@@ -2,8 +2,8 @@ PASSES :=
.PHONY: clean
-LLVM_INSTALL:=/home/psrivas2/Hetero/HPVM/Code/trunk/llvm-install
-LIBCLC:=/home/psrivas2/Hetero/HPVM/Code/trunk/libclc
+LLVM_INSTALL:=/home/psrivas2/Hetero/VISC/Code/trunk/llvm-install
+LIBCLC:=/home/psrivas2/Hetero/VISC/Code/trunk/libclc
HOST:=gemm_opencl
KERNELS:=matrixMul
LLVM_CC:=$(LLVM_INSTALL)/bin/clang
diff --git a/hpvm/test/unitTests/temp/query2D.ll b/hpvm/test/unitTests/temp/query2D.ll
index 48358a3527..c994c2a3ff 100644
--- a/hpvm/test/unitTests/temp/query2D.ll
+++ b/hpvm/test/unitTests/temp/query2D.ll
@@ -1,5 +1,5 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG -o %t.ll -S < %s
-; RUN: llvm-link %t.ll ~/current-src/projects/hpvm-rt/hpvm-rt.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll ~/current-src/projects/visc-rt/visc-rt.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -lrt -o %t.bin
; RUN: %t.bin 5
; ModuleID = '/home/psrivas2/current-test/unitTests/query2D.ll'
@@ -12,46 +12,46 @@ target triple = "x86_64-unknown-linux-gnu"
@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #0
+declare i8* @llvm.visc.createNode(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode1D(i8*, i32) #0
+declare i8* @llvm.visc.createNode1D(i8*, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode2D(i8*, i32, i32) #0
+declare i8* @llvm.visc.createNode2D(i8*, i32, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32) #0
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*) #0
+declare i8* @llvm.visc.launch(i8*, i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #0
+declare void @llvm.visc.wait(i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32)
+declare void @llvm.visc.bind.input(i8*, i32, i32)
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32)
+declare void @llvm.visc.bind.output(i8*, i32, i32)
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.getNode() #0
+declare i8* @llvm.visc.getNode() #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.getParentNode(i8*) #0
+declare i8* @llvm.visc.getParentNode(i8*) #0
; Function Attrs: nounwind
-declare i32 @llvm.hpvm.getNumDims(i8*) #0
+declare i32 @llvm.visc.getNumDims(i8*) #0
; Function Attrs: nounwind
-declare i32 @llvm.hpvm.getNumNodeInstances.x(i8*) #0
+declare i32 @llvm.visc.getNumNodeInstances.x(i8*) #0
; Function Attrs: nounwind uwtable
define i32 @main(i32 %argc, i8** nocapture %argv) #1 {
@@ -61,25 +61,25 @@ entry:
%0 = load i8** %arrayidx, align 8, !tbaa !0
%call.i = tail call i64 @strtol(i8* nocapture %0, i8** null, i32 10) #0
%conv.i = trunc i64 %call.i to i32
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%1 = bitcast %struct.arg* %in.addr to i32*
store i32 %conv.i, i32* %1
%args = bitcast %struct.arg* %in.addr to i8*
- %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
+ %graphID = call i8* @llvm.visc.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
%call1 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %conv.i) #0
- call void @llvm.hpvm.wait(i8* %graphID)
+ call void @llvm.visc.wait(i8* %graphID)
%2 = getelementptr %struct.arg* %in.addr, i32 0, i32 1
%outputstruct = load %rtype* %2
%output = extractvalue %rtype %outputstruct, 0
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.cleanup()
%call2 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %output) #0
ret i32 0
}
define %rtype @producer(i32 %id) {
%sum = add i32 4, %id
- %this_node = call i8* @llvm.hpvm.getNode()
- %dim = call i32 @llvm.hpvm.getNumNodeInstances.x(i8* %this_node)
+ %this_node = call i8* @llvm.visc.getNode()
+ %dim = call i32 @llvm.visc.getNumNodeInstances.x(i8* %this_node)
%sum2 = add i32 %sum, %dim
%output = insertvalue %rtype undef, i32 %sum2, 0
ret %rtype %output
@@ -92,11 +92,11 @@ define %rtype @consumer(i32 %id) {
}
define %rtype @Root(i32 %dimension) {
- %p_node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%rtype (i32)* @producer to i8*), i32 %dimension, i32 %dimension)
- %c_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
- %edge = call i8* @llvm.hpvm.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
- call void @llvm.hpvm.bind.input(i8* %p_node, i32 0, i32 0)
- call void @llvm.hpvm.bind.output(i8* %c_node, i32 0, i32 0)
+ %p_node = call i8* @llvm.visc.createNode2D(i8* bitcast (%rtype (i32)* @producer to i8*), i32 %dimension, i32 %dimension)
+ %c_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
+ %edge = call i8* @llvm.visc.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %p_node, i32 0, i32 0)
+ call void @llvm.visc.bind.output(i8* %c_node, i32 0, i32 0)
ret %rtype zeroinitializer
}
diff --git a/hpvm/test/unitTests/temp/query3D.ll b/hpvm/test/unitTests/temp/query3D.ll
index d2ff16ef56..438fe60a3b 100644
--- a/hpvm/test/unitTests/temp/query3D.ll
+++ b/hpvm/test/unitTests/temp/query3D.ll
@@ -1,5 +1,5 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG -o %t.ll -S < %s
-; RUN: llvm-link %t.ll ~/current-src/projects/hpvm-rt/hpvm-rt.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll ~/current-src/projects/visc-rt/visc-rt.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -lrt -o %t.bin
; RUN: %t.bin 5
; ModuleID = '/home/psrivas2/current-test/unitTests/query3D.ll'
@@ -12,57 +12,57 @@ target triple = "x86_64-unknown-linux-gnu"
@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #0
+declare i8* @llvm.visc.createNode(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode1D(i8*, i32) #0
+declare i8* @llvm.visc.createNode1D(i8*, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode2D(i8*, i32, i32) #0
+declare i8* @llvm.visc.createNode2D(i8*, i32, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode3D(i8*, i32, i32, i32) #0
