Skip to content
Snippets Groups Projects
Commit 8914fe62 authored by Hashim Sharif's avatar Hashim Sharif
Browse files

Adding Lenet Promise version

parent 1bc92263
No related branches found
No related tags found
No related merge requests found
Hide whitespace changes
Inline Side-by-side
llvm/test/VISC/DNN_Benchmarks/benchmarks/lenet_front/Makefile
+ 20
− 12
View file @ 8914fe62
DNN_BENCHMARK_ROOT = $(LLVM_SRC_ROOT)/test/VISC/DNN_Benchmarks DNN_BENCHMARK_ROOT = $(LLVM_SRC_ROOT)/test/VISC/DNN_Benchmarks
# NOTE: can configure build directory # NOTE: can configure build directory
HPVM_BUILD_DIR = $(LLVM_SRC_ROOT)/../build_fresh/ HPVM_BUILD_DIR = $(LLVM_SRC_ROOT)/../build_hpvm/
CC = $(HPVM_BUILD_DIR)/bin/clang++ CC = $(HPVM_BUILD_DIR)/bin/clang++
OPT = $(HPVM_BUILD_DIR)/bin/opt OPT = $(HPVM_BUILD_DIR)/bin/opt
...@@ -26,6 +26,11 @@ HPVM_LIB_DIR = $(HPVM_BUILD_DIR)/lib ...@@ -26,6 +26,11 @@ HPVM_LIB_DIR = $(HPVM_BUILD_DIR)/lib
VISC_OPTFLAGS = -load $(HPVM_LIB_DIR)/LLVMBuildDFG.so -load $(HPVM_LIB_DIR)/LLVMInPlaceDFGAnalysis.so -load $(HPVM_LIB_DIR)/LLVMDFG2LLVM_CUDNN.so -load $(HPVM_LIB_DIR)/LLVMDFG2LLVM_X86.so -load $(HPVM_LIB_DIR)/LLVMClearDFG.so -inplace -dfg2llvm-cudnn -dfg2llvm-x86 -clearDFG VISC_OPTFLAGS = -load $(HPVM_LIB_DIR)/LLVMBuildDFG.so -load $(HPVM_LIB_DIR)/LLVMInPlaceDFGAnalysis.so -load $(HPVM_LIB_DIR)/LLVMDFG2LLVM_CUDNN.so -load $(HPVM_LIB_DIR)/LLVMDFG2LLVM_X86.so -load $(HPVM_LIB_DIR)/LLVMClearDFG.so -inplace -dfg2llvm-cudnn -dfg2llvm-x86 -clearDFG
VISC_OPTFLAGS2 = -load $(HPVM_LIB_DIR)/LLVMBuildDFG.so -load $(HPVM_LIB_DIR)/LLVMInPlaceDFGAnalysis.so -load $(HPVM_LIB_DIR)/LLVMDFG2LLVM_PROMISE.so -load $(HPVM_LIB_DIR)/LLVMDFG2LLVM_X86.so -load $(HPVM_LIB_DIR)/LLVMFuseHPVMTensorNodes.so -load $(HPVM_LIB_DIR)/LLVMClearDFG.so -inplace -hpvm-fuse -dfg2llvm-promise -dfg2llvm-x86 -clearDFG
TARGET = $(BUILD_DIR)/$(APP).opt.bc TARGET = $(BUILD_DIR)/$(APP).opt.bc
SOURCES = $(SRC_DIR)/$(APP).cpp SOURCES = $(SRC_DIR)/$(APP).cpp
