diff --git a/hpvm/test/README.md b/hpvm/test/README.md
index e709ef04195c90e0f91c2a4b4a4b1d2f0b716d1f..94103affb668afc29d32e52d85d0d60182bd16d8 100644
--- a/hpvm/test/README.md
+++ b/hpvm/test/README.md
@@ -22,9 +22,9 @@ TODO
## Pipeline
```
make TARGET={seq, gpu}
- ./pipeline-{seq, gpu} datasets/big/input/formula1_scaled.mp4
+ ./pipeline-{seq, gpu} datasets/formula1_scaled.mp4
```
## Your own project
See `template/` for an example Makefile and config.
-Include `visc.h` to use HPVM intrinsics, found in the `test/include/visc.h`.
+Include `visc.h` to use HPVM C api functions, found in the `test/include/visc.h`.
diff --git a/hpvm/test/parboil/README.md b/hpvm/test/parboil/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..1166e4f10f6a6e29e4f5d40871674c27da975acc
--- /dev/null
+++ b/hpvm/test/parboil/README.md
@@ -0,0 +1,8 @@
+# Current Benchmark Compatability
+
+| Benchmark | Version | Supported on CPU | Supported on GPU |
+| :-------- | :------ | :--------------: | :--------------: |
+| sgemm | visc | ✔ | ✔ |
+| stencil | visc | ✔ | ✔ |
+| spmv | visc | ✔ | ✘ |
+| lbm | visc | ✔ | ✘ |
diff --git a/hpvm/test/parboil/common/Makefile.conf.example-ati b/hpvm/test/parboil/common/Makefile.conf.example-ati
deleted file mode 100644
index e7dacca07360035cc3d07404ae0358c310ad4507..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/Makefile.conf.example-ati
+++ /dev/null
@@ -1,3 +0,0 @@
-OPENCL_PATH=/opt/ati/
-OPENCL_LIB_PATH=/opt/ati/lib/x86_64
-
diff --git a/hpvm/test/parboil/common/Makefile.conf.example-mcuda b/hpvm/test/parboil/common/Makefile.conf.example-mcuda
deleted file mode 100644
index 36c110c015b6cde1aafc01cc3d6623c3bf905887..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/Makefile.conf.example-mcuda
+++ /dev/null
@@ -1 +0,0 @@
-MCUDA_PATH=/usr/local/mcuda
diff --git a/hpvm/test/parboil/common/Makefile.conf.example-nvidia b/hpvm/test/parboil/common/Makefile.conf.example-nvidia
deleted file mode 100644
index bdf84b6329487cba5c1c38dd845b9f63af4c28b5..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/Makefile.conf.example-nvidia
+++ /dev/null
@@ -1,5 +0,0 @@
-CUDA_PATH=/usr/local/cuda
-CUDA_LIB_PATH=/usr/local/cuda/lib64
-OPENCL_PATH=/usr/local/cuda
-OPENCL_LIB_PATH=/usr/lib
-
diff --git a/hpvm/test/parboil/common/mk/c.mk b/hpvm/test/parboil/common/mk/c.mk
deleted file mode 100644
index 9334ba5c8f76e56490b96ec1527bc33e7d8248f0..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/mk/c.mk
+++ /dev/null
@@ -1,86 +0,0 @@
-# (c) 2007 The Board of Trustees of the University of Illinois.
-
-# Default language wide options
-
-LANG_CFLAGS=-I$(PARBOIL_ROOT)/common/include -I/usr/local/cuda/include
-LANG_CXXFLAGS=$(LANG_CFLAGS)
-LANG_LDFLAGS=
-
-CFLAGS=$(LANG_CFLAGS) $(PLATFORM_CFLAGS) $(APP_CFLAGS)
-CXXFLAGS=$(LANG_CXXFLAGS) $(PLATFORM_CXXFLAGS) $(APP_CXXFLAGS)
-LDFLAGS=$(LANG_LDFLAGS) $(PLATFORM_LDFLAGS) $(APP_LDFLAGS)
-
-# Rules common to all makefiles
-
-########################################
-# Functions
-########################################
-
-# Add BUILDDIR as a prefix to each element of $1
-INBUILDDIR=$(addprefix $(BUILDDIR)/,$(1))
-
-# Add SRCDIR as a prefix to each element of $1
-INSRCDIR=$(addprefix $(SRCDIR)/,$(1))
-
-
-########################################
-# Environment variable check
-########################################
-
-# The second-last directory in the $(BUILDDIR) path
-# must have the name "build". This reduces the risk of terrible
-# accidents if paths are not set up correctly.
-ifeq ("$(notdir $(BUILDDIR))", "")
-$(error $$BUILDDIR is not set correctly)
-endif
-
-ifneq ("$(notdir $(patsubst %/,%,$(dir $(BUILDDIR))))", "build")
-$(error $$BUILDDIR is not set correctly)
-endif
-
-.PHONY: run
-
-########################################
-# Derived variables
-########################################
-
-ifeq ($(DEBUGGER),)
-DEBUGGER=gdb
-endif
-
-OBJS = $(call INBUILDDIR,$(SRCDIR_OBJS))
-
-########################################
-# Rules
-########################################
-
-default: $(BUILDDIR) $(BIN)
-
-run:
- @$(BIN) $(ARGS)
-
-debug:
- @$(DEBUGGER) --args $(BIN) $(ARGS)
-
-clean :
- rm -f $(BUILDDIR)/*
- if [ -d $(BUILDDIR) ]; then rmdir $(BUILDDIR); fi
-
-$(BIN) : $(OBJS) $(BUILDDIR)/parboil.o
- $(CXX) $^ -o $@ $(LDFLAGS)
-
-$(BUILDDIR) :
- mkdir -p $(BUILDDIR)
-
-$(BUILDDIR)/%.o : $(SRCDIR)/%.c
- $(CC) $(CFLAGS) -c $< -o $@
-
-$(BUILDDIR)/parboil.o: $(PARBOIL_ROOT)/common/src/parboil.c
- $(CC) $(CFLAGS) -c $< -o $@
-
-$(BUILDDIR)/%.o : $(SRCDIR)/%.cc
- $(CXX) $(CXXFLAGS) -c $< -o $@
-
-$(BUILDDIR)/%.o : $(SRCDIR)/%.cpp
- $(CXX) $(CXXFLAGS) -c $< -o $@
-
diff --git a/hpvm/test/parboil/common/mk/cuda.mk b/hpvm/test/parboil/common/mk/cuda.mk
deleted file mode 100644
index 00f205b2bb0dca42ea140f1d5dd4fa5f149fa178..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/mk/cuda.mk
+++ /dev/null
@@ -1,109 +0,0 @@
-# (c) 2007 The Board of Trustees of the University of Illinois.
-
-# Default language wide options
-
-# CUDA specific
-LANG_CFLAGS=-I$(PARBOIL_ROOT)/common/include -I$(CUDA_PATH)/include
-LANG_CXXFLAGS=$(LANG_CFLAGS)
-LANG_LDFLAGS=-L$(CUDA_LIB_PATH)
-
-LANG_CUDACFLAGS=$(LANG_CFLAGS)
-
-CFLAGS=$(APP_CFLAGS) $(LANG_CFLAGS) $(PLATFORM_CFLAGS)
-CXXFLAGS=$(APP_CXXFLAGS) $(LANG_CXXFLAGS) $(PLATFORM_CXXFLAGS)
-
-CUDACFLAGS=$(LANG_CUDACFLAGS) $(PLATFORM_CUDACFLAGS) $(APP_CUDACFLAGS)
-CUDALDFLAGS=$(LANG_LDFLAGS) $(PLATFORM_CUDALDFLAGS) $(APP_CUDALDFLAGS)
-
-# Rules common to all makefiles
-
-########################################
-# Functions
-########################################
-
-# Add BUILDDIR as a prefix to each element of $1
-INBUILDDIR=$(addprefix $(BUILDDIR)/,$(1))
-
-# Add SRCDIR as a prefix to each element of $1
-INSRCDIR=$(addprefix $(SRCDIR)/,$(1))
-
-
-########################################
-# Environment variable check
-########################################
-
-# The second-last directory in the $(BUILDDIR) path
-# must have the name "build". This reduces the risk of terrible
-# accidents if paths are not set up correctly.
-ifeq ("$(notdir $(BUILDDIR))", "")
-$(error $$BUILDDIR is not set correctly)
-endif
-
-ifneq ("$(notdir $(patsubst %/,%,$(dir $(BUILDDIR))))", "build")
-$(error $$BUILDDIR is not set correctly)
-endif
-
-.PHONY: run
-
-ifeq ($(CUDA_PATH),)
-FAILSAFE=no_cuda
-else
-FAILSAFE=
-endif
-
-########################################
-# Derived variables
-########################################
-
-ifeq ($(DEBUGGER),)
-DEBUGGER=gdb
-endif
-
-OBJS = $(call INBUILDDIR,$(SRCDIR_OBJS))
-
-########################################
-# Rules
-########################################
-
-default: $(FAILSAFE) $(BUILDDIR) $(BIN)
-
-run:
- @echo "Resolving CUDA runtime library..."
- @$(shell echo $(RUNTIME_ENV)) LD_LIBRARY_PATH=$(CUDA_LIB_PATH) ldd $(BIN) | grep cuda
- $(shell echo $(RUNTIME_ENV)) LD_LIBRARY_PATH=$(CUDA_LIB_PATH) ./$(BIN) $(ARGS)
-
-debug:
- @echo "Resolving CUDA runtime library..."
- @$(shell echo $(RUNTIME_ENV)) LD_LIBRARY_PATH=$(CUDA_LIB_PATH) ldd $(BIN) | grep cuda
- @$(shell echo $(RUNTIME_ENV)) LD_LIBRARY_PATH=$(CUDA_LIB_PATH) $(DEBUGGER) --args $(BIN) $(ARGS)
-
-clean :
- rm -rf $(BUILDDIR)/*
- if [ -d $(BUILDDIR) ]; then rmdir $(BUILDDIR); fi
-
-$(BIN) : $(OBJS) $(BUILDDIR)/parboil_cuda.o
- $(CUDALINK) $^ -o $@ $(CUDALDFLAGS)
-
-$(BUILDDIR) :
- mkdir -p $(BUILDDIR)
-
-$(BUILDDIR)/%.o : $(SRCDIR)/%.c
- $(CC) $(CFLAGS) -c $< -o $@
-
-$(BUILDDIR)/parboil_cuda.o: $(PARBOIL_ROOT)/common/src/parboil_cuda.c
- $(CC) $(CFLAGS) -c $< -o $@
-
-$(BUILDDIR)/%.o : $(SRCDIR)/%.cc
- $(CXX) $(CXXFLAGS) -c $< -o $@
-
-$(BUILDDIR)/%.o : $(SRCDIR)/%.cpp
- $(CXX) $(CXXFLAGS) -c $< -o $@
-
-$(BUILDDIR)/%.o : $(SRCDIR)/%.cu
- $(CUDACC) $< $(CUDACFLAGS) -c -o $@
-
-no_cuda:
- @echo "CUDA_PATH is not set. Open $(CUDA_ROOT)/common/Makefile.conf to set default value."