+declare i8* @llvm.visc.createNode3D(i8*, i32, i32, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32) #0
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*) #0
+declare i8* @llvm.visc.launch(i8*, i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #0
+declare void @llvm.visc.wait(i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32)
+declare void @llvm.visc.bind.input(i8*, i32, i32)
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32)
+declare void @llvm.visc.bind.output(i8*, i32, i32)
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.getNode() #0
+declare i8* @llvm.visc.getNode() #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.getParentNode(i8*) #0
+declare i8* @llvm.visc.getParentNode(i8*) #0
; Function Attrs: nounwind
-declare i32 @llvm.hpvm.getNumDims(i8*) #0
+declare i32 @llvm.visc.getNumDims(i8*) #0
; Function Attrs: nounwind
-declare i32 @llvm.hpvm.getNumNodeInstances.x(i8*) #0
+declare i32 @llvm.visc.getNumNodeInstances.x(i8*) #0
; Function Attrs: nounwind
-declare i32 @llvm.hpvm.getNumNodeInstances.y(i8*) #0
+declare i32 @llvm.visc.getNumNodeInstances.y(i8*) #0
; Function Attrs: nounwind uwtable
define i32 @main(i32 %argc, i8** nocapture %argv) #1 {
entry:
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%in.addr = alloca %struct.arg
%arrayidx = getelementptr inbounds i8** %argv, i64 1
%0 = load i8** %arrayidx, align 8, !tbaa !0
@@ -71,21 +71,21 @@ entry:
%1 = bitcast %struct.arg* %in.addr to i32*
store i32 %conv.i, i32* %1
%args = bitcast %struct.arg* %in.addr to i8*
- %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
+ %graphID = call i8* @llvm.visc.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
%call1 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %conv.i) #0
- call void @llvm.hpvm.wait(i8* %graphID)
+ call void @llvm.visc.wait(i8* %graphID)
%2 = getelementptr %struct.arg* %in.addr, i32 0, i32 1
%outputstruct = load %rtype* %2
%output = extractvalue %rtype %outputstruct, 0
%call2 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %output) #0
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.cleanup()
ret i32 0
}
define %rtype @producer(i32 %id) {
%sum = add i32 4, %id
- %this_node = call i8* @llvm.hpvm.getNode()
- %dim = call i32 @llvm.hpvm.getNumNodeInstances.y(i8* %this_node)
+ %this_node = call i8* @llvm.visc.getNode()
+ %dim = call i32 @llvm.visc.getNumNodeInstances.y(i8* %this_node)
%sum2 = add i32 %sum, %dim
%output = insertvalue %rtype undef, i32 %sum2, 0
ret %rtype %output
@@ -98,11 +98,11 @@ define %rtype @consumer(i32 %id) {
}
define %rtype @Root(i32 %dimension) {
- %p_node = call i8* @llvm.hpvm.createNode3D(i8* bitcast (%rtype (i32)* @producer to i8*), i32 %dimension, i32 10, i32 30)
- %c_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
- %edge = call i8* @llvm.hpvm.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
- call void @llvm.hpvm.bind.input(i8* %p_node, i32 0, i32 0)
- call void @llvm.hpvm.bind.output(i8* %c_node, i32 0, i32 0)
+ %p_node = call i8* @llvm.visc.createNode3D(i8* bitcast (%rtype (i32)* @producer to i8*), i32 %dimension, i32 10, i32 30)
+ %c_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
+ %edge = call i8* @llvm.visc.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %p_node, i32 0, i32 0)
+ call void @llvm.visc.bind.output(i8* %c_node, i32 0, i32 0)
ret %rtype zeroinitializer
}
diff --git a/hpvm/test/unitTests/temp/queryNodeInst.ll b/hpvm/test/unitTests/temp/queryNodeInst.ll
index 4e3dd75530..24d6a3f0d3 100644
--- a/hpvm/test/unitTests/temp/queryNodeInst.ll
+++ b/hpvm/test/unitTests/temp/queryNodeInst.ll
@@ -1,5 +1,5 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG -o %t.ll -S < %s
-; RUN: llvm-link %t.ll ~/current-src/projects/hpvm-rt/hpvm-rt.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll ~/current-src/projects/visc-rt/visc-rt.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -lrt -o %t.bin
; RUN: %t.bin 5
; ModuleID = '/home/psrivas2/current-test/unitTests/twoNode.ll'
@@ -12,40 +12,40 @@ target triple = "x86_64-unknown-linux-gnu"
@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #0
+declare i8* @llvm.visc.createNode(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32) #0
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*) #0
+declare i8* @llvm.visc.launch(i8*, i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #0
+declare void @llvm.visc.wait(i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32)
+declare void @llvm.visc.bind.input(i8*, i32, i32)
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32)
+declare void @llvm.visc.bind.output(i8*, i32, i32)
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.getNode() #0
+declare i8* @llvm.visc.getNode() #0
; Function Attrs: nounwind
-declare i32 @llvm.hpvm.getNumDims(i8*) #0
+declare i32 @llvm.visc.getNumDims(i8*) #0
; Function Attrs: nounwind uwtable
define i32 @main(i32 %argc, i8** nocapture %argv) #1 {
entry:
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%in.addr = alloca %struct.arg
%arrayidx = getelementptr inbounds i8** %argv, i64 1
%0 = load i8** %arrayidx, align 8, !tbaa !0
@@ -54,21 +54,21 @@ entry:
%1 = bitcast %struct.arg* %in.addr to i32*
store i32 %conv.i, i32* %1
%args = bitcast %struct.arg* %in.addr to i8*
- %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
+ %graphID = call i8* @llvm.visc.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
%call1 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %conv.i) #0
- call void @llvm.hpvm.wait(i8* %graphID)