VISC_RT_PATH = $(LLVM_SRC_ROOT)/../build/projects/visc-rt/visc-rt.ll VISC_RT_PATH = $(LLVM_SRC_ROOT)/../build/projects/visc-rt/visc-rt.ll
...@@ -36,17 +41,20 @@ default: $(BUILD_DIR) $(TARGET) ...@@ -36,17 +41,20 @@ default: $(BUILD_DIR) $(TARGET)
$(BUILD_DIR)/%.ll: $(SRC_DIR)/%.cpp $(BUILD_DIR)/%.ll: $(SRC_DIR)/%.cpp
$(CC) $(CC_FLAGS) -emit-llvm -S -o $@ $< $(CC) $(CC_FLAGS) -emit-llvm src/$(APP).cpp -S -o $(BUILD_DIR)/$(APP).ll
$(CC) $(CC_FLAGS) -emit-llvm src/$(APP)_promise.cpp -S -o $(BUILD_DIR)/$(APP)_promise.ll
#-visc-timers-gen
$(BUILD_DIR)/%.visc.ll: $(BUILD_DIR)/%.ll
$(OPT) -load LLVMGenVISC.so -genvisc -globaldce $< -S -o $@ $(BUILD_DIR)/%.opt.bc: $(BUILD_DIR)/%.ll
$(OPT) -load LLVMGenVISC.so -genvisc -globaldce $(BUILD_DIR)/$(APP).ll -S -o $(BUILD_DIR)/$(APP).visc.ll
$(BUILD_DIR)/%.opt.bc: $(BUILD_DIR)/%.visc.ll $(OPT) -load LLVMGenVISC.so -genvisc -globaldce $(BUILD_DIR)/$(APP)_promise.ll -S -o $(BUILD_DIR)/$(APP)_promise.visc.ll
$(OPT) $(VISC_OPTFLAGS) $< -o $@ $(OPT) $(VISC_OPTFLAGS) $(BUILD_DIR)/$(APP).visc.ll -o $(BUILD_DIR)/$(APP)_cudnn.bc
$(LLVM_LINK) $@ $(VISC_RT_PATH) -o $(BUILD_DIR)/lenet_linked.bc $(OPT) $(VISC_OPTFLAGS2) $(BUILD_DIR)/$(APP)_promise.visc.ll -o $(BUILD_DIR)/$(APP)_promise.bc
$(CC) $(BUILD_DIR)/lenet_linked.bc $(TENSOR_LIB_DIR) -o $(BUILD_DIR)/lenet_linked $(LINKER_FLAGS) $(LLVM_LINK) $(BUILD_DIR)/$(APP)_cudnn.bc $(VISC_RT_PATH) -o $(BUILD_DIR)/$(APP)_cudnn_linked.bc
$(CC) $(BUILD_DIR)/lenet_linked.bc $(TENSOR_AUTOTUNER_DIR) -o $(BUILD_DIR)/lenet_tune $(LINKER_FLAGS) $(LLVM_LINK) $(BUILD_DIR)/$(APP)_promise.bc $(VISC_RT_PATH) -o $(BUILD_DIR)/$(APP)_promise_linked.bc
$(CC) $(BUILD_DIR)/$(BUILD_DIR)/$(APP)_cudnn_linked.bc $(TENSOR_LIB_DIR) -o $(BUILD_DIR)/$(APP)_cudnn_linked $(LINKER_FLAGS)
$(CC) $(BUILD_DIR)/$(BUILD_DIR)/$(APP)_promise_linked.bc $(TENSOR_LIB_DIR) -o $(BUILD_DIR)/$(APP)_promise_linked $(LINKER_FLAGS)
$(CC) $(BUILD_DIR)/$(BUILD_DIR)/$(APP)_cudnn_linked.bc $(TENSOR_AUTOTUNER_DIR) -o $(BUILD_DIR)/lenet_tune $(LINKER_FLAGS)
$(BUILD_DIR): $(BUILD_DIR):
mkdir -p $@ mkdir -p $@
......