- @echo "You may use $(PLATFORM_MK) if you want a platform specific configurations."
- @exit 1
-
diff --git a/hpvm/test/parboil/common/mk/opencl.mk b/hpvm/test/parboil/common/mk/opencl.mk
deleted file mode 100644
index 35b4d1bc90c6fe776ca7e376e6883dd3bef687c9..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/mk/opencl.mk
+++ /dev/null
@@ -1,129 +0,0 @@
-# (c) 2007 The Board of Trustees of the University of Illinois.
-
-# Default language wide options
-
-LANG_CFLAGS=-I$(PARBOIL_ROOT)/common/include -I$(OPENCL_PATH)/include
-LANG_CXXFLAGS=$(LANG_CFLAGS)
-LANG_LDFLAGS=-lOpenCL -L$(OPENCL_LIB_PATH) -lrt
-
-CFLAGS=$(LANG_CFLAGS) $(PLATFORM_CFLAGS) $(APP_CFLAGS)
-CXXFLAGS=$(LANG_CXXFLAGS) $(PLATFORM_CXXFLAGS) $(APP_CXXFLAGS)
-LDFLAGS=$(LANG_LDFLAGS) $(PLATFORM_LDFLAGS) $(APP_LDFLAGS)
-
-
-LLVM_INSTALL:=$(LLVM_SRC_ROOT)/../build
-LIBCLC:=$(LLVM_SRC_ROOT)/../../libclc-install
-LLVM_CC:=$(LLVM_INSTALL)/bin/clang
-LLVM_LINK:=$(LLVM_INSTALL)/bin/llvm-link
-
-
-# Rules common to all makefiles
-
-########################################
-# Functions
-########################################
-
-# Add BUILDDIR as a prefix to each element of $1
-INBUILDDIR=$(addprefix $(BUILDDIR)/,$(1))
-
-# Add SRCDIR as a prefix to each element of $1
-INSRCDIR=$(addprefix $(SRCDIR)/,$(1))
-
-
-########################################
-# Environment variable check
-########################################
-
-# The second-last directory in the $(BUILDDIR) path
-# must have the name "build". This reduces the risk of terrible
-# accidents if paths are not set up correctly.
-ifeq ("$(notdir $(BUILDDIR))", "")
-$(error $$BUILDDIR is not set correctly)
-endif
-
-ifneq ("$(notdir $(patsubst %/,%,$(dir $(BUILDDIR))))", "build")
-$(error $$BUILDDIR is not set correctly)
-endif
-
-.PHONY: run
-.PRECIOUS: $(BUILDDIR)/%.ll
-
-ifeq ($(OPENCL_PATH),)
-FAILSAFE=no_opencl
-else
-FAILSAFE=
-endif
-
-########################################
-# Derived variables
-########################################
-
-OBJS = $(call INBUILDDIR,$(SRCDIR_OBJS))
-KERNEL = $(call INBUILDDIR,$(KERNEL_OBJS))
-
-ifeq ($(DEBUGGER),)
-DEBUGGER=gdb
-endif
-
-########################################
-# Rules
-########################################
-
-default: $(FAILSAFE) $(BUILDDIR) $(BIN) $(KERNEL)
-
-run : $(RUNDIR)
- echo "Resolving OpenCL library..."
- $(shell echo $(RUNTIME_ENV)) LD_LIBRARY_PATH=$(OPENCL_LIB_PATH) ldd ./$(BIN) | grep OpenCL
- $(shell echo $(RUNTIME_ENV)) LD_LIBRARY_PATH=$(OPENCL_LIB_PATH) ./$(BIN) $(ARGS)
- $(TOOL) $(OUTPUT) $(REF_OUTPUT)
-
-debug:
- @echo "Resolving OpenCL library..."
- @$(shell echo $(RUNTIME_ENV)) LD_LIBRARY_PATH=$(OPENCL_LIB_PATH) ldd $(BIN) | grep OpenCL
- @$(shell echo $(RUNTIME_ENV)) LD_LIBRARY_PATH=$(OPENCL_LIB_PATH) $(DEBUGGER) --args $(BIN) $(ARGS)
-
-clean :
- rm -f $(BUILDDIR)/*
- if [ -a $(BIN) ]; then rm $(BIN); fi
- if [ -d $(BUILDDIR) ]; then rm -rf $(BUILDDIR); fi
- if [ -d $(RUNDIR) ]; then rm -rf $(RUNDIR); fi
-
-$(BIN) : $(OBJS) $(BUILDDIR)/parboil_opencl.o
- $(CXX) $^ -o $@ $(LDFLAGS)
-
-$(RUNDIR) :
- mkdir -p $(RUNDIR)
-
-$(BUILDDIR) :
- mkdir -p $(BUILDDIR)
-
-$(BUILDDIR)/%.o : $(SRCDIR)/%.c
- $(CC) $(CFLAGS) -c $< -o $@
-
-$(BUILDDIR)/parboil_opencl.o : $(PARBOIL_ROOT)/common/src/parboil_opencl.c
- $(CC) $(CFLAGS) -c $< -o $@
-
-$(BUILDDIR)/%.o : $(SRCDIR)/%.cc
- $(CXX) $(CXXFLAGS) -c $< -o $@
-
-$(BUILDDIR)/%.o : $(SRCDIR)/%.cpp
- $(CXX) $(CXXFLAGS) -c $< -o $@
-
-$(BUILDDIR)/%.nvptx.s : $(BUILDDIR)/%.linked.bc
- $(LLVM_CC) -O3 -target nvptx64-nvidia-nvcl $< -S -o $@
-
-$(BUILDDIR)/%.linked.bc : $(BUILDDIR)/%.ll
- $(LLVM_LINK) $(LIBCLC)/lib/clc/nvptx64--nvidiacl.bc $< -o $@
-
-$(BUILDDIR)/%.ll : $(SRCDIR)/%.cl
- $(LLVM_CC) $(CFLAGS) -Dcl_clang_storage_class_specifiers -isystem $(LIBCLC)/include -include clc/clc.h -target nvptx64-nvidia-nvcl $< -O3 -emit-llvm -S -o $@
-
-$(BUILDDIR)/%.ir : $(SRCDIR)/%.cl
- cd $(SRCDIR); ioc64 -input=kernel.cl -ir=kernel.ir
- cp $(SRCDIR)/kernel.ir $@
-
-no_opencl:
- @echo "OPENCL_PATH is not set. Open $(PARBOIL_ROOT)/common/Makefile.conf to set default value."
- @echo "You may use $(PLATFORM_MK) if you want a platform specific configurations."
- @exit 1
-
diff --git a/hpvm/test/parboil/common/platform/c.default.mk b/hpvm/test/parboil/common/platform/c.default.mk
deleted file mode 100644
index d9058f11b1d913189a550f060f5d5dc9e08af2c9..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/platform/c.default.mk
+++ /dev/null
@@ -1,17 +0,0 @@
-# (c) 2007 The Board of Trustees of the University of Illinois.
-
-# Rules common to all makefiles
-
-# Commands to build objects from source file using C compiler
-# with gcc
-
-# gcc (default)
-CC = gcc
-PLATFORM_CFLAGS =
-
-CXX = g++
-PLATFORM_CXXFLAGS =
-
-LINKER = g++
-PLATFORM_LDFLAGS = -lm -lpthread
-
diff --git a/hpvm/test/parboil/common/platform/c.gcc.mk b/hpvm/test/parboil/common/platform/c.gcc.mk
deleted file mode 100644
index d9058f11b1d913189a550f060f5d5dc9e08af2c9..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/platform/c.gcc.mk
+++ /dev/null
@@ -1,17 +0,0 @@
-# (c) 2007 The Board of Trustees of the University of Illinois.
-
-# Rules common to all makefiles
-
-# Commands to build objects from source file using C compiler
-# with gcc
-
-# gcc (default)
-CC = gcc
-PLATFORM_CFLAGS =
-
-CXX = g++
-PLATFORM_CXXFLAGS =
-
-LINKER = g++
-PLATFORM_LDFLAGS = -lm -lpthread
-
diff --git a/hpvm/test/parboil/common/platform/cuda.arch20.mk b/hpvm/test/parboil/common/platform/cuda.arch20.mk
deleted file mode 100644
index b08376c93195fda7ad477e89a9a93b0212ca1161..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/platform/cuda.arch20.mk
+++ /dev/null
@@ -1,23 +0,0 @@
-# (c) 2007 The Board of Trustees of the University of Illinois.
-
-# Cuda-related definitions common to all benchmarks
-
-########################################
-# Variables
-########################################
-
-# c.default is the base along with CUDA configuration in this setting
-include $(PARBOIL_ROOT)/common/platform/c.default.mk
-
-# Paths
-CUDAHOME=/usr/local/cuda
-
-# Programs
-CUDACC=$(CUDAHOME)/bin/nvcc
-CUDALINK=$(CUDAHOME)/bin/nvcc
-
-# Flags
-PLATFORM_CUDACFLAGS=-code=sm_20
-PLATFORM_CUDALDFLAGS=-lm -lpthread
-
-
diff --git a/hpvm/test/parboil/common/platform/cuda.default.mk b/hpvm/test/parboil/common/platform/cuda.default.mk
deleted file mode 100644
index 237da5aa3e601ed1fc53cb7fc5fb8c986c738eec..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/platform/cuda.default.mk
+++ /dev/null
@@ -1,23 +0,0 @@
-# (c) 2007 The Board of Trustees of the University of Illinois.
-
-# Cuda-related definitions common to all benchmarks
-
-########################################
-# Variables
-########################################
-
-# c.default is the base along with CUDA configuration in this setting
-include $(PARBOIL_ROOT)/common/platform/c.default.mk
-
-# Paths
-CUDAHOME=/usr/local/cuda
-
-# Programs
-CUDACC=$(CUDAHOME)/bin/nvcc
-CUDALINK=$(CUDAHOME)/bin/nvcc
-
-# Flags
-PLATFORM_CUDACFLAGS=-O3
-PLATFORM_CUDALDFLAGS=-lm -lpthread
-
-
diff --git a/hpvm/test/parboil/common/platform/cuda.mcuda.mk b/hpvm/test/parboil/common/platform/cuda.mcuda.mk
deleted file mode 100644
index 05a6fee9a9a0328d6ff415afdb4b9acbf91580fa..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/platform/cuda.mcuda.mk
+++ /dev/null
@@ -1,25 +0,0 @@
-# (c) 2011 The Board of Trustees of the University of Illinois.