+ call void @llvm.visc.wait(i8* %graphID)
%2 = getelementptr %struct.arg* %in.addr, i32 0, i32 1
%outputstruct = load %rtype* %2
%output = extractvalue %rtype %outputstruct, 0
%call2 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %output) #0
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.cleanup()
ret i32 0
}
define %rtype @producer(i32 %id) {
%sum = add i32 4, %id
- %this_node = call i8* @llvm.hpvm.getNode()
- %numDim = call i32 @llvm.hpvm.getNumDims(i8* %this_node)
+ %this_node = call i8* @llvm.visc.getNode()
+ %numDim = call i32 @llvm.visc.getNumDims(i8* %this_node)
%sum2 = add i32 %sum, %numDim
%output = insertvalue %rtype undef, i32 %sum, 0
ret %rtype %output
@@ -81,11 +81,11 @@ define %rtype @consumer(i32 %id) {
}
define %rtype @Root(i32 %id) {
- %p_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype (i32)* @producer to i8*))
- %c_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
- %edge = call i8* @llvm.hpvm.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
- call void @llvm.hpvm.bind.input(i8* %p_node, i32 0, i32 0)
- call void @llvm.hpvm.bind.output(i8* %c_node, i32 0, i32 0)
+ %p_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype (i32)* @producer to i8*))
+ %c_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
+ %edge = call i8* @llvm.visc.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %p_node, i32 0, i32 0)
+ call void @llvm.visc.bind.output(i8* %c_node, i32 0, i32 0)
ret %rtype zeroinitializer
}
diff --git a/hpvm/test/unitTests/temp/queryNumDim.ll b/hpvm/test/unitTests/temp/queryNumDim.ll
index caa0978dab..500e2ff41b 100644
--- a/hpvm/test/unitTests/temp/queryNumDim.ll
+++ b/hpvm/test/unitTests/temp/queryNumDim.ll
@@ -1,5 +1,5 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG -o %t.ll -S < %s
-; RUN: llvm-link %t.ll ~/current-src/projects/hpvm-rt/hpvm-rt.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll ~/current-src/projects/visc-rt/visc-rt.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -lrt -o %t.bin
; RUN: %t.bin 5
; ModuleID = '/home/psrivas2/current-test/unitTests/twoNode.ll'
@@ -12,42 +12,42 @@ target triple = "x86_64-unknown-linux-gnu"
@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #0
+declare i8* @llvm.visc.createNode(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32) #0
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*) #0
+declare i8* @llvm.visc.launch(i8*, i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #0
+declare void @llvm.visc.wait(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.getNode() #0
+declare i8* @llvm.visc.getNode() #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.getParentNode(i8*) #0
+declare i8* @llvm.visc.getParentNode(i8*) #0
; Function Attrs: nounwind
-declare i32 @llvm.hpvm.getNumDims(i8*) #0
+declare i32 @llvm.visc.getNumDims(i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32)
+declare void @llvm.visc.bind.input(i8*, i32, i32)
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32)
+declare void @llvm.visc.bind.output(i8*, i32, i32)
; Function Attrs: nounwind uwtable
define i32 @main(i32 %argc, i8** nocapture %argv) #1 {
entry:
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%in.addr = alloca %struct.arg
%arrayidx = getelementptr inbounds i8** %argv, i64 1
%0 = load i8** %arrayidx, align 8, !tbaa !0
@@ -56,21 +56,21 @@ entry:
%1 = bitcast %struct.arg* %in.addr to i32*
store i32 %conv.i, i32* %1
%args = bitcast %struct.arg* %in.addr to i8*
- %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
+ %graphID = call i8* @llvm.visc.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
%call1 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %conv.i) #0
- call void @llvm.hpvm.wait(i8* %graphID)
+ call void @llvm.visc.wait(i8* %graphID)
%2 = getelementptr %struct.arg* %in.addr, i32 0, i32 1
%outputstruct = load %rtype* %2
%output = extractvalue %rtype %outputstruct, 0
%call2 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %output) #0
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.cleanup()
ret i32 0
}
define %rtype @producer(i32 %id) {
%sum = add i32 4, %id
- %this_node = call i8* @llvm.hpvm.getNode()
- %numDim = call i32 @llvm.hpvm.getNumDims(i8* %this_node)
+ %this_node = call i8* @llvm.visc.getNode()
+ %numDim = call i32 @llvm.visc.getNumDims(i8* %this_node)
%sum2 = add i32 %sum, %numDim
%output = insertvalue %rtype undef, i32 %sum, 0
ret %rtype %output
@@ -83,11 +83,11 @@ define %rtype @consumer(i32 %id) {
}
define %rtype @Root(i32 %id) {
- %p_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype (i32)* @producer to i8*))
- %c_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
- %edge = call i8* @llvm.hpvm.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
- call void @llvm.hpvm.bind.input(i8* %p_node, i32 0, i32 0)
- call void @llvm.hpvm.bind.output(i8* %c_node, i32 0, i32 0)
+ %p_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype (i32)* @producer to i8*))
+ %c_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
+ %edge = call i8* @llvm.visc.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %p_node, i32 0, i32 0)
+ call void @llvm.visc.bind.output(i8* %c_node, i32 0, i32 0)
ret %rtype zeroinitializer
}
diff --git a/hpvm/test/unitTests/temp/queryNumNodeInst.ll b/hpvm/test/unitTests/temp/queryNumNodeInst.ll
index 07418ff725..48add92f16 100644
--- a/hpvm/test/unitTests/temp/queryNumNodeInst.ll
+++ b/hpvm/test/unitTests/temp/queryNumNodeInst.ll
@@ -1,5 +1,5 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG -o %t.ll -S < %s