llvm/test/VISC/DNN_Benchmarks/benchmarks/lenet_front/src/lenet_promise.cpp 0 → 100644
+ 373
− 0
View file @ 8914fe62
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/stat.h>
#include <cstring>
#include <visc.h>
#include <tensorTypes.h>
#include <tensorUtils.h>
void var_0_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(2, t1, t2, 0);
void *r = __visc__tensor_convolution(t1, t2, 2, 2, 1, 1);
__visc__return(2, r, (size_t) 0);
}
void var_1_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(2, t1, t2, 0);
void *r = __visc__tensor_add(t1, t2);
__visc__return(2, r, (size_t) 0);
}
void var_2_node(void* t1, size_t bytes_t1) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(1, t1, 0);
void* r = __visc__tensor_relu(t1);
__visc__return(2, r, (size_t) 0);
}
void var_3_node(void* t1, size_t bytes_t1) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(1, t1, 0);
void* r = __visc__tensor_pool_max(t1, 2, 2, 0, 0, 2, 2);
__visc__return(2, r, (size_t) 0);
}
void var_4_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(2, t1, t2, 0);
void *r = __visc__tensor_convolution(t1, t2, 2, 2, 1, 1);
__visc__return(2, r, (size_t) 0);
}
void var_5_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(2, t1, t2, 0);
void *r = __visc__tensor_add(t1, t2);
__visc__return(2, r, (size_t) 0);
}
void var_6_node(void* t1, size_t bytes_t1) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(1, t1, 0);
void* r = __visc__tensor_relu(t1);
__visc__return(2, r, (size_t) 0);
}
void var_7_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(2, t1, t2, 0);
void *r = __visc__tensor_convolution(t1, t2, 1, 1, 2, 2);
__visc__return(2, r, (size_t) 0);
}
void var_8_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(2, t1, t2, 0);
void *r = __visc__tensor_add(t1, t2);
__visc__return(2, r, (size_t) 0);
}
void var_9_node(void* t1, size_t bytes_t1) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(1, t1, 0);
void* r = __visc__tensor_relu(t1);
__visc__return(2, r, (size_t) 0);
}
void var_10_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(2, t1, t2, 0);
void *r = __visc__tensor_mul(t1, t2);
__visc__return(2, r, (size_t) 0);
}
void var_11_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(2, t1, t2, 0);
void *r = __visc__tensor_add(t1, t2);
__visc__return(2, r, (size_t) 0);
}
void var_12_node(void* t1, size_t bytes_t1) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(1, t1, 0);
void* r = __visc__tensor_relu(t1);
__visc__return(2, r, (size_t) 0);
}
void var_13_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(2, t1, t2, 0);
void *r = __visc__tensor_mul(t1, t2);
__visc__return(2, r, (size_t) 0);
}
void var_14_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(2, t1, t2, 0);
void *r = __visc__tensor_add(t1, t2);
__visc__return(2, r, (size_t) 0);
}
void var_15_node(void* t1, size_t bytes_t1) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(1, t1, 0);
void* r = __visc__tensor_relu(t1);
__visc__return(2, r, (size_t) 0);
}
void var_16_node(void* t1, size_t bytes_t1) {
__visc__hint(visc::PROMISE_TARGET);
__visc__attributes(1, t1, 0);
void* r = __visc__tensor_softmax(t1);
__visc__return(2, r, (size_t) 0);
}
void root(void* input, size_t input_bytes,
void* conv2d_1_w, size_t conv2d_1_w_bytes,
void* conv2d_1_b, size_t conv2d_1_b_bytes,
void* conv2d_2_w, size_t conv2d_2_w_bytes,
void* conv2d_2_b, size_t conv2d_2_b_bytes,
void* conv2d_3_w, size_t conv2d_3_w_bytes,