-
-# Cuda-related definitions common to all benchmarks
-
-########################################
-# Variables
-########################################
-
-# c.default is the base along with CUDA configuration in this setting
-include $(PARBOIL_ROOT)/common/platform/c.default.mk
-
-# Paths
-CUDA_PATH=$(MCUDA_PATH)/include
-
-# Programs
-CUDACC=$(MCUDA_PATH)/bin/mcc_xmm
-CUDALINK=$(LINKER)
-
-# Flags
-PLATFORM_CUDACFLAGS=-O3
-PLATFORM_CFLAGS=-O3 -I$(MCUDA_PATH)/include -D__MCUDA__
-PLATFORM_CXXFLAGS=-O3 -I$(MCUDA_PATH)/include -D__MCUDA__
-PLATFORM_CUDALDFLAGS=-lm -lpthread -L$(MCUDA_PATH)/lib -lmcuda
-
-
diff --git a/hpvm/test/parboil/common/platform/cuda.nvcc.mk b/hpvm/test/parboil/common/platform/cuda.nvcc.mk
deleted file mode 100644
index 237da5aa3e601ed1fc53cb7fc5fb8c986c738eec..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/platform/cuda.nvcc.mk
+++ /dev/null
@@ -1,23 +0,0 @@
-# (c) 2007 The Board of Trustees of the University of Illinois.
-
-# Cuda-related definitions common to all benchmarks
-
-########################################
-# Variables
-########################################
-
-# c.default is the base along with CUDA configuration in this setting
-include $(PARBOIL_ROOT)/common/platform/c.default.mk
-
-# Paths
-CUDAHOME=/usr/local/cuda
-
-# Programs
-CUDACC=$(CUDAHOME)/bin/nvcc
-CUDALINK=$(CUDAHOME)/bin/nvcc
-
-# Flags
-PLATFORM_CUDACFLAGS=-O3
-PLATFORM_CUDALDFLAGS=-lm -lpthread
-
-
diff --git a/hpvm/test/parboil/common/platform/opencl.default.mk b/hpvm/test/parboil/common/platform/opencl.default.mk
deleted file mode 100644
index a6ac4645dceb8097c304784dd66280d4f53d1f4f..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/platform/opencl.default.mk
+++ /dev/null
@@ -1,23 +0,0 @@
-# (c) 2007 The Board of Trustees of the University of Illinois.
-
-# Rules common to all makefiles
-
-# Commands to build objects from source file using C compiler
-# with gcc
-
-# Uncomment below two lines and configure if you want to use a platform
-# other than global one
-
-#OPENCL_PATH=/scr/hskim/ati-stream-sdk-v2.3-lnx64
-#OPENCL_LIB_PATH=$(OPENCL_PATH)/lib/x86_64
-
-# gcc (default)
-CC = clang
-PLATFORM_CFLAGS =
-
-CXX = clang++
-PLATFORM_CXXFLAGS =
-
-LINKER = clang++
-PLATFORM_LDFLAGS = -lm -lpthread
-
diff --git a/hpvm/test/parboil/common/platform/opencl.gcc.mk b/hpvm/test/parboil/common/platform/opencl.gcc.mk
deleted file mode 100644
index d9058f11b1d913189a550f060f5d5dc9e08af2c9..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/platform/opencl.gcc.mk
+++ /dev/null
@@ -1,17 +0,0 @@
-# (c) 2007 The Board of Trustees of the University of Illinois.
-
-# Rules common to all makefiles
-
-# Commands to build objects from source file using C compiler
-# with gcc
-
-# gcc (default)
-CC = gcc
-PLATFORM_CFLAGS =
-
-CXX = g++
-PLATFORM_CXXFLAGS =
-
-LINKER = g++
-PLATFORM_LDFLAGS = -lm -lpthread
-
diff --git a/hpvm/test/parboil/common/src/parboil_cuda.c b/hpvm/test/parboil/common/src/parboil_cuda.c
deleted file mode 100644
index 9fd64661643c9afec5cb470beaa516d545017bd3..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/src/parboil_cuda.c
+++ /dev/null
@@ -1,863 +0,0 @@
-/*
- * (c) 2007 The Board of Trustees of the University of Illinois.
- */
-
-#include <parboil.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#ifndef __MCUDA__
-#include <cuda_runtime_api.h>
-#else
-#include <mcuda.h>
-#endif
-
-#if _POSIX_VERSION >= 200112L
-#include <sys/time.h>
-#endif
-
-#define true 1
-
-/* Free an array of owned strings. */
-static void free_string_array(char **string_array) {
- char **p;
-
- if (!string_array)
- return;
- for (p = string_array; *p; p++)
- free(*p);
- free(string_array);
-}
-
-/* Parse a comma-delimited list of strings into an
- * array of strings. */
-static char **read_string_array(char *in) {
- char **ret;
- int i;
- int count; /* Number of items in the input */
- char *substring; /* Current substring within 'in' */
-
- /* Count the number of items in the string */
- count = 1;
- for (i = 0; in[i]; i++)
- if (in[i] == ',')
- count++;
-
- /* Allocate storage */
- ret = (char **)malloc((count + 1) * sizeof(char *));
-
- /* Create copies of the strings from the list */
- substring = in;
- for (i = 0; i < count; i++) {
- char *substring_end;
- int substring_length;
-
- /* Find length of substring */
- for (substring_end = substring;
- (*substring_end != ',') && (*substring_end != 0); substring_end++)
- ;
-
- substring_length = substring_end - substring;
-
- /* Allocate memory and copy the substring */
- ret[i] = (char *)malloc(substring_length + 1);
- memcpy(ret[i], substring, substring_length);
- ret[i][substring_length] = 0;
-
- /* go to next substring */
- substring = substring_end + 1;
- }
- ret[i] = NULL; /* Write the sentinel value */
-
- return ret;
-}
-
-struct argparse {
- int argc; /* Number of arguments. Mutable. */
- char **argv; /* Argument values. Immutable. */
-
- int argn; /* Current argument number. */
- char **argv_get; /* Argument value being read. */
- char **argv_put; /* Argument value being written.
- * argv_put <= argv_get. */
-};
-
-static void initialize_argparse(struct argparse *ap, int argc, char **argv) {
- ap->argc = argc;
- ap->argn = 0;
- ap->argv_get = ap->argv_put = ap->argv = argv;
-}
-
-static void finalize_argparse(struct argparse *ap) {
- /* Move the remaining arguments */
- for (; ap->argn < ap->argc; ap->argn++)
- *ap->argv_put++ = *ap->argv_get++;
-}
-
-/* Delete the current argument. */
-static void delete_argument(struct argparse *ap) {
- if (ap->argn >= ap->argc) {
- fprintf(stderr, "delete_argument\n");
- }
- ap->argc--;
- ap->argv_get++;
-}
-
-/* Go to the next argument. Also, move the current argument to its
- * final location in argv. */
-static void next_argument(struct argparse *ap) {
- if (ap->argn >= ap->argc) {
- fprintf(stderr, "next_argument\n");
- }
- /* Move argument to its new location. */
- *ap->argv_put++ = *ap->argv_get++;
- ap->argn++;
-}
-
-static int is_end_of_arguments(struct argparse *ap) {
- return ap->argn == ap->argc;
-}
-
-static char *get_argument(struct argparse *ap) { return *ap->argv_get; }
-
-static char *consume_argument(struct argparse *ap) {
- char *ret = get_argument(ap);
- delete_argument(ap);
- return ret;
-}
-
-struct pb_Parameters *pb_ReadParameters(int *_argc, char **argv) {
- char *err_message;
- struct argparse ap;
- struct pb_Parameters *ret =
- (struct pb_Parameters *)malloc(sizeof(struct pb_Parameters));
-
- /* Initialize the parameters structure */
- ret->outFile = NULL;
- ret->inpFiles = (char **)malloc(sizeof(char *));
- ret->inpFiles[0] = NULL;
-
- /* Each argument */
- initialize_argparse(&ap, *_argc, argv);
- while (!is_end_of_arguments(&ap)) {
- char *arg = get_argument(&ap);
-
- /* Single-character flag */
- if ((arg[0] == '-') && (arg[1] != 0) && (arg[2] == 0)) {
- delete_argument(&ap); /* This argument is consumed here */
-
- switch (arg[1]) {
- case 'o': /* Output file name */
- if (is_end_of_arguments(&ap)) {
- err_message = "Expecting file name after '-o'\n";
- goto error;
- }
- free(ret->outFile);
- ret->outFile = strdup(consume_argument(&ap));
- break;
- case 'i': /* Input file name */
- if (is_end_of_arguments(&ap)) {
- err_message = "Expecting file name after '-i'\n";
- goto error;
- }
- ret->inpFiles = read_string_array(consume_argument(&ap));
- break;
- case '-': /* End of options */
- goto end_of_options;
- default:
- err_message = "Unexpected command-line parameter\n";
- goto error;
- }
- } else {
- /* Other parameters are ignored */
- next_argument(&ap);
- }
- } /* end for each argument */
-
-end_of_options:
- *_argc = ap.argc; /* Save the modified argc value */
- finalize_argparse(&ap);
-
- return ret;
-
-error:
- fputs(err_message, stderr);
- pb_FreeParameters(ret);
- return NULL;
-}
-
-void pb_FreeParameters(struct pb_Parameters *p) {
- char **cpp;
-
- free(p->outFile);
- free_string_array(p->inpFiles);
- free(p);
-}
-
-int pb_Parameters_CountInputs(struct pb_Parameters *p) {
- int n;
-
- for (n = 0; p->inpFiles[n]; n++)
- ;
- return n;
-}
-
-/*****************************************************************************/
-/* Timer routines */
-
-static int is_async(enum pb_TimerID timer) {
- return (timer == pb_TimerID_KERNEL) || (timer == pb_TimerID_COPY_ASYNC);
-}
-
-static int is_blocking(enum pb_TimerID timer) {
- return (timer == pb_TimerID_COPY) || (timer == pb_TimerID_NONE);
-}
-
-#define INVALID_TIMERID pb_TimerID_LAST
-
-static int asyncs_outstanding(struct pb_TimerSet *timers) {
- return (timers->async_markers != NULL) &&
- (timers->async_markers->timerID != INVALID_TIMERID);
-}
-
-static struct pb_async_time_marker_list *
-get_last_async(struct pb_TimerSet *timers) {
- /* Find the last event recorded thus far */
- struct pb_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)
- last_event = last_event->next;
- return last_event;
- } else
- return NULL;
-}
-
-static void insert_marker(struct pb_TimerSet *tset, enum pb_TimerID timer) {
- struct pb_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 pb_async_time_marker_list *)malloc(
- sizeof(struct pb_async_time_marker_list));
- (*new_event)->marker = malloc(sizeof(cudaEvent_t));
- cudaEventCreate((*new_event)->marker);
- (*new_event)->next = NULL;
- }
-
- /* valid event handle now aquired: insert the event record */
- (*new_event)->label = NULL;
- (*new_event)->timerID = timer;
- cudaEventRecord(*((cudaEvent_t *)((*new_event)->marker)), 0);
-}
-
-static void insert_submarker(struct pb_TimerSet *tset, char *label,
- enum pb_TimerID timer) {
- struct pb_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 pb_async_time_marker_list *)malloc(
- sizeof(struct pb_async_time_marker_list));
- (*new_event)->marker = malloc(sizeof(cudaEvent_t));