-; RUN: llvm-link %t.ll ~/current-src/projects/hpvm-rt/hpvm-rt.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll ~/current-src/projects/visc-rt/visc-rt.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -lrt -o %t.bin
; RUN: %t.bin 5
; ModuleID = '/home/psrivas2/current-test/unitTests/twoNode.ll'
@@ -12,48 +12,48 @@ target triple = "x86_64-unknown-linux-gnu"
@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #0
+declare i8* @llvm.visc.createNode(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode1D(i8*, i32) #0
+declare i8* @llvm.visc.createNode1D(i8*, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32) #0
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*) #0
+declare i8* @llvm.visc.launch(i8*, i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #0
+declare void @llvm.visc.wait(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.getNode() #0
+declare i8* @llvm.visc.getNode() #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.getParentNode(i8*) #0
+declare i8* @llvm.visc.getParentNode(i8*) #0
; Function Attrs: nounwind
-declare i32 @llvm.hpvm.getNumDims(i8*) #0
+declare i32 @llvm.visc.getNumDims(i8*) #0
; Function Attrs: nounwind
-declare i32 @llvm.hpvm.getNumNodeInstances.x(i8*) #0
+declare i32 @llvm.visc.getNumNodeInstances.x(i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32)
+declare void @llvm.visc.bind.input(i8*, i32, i32)
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32)
+declare void @llvm.visc.bind.output(i8*, i32, i32)
; Function Attrs: nounwind uwtable
define i32 @main(i32 %argc, i8** nocapture %argv) #1 {
entry:
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%in.addr = alloca %struct.arg
%arrayidx = getelementptr inbounds i8** %argv, i64 1
%0 = load i8** %arrayidx, align 8, !tbaa !0
@@ -62,21 +62,21 @@ entry:
%1 = bitcast %struct.arg* %in.addr to i32*
store i32 %conv.i, i32* %1
%args = bitcast %struct.arg* %in.addr to i8*
- %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
+ %graphID = call i8* @llvm.visc.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
%call1 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %conv.i) #0
- call void @llvm.hpvm.wait(i8* %graphID)
+ call void @llvm.visc.wait(i8* %graphID)
%2 = getelementptr %struct.arg* %in.addr, i32 0, i32 1
%outputstruct = load %rtype* %2
%output = extractvalue %rtype %outputstruct, 0
%call2 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %output) #0
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.cleanup()
ret i32 0
}
define %rtype @producer(i32 %id) {
%sum = add i32 4, %id
- %this_node = call i8* @llvm.hpvm.getNode()
- %dim = call i32 @llvm.hpvm.getNumNodeInstances.x(i8* %this_node)
+ %this_node = call i8* @llvm.visc.getNode()
+ %dim = call i32 @llvm.visc.getNumNodeInstances.x(i8* %this_node)
%sum2 = add i32 %sum, %dim
%output = insertvalue %rtype undef, i32 %sum2, 0
ret %rtype %output
@@ -89,11 +89,11 @@ define %rtype @consumer(i32 %id) {
}
define %rtype @Root(i32 %dimension) {
- %p_node = call i8* @llvm.hpvm.createNode1D(i8* bitcast (%rtype (i32)* @producer to i8*), i32 %dimension)
- %c_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
- %edge = call i8* @llvm.hpvm.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
- call void @llvm.hpvm.bind.input(i8* %p_node, i32 0, i32 0)
- call void @llvm.hpvm.bind.output(i8* %c_node, i32 0, i32 0)
+ %p_node = call i8* @llvm.visc.createNode1D(i8* bitcast (%rtype (i32)* @producer to i8*), i32 %dimension)
+ %c_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
+ %edge = call i8* @llvm.visc.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %p_node, i32 0, i32 0)
+ call void @llvm.visc.bind.output(i8* %c_node, i32 0, i32 0)
ret %rtype zeroinitializer
}
diff --git a/hpvm/test/unitTests/temp/singleNode.ll b/hpvm/test/unitTests/temp/singleNode.ll
index 99e5318131..20713e955f 100644
--- a/hpvm/test/unitTests/temp/singleNode.ll
+++ b/hpvm/test/unitTests/temp/singleNode.ll
@@ -1,5 +1,5 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG -o %t.ll -S < %s
-; RUN: llvm-link %t.ll ~/current-src/projects/hpvm-rt/hpvm-rt.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll ~/current-src/projects/visc-rt/visc-rt.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -lrt -o %t.bin
; RUN: %t.bin 5
; ModuleID = '/home/psrivas2/current-test/unitTests/singleNode.ll'
@@ -12,43 +12,43 @@ target triple = "x86_64-unknown-linux-gnu"
@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #0
+declare i8* @llvm.visc.createNode(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32) #0
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*) #0
+declare i8* @llvm.visc.launch(i8*, i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #0
+declare void @llvm.visc.wait(i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32)
+declare void @llvm.visc.bind.input(i8*, i32, i32)
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32)
+declare void @llvm.visc.bind.output(i8*, i32, i32)
; Function Attrs: nounwind uwtable
define i32 @main(i32 %argc, i8** nocapture %argv) #1 {
entry:
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%in.addr = alloca %struct.arg
%arrayidx = getelementptr inbounds i8** %argv, i64 1
%0 = load i8** %arrayidx, align 8, !tbaa !0
%call.i = tail call i64 @strtol(i8* nocapture %0, i8** null, i32 10) #0
%conv.i = trunc i64 %call.i to i32
%args = bitcast %struct.arg* %in.addr to i8*
- %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%rtype ()* @Root to i8*), i8* %args)
+ %graphID = call i8* @llvm.visc.launch(i8* bitcast (%rtype ()* @Root to i8*), i8* %args)
%call1 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %conv.i) #0
- call void @llvm.hpvm.wait(i8* %graphID)
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.wait(i8* %graphID)
+ call void @llvm.visc.cleanup()
ret i32 0
}
@@ -59,8 +59,8 @@ define %rtype @foo() {
}
define %rtype @Root() {
- %node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype ()* @foo to i8*))
- call void @llvm.hpvm.bind.output(i8* %node, i32 0, i32 0)
+ %node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype ()* @foo to i8*))
+ call void @llvm.visc.bind.output(i8* %node, i32 0, i32 0)
ret %rtype zeroinitializer
}
diff --git a/hpvm/test/unitTests/temp/singleNodeStream.ll b/hpvm/test/unitTests/temp/singleNodeStream.ll
index aa0243603c..fce75df671 100644
--- a/hpvm/test/unitTests/temp/singleNodeStream.ll
+++ b/hpvm/test/unitTests/temp/singleNodeStream.ll
@@ -1,5 +1,5 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG -o %t.ll -S < %s
-; RUN: llvm-link %t.ll ~/current-src/projects/hpvm-rt/hpvm-rt.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll ~/current-src/projects/visc-rt/visc-rt.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -lrt -o %t.bin
; RUN: %t.bin 5
; ModuleID = '/home/psrivas2/current-test/unitTests/twoNodeConnect.ll'
@@ -14,39 +14,39 @@ target triple = "x86_64-unknown-linux-gnu"
@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #0
+declare i8* @llvm.visc.createNode(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32, i1) #0
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32, i1) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #0
+declare i8* @llvm.visc.launch(i8*, i8*, i1) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.push(i8*, i8*) #0
+declare void @llvm.visc.push(i8*, i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.pop(i8*) #0
+declare i8* @llvm.visc.pop(i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #0
+declare void @llvm.visc.wait(i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1)
+declare void @llvm.visc.bind.input(i8*, i32, i32, i1)
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32, i1)
+declare void @llvm.visc.bind.output(i8*, i32, i32, i1)
; Function Attrs: nounwind uwtable
define i32 @main(i32 %argc, i8** nocapture %argv) #1 {
entry:
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%in.addr = alloca %struct.arg
%num = alloca i32
%arrayidx = getelementptr inbounds i8** %argv, i64 1
@@ -60,27 +60,27 @@ entry:
%args = bitcast %struct.arg* %in.addr to i8*
; Launch the pipeline
- %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%rptype (i32*, i64)* @Root to i8*), i8* %args, i1 1)
+ %graphID = call i8* @llvm.visc.launch(i8* bitcast (%rptype (i32*, i64)* @Root to i8*), i8* %args, i1 1)
; Push arguments into the pipeline
- call void @llvm.hpvm.push(i8* %graphID, i8* %args)
- call void @llvm.hpvm.push(i8* %graphID, i8* %args)
- call void @llvm.hpvm.push(i8* %graphID, i8* %args)
- call void @llvm.hpvm.push(i8* %graphID, i8* %args)
+ call void @llvm.visc.push(i8* %graphID, i8* %args)
+ call void @llvm.visc.push(i8* %graphID, i8* %args)
+ call void @llvm.visc.push(i8* %graphID, i8* %args)
+ call void @llvm.visc.push(i8* %graphID, i8* %args)
; Pop out arguments and read the output
- %graph_output = call i8* @llvm.hpvm.pop(i8* %graphID)
- %graph_output1 = call i8* @llvm.hpvm.pop(i8* %graphID)
- %graph_output2 = call i8* @llvm.hpvm.pop(i8* %graphID)
- %graph_output3 = call i8* @llvm.hpvm.pop(i8* %graphID)
+ %graph_output = call i8* @llvm.visc.pop(i8* %graphID)
+ %graph_output1 = call i8* @llvm.visc.pop(i8* %graphID)
+ %graph_output2 = call i8* @llvm.visc.pop(i8* %graphID)
+ %graph_output3 = call i8* @llvm.visc.pop(i8* %graphID)
%output.addr = bitcast i8* %graph_output to %rptype*
%outputstruct = load %rptype* %output.addr
%output = extractvalue %rptype %outputstruct, 0
%output_val = load i32* %output
%call2 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([18 x i8]* @out.str, i64 0, i64 0), i32 %output_val) #0
- call void @llvm.hpvm.wait(i8* %graphID)
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.wait(i8* %graphID)
+ call void @llvm.visc.cleanup()
ret i32 0
}
@@ -95,11 +95,11 @@ define %rptype @producer(i32* %id, i64 %size) {
}
define %rptype @Root(i32* %id, i64 %size) {
- %p_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rptype (i32*, i64)* @producer to i8*))
- call void @llvm.hpvm.bind.input(i8* %p_node, i32 0, i32 0, i1 1)
- call void @llvm.hpvm.bind.input(i8* %p_node, i32 1, i32 1, i1 1)
- call void @llvm.hpvm.bind.output(i8* %p_node, i32 0, i32 0, i1 1)
- call void @llvm.hpvm.bind.output(i8* %p_node, i32 1, i32 1, i1 1)
+ %p_node = call i8* @llvm.visc.createNode(i8* bitcast (%rptype (i32*, i64)* @producer to i8*))
+ call void @llvm.visc.bind.input(i8* %p_node, i32 0, i32 0, i1 1)
+ call void @llvm.visc.bind.input(i8* %p_node, i32 1, i32 1, i1 1)
+ call void @llvm.visc.bind.output(i8* %p_node, i32 0, i32 0, i1 1)
+ call void @llvm.visc.bind.output(i8* %p_node, i32 1, i32 1, i1 1)
ret %rptype zeroinitializer
}
diff --git a/hpvm/test/unitTests/temp/twoLaunch.ll b/hpvm/test/unitTests/temp/twoLaunch.ll
index ee602f58d8..48c973a7e6 100644
--- a/hpvm/test/unitTests/temp/twoLaunch.ll
+++ b/hpvm/test/unitTests/temp/twoLaunch.ll
@@ -1,5 +1,5 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG -o %t.ll -S < %s
-; RUN: llvm-link %t.ll ~/current-src/projects/hpvm-rt/hpvm-rt.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll ~/current-src/projects/visc-rt/visc-rt.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -lrt -o %t.bin
; RUN: %t.bin 5
; ModuleID = '/home/psrivas2/current-test/unitTests/singleNode.ll'