void* conv2d_3_b, size_t conv2d_3_b_bytes,
void* dense_1_w, size_t dense_1_w_bytes,
void* dense_1_b, size_t dense_1_b_bytes,
void* dense_2_w, size_t dense_2_w_bytes,
void* dense_2_b, size_t dense_2_b_bytes){
__visc__hint(visc::CPU_TARGET);
__visc__attributes(11, input, conv2d_1_w, conv2d_1_b, conv2d_2_w, conv2d_2_b, conv2d_3_w, conv2d_3_b, dense_1_w, dense_1_b, dense_2_w, dense_2_b, 0);
void* var_0 = __visc__createNodeND(0, var_0_node);
__visc__bindIn(var_0, 0, 0, 0);
__visc__bindIn(var_0, 1, 1, 0);
__visc__bindIn(var_0, 2, 2, 0);
__visc__bindIn(var_0, 3, 3, 0);
void* var_1 = __visc__createNodeND(0, var_1_node);
__visc__edge(var_0, var_1, 1, 0, 0, 0);
__visc__edge(var_0, var_1, 1, 1, 1, 0);
__visc__bindIn(var_1, 4, 2, 0);
__visc__bindIn(var_1, 5, 3, 0);
void* var_2 = __visc__createNodeND(0, var_2_node);
__visc__edge(var_1, var_2, 1, 0, 0, 0);
__visc__edge(var_1, var_2, 1, 1, 1, 0);
void* var_3 = __visc__createNodeND(0, var_3_node);
__visc__edge(var_2, var_3, 1, 0, 0, 0);
__visc__edge(var_2, var_3, 1, 1, 1, 0);
void* var_4 = __visc__createNodeND(0, var_4_node);
__visc__edge(var_3, var_4, 1, 0, 0, 0);
__visc__edge(var_3, var_4, 1, 1, 1, 0);
__visc__bindIn(var_4, 6, 2, 0);
__visc__bindIn(var_4, 7, 3, 0);
void* var_5 = __visc__createNodeND(0, var_5_node);
__visc__edge(var_4, var_5, 1, 0, 0, 0);
__visc__edge(var_4, var_5, 1, 1, 1, 0);
__visc__bindIn(var_5, 8, 2, 0);
__visc__bindIn(var_5, 9, 3, 0);
void* var_6 = __visc__createNodeND(0, var_6_node);
__visc__edge(var_5, var_6, 1, 0, 0, 0);
__visc__edge(var_5, var_6, 1, 1, 1, 0);
void* var_7 = __visc__createNodeND(0, var_7_node);
__visc__edge(var_6, var_7, 1, 0, 0, 0);
__visc__edge(var_6, var_7, 1, 1, 1, 0);
__visc__bindIn(var_7, 10, 2, 0);
__visc__bindIn(var_7, 11, 3, 0);
void* var_8 = __visc__createNodeND(0, var_8_node);
__visc__edge(var_7, var_8, 1, 0, 0, 0);
__visc__edge(var_7, var_8, 1, 1, 1, 0);
__visc__bindIn(var_8, 12, 2, 0);
__visc__bindIn(var_8, 13, 3, 0);
void* var_9 = __visc__createNodeND(0, var_9_node);
__visc__edge(var_8, var_9, 1, 0, 0, 0);
__visc__edge(var_8, var_9, 1, 1, 1, 0);
void* var_10 = __visc__createNodeND(0, var_10_node);
__visc__edge(var_9, var_10, 1, 0, 0, 0);
__visc__edge(var_9, var_10, 1, 1, 1, 0);
__visc__bindIn(var_10, 14, 2, 0);
__visc__bindIn(var_10, 15, 3, 0);
void* var_11 = __visc__createNodeND(0, var_11_node);
__visc__edge(var_10, var_11, 1, 0, 0, 0);
__visc__edge(var_10, var_11, 1, 1, 1, 0);
__visc__bindIn(var_11, 16, 2, 0);
__visc__bindIn(var_11, 17, 3, 0);
void* var_12 = __visc__createNodeND(0, var_12_node);
__visc__edge(var_11, var_12, 1, 0, 0, 0);
__visc__edge(var_11, var_12, 1, 1, 1, 0);
void* var_13 = __visc__createNodeND(0, var_13_node);
__visc__edge(var_12, var_13, 1, 0, 0, 0);
__visc__edge(var_12, var_13, 1, 1, 1, 0);
__visc__bindIn(var_13, 18, 2, 0);
__visc__bindIn(var_13, 19, 3, 0);
void* var_14 = __visc__createNodeND(0, var_14_node);
__visc__edge(var_13, var_14, 1, 0, 0, 0);
__visc__edge(var_13, var_14, 1, 1, 1, 0);
__visc__bindIn(var_14, 20, 2, 0);
__visc__bindIn(var_14, 21, 3, 0);
void* var_15 = __visc__createNodeND(0, var_15_node);
__visc__edge(var_14, var_15, 1, 0, 0, 0);
__visc__edge(var_14, var_15, 1, 1, 1, 0);
void* var_16 = __visc__createNodeND(0, var_16_node);
__visc__edge(var_15, var_16, 1, 0, 0, 0);
__visc__edge(var_15, var_16, 1, 1, 1, 0);
__visc__bindOut(var_16, 0, 0, 0);
__visc__bindOut(var_16, 1, 1, 0);
}
struct ret_t {
void* tensor;