- cudaEventCreate((*new_event)->marker);
-
- (*new_event)->next = NULL;
- }
-
- /* valid event handle now aquired: insert the event record */
- (*new_event)->label = label;
- (*new_event)->timerID = timer;
- cudaEventRecord(*((cudaEvent_t *)((*new_event)->marker)), 0);
-}
-
-/* Assumes that all recorded events have completed */
-static pb_Timestamp record_async_times(struct pb_TimerSet *tset) {
- struct pb_async_time_marker_list *next_interval = NULL;
- struct pb_async_time_marker_list *last_marker = get_last_async(tset);
- pb_Timestamp total_async_time = 0;
- enum pb_TimerID timer;
- for (next_interval = tset->async_markers; next_interval != last_marker;
- next_interval = next_interval->next) {
- float interval_time_ms;
- cudaEventElapsedTime(&interval_time_ms,
- *((cudaEvent_t *)next_interval->marker),
- *((cudaEvent_t *)next_interval->next->marker));
- pb_Timestamp interval = (pb_Timestamp)(interval_time_ms * 1e3);
- tset->timers[next_interval->timerID].elapsed += interval;
- if (next_interval->label != NULL) {
- struct pb_SubTimer *subtimer =
- tset->sub_timer_list[next_interval->timerID]->subtimer_list;
- while (subtimer != NULL) {
- if (strcmp(subtimer->label, next_interval->label) == 0) {
- subtimer->timer.elapsed += interval;
- break;
- }
- subtimer = subtimer->next;
- }
- }
- total_async_time += interval;
- next_interval->timerID = INVALID_TIMERID;
- }
-
- if (next_interval != NULL)
- next_interval->timerID = INVALID_TIMERID;
-
- return total_async_time;
-}
-
-static void accumulate_time(pb_Timestamp *accum, pb_Timestamp start,
- pb_Timestamp end) {
-#if _POSIX_VERSION >= 200112L
- *accum += end - start;
-#else
-#error "Timestamps not implemented for this system"
-#endif
-}
-
-#if _POSIX_VERSION >= 200112L
-static pb_Timestamp get_time() {
- struct timeval tv;
- gettimeofday(&tv, NULL);
- return (pb_Timestamp)(tv.tv_sec * 1000000LL + tv.tv_usec);
-}
-#else
-#error "no supported time libraries are available on this platform"
-#endif
-
-void pb_ResetTimer(struct pb_Timer *timer) {
- timer->state = pb_Timer_STOPPED;
-
-#if _POSIX_VERSION >= 200112L
- timer->elapsed = 0;
-#else
-#error "pb_ResetTimer: not implemented for this system"
-#endif
-}
-
-void pb_StartTimer(struct pb_Timer *timer) {
- if (timer->state != pb_Timer_STOPPED) {
- fputs("Ignoring attempt to start a running timer\n", stderr);
- return;
- }
-
- timer->state = pb_Timer_RUNNING;
-
-#if _POSIX_VERSION >= 200112L
- {
- struct timeval tv;
- gettimeofday(&tv, NULL);
- timer->init = tv.tv_sec * 1000000LL + tv.tv_usec;
- }
-#else
-#error "pb_StartTimer: not implemented for this system"
-#endif
-}
-
-void pb_StartTimerAndSubTimer(struct pb_Timer *timer,
- struct pb_Timer *subtimer) {
-
- unsigned int numNotStopped = 0x3; // 11
- if (timer->state != pb_Timer_STOPPED) {
- fputs("Warning: Timer was not stopped\n", stderr);
- numNotStopped &= 0x1; // Zero out 2^1
- }
- if (subtimer->state != pb_Timer_STOPPED) {
- fputs("Warning: Subtimer was not stopped\n", stderr);
- numNotStopped &= 0x2; // Zero out 2^0
- }
- if (numNotStopped == 0x0) {
- fputs("Ignoring attempt to start running timer and subtimer\n", stderr);
- return;
- }
-
- timer->state = pb_Timer_RUNNING;
- subtimer->state = pb_Timer_RUNNING;
-
-#if _POSIX_VERSION >= 200112L
- {
- struct timeval tv;
- gettimeofday(&tv, NULL);
-
- if (numNotStopped & 0x2) {
- timer->init = tv.tv_sec * 1000000LL + tv.tv_usec;
- }
-
- if (numNotStopped & 0x1) {
- subtimer->init = tv.tv_sec * 1000000LL + tv.tv_usec;
- }
- }
-#else
-#error "pb_StartTimer: not implemented for this system"
-#endif
-}
-
-void pb_StopTimer(struct pb_Timer *timer) {
- pb_Timestamp fini;
-
- if (timer->state != pb_Timer_RUNNING) {
- fputs("Ignoring attempt to stop a stopped timer\n", stderr);
- return;
- }
-
- timer->state = pb_Timer_STOPPED;
-
-#if _POSIX_VERSION >= 200112L
- {
- struct timeval tv;
- gettimeofday(&tv, NULL);
- fini = tv.tv_sec * 1000000LL + tv.tv_usec;
- }
-#else
-#error "pb_StopTimer: not implemented for this system"
-#endif
-
- accumulate_time(&timer->elapsed, timer->init, fini);
- timer->init = fini;
-}
-
-void pb_StopTimerAndSubTimer(struct pb_Timer *timer,
- struct pb_Timer *subtimer) {
-
- pb_Timestamp fini;
-
- unsigned int numNotRunning = 0x3; // 11
- if (timer->state != pb_Timer_RUNNING) {
- fputs("Warning: Timer was not running\n", stderr);
- numNotRunning &= 0x1; // Zero out 2^1
- }
- if (subtimer->state != pb_Timer_RUNNING) {
- fputs("Warning: Subtimer was not running\n", stderr);
- numNotRunning &= 0x2; // Zero out 2^0
- }
- if (numNotRunning == 0x0) {
- fputs("Ignoring attempt to stop stopped timer and subtimer\n", stderr);
- return;
- }
-
- timer->state = pb_Timer_STOPPED;
- subtimer->state = pb_Timer_STOPPED;
-
-#if _POSIX_VERSION >= 200112L
- {
- struct timeval tv;
- gettimeofday(&tv, NULL);
- fini = tv.tv_sec * 1000000LL + tv.tv_usec;
- }
-#else
-#error "pb_StopTimer: not implemented for this system"
-#endif
-
- if (numNotRunning & 0x2) {
- accumulate_time(&timer->elapsed, timer->init, fini);
- timer->init = fini;
- }
-
- if (numNotRunning & 0x1) {
- accumulate_time(&subtimer->elapsed, subtimer->init, fini);
- subtimer->init = fini;
- }
-}
-
-/* Get the elapsed time in seconds. */
-double pb_GetElapsedTime(struct pb_Timer *timer) {
- double ret;
-
- if (timer->state != pb_Timer_STOPPED) {
- fputs("Elapsed time from a running timer is inaccurate\n", stderr);
- }
-
-#if _POSIX_VERSION >= 200112L
- ret = timer->elapsed / 1e6;
-#else
-#error "pb_GetElapsedTime: not implemented for this system"
-#endif
- return ret;
-}
-
-void pb_InitializeTimerSet(struct pb_TimerSet *timers) {
- int n;
-
- timers->wall_begin = get_time();
- timers->current = pb_TimerID_NONE;
-
- timers->async_markers = NULL;
-
- for (n = 0; n < pb_TimerID_LAST; n++) {
- pb_ResetTimer(&timers->timers[n]);
- timers->sub_timer_list[n] = NULL;
- }
-}
-
-void pb_SetOpenCL(void *clContextPtr, void *clCommandQueuePtr) {}
-
-void pb_AddSubTimer(struct pb_TimerSet *timers, char *label,
- enum pb_TimerID pb_Category) {
-
- struct pb_SubTimer *subtimer =
- (struct pb_SubTimer *)malloc(sizeof(struct pb_SubTimer));
-
- int len = strlen(label);
-
- subtimer->label = (char *)malloc(sizeof(char) * (len + 1));
- sprintf(subtimer->label, "%s\0", label);
-
- pb_ResetTimer(&subtimer->timer);
- subtimer->next = NULL;
-
- struct pb_SubTimerList *subtimerlist = timers->sub_timer_list[pb_Category];
- if (subtimerlist == NULL) {
- subtimerlist =
- (struct pb_SubTimerList *)malloc(sizeof(struct pb_SubTimerList));
- subtimerlist->subtimer_list = subtimer;
- timers->sub_timer_list[pb_Category] = subtimerlist;
- } else {
- // Append to list
- struct pb_SubTimer *element = subtimerlist->subtimer_list;
- while (element->next != NULL) {
- element = element->next;
- }
- element->next = subtimer;
- }
-}
-
-void pb_SwitchToTimer(struct pb_TimerSet *timers, enum pb_TimerID timer) {
- /* Stop the currently running timer */
- if (timers->current != pb_TimerID_NONE) {
- struct pb_SubTimerList *subtimerlist =
- timers->sub_timer_list[timers->current];
- struct pb_SubTimer *currSubTimer =
- (subtimerlist != NULL) ? subtimerlist->current : NULL;
-
- if (!is_async(timers->current)) {
- if (timers->current != timer) {
- if (currSubTimer != NULL) {
- pb_StopTimerAndSubTimer(&timers->timers[timers->current],
- &currSubTimer->timer);
- } else {
- pb_StopTimer(&timers->timers[timers->current]);
- }
- } else {
- if (currSubTimer != NULL) {
- pb_StopTimer(&currSubTimer->timer);
- }
- }
- } else {
- insert_marker(timers, timer);
- if (!is_async(timer)) { // if switching to async too, keep driver going
- pb_StopTimer(&timers->timers[pb_TimerID_DRIVER]);
- }
- }
- }
-
- pb_Timestamp currentTime = get_time();
-
- /* The only cases we check for asynchronous task completion is
- * when an overlapping CPU operation completes, or the next
- * segment blocks on completion of previous async operations */
- if (asyncs_outstanding(timers) &&
- (!is_async(timers->current) || is_blocking(timer))) {
-
- struct pb_async_time_marker_list *last_event = get_last_async(timers);
- /* cudaSuccess if completed */
- cudaError_t async_done =
- cudaEventQuery(*((cudaEvent_t *)last_event->marker));
-
- if (is_blocking(timer)) {
- /* Async operations completed after previous CPU operations:
- * overlapped time is the total CPU time since this set of async
- * operations were first issued */
-
- // timer to switch to is COPY or NONE
- if (async_done != cudaSuccess)
- accumulate_time(&(timers->timers[pb_TimerID_OVERLAP].elapsed),
- timers->async_begin, currentTime);
-
- /* Wait on async operation completion */
- cudaEventSynchronize(*((cudaEvent_t *)last_event->marker));
- pb_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 == cudaSuccess)
- timers->timers[pb_TimerID_OVERLAP].elapsed += total_async_time;
-
- } else
- /* implies (!is_async(timers->current) && asyncs_outstanding(timers)) */
- // i.e. Current Not Async (not KERNEL/COPY_ASYNC) but there are
- // outstanding so something is deeper in stack
- if (async_done == cudaSuccess) {
- /* Async operations completed before previous CPU operations:
- * overlapped time is the total async time */
- timers->timers[pb_TimerID_OVERLAP].elapsed += record_async_times(timers);
- }
- }
-
- /* Start the new timer */
- if (timer != pb_TimerID_NONE) {
- if (!is_async(timer)) {
- pb_StartTimer(&timers->timers[timer]);
- } else {
- // toSwitchTo Is Async (KERNEL/COPY_ASYNC)
- if (!asyncs_outstanding(timers)) {
- /* No asyncs outstanding, insert a fresh async marker */
-
- insert_marker(timers, timer);
- timers->async_begin = currentTime;
- } else if (!is_async(timers->current)) {
- /* Previous asyncs still in flight, but a previous SwitchTo
- * already marked the end of the most recent async operation,
- * so we can rename that marker as the beginning of this async
- * operation */
-
- struct pb_async_time_marker_list *last_event = get_last_async(timers);
- last_event->label = NULL;
- last_event->timerID = timer;
- }
- if (!is_async(timers->current)) {
- pb_StartTimer(&timers->timers[pb_TimerID_DRIVER]);
- }
- }
- }
- timers->current = timer;
-}
-
-void pb_SwitchToSubTimer(struct pb_TimerSet *timers, char *label,
- enum pb_TimerID category) {
- struct pb_SubTimerList *subtimerlist =
- timers->sub_timer_list[timers->current];
- struct pb_SubTimer *curr =
- (subtimerlist != NULL) ? subtimerlist->current : NULL;
-
- if (timers->current != pb_TimerID_NONE) {
- if (!is_async(timers->current)) {
- if (timers->current != category) {
- if (curr != NULL) {
- pb_StopTimerAndSubTimer(&timers->timers[timers->current],
- &curr->timer);
- } else {
- pb_StopTimer(&timers->timers[timers->current]);
- }
- } else {
- if (curr != NULL) {
- pb_StopTimer(&curr->timer);
- }
- }
- } else {
- insert_submarker(timers, label, category);
- if (!is_async(category)) { // if switching to async too, keep driver going
- pb_StopTimer(&timers->timers[pb_TimerID_DRIVER]);
- }
- }
- }
-
- pb_Timestamp currentTime = get_time();
-
- /* The only cases we check for asynchronous task completion is
- * when an overlapping CPU operation completes, or the next
- * segment blocks on completion of previous async operations */
- if (asyncs_outstanding(timers) &&
- (!is_async(timers->current) || is_blocking(category))) {
-
- struct pb_async_time_marker_list *last_event = get_last_async(timers);
- /* cudaSuccess if completed */
- cudaError_t async_done =
- cudaEventQuery(*((cudaEvent_t *)last_event->marker));
-
- if (is_blocking(category)) {
- /* Async operations completed after previous CPU operations:
- * overlapped time is the total CPU time since this set of async
- * operations were first issued */
-
- // timer to switch to is COPY or NONE
- // if it hasn't already finished, then just take now and use that as the
- // elapsed time in OVERLAP anything happening after now isn't OVERLAP
- // 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 != cudaSuccess)
- accumulate_time(&(timers->timers[pb_TimerID_OVERLAP].elapsed),
- timers->async_begin, currentTime);
-
- /* Wait on async operation completion */
- cudaEventSynchronize(*((cudaEvent_t *)last_event->marker));
- pb_Timestamp total_async_time = record_async_times(timers);
-
- /* Async operations completed before previous CPU operations:
- * overlapped time is the total async time */
- // If it did finish, then accumulate all the async time that did happen
- // into OVERLAP the immediately preceding EventSynchronize theoretically
- // didn't have any effect since it was already completed.
- if (async_done == cudaSuccess)
- timers->timers[pb_TimerID_OVERLAP].elapsed += total_async_time;
-
- } else
- /* implies (!is_async(timers->current) && asyncs_outstanding(timers)) */
- // i.e. Current Not Async (not KERNEL/COPY_ASYNC) but there are
- // outstanding so something is deeper in stack
- if (async_done == cudaSuccess) {
- /* Async operations completed before previous CPU operations:
- * overlapped time is the total async time */
- timers->timers[pb_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 pb_SubTimer *subtimer = NULL;
-
- if (label != NULL) {
- subtimer = subtimerlist->subtimer_list;
- while (subtimer != NULL) {
- if (strcmp(subtimer->label, label) == 0) {
- break;
- } else {
- subtimer = subtimer->next;
- }
- }
- }
-
- /* Start the new timer */
- if (category != pb_TimerID_NONE) {
- if (!is_async(category)) {
-
- if (subtimerlist != NULL) {
- subtimerlist->current = subtimer;
- }
-
- if (category != timers->current && subtimer != NULL) {
- pb_StartTimerAndSubTimer(&timers->timers[category], &subtimer->timer);
- } else if (subtimer != NULL) {
- pb_StartTimer(&subtimer->timer);
- } else {
- pb_StartTimer(&timers->timers[category]);
- }
- } else {
- if (subtimerlist != NULL) {
- subtimerlist->current = subtimer;
- }
-
- // toSwitchTo Is Async (KERNEL/COPY_ASYNC)
- if (!asyncs_outstanding(timers)) {
- /* No asyncs outstanding, insert a fresh async marker */
- insert_submarker(timers, label, category);
- timers->async_begin = currentTime;
- } else if (!is_async(timers->current)) {
- /* Previous asyncs still in flight, but a previous SwitchTo
- * already marked the end of the most recent async operation,
- * so we can rename that marker as the beginning of this async
- * operation */
-
- struct pb_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
-
- // toSwitchto is already asynchronous, but if current/prev state is async
- // too, then DRIVER is already running
- if (!is_async(timers->current)) {
- pb_StartTimer(&timers->timers[pb_TimerID_DRIVER]);
- }
- }
- }
-
- timers->current = category;
-}
-
-void pb_PrintTimerSet(struct pb_TimerSet *timers) {
- pb_Timestamp wall_end = get_time();
-
- struct pb_Timer *t = timers->timers;
- struct pb_SubTimer *sub = NULL;
-
- int maxSubLength;
-
- // const char *categories[] = {
- // "IO", "Kernel", "Copy", "Driver", "Copy Async", "Compute"
- // };
- const char *categories[] = {
- "IO", "Kernel", "Copy", "Driver",
- "Copy Async", "Compute", "Overlap", "Init_Ctx",
- "Clear_Ctx", "Copy_Scalar", "Copy_Ptr", "Mem_Free",
- "Read_Output", "Setup", "Mem_Track", "Mem_Untrack",
- "Misc", "Pthread_Create", "Arg_Unpack", "Computation",
- "Output_Pack", "Output_Unpack"};
-
- const int maxCategoryLength = 10;
-
- int i;
- for (i = 1; i < pb_TimerID_LAST;
- ++i) { // exclude NONE and OVRELAP from this format
- if (pb_GetElapsedTime(&t[i]) != 0 || true) {
-
- // Print Category Timer
- printf("%-*s: %f\n", maxCategoryLength, categories[i - 1],
- pb_GetElapsedTime(&t[i]));
-
- if (timers->sub_timer_list[i] != NULL) {
- sub = timers->sub_timer_list[i]->subtimer_list;
- maxSubLength = 0;
- while (sub != NULL) {
- // Find longest SubTimer label
- if (strlen(sub->label) > maxSubLength) {
- maxSubLength = strlen(sub->label);
- }
- sub = sub->next;
- }
-
- // Fit to Categories
- if (maxSubLength <= maxCategoryLength) {
- maxSubLength = maxCategoryLength;
- }
-
- sub = timers->sub_timer_list[i]->subtimer_list;
-
- // Print SubTimers
- while (sub != NULL) {
- printf(" -%-*s: %f\n", maxSubLength, sub->label,
- pb_GetElapsedTime(&sub->timer));
- sub = sub->next;
- }
- }
- }
- }
-
- if (pb_GetElapsedTime(&t[pb_TimerID_OVERLAP]) != 0)
- printf("CPU/Kernel Overlap: %f\n",
- pb_GetElapsedTime(&t[pb_TimerID_OVERLAP]));
-
- float walltime = (wall_end - timers->wall_begin) / 1e6;
- printf("Timer Wall Time: %f\n", walltime);
-}
-
-void pb_DestroyTimerSet(struct pb_TimerSet *timers) {
- /* clean up all of the async event markers */
- struct pb_async_time_marker_list **event = &(timers->async_markers);
- while (*event != NULL) {
- cudaEventSynchronize(*((cudaEvent_t *)(*event)->marker));
- cudaEventDestroy(*((cudaEvent_t *)(*event)->marker));
- free((*event)->marker);
- struct pb_async_time_marker_list **next = &((*event)->next);
- free(*event);
- (*event) = NULL;
- event = next;
- }
-
- int i = 0;
- for (i = 0; i < pb_TimerID_LAST; ++i) {
- if (timers->sub_timer_list[i] != NULL) {
- struct pb_SubTimer *subtimer = timers->sub_timer_list[i]->subtimer_list;
- struct pb_SubTimer *prev = NULL;
- while (subtimer != NULL) {
- free(subtimer->label);
- prev = subtimer;
- subtimer = subtimer->next;
- free(prev);
- }
- free(timers->sub_timer_list[i]);
- }
- }
-}
diff --git a/hpvm/test/parboil/common/src/parboil_opencl.c b/hpvm/test/parboil/common/src/parboil_opencl.c
deleted file mode 100644
index d493992acee859186d58330a9988ef7ef2571f73..0000000000000000000000000000000000000000
--- a/hpvm/test/parboil/common/src/parboil_opencl.c
+++ /dev/null
@@ -1,1051 +0,0 @@
-/*
- * (c) 2007 The Board of Trustees of the University of Illinois.
- */
-
-#include <CL/cl.h>
-#include <assert.h>
-#include <parboil.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-
-#if _POSIX_VERSION >= 200112L
-#include <time.h>
-#endif
-
-#define BILLION 1000000000LL
-#define true 1
-
-cl_context *clContextPtr;
-cl_command_queue *clCommandQueuePtr;
-
-/* Free an array of owned strings. */
-static void free_string_array(char **string_array) {
- char **p;
-
- if (!string_array)
- return;
- for (p = string_array; *p; p++)
- free(*p);
- free(string_array);
-}
-
-/* Parse a comma-delimited list of strings into an
- * array of strings. */
-static char **read_string_array(char *in) {
- char **ret;
- int i;
- int count; /* Number of items in the input */
- char *substring; /* Current substring within 'in' */
-
- /* Count the number of items in the string */
- count = 1;
- for (i = 0; in[i]; i++)
- if (in[i] == ',')
- count++;
-
- /* Allocate storage */
- ret = (char **)malloc((count + 1) * sizeof(char *));
-
- /* Create copies of the strings from the list */
- substring = in;
- for (i = 0; i < count; i++) {
- char *substring_end;
- int substring_length;
-
- /* Find length of substring */
- for (substring_end = substring;
- (*substring_end != ',') && (*substring_end != 0); substring_end++)
- ;
-
- substring_length = substring_end - substring;
-
- /* Allocate memory and copy the substring */
- ret[i] = (char *)malloc(substring_length + 1);
- memcpy(ret[i], substring, substring_length);
- ret[i][substring_length] = 0;
-
- /* go to next substring */
- substring = substring_end + 1;
- }
- ret[i] = NULL; /* Write the sentinel value */
-
- return ret;
-}
-
-struct argparse {
- int argc; /* Number of arguments. Mutable. */
- char **argv; /* Argument values. Immutable. */
-
- int argn; /* Current argument number. */
- char **argv_get; /* Argument value being read. */
- char **argv_put; /* Argument value being written.
- * argv_put <= argv_get. */
-};
-
-static void initialize_argparse(struct argparse *ap, int argc, char **argv) {
- ap->argc = argc;
- ap->argn = 0;
- ap->argv_get = ap->argv_put = ap->argv = argv;
-}
-
-static void finalize_argparse(struct argparse *ap) {
- /* Move the remaining arguments */
- for (; ap->argn < ap->argc; ap->argn++)
- *ap->argv_put++ = *ap->argv_get++;
-}
-
-/* Delete the current argument. */
-static void delete_argument(struct argparse *ap) {
- if (ap->argn >= ap->argc) {
- // fprintf(stderr, "delete_argument\n");
- }
- ap->argc--;
- ap->argv_get++;
-}
-
-/* Go to the next argument. Also, move the current argument to its
- * final location in argv. */
-static void next_argument(struct argparse *ap) {
- if (ap->argn >= ap->argc) {
- // fprintf(stderr, "next_argument\n");
- }
- /* Move argument to its new location. */
- *ap->argv_put++ = *ap->argv_get++;
- ap->argn++;
-}
-
-static int is_end_of_arguments(struct argparse *ap) {
- return ap->argn == ap->argc;
-}
-
-static char *get_argument(struct argparse *ap) { return *ap->argv_get; }
-
-static char *consume_argument(struct argparse *ap) {
- char *ret = get_argument(ap);
- delete_argument(ap);
- return ret;
-}
-
-struct pb_Parameters *pb_ReadParameters(int *_argc, char **argv) {
- char *err_message;
- struct argparse ap;
- struct pb_Parameters *ret =
- (struct pb_Parameters *)malloc(sizeof(struct pb_Parameters));
-
- /* Initialize the parameters structure */
- ret->outFile = NULL;
- ret->inpFiles = (char **)malloc(sizeof(char *));
- ret->inpFiles[0] = NULL;
-
- /* Each argument */
- initialize_argparse(&ap, *_argc, argv);
- while (!is_end_of_arguments(&ap)) {
- char *arg = get_argument(&ap);
-
- /* Single-character flag */
- if ((arg[0] == '-') && (arg[1] != 0) && (arg[2] == 0)) {
- delete_argument(&ap); /* This argument is consumed here */
-
- switch (arg[1]) {
- case 'o': /* Output file name */
- if (is_end_of_arguments(&ap)) {
- err_message = "Expecting file name after '-o'\n";
- goto error;
- }
- free(ret->outFile);
- ret->outFile = strdup(consume_argument(&ap));
- break;
- case 'i': /* Input file name */
- if (is_end_of_arguments(&ap)) {
- err_message = "Expecting file name after '-i'\n";
- goto error;
- }
- ret->inpFiles = read_string_array(consume_argument(&ap));
- break;
- case '-': /* End of options */
- goto end_of_options;
- default:
- err_message = "Unexpected command-line parameter\n";
- goto error;
- }
- } else {
- /* Other parameters are ignored */
- next_argument(&ap);
- }
- } /* end for each argument */
-
-end_of_options:
- *_argc = ap.argc; /* Save the modified argc value */
- finalize_argparse(&ap);
-
- return ret;
-
-error:
- fputs(err_message, stderr);
- pb_FreeParameters(ret);
- return NULL;
-}
-
-void pb_FreeParameters(struct pb_Parameters *p) {
- char **cpp;
-
- free(p->outFile);
- free_string_array(p->inpFiles);
- free(p);
-}
-
-int pb_Parameters_CountInputs(struct pb_Parameters *p) {
- int n;
-
- for (n = 0; p->inpFiles[n]; n++)
- ;
- return n;
-}
-
-/*****************************************************************************/
-/* Timer routines */
-
-static int is_async(enum pb_TimerID timer) {
-#ifndef OPENCL_CPU
- return (timer == pb_TimerID_KERNEL) || (timer == pb_TimerID_COPY_ASYNC);
-#else
- return (timer == pb_TimerID_COPY_ASYNC);
-#endif
-}
-
-static int is_blocking(enum pb_TimerID timer) {
- return (timer == pb_TimerID_COPY) || (timer == pb_TimerID_NONE);
-}
-
-#define INVALID_TIMERID pb_TimerID_LAST
-
-static int asyncs_outstanding(struct pb_TimerSet *timers) {
- return (timers->async_markers != NULL) &&
- (timers->async_markers->timerID != INVALID_TIMERID);
-}
-
-static struct pb_async_time_marker_list *
-get_last_async(struct pb_TimerSet *timers) {
- /* Find the last event recorded thus far */
- struct pb_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)
- last_event = last_event->next;
- return last_event;
- } else
- return NULL;
-}
-
-static void insert_marker(struct pb_TimerSet *tset, enum pb_TimerID timer) {
- cl_int ciErrNum = CL_SUCCESS;
- struct pb_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 pb_async_time_marker_list *)malloc(
- sizeof(struct pb_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'
-the event #if ( __OPENCL_VERSION__ >= CL_VERSION_1_1 ) fprintf(stderr, "Creating
-Marker [%d]\n", timer);
- *((cl_event *)((*new_event)->marker)) = clCreateUserEvent(*clContextPtr,
-&ciErrNum); if (ciErrNum != CL_SUCCESS) { fprintf(stderr, "Error Creating User
-Event Object!\n");
- }
- ciErrNum = clSetUserEventStatus(*((cl_event *)((*new_event)->marker)),
-CL_QUEUED); if (ciErrNum != CL_SUCCESS) { fprintf(stderr, "Error Setting User
-Event Status!\n");
- }
-#endif
-*/
- (*new_event)->next = NULL;
- }
-
- /* valid event handle now aquired: insert the event record */
- (*new_event)->label = NULL;
- (*new_event)->timerID = timer;
- ciErrNum =
- clEnqueueMarker(*clCommandQueuePtr, (cl_event *)(*new_event)->marker);
- if (ciErrNum != CL_SUCCESS) {
- fprintf(stderr, "Error Enqueueing Marker!\n");
- }
-}
-
-static void insert_submarker(struct pb_TimerSet *tset, char *label,
- enum pb_TimerID timer) {
- cl_int ciErrNum = CL_SUCCESS;
- struct pb_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 pb_async_time_marker_list *)malloc(
- sizeof(struct pb_async_time_marker_list));
- (*new_event)->marker = calloc(1, sizeof(cl_event));
- /*
-#if ( __OPENCL_VERSION__ >= CL_VERSION_1_1 )
-fprintf(stderr, "Creating SubMarker %s[%d]\n", label, timer);
- *((cl_event *)((*new_event)->marker)) = clCreateUserEvent(*clContextPtr,
-&ciErrNum); if (ciErrNum != CL_SUCCESS) { fprintf(stderr, "Error Creating User
-Event Object!\n");
- }
- ciErrNum = clSetUserEventStatus(*((cl_event *)((*new_event)->marker)),
-CL_QUEUED); if (ciErrNum != CL_SUCCESS) { fprintf(stderr, "Error Setting User
-Event Status!\n");
- }
-#endif
-*/
- (*new_event)->next = NULL;
- }
-
- /* valid event handle now aquired: insert the event record */
- (*new_event)->label = label;
- (*new_event)->timerID = timer;
- ciErrNum =
- clEnqueueMarker(*clCommandQueuePtr, (cl_event *)(*new_event)->marker);
- if (ciErrNum != CL_SUCCESS) {
- fprintf(stderr, "Error Enqueueing Marker!\n");
- }
-}
-
-/* Assumes that all recorded events have completed */
-static pb_Timestamp record_async_times(struct pb_TimerSet *tset) {
- struct pb_async_time_marker_list *next_interval = NULL;
- struct pb_async_time_marker_list *last_marker = get_last_async(tset);
- pb_Timestamp total_async_time = 0;
- enum pb_TimerID timer;
-
- for (next_interval = tset->async_markers; next_interval != last_marker;
- next_interval = next_interval->next) {
- cl_ulong command_start = 0, command_end = 0;
- cl_int ciErrNum = CL_SUCCESS;
-
- ciErrNum = clGetEventProfilingInfo(*((cl_event *)next_interval->marker),
- CL_PROFILING_COMMAND_END,
- sizeof(cl_ulong), &command_start, NULL);
- if (ciErrNum != CL_SUCCESS) {
- fprintf(stderr, "Error getting first EventProfilingInfo: %d\n", ciErrNum);
- }
-
- ciErrNum = clGetEventProfilingInfo(
- *((cl_event *)next_interval->next->marker), CL_PROFILING_COMMAND_END,
- sizeof(cl_ulong), &command_end, NULL);
- if (ciErrNum != CL_SUCCESS) {
- fprintf(stderr, "Error getting second EventProfilingInfo: %d\n",
- ciErrNum);
- }
-
- pb_Timestamp interval =
- (pb_Timestamp)(((double)(command_end - command_start)));
- tset->timers[next_interval->timerID].elapsed += interval;
- if (next_interval->label != NULL) {
- struct pb_SubTimer *subtimer =
- tset->sub_timer_list[next_interval->timerID]->subtimer_list;
- while (subtimer != NULL) {
- if (strcmp(subtimer->label, next_interval->label) == 0) {
- subtimer->timer.elapsed += interval;
- break;
- }
- subtimer = subtimer->next;
- }
- }
- total_async_time += interval;
- next_interval->timerID = INVALID_TIMERID;
- }
-
- if (next_interval != NULL)
- next_interval->timerID = INVALID_TIMERID;
-
- return total_async_time;
-}
-
-static void accumulate_time(pb_Timestamp *accum, pb_Timestamp start,
- pb_Timestamp end) {
-#if _POSIX_VERSION >= 200112L
- *accum += end - start;
-#else
-#error "Timestamps not implemented for this system"
-#endif
-}
-
-#if _POSIX_VERSION >= 200112L
-static pb_Timestamp get_time() {
- struct timespec tv;
- clock_gettime(CLOCK_MONOTONIC, &tv);
- return (pb_Timestamp)(tv.tv_sec * BILLION + tv.tv_nsec);
-}
-#else
-#error "no supported time libraries are available on this platform"
-#endif
-
-void pb_ResetTimer(struct pb_Timer *timer) {
- timer->state = pb_Timer_STOPPED;
-
-#if _POSIX_VERSION >= 200112L
- timer->elapsed = 0;
-#else
-#error "pb_ResetTimer: not implemented for this system"
-#endif
-}
-
-void pb_StartTimer(struct pb_Timer *timer) {
- if (timer->state != pb_Timer_STOPPED) {
- fputs("Ignoring attempt to start a running timer\n", stderr);
- return;
- }
-
- timer->state = pb_Timer_RUNNING;
-
-#if _POSIX_VERSION >= 200112L
- {
- struct timespec tv;
- clock_gettime(CLOCK_MONOTONIC, &tv);
- timer->init = tv.tv_sec * BILLION + tv.tv_nsec;
- }
-#else
-#error "pb_StartTimer: not implemented for this system"
-#endif
-}
-
-void pb_StartTimerAndSubTimer(struct pb_Timer *timer,
- struct pb_Timer *subtimer) {
-
- unsigned int numNotStopped = 0x3; // 11
- if (timer->state != pb_Timer_STOPPED) {
- fputs("Warning: Timer was not stopped\n", stderr);
- numNotStopped &= 0x1; // Zero out 2^1
- }
- if (subtimer->state != pb_Timer_STOPPED) {
- fputs("Warning: Subtimer was not stopped\n", stderr);
- numNotStopped &= 0x2; // Zero out 2^0
- }
- if (numNotStopped == 0x0) {
- fputs("Ignoring attempt to start running timer and subtimer\n", stderr);
- return;
- }
-
- timer->state = pb_Timer_RUNNING;
- subtimer->state = pb_Timer_RUNNING;
-
-#if _POSIX_VERSION >= 200112L
- {
- struct timespec tv;
- clock_gettime(CLOCK_MONOTONIC, &tv);
-
- if (numNotStopped & 0x2) {
- timer->init = tv.tv_sec * BILLION + tv.tv_nsec;
- }
-
- if (numNotStopped & 0x1) {
- subtimer->init = tv.tv_sec * BILLION + tv.tv_nsec;
- }
- }
-#else
-#error "pb_StartTimer: not implemented for this system"
-#endif
-}
-
-void pb_StopTimer(struct pb_Timer *timer) {
- pb_Timestamp fini;
-
- if (timer->state != pb_Timer_RUNNING) {
- fputs("Ignoring attempt to stop a stopped timer\n", stderr);
- return;
- }
-
- timer->state = pb_Timer_STOPPED;
-
-#if _POSIX_VERSION >= 200112L
- {
- struct timespec tv;
- clock_gettime(CLOCK_MONOTONIC, &tv);
- fini = tv.tv_sec * BILLION + tv.tv_nsec;
- }
-#else
-#error "pb_StopTimer: not implemented for this system"
-#endif
-
- accumulate_time(&timer->elapsed, timer->init, fini);
- timer->init = fini;
-}
-
-void pb_StopTimerAndSubTimer(struct pb_Timer *timer,
- struct pb_Timer *subtimer) {
-
- pb_Timestamp fini;
-
- unsigned int numNotRunning = 0x3; // 11
- if (timer->state != pb_Timer_RUNNING) {
- fputs("Warning: Timer was not running\n", stderr);
- numNotRunning &= 0x1; // Zero out 2^1
- }
- if (subtimer->state != pb_Timer_RUNNING) {
- fputs("Warning: Subtimer was not running\n", stderr);
- numNotRunning &= 0x2; // Zero out 2^0
- }
- if (numNotRunning == 0x0) {
- fputs("Ignoring attempt to stop stopped timer and subtimer\n", stderr);
- return;
- }
-
- timer->state = pb_Timer_STOPPED;
- subtimer->state = pb_Timer_STOPPED;
-
-#if _POSIX_VERSION >= 200112L
- {
- struct timespec tv;
- clock_gettime(CLOCK_MONOTONIC, &tv);
- fini = tv.tv_sec * BILLION + tv.tv_nsec;
- }
-#else
-#error "pb_StopTimer: not implemented for this system"
-#endif
-
- if (numNotRunning & 0x2) {
- accumulate_time(&timer->elapsed, timer->init, fini);
- timer->init = fini;
- }
-
- if (numNotRunning & 0x1) {
- accumulate_time(&subtimer->elapsed, subtimer->init, fini);
- subtimer->init = fini;
- }
-}
-
-/* Get the elapsed time in seconds. */
-double pb_GetElapsedTime(struct pb_Timer *timer) {
- double ret;
-
- if (timer->state != pb_Timer_STOPPED) {
- fputs("Elapsed time from a running timer is inaccurate\n", stderr);
- }
-
-#if _POSIX_VERSION >= 200112L
- ret = timer->elapsed / 1e9;
-#else
-#error "pb_GetElapsedTime: not implemented for this system"
-#endif
- return ret;
-}
-
-void pb_InitializeTimerSet(struct pb_TimerSet *timers) {
- int n;
-
- timers->wall_begin = get_time();
- timers->current = pb_TimerID_NONE;
-
- timers->async_markers = NULL;
-
- for (n = 0; n < pb_TimerID_LAST; n++) {
- pb_ResetTimer(&timers->timers[n]);
- timers->sub_timer_list[n] = NULL;
- }
-}
-
-void pb_SetOpenCL(void *p_clContextPtr, void *p_clCommandQueuePtr) {
- clContextPtr = ((cl_context *)p_clContextPtr);
- clCommandQueuePtr = ((cl_command_queue *)p_clCommandQueuePtr);
-}
-
-static char *LoadProgSource(const char *Filename, size_t *szFinalLength) {
- // locals
- FILE *pFileStream = NULL;
- size_t szSourceLength;
-
- // open the OpenCL source code file
- pFileStream = fopen(Filename, "rb");
- if (pFileStream == 0) {
- return NULL;
- }
-
- // get the length of the source code
- fseek(pFileStream, 0, SEEK_END);
- szSourceLength = ftell(pFileStream);
- fseek(pFileStream, 0, SEEK_SET);
-
- // allocate a buffer for the source code string and read it in
- char *cSourceString = (char *)malloc(szSourceLength + 1);
- if (fread((cSourceString), szSourceLength, 1, pFileStream) != 1) {
- fclose(pFileStream);
- free(cSourceString);
- return 0;
- }
-
- // close the file and return the total length of the combined (preamble +
- // source) string
- fclose(pFileStream);
- if (szFinalLength != 0) {
- *szFinalLength = szSourceLength;
- }
- cSourceString[szSourceLength] = '\0';
-
- return cSourceString;
-}
-
-static inline void checkErr(cl_int err, cl_int success, const char *name) {
- if (err != success) {
- printf("ERROR: %s\n", name);
- exit(EXIT_FAILURE);
- }
-}
-
-void pb_CreateAndBuildKernelFromBinary(const char *file, const char *kernel,
- void *clContextPtr, void *clDevicePtr,
- void *clProgramPtr, void *clKernelPtr) {
- size_t kernelLength;
- char *programSource = LoadProgSource(file, &kernelLength);
- checkErr(programSource != NULL, 1 /*bool true*/,
- "Failure to load Program Binary");
-
- cl_int binaryStatus;
- cl_int errcode;
- cl_device_id clDevice = *(cl_device_id *)clDevicePtr;
- cl_context clContext = *(cl_context *)clContextPtr;
- cl_program clProgram = clCreateProgramWithBinary(
- clContext, 1, &clDevice, &kernelLength,
- (const unsigned char **)&programSource, &binaryStatus, &errcode);
- checkErr(errcode, CL_SUCCESS, "Failure to create program from binary");
-
- // printf("Building kernel - %s, from file %s\n", kernel, file);
- errcode = clBuildProgram(clProgram, 0, NULL, NULL, NULL, NULL);
- // If build fails, get build log from device
- if (errcode != CL_SUCCESS) {
- printf("ERROR: Failure to build program\n");
- size_t len = 0;
- errcode = clGetProgramBuildInfo(clProgram, clDevice, CL_PROGRAM_BUILD_LOG,
- 0, NULL, &len);
- printf("LOG LENGTH: %lu\n", len);
- checkErr(errcode, CL_SUCCESS,
- "Failure to collect program build log length");
- char *log = (char *)malloc(len * sizeof(char));
- errcode = clGetProgramBuildInfo(clProgram, clDevice, CL_PROGRAM_BUILD_LOG,
- len, log, NULL);
- checkErr(errcode, CL_SUCCESS, "Failure to collect program build log");
-
- printf("Device Build Log: %s\n", log);
- free(log);
- exit(EXIT_FAILURE);
- }
-
- cl_kernel clKernel = clCreateKernel(clProgram, kernel, &errcode);
- checkErr(errcode, CL_SUCCESS, "Failure to create kernel");
-
- *(cl_program *)clProgramPtr = clProgram;
- *(cl_kernel *)clKernelPtr = clKernel;
-
- free(programSource);
-}
-
-void pb_AddSubTimer(struct pb_TimerSet *timers, char *label,
- enum pb_TimerID pb_Category) {
-
- struct pb_SubTimer *subtimer =
- (struct pb_SubTimer *)malloc(sizeof(struct pb_SubTimer));
-
- int len = strlen(label);
-
- subtimer->label = (char *)malloc(sizeof(char) * (len + 1));
- sprintf(subtimer->label, "%s\0", label);
-
- pb_ResetTimer(&subtimer->timer);
- subtimer->next = NULL;
-
- struct pb_SubTimerList *subtimerlist = timers->sub_timer_list[pb_Category];
- if (subtimerlist == NULL) {
- subtimerlist =
- (struct pb_SubTimerList *)calloc(1, sizeof(struct pb_SubTimerList));
- subtimerlist->subtimer_list = subtimer;
- timers->sub_timer_list[pb_Category] = subtimerlist;
- } else {
- // Append to list
- struct pb_SubTimer *element = subtimerlist->subtimer_list;
- while (element->next != NULL) {
- element = element->next;
- }
- element->next = subtimer;
- }
-}
-
-void pb_SwitchToTimer(struct pb_TimerSet *timers, enum pb_TimerID timer) {
- /* Stop the currently running timer */
- if (timers->current != pb_TimerID_NONE) {
- struct pb_SubTimerList *subtimerlist =
- timers->sub_timer_list[timers->current];
- struct pb_SubTimer *currSubTimer =
- (subtimerlist != NULL) ? subtimerlist->current : NULL;
-
- if (!is_async(timers->current)) {
- if (timers->current != timer) {
- if (currSubTimer != NULL) {
- pb_StopTimerAndSubTimer(&timers->timers[timers->current],
- &currSubTimer->timer);
- } else {
- pb_StopTimer(&timers->timers[timers->current]);
- }
- } else {
- if (currSubTimer != NULL) {
- pb_StopTimer(&currSubTimer->timer);
- }
- }
- } else {
- insert_marker(timers, timer);
- if (!is_async(timer)) { // if switching to async too, keep driver going
- pb_StopTimer(&timers->timers[pb_TimerID_DRIVER]);
- }
- }
- }
-
- pb_Timestamp currentTime = get_time();
-
- /* The only cases we check for asynchronous task completion is
- * when an overlapping CPU operation completes, or the next
- * segment blocks on completion of previous async operations */
- if (asyncs_outstanding(timers) &&
- (!is_async(timers->current) || is_blocking(timer))) {
-
- struct pb_async_time_marker_list *last_event = get_last_async(timers);
- /* CL_COMPLETE if completed */
-
- cl_int ciErrNum = CL_SUCCESS;
- cl_int async_done = CL_COMPLETE;
-
- ciErrNum = clGetEventInfo(*((cl_event *)last_event->marker),
- CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(cl_int),
- &async_done, NULL);
- if (ciErrNum != CL_SUCCESS) {
- fprintf(stderr, "Error Querying EventInfo!\n");
- }
-
- if (is_blocking(timer)) {
- /* Async operations completed after previous CPU operations:
- * overlapped time is the total CPU time since this set of async
- * operations were first issued */
-
- // timer to switch to is COPY or NONE
- if (async_done != CL_COMPLETE) {
- accumulate_time(&(timers->timers[pb_TimerID_OVERLAP].elapsed),
- timers->async_begin, currentTime);
- }
-
- /* Wait on async operation completion */
- ciErrNum = clWaitForEvents(1, (cl_event *)last_event->marker);
- if (ciErrNum != CL_SUCCESS) {
- fprintf(stderr, "Error Waiting for Events!\n");
- }
-
- pb_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[pb_TimerID_OVERLAP].elapsed += total_async_time;
- }
-
- } else
- /* implies (!is_async(timers->current) && asyncs_outstanding(timers)) */
- // i.e. Current Not Async (not KERNEL/COPY_ASYNC) but there are
- // outstanding so something is deeper in stack
- if (async_done == CL_COMPLETE) {
- /* Async operations completed before previous CPU operations:
- * overlapped time is the total async time */
- timers->timers[pb_TimerID_OVERLAP].elapsed += record_async_times(timers);
- }
- }
-
- /* Start the new timer */
- if (timer != pb_TimerID_NONE) {
- if (!is_async(timer)) {
- pb_StartTimer(&timers->timers[timer]);
- } else {
- // toSwitchTo Is Async (KERNEL/COPY_ASYNC)
- if (!asyncs_outstanding(timers)) {
- /* No asyncs outstanding, insert a fresh async marker */
-
- insert_marker(timers, timer);
- timers->async_begin = currentTime;
- } else if (!is_async(timers->current)) {
- /* Previous asyncs still in flight, but a previous SwitchTo
- * already marked the end of the most recent async operation,
- * so we can rename that marker as the beginning of this async
- * operation */
-
- struct pb_async_time_marker_list *last_event = get_last_async(timers);
- last_event->label = NULL;
- last_event->timerID = timer;
- }
- if (!is_async(timers->current)) {
- pb_StartTimer(&timers->timers[pb_TimerID_DRIVER]);
- }
- }
- }
- timers->current = timer;
-}
-
-void pb_SwitchToSubTimer(struct pb_TimerSet *timers, char *label,
- enum pb_TimerID category) {
- struct pb_SubTimerList *subtimerlist =
- timers->sub_timer_list[timers->current];
- struct pb_SubTimer *curr =
- (subtimerlist != NULL) ? subtimerlist->current : NULL;
-
- if (timers->current != pb_TimerID_NONE) {
- if (!is_async(timers->current)) {
- if (timers->current != category) {
- if (curr != NULL) {
- pb_StopTimerAndSubTimer(&timers->timers[timers->current],
- &curr->timer);
- } else {
- pb_StopTimer(&timers->timers[timers->current]);
- }
- } else {
- if (curr != NULL) {
- pb_StopTimer(&curr->timer);
- }
- }
- } else {
- insert_submarker(timers, label, category);
- if (!is_async(category)) { // if switching to async too, keep driver going
- pb_StopTimer(&timers->timers[pb_TimerID_DRIVER]);
- }
- }
- }
-
- pb_Timestamp currentTime = get_time();
-
- /* The only cases we check for asynchronous task completion is
- * when an overlapping CPU operation completes, or the next
- * segment blocks on completion of previous async operations */
- if (asyncs_outstanding(timers) &&
- (!is_async(timers->current) || is_blocking(category))) {
-
- struct pb_async_time_marker_list *last_event = get_last_async(timers);
- /* CL_COMPLETE if completed */
-
- cl_int ciErrNum = CL_SUCCESS;
- cl_int async_done = CL_COMPLETE;
-
- ciErrNum = clGetEventInfo(*((cl_event *)last_event->marker),
- CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(cl_int),
- &async_done, NULL);
- if (ciErrNum != CL_SUCCESS) {
- fprintf(stderr, "Error Querying EventInfo!\n");
- }
-
- if (is_blocking(category)) {
- /* Async operations completed after previous CPU operations:
- * overlapped time is the total CPU time since this set of async
- * operations were first issued */
-
- // timer to switch to is COPY or NONE
- // if it hasn't already finished, then just take now and use that as the
- // elapsed time in OVERLAP anything happening after now isn't OVERLAP
- // 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[pb_TimerID_OVERLAP].elapsed),
- timers->async_begin, currentTime);
-
- /* Wait on async operation completion */
- ciErrNum = clWaitForEvents(1, (cl_event *)last_event->marker);
- if (ciErrNum != CL_SUCCESS) {
- fprintf(stderr, "Error Waiting for Events!\n");
- }
- pb_Timestamp total_async_time = record_async_times(timers);
-
- /* Async operations completed before previous CPU operations:
- * overlapped time is the total async time */
- // If it did finish, then accumulate all the async time that did happen
- // 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[pb_TimerID_OVERLAP].elapsed += total_async_time;
-
- } else
- /* implies (!is_async(timers->current) && asyncs_outstanding(timers)) */
- // i.e. Current Not Async (not KERNEL/COPY_ASYNC) but there are
- // outstanding so something is deeper in stack
- if (async_done == CL_COMPLETE /*cudaSuccess*/) {
- /* Async operations completed before previous CPU operations:
- * overlapped time is the total async time */
- timers->timers[pb_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 pb_SubTimer *subtimer = NULL;
-
- if (label != NULL) {
- subtimer = subtimerlist->subtimer_list;
- while (subtimer != NULL) {
- if (strcmp(subtimer->label, label) == 0) {
- break;
- } else {
- subtimer = subtimer->next;
- }
- }
- }
-
- /* Start the new timer */
- if (category != pb_TimerID_NONE) {
- if (!is_async(category)) {
- if (subtimerlist != NULL) {
- subtimerlist->current = subtimer;
- }
-
- if (category != timers->current && subtimer != NULL) {
- pb_StartTimerAndSubTimer(&timers->timers[category], &subtimer->timer);
- } else if (subtimer != NULL) {
- pb_StartTimer(&subtimer->timer);
- } else {
- pb_StartTimer(&timers->timers[category]);
- }
- } else {
- if (subtimerlist != NULL) {
- subtimerlist->current = subtimer;
- }
-
- // toSwitchTo Is Async (KERNEL/COPY_ASYNC)
- if (!asyncs_outstanding(timers)) {
- /* No asyncs outstanding, insert a fresh async marker */
- insert_submarker(timers, label, category);
- timers->async_begin = currentTime;
- } else if (!is_async(timers->current)) {
- /* Previous asyncs still in flight, but a previous SwitchTo
- * already marked the end of the most recent async operation,
- * so we can rename that marker as the beginning of this async
- * operation */
-
- struct pb_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
-
- // toSwitchto is already asynchronous, but if current/prev state is async
- // too, then DRIVER is already running
- if (!is_async(timers->current)) {
- pb_StartTimer(&timers->timers[pb_TimerID_DRIVER]);
- }
- }
- }
-
- timers->current = category;
-}
-
-void pb_PrintTimerSet(struct pb_TimerSet *timers) {
- printf("Printing Parboil Timer: Default\n");
- pb_Timestamp wall_end = get_time();
-
- struct pb_Timer *t = timers->timers;
- struct pb_SubTimer *sub = NULL;
-
- int maxSubLength;
-
- // const char *categories[] = {
- // "IO", "Kernel", "Copy", "Driver", "Copy Async", "Compute"
- // };
- const char *categories[] = {
- "IO", "Kernel", "Copy", "Driver",
- "Copy Async", "Compute", "Overlap", "Init_Ctx",
- "Clear_Ctx", "Copy_Scalar", "Copy_Ptr", "Mem_Free",
- "Read_Output", "Setup", "Mem_Track", "Mem_Untrack",
- "Misc", "Pthread_Create", "Arg_Pack", "Arg_Unpack",
- "Computation", "Output_Pack", "Output_Unpack"};
-
- const int maxCategoryLength = 20;
-
- int i;
- for (i = 1; i < pb_TimerID_LAST;
- ++i) { // exclude NONE and OVRELAP from this format
- if (pb_GetElapsedTime(&t[i]) != 0 || true) {
-
- // Print Category Timer
- printf("%-*s: %.9f\n", maxCategoryLength, categories[i - 1],
- pb_GetElapsedTime(&t[i]));
-
- if (timers->sub_timer_list[i] != NULL) {
- sub = timers->sub_timer_list[i]->subtimer_list;
- maxSubLength = 0;
- while (sub != NULL) {
- // Find longest SubTimer label
- if (strlen(sub->label) > maxSubLength) {
- maxSubLength = strlen(sub->label);
- }
- sub = sub->next;
- }
-
- // Fit to Categories
- if (maxSubLength <= maxCategoryLength) {
- maxSubLength = maxCategoryLength;
- }
-
- sub = timers->sub_timer_list[i]->subtimer_list;
-
- // Print SubTimers
- while (sub != NULL) {
- printf(" -%-*s: %.9f\n", maxSubLength, sub->label,
- pb_GetElapsedTime(&sub->timer));
- sub = sub->next;
- }
- }
- }
- }
-
- if (pb_GetElapsedTime(&t[pb_TimerID_OVERLAP]) != 0)
- printf("CPU/Kernel Overlap: %.9f\n",
- pb_GetElapsedTime(&t[pb_TimerID_OVERLAP]));
-
- float walltime = (wall_end - timers->wall_begin) / 1e9;
- printf("Timer Wall Time: %.9f\n", walltime);
-}
-
-void pb_DestroyTimerSet(struct pb_TimerSet *timers) {
- /* clean up all of the async event markers */
- struct pb_async_time_marker_list *event = timers->async_markers;
- while (event != NULL) {
-
- cl_int ciErrNum = CL_SUCCESS;
- ciErrNum = clWaitForEvents(1, (cl_event *)(event)->marker);
- if (ciErrNum != CL_SUCCESS) {
- // fprintf(stderr, "Error Waiting for Events!\n");
- }
-
- ciErrNum = clReleaseEvent(*((cl_event *)(event)->marker));
- if (ciErrNum != CL_SUCCESS) {
- fprintf(stderr, "Error Release Events!\n");
- }
-
- free((event)->marker);
- struct pb_async_time_marker_list *next = ((event)->next);
-
- free(event);
-
- // (*event) = NULL;
- event = next;
- }
-
- int i = 0;
- for (i = 0; i < pb_TimerID_LAST; ++i) {
- if (timers->sub_timer_list[i] != NULL) {
- struct pb_SubTimer *subtimer = timers->sub_timer_list[i]->subtimer_list;
- struct pb_SubTimer *prev = NULL;
- while (subtimer != NULL) {
- free(subtimer->label);
- prev = subtimer;
- subtimer = subtimer->next;
- free(prev);
- }
- free(timers->sub_timer_list[i]);
- }
- }
-}