@@ -12,33 +12,33 @@ target triple = "x86_64-unknown-linux-gnu"
@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #0
+declare i8* @llvm.visc.createNode(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32) #0
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*) #0
+declare i8* @llvm.visc.launch(i8*, i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #0
+declare void @llvm.visc.wait(i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32)
+declare void @llvm.visc.bind.input(i8*, i32, i32)
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32)
+declare void @llvm.visc.bind.output(i8*, i32, i32)
; Function Attrs: nounwind uwtable
define i32 @main(i32 %argc, i8** nocapture %argv) #1 {
entry:
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%in.addr_1 = alloca %struct.arg
%in.addr_2= alloca %struct.arg
%arrayidx = getelementptr inbounds i8** %argv, i64 1
@@ -47,12 +47,12 @@ entry:
%conv.i = trunc i64 %call.i to i32
%args_1 = bitcast %struct.arg* %in.addr_1 to i8*
%args_2 = bitcast %struct.arg* %in.addr_2 to i8*
- %graphID_1 = call i8* @llvm.hpvm.launch(i8* bitcast (%rtype ()* @Root_1 to i8*), i8* %args_1)
- %graphID_2 = call i8* @llvm.hpvm.launch(i8* bitcast (%rtype ()* @Root_2 to i8*), i8* %args_2)
+ %graphID_1 = call i8* @llvm.visc.launch(i8* bitcast (%rtype ()* @Root_1 to i8*), i8* %args_1)
+ %graphID_2 = call i8* @llvm.visc.launch(i8* bitcast (%rtype ()* @Root_2 to i8*), i8* %args_2)
%call1 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %conv.i) #0
- call void @llvm.hpvm.wait(i8* %graphID_1)
- call void @llvm.hpvm.wait(i8* %graphID_2)
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.wait(i8* %graphID_1)
+ call void @llvm.visc.wait(i8* %graphID_2)
+ call void @llvm.visc.cleanup()
ret i32 0
}
@@ -70,14 +70,14 @@ define %rtype @foo_2() {
}
define %rtype @Root_1() {
- %node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype ()* @foo_1 to i8*))
- call void @llvm.hpvm.bind.output(i8* %node, i32 0, i32 0)
+ %node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype ()* @foo_1 to i8*))
+ call void @llvm.visc.bind.output(i8* %node, i32 0, i32 0)
ret %rtype zeroinitializer
}
define %rtype @Root_2() {
- %node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype ()* @foo_2 to i8*))
- call void @llvm.hpvm.bind.output(i8* %node, i32 0, i32 0)
+ %node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype ()* @foo_2 to i8*))
+ call void @llvm.visc.bind.output(i8* %node, i32 0, i32 0)
ret %rtype zeroinitializer
}
diff --git a/hpvm/test/unitTests/temp/twoNode.ll b/hpvm/test/unitTests/temp/twoNode.ll
index 74e4c64d59..5e2899830b 100644
--- a/hpvm/test/unitTests/temp/twoNode.ll
+++ b/hpvm/test/unitTests/temp/twoNode.ll
@@ -1,5 +1,5 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG -o %t.ll -S < %s
-; RUN: llvm-link %t.ll ~/current-src/projects/hpvm-rt/hpvm-rt.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll ~/current-src/projects/visc-rt/visc-rt.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -lrt -o %t.bin
; RUN: %t.bin 5
; ModuleID = '/home/psrivas2/current-test/unitTests/twoNode.ll'
@@ -11,33 +11,33 @@ target triple = "x86_64-unknown-linux-gnu"
@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #0
+declare i8* @llvm.visc.createNode(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32) #0
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*) #0
+declare i8* @llvm.visc.launch(i8*, i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #0
+declare void @llvm.visc.wait(i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32)
+declare void @llvm.visc.bind.input(i8*, i32, i32)
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32)
+declare void @llvm.visc.bind.output(i8*, i32, i32)
; Function Attrs: nounwind uwtable
define i32 @main(i32 %argc, i8** nocapture %argv) #1 {
entry:
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%in.addr = alloca %struct.arg
%arrayidx = getelementptr inbounds i8** %argv, i64 1
%0 = load i8** %arrayidx, align 8, !tbaa !0
@@ -46,10 +46,10 @@ entry:
%1 = bitcast %struct.arg* %in.addr to i32*
store i32 %conv.i, i32* %1
%args = bitcast %struct.arg* %in.addr to i8*
- %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
+ %graphID = call i8* @llvm.visc.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
%call1 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %conv.i) #0
- call void @llvm.hpvm.wait(i8* %graphID)
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.wait(i8* %graphID)
+ call void @llvm.visc.cleanup()
ret i32 0
}
@@ -66,10 +66,10 @@ define %rtype @consumer(i32 %id) {
}
define %rtype @Root(i32 %id) {
- %p_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype ()* @producer to i8*))
- %c_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
- %edge = call i8* @llvm.hpvm.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
- call void @llvm.hpvm.bind.output(i8* %c_node, i32 0, i32 0)
+ %p_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype ()* @producer to i8*))
+ %c_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
+ %edge = call i8* @llvm.visc.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
+ call void @llvm.visc.bind.output(i8* %c_node, i32 0, i32 0)
ret %rtype zeroinitializer
}
diff --git a/hpvm/test/unitTests/temp/twoNodeConnect.ll b/hpvm/test/unitTests/temp/twoNodeConnect.ll
index 6b23ad691b..06652b94e0 100644
--- a/hpvm/test/unitTests/temp/twoNodeConnect.ll
+++ b/hpvm/test/unitTests/temp/twoNodeConnect.ll
@@ -1,5 +1,5 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG -o %t.ll -S < %s
-; RUN: llvm-link %t.ll ~/current-src/projects/hpvm-rt/hpvm-rt.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll ~/current-src/projects/visc-rt/visc-rt.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -lrt -o %t.bin
; RUN: %t.bin 5
; ModuleID = '/home/psrivas2/current-test/unitTests/twoNodeConnect.ll'
@@ -11,33 +11,33 @@ target triple = "x86_64-unknown-linux-gnu"
@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #0
+declare i8* @llvm.visc.createNode(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32) #0
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*) #0
+declare i8* @llvm.visc.launch(i8*, i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #0
+declare void @llvm.visc.wait(i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32)
+declare void @llvm.visc.bind.input(i8*, i32, i32)
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32)
+declare void @llvm.visc.bind.output(i8*, i32, i32)
; Function Attrs: nounwind uwtable
define i32 @main(i32 %argc, i8** nocapture %argv) #1 {
entry:
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%in.addr = alloca %struct.arg
%arrayidx = getelementptr inbounds i8** %argv, i64 1
%0 = load i8** %arrayidx, align 8, !tbaa !0
@@ -46,14 +46,14 @@ entry:
%1 = bitcast %struct.arg* %in.addr to i32*
store i32 %conv.i, i32* %1
%args = bitcast %struct.arg* %in.addr to i8*
- %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
+ %graphID = call i8* @llvm.visc.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
%call1 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %conv.i) #0
- call void @llvm.hpvm.wait(i8* %graphID)
+ call void @llvm.visc.wait(i8* %graphID)
%2 = getelementptr %struct.arg* %in.addr, i32 0, i32 1
%outputstruct = load %rtype* %2
%output = extractvalue %rtype %outputstruct, 0
%call2 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %output) #0
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.cleanup()
ret i32 0
}
@@ -70,11 +70,11 @@ define %rtype @consumer(i32 %id) {
}
define %rtype @Root(i32 %id) {
- %p_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype (i32)* @producer to i8*))
- %c_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
- %edge = call i8* @llvm.hpvm.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
- call void @llvm.hpvm.bind.input(i8* %p_node, i32 0, i32 0)
- call void @llvm.hpvm.bind.output(i8* %c_node, i32 0, i32 0)
+ %p_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype (i32)* @producer to i8*))
+ %c_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
+ %edge = call i8* @llvm.visc.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %p_node, i32 0, i32 0)
+ call void @llvm.visc.bind.output(i8* %c_node, i32 0, i32 0)
ret %rtype zeroinitializer
}
diff --git a/hpvm/test/unitTests/temp/twoNodeQuery.ll b/hpvm/test/unitTests/temp/twoNodeQuery.ll
index 247d1830da..2e1ea0dba4 100644
--- a/hpvm/test/unitTests/temp/twoNodeQuery.ll
+++ b/hpvm/test/unitTests/temp/twoNodeQuery.ll
@@ -1,5 +1,5 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG -o %t.ll -S < %s
-; RUN: llvm-link %t.ll ~/current-src/projects/hpvm-rt/hpvm-rt.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll ~/current-src/projects/visc-rt/visc-rt.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -lrt -o %t.bin
; RUN: %t.bin 5
; ModuleID = '/home/psrivas2/current-test/unitTests/twoNodeQuery.ll'
@@ -11,42 +11,42 @@ target triple = "x86_64-unknown-linux-gnu"
@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #0
+declare i8* @llvm.visc.createNode(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32) #0
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*) #0
+declare i8* @llvm.visc.launch(i8*, i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #0
+declare void @llvm.visc.wait(i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32)
+declare void @llvm.visc.bind.input(i8*, i32, i32)
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32)
+declare void @llvm.visc.bind.output(i8*, i32, i32)
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.getNode() #0
+declare i8* @llvm.visc.getNode() #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.getParentNode(i8*) #0
+declare i8* @llvm.visc.getParentNode(i8*) #0
; Function Attrs: nounwind
-declare i32 @llvm.hpvm.getNumDims(i8*) #0
+declare i32 @llvm.visc.getNumDims(i8*) #0
; Function Attrs: nounwind uwtable
define i32 @main(i32 %argc, i8** nocapture %argv) #1 {
entry:
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%in.addr = alloca %struct.arg
%arrayidx = getelementptr inbounds i8** %argv, i64 1
%0 = load i8** %arrayidx, align 8, !tbaa !0
@@ -55,21 +55,21 @@ entry:
%1 = bitcast %struct.arg* %in.addr to i32*
store i32 %conv.i, i32* %1
%args = bitcast %struct.arg* %in.addr to i8*
- %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
+ %graphID = call i8* @llvm.visc.launch(i8* bitcast (%rtype (i32)* @Root to i8*), i8* %args)
%call1 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %conv.i) #0
- call void @llvm.hpvm.wait(i8* %graphID)
+ call void @llvm.visc.wait(i8* %graphID)
%2 = getelementptr %struct.arg* %in.addr, i32 0, i32 1
%outputstruct = load %rtype* %2
%output = extractvalue %rtype %outputstruct, 0
%call2 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %output) #0
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.cleanup()
ret i32 0
}
define %rtype @producer(i32 %id) {
%sum = add i32 4, %id
- %this_node = call i8* @llvm.hpvm.getNode()
- %numDim = call i32 @llvm.hpvm.getNumDims(i8* %this_node)
+ %this_node = call i8* @llvm.visc.getNode()
+ %numDim = call i32 @llvm.visc.getNumDims(i8* %this_node)
%sum2 = add i32 %sum, %numDim
%output = insertvalue %rtype undef, i32 %sum, 0
ret %rtype %output
@@ -82,11 +82,11 @@ define %rtype @consumer(i32 %id) {
}
define %rtype @Root(i32 %id) {
- %p_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype (i32)* @producer to i8*))
- %c_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
- %edge = call i8* @llvm.hpvm.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
- call void @llvm.hpvm.bind.input(i8* %p_node, i32 0, i32 0)
- call void @llvm.hpvm.bind.output(i8* %c_node, i32 0, i32 0)
+ %p_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype (i32)* @producer to i8*))
+ %c_node = call i8* @llvm.visc.createNode(i8* bitcast (%rtype (i32)* @consumer to i8*))
+ %edge = call i8* @llvm.visc.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0)
+ call void @llvm.visc.bind.input(i8* %p_node, i32 0, i32 0)
+ call void @llvm.visc.bind.output(i8* %c_node, i32 0, i32 0)
ret %rtype zeroinitializer
}
diff --git a/hpvm/test/unitTests/temp/twoNodeStream.ll b/hpvm/test/unitTests/temp/twoNodeStream.ll
index f9820abd19..6e99259518 100644
--- a/hpvm/test/unitTests/temp/twoNodeStream.ll
+++ b/hpvm/test/unitTests/temp/twoNodeStream.ll
@@ -1,5 +1,5 @@
; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -dfg2llvm-x86 -clearDFG -o %t.ll -S < %s
-; RUN: llvm-link %t.ll ~/current-src/projects/hpvm-rt/hpvm-rt.ll -S -o %t.linked.ll
+; RUN: llvm-link %t.ll ~/current-src/projects/visc-rt/visc-rt.ll -S -o %t.linked.ll
; RUN: clang++ -O3 %t.linked.ll -lpthread -lOpenCL -lrt -o %t.bin
; RUN: %t.bin 5
; ModuleID = '/home/psrivas2/current-test/unitTests/twoNodeConnect.ll'
@@ -14,39 +14,39 @@ target triple = "x86_64-unknown-linux-gnu"
@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
; Function Attrs: nounwind
-declare void @llvm.hpvm.init() #1
+declare void @llvm.visc.init() #1
; Function Attrs: nounwind
-declare void @llvm.hpvm.cleanup() #1
+declare void @llvm.visc.cleanup() #1
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createNode(i8*) #0
+declare i8* @llvm.visc.createNode(i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32, i1) #0
+declare i8* @llvm.visc.createEdge(i8*, i8*, i1, i32, i32, i1) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #0
+declare i8* @llvm.visc.launch(i8*, i8*, i1) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.push(i8*, i8*) #0
+declare void @llvm.visc.push(i8*, i8*) #0
; Function Attrs: nounwind
-declare i8* @llvm.hpvm.pop(i8*) #0
+declare i8* @llvm.visc.pop(i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.wait(i8*) #0
+declare void @llvm.visc.wait(i8*) #0
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1)
+declare void @llvm.visc.bind.input(i8*, i32, i32, i1)
; Function Attrs: nounwind
-declare void @llvm.hpvm.bind.output(i8*, i32, i32, i1)
+declare void @llvm.visc.bind.output(i8*, i32, i32, i1)
; Function Attrs: nounwind uwtable
define i32 @main(i32 %argc, i8** nocapture %argv) #1 {
entry:
- call void @llvm.hpvm.init()
+ call void @llvm.visc.init()
%in.addr = alloca %struct.arg
%num = alloca i32
%arrayidx = getelementptr inbounds i8** %argv, i64 1
@@ -60,21 +60,21 @@ entry:
%args = bitcast %struct.arg* %in.addr to i8*
; Launch the pipeline
- %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%rctype (i32*, i64)* @Root to i8*), i8* %args, i1 1)
+ %graphID = call i8* @llvm.visc.launch(i8* bitcast (%rctype (i32*, i64)* @Root to i8*), i8* %args, i1 1)
; Push arguments into the pipeline
- call void @llvm.hpvm.push(i8* %graphID, i8* %args)
+ call void @llvm.visc.push(i8* %graphID, i8* %args)
; Pop out arguments and read the output
- %graph_output = call i8* @llvm.hpvm.pop(i8* %graphID)
+ %graph_output = call i8* @llvm.visc.pop(i8* %graphID)
%output.addr = bitcast i8* %graph_output to %rctype*
%outputstruct = load %rctype* %output.addr
%output = extractvalue %rctype %outputstruct, 0
%output_val = load i32* %output
%call2 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([18 x i8]* @out.str, i64 0, i64 0), i32 %output_val) #0
- call void @llvm.hpvm.wait(i8* %graphID)
- call void @llvm.hpvm.cleanup()
+ call void @llvm.visc.wait(i8* %graphID)
+ call void @llvm.visc.cleanup()
ret i32 0
}
@@ -97,14 +97,14 @@ define %rctype @consumer(i32* %id, i64 %size) {
}
define %rctype @Root(i32* %id, i64 %size) {
- %p_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rptype (i32*, i64)* @producer to i8*))
- %c_node = call i8* @llvm.hpvm.createNode(i8* bitcast (%rctype (i32*, i64)* @consumer to i8*))
- %edge = call i8* @llvm.hpvm.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0, i1 1)
- %edge2 = call i8* @llvm.hpvm.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 1, i32 1, i1 1)
- call void @llvm.hpvm.bind.input(i8* %p_node, i32 0, i32 0, i1 1)
- call void @llvm.hpvm.bind.input(i8* %p_node, i32 1, i32 1, i1 0)
- call void @llvm.hpvm.bind.output(i8* %c_node, i32 0, i32 0, i1 1)
- call void @llvm.hpvm.bind.output(i8* %c_node, i32 1, i32 1, i1 1)
+ %p_node = call i8* @llvm.visc.createNode(i8* bitcast (%rptype (i32*, i64)* @producer to i8*))
+ %c_node = call i8* @llvm.visc.createNode(i8* bitcast (%rctype (i32*, i64)* @consumer to i8*))
+ %edge = call i8* @llvm.visc.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 0, i32 0, i1 1)
+ %edge2 = call i8* @llvm.visc.createEdge(i8* %p_node, i8* %c_node, i1 false, i32 1, i32 1, i1 1)
+ call void @llvm.visc.bind.input(i8* %p_node, i32 0, i32 0, i1 1)
+ call void @llvm.visc.bind.input(i8* %p_node, i32 1, i32 1, i1 0)
+ call void @llvm.visc.bind.output(i8* %c_node, i32 0, i32 0, i1 1)
+ call void @llvm.visc.bind.output(i8* %c_node, i32 1, i32 1, i1 1)
ret %rctype zeroinitializer
}
--
GitLab