size_t bytes;
};
typedef struct __attribute__((__packed__)) {
void* input;
size_t input_bytes;
void* conv2d_1_w;
size_t conv2d_1_w_bytes;
void* conv2d_1_b;
size_t conv2d_1_b_bytes;
void* conv2d_2_w;
size_t conv2d_2_w_bytes;
void* conv2d_2_b;
size_t conv2d_2_b_bytes;
void* conv2d_3_w;
size_t conv2d_3_w_bytes;
void* conv2d_3_b;
size_t conv2d_3_b_bytes;
void* dense_1_w;
size_t dense_1_w_bytes;
void* dense_1_b;
size_t dense_1_b_bytes;
void* dense_2_w;
size_t dense_2_w_bytes;
void* dense_2_b;
size_t dense_2_b_bytes;
struct ret_t r;
}
RootIn;
int main(){
std::string dir_prefix = std::string("../../../../../../projects/hpvm-tensor-rt/model_params/lenet_hpvm/");
std::string input_path = dir_prefix + std::string("input.bin");
void* input = readTrainedWeights(input_path.c_str(), 0,10000,1,28,28);
std::string labels_path = dir_prefix + std::string("labels.bin");
uint8_t* labels = readLabels(labels_path.c_str(),10000);
std::string conv2d_1_w_path = dir_prefix + std::string("conv2d_1_w.bin");
void* conv2d_1_w = readTrainedWeights(conv2d_1_w_path.c_str(), 0,32,1,5,5);
std::string conv2d_1_b_path = dir_prefix + std::string("conv2d_1_b.bin");
void* conv2d_1_b = readTrainedWeights(conv2d_1_b_path.c_str(), 0,1,32,1,1);
std::string conv2d_2_w_path = dir_prefix + std::string("conv2d_2_w.bin");
void* conv2d_2_w = readTrainedWeights(conv2d_2_w_path.c_str(), 0,64,32,5,5);
std::string conv2d_2_b_path = dir_prefix + std::string("conv2d_2_b.bin");
void* conv2d_2_b = readTrainedWeights(conv2d_2_b_path.c_str(), 0,1,64,1,1);
std::string conv2d_3_w_path = dir_prefix + std::string("conv2d_3_w.bin");
void* conv2d_3_w = readTrainedWeights(conv2d_3_w_path.c_str(), 0,64,64,3,3);
std::string conv2d_3_b_path = dir_prefix + std::string("conv2d_3_b.bin");
void* conv2d_3_b = readTrainedWeights(conv2d_3_b_path.c_str(), 0,1,64,1,1);
std::string dense_1_w_path = dir_prefix + std::string("dense_1_w.bin");
void* dense_1_w = readTrainedWeights(dense_1_w_path.c_str(), 0,1,1,3136,1024);
std::string dense_1_b_path = dir_prefix + std::string("dense_1_b.bin");
void* dense_1_b = readTrainedWeights(dense_1_b_path.c_str(), 0,1,1024,1,1);
std::string dense_2_w_path = dir_prefix + std::string("dense_2_w.bin");
void* dense_2_w = readTrainedWeights(dense_2_w_path.c_str(), 0,1,1,1024,10);
std::string dense_2_b_path = dir_prefix + std::string("dense_2_b.bin");
void* dense_2_b = readTrainedWeights(dense_2_b_path.c_str(), 0,1,10,1,1);
__visc__init();
RootIn* args = static_cast<RootIn*>(malloc(sizeof(RootIn)));
args->input = input;
args->input_bytes = 0;
args->conv2d_1_w = conv2d_1_w;
args->conv2d_1_w_bytes = 0;
args->conv2d_1_b = conv2d_1_b;
args->conv2d_1_b_bytes = 0;
args->conv2d_2_w = conv2d_2_w;
args->conv2d_2_w_bytes = 0;
args->conv2d_2_b = conv2d_2_b;
args->conv2d_2_b_bytes = 0;
args->conv2d_3_w = conv2d_3_w;
args->conv2d_3_w_bytes = 0;
args->conv2d_3_b = conv2d_3_b;
args->conv2d_3_b_bytes = 0;
args->dense_1_w = dense_1_w;
args->dense_1_w_bytes = 0;
args->dense_1_b = dense_1_b;
args->dense_1_b_bytes = 0;
args->dense_2_w = dense_2_w;
args->dense_2_w_bytes = 0;
args->dense_2_b = dense_2_b;
args->dense_2_b_bytes = 0;
void* dfg = __visc__launch(0, root, (void*) args);
__visc__wait(dfg);
void *result = static_cast<RootIn*>(args)->input;
hpvm_request_tensor(result, 0);
__visc__cleanup();
computeAccuracy2(labels, 10000, result);
return 0;
}
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment