Skip to content
Snippets Groups Projects
Commit 93538c03 authored by Yifan Zhao's avatar Yifan Zhao
Browse files

All dnn_benchmarks read model_params from build dir

parent f097e39a
No related branches found
No related tags found
No related merge requests found
Expand all Hide whitespace changes
Inline Side-by-side
Showing
with 12242 additions and 11404 deletions
hpvm/test/dnn_benchmarks/CMakeLists.txt
+ 15
− 1
View file @ 93538c03
......@@ -8,12 +8,26 @@ set(LLVM_OPT "${LLVM_BIN_DIR}/opt")
set(LLVM_LINK "${LLVM_BIN_DIR}/llvm-link")
set(CMAKE_CXX_COMPILER "${LLVM_BIN_DIR}/clang++")
# Configure config.h which tells the benchmarks where's the build directory.
# We can also use the one in tensor_runtime, but we're avoiding that trying to
# decouple things.
set(MODEL_PARAMS_DIR "${CMAKE_BINARY_DIR}/model_params/")
configure_file(
"include/config.h.in"
"${CMAKE_CURRENT_BINARY_DIR}/include/config.h"
)
set(CONFIG_INCLUDE_DIR "${CMAKE_CURRENT_BINARY_DIR}/include")
# Directories to include
set(HPVM_PROJECTS ${PROJECT_SOURCE_DIR}/tools/hpvm/projects)
set(HPVM_INCLUDE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/../dnn_benchmarks/include)
set(TENSOR_INCLUDE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/include)
set(TENSOR_RT_INCLUDE_DIR ${HPVM_PROJECTS}/hpvm-tensor-rt/tensor_runtime/include)
set(INCLUDES ${HPVM_INCLUDE_DIR} ${TENSOR_INCLUDE_DIR} ${TENSOR_RT_INCLUDE_DIR} ${CUDA_INCLUDE_DIRS})
set(
INCLUDES
${CONFIG_INCLUDE_DIR} ${HPVM_INCLUDE_DIR}
${TENSOR_INCLUDE_DIR} ${TENSOR_RT_INCLUDE_DIR} ${CUDA_INCLUDE_DIRS}
)
foreach(dir ${INCLUDES})
list(APPEND INCLUDE_COMPILER_STRINGS "-I${dir}")
endforeach()
......
hpvm/test/dnn_benchmarks/benchmarks/alexnet/alexnet.cpp
+ 334
− 336
View file @ 93538c03
This diff is collapsed.
hpvm/test/dnn_benchmarks/benchmarks/alexnet2/alexnet2.cpp
+ 381
− 381
View file @ 93538c03
This diff is collapsed.
hpvm/test/dnn_benchmarks/benchmarks/lenet_mnist/lenet_mnist.cpp
+ 248
− 250
View file @ 93538c03
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/stat.h>
#include <cstring>
#include <hpvm.h>
#include <tensorTypes.h>
#include <tensorUtils.h>
void var_0_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void *r = __hpvm__tensor_convolution(t1, t2, 2, 2, 1, 1);
__hpvm__return(2, r, (size_t) 0);
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/stat.h>
#include <cstring>
#include <hpvm.h>
#include <tensorTypes.h>
#include <tensorUtils.h>
void var_0_node(void *t1, size_t bytes_t1, void *t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void *r = __hpvm__tensor_convolution(t1, t2, 2, 2, 1, 1);
__hpvm__return(2, r, (size_t)0);
}
void var_1_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void var_1_node(void *t1, size_t bytes_t1, void *t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void *r = __hpvm__tensor_add(t1, t2);
__hpvm__return(2, r, (size_t) 0);
void *r = __hpvm__tensor_add(t1, t2);
__hpvm__return(2, r, (size_t)0);
}
void var_2_node(void* t1, size_t bytes_t1) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(1, t1, 0);
void var_2_node(void *t1, size_t bytes_t1) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(1, t1, 0);
void* r = __hpvm__tensor_tanh(t1);
__hpvm__return(2, r, (size_t) 0);
void *r = __hpvm__tensor_tanh(t1);
__hpvm__return(2, r, (size_t)0);
}
void var_3_node(void* t1, size_t bytes_t1) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(1, t1, 0);
void var_3_node(void *t1, size_t bytes_t1) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(1, t1, 0);
void* r = __hpvm__tensor_pool_max(t1, 2, 2, 0, 0, 2, 2);
__hpvm__return(2, r, (size_t) 0);
void *r = __hpvm__tensor_pool_max(t1, 2, 2, 0, 0, 2, 2);
__hpvm__return(2, r, (size_t)0);
}
void var_4_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void var_4_node(void *t1, size_t bytes_t1, void *t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void *r = __hpvm__tensor_convolution(t1, t2, 2, 2, 1, 1);
__hpvm__return(2, r, (size_t) 0);
void *r = __hpvm__tensor_convolution(t1, t2, 2, 2, 1, 1);
__hpvm__return(2, r, (size_t)0);
}
void var_5_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void var_5_node(void *t1, size_t bytes_t1, void *t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void *r = __hpvm__tensor_add(t1, t2);
__hpvm__return(2, r, (size_t) 0);
void *r = __hpvm__tensor_add(t1, t2);
__hpvm__return(2, r, (size_t)0);
}
void var_6_node(void* t1, size_t bytes_t1) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(1, t1, 0);
void var_6_node(void *t1, size_t bytes_t1) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(1, t1, 0);
void* r = __hpvm__tensor_tanh(t1);
__hpvm__return(2, r, (size_t) 0);
void *r = __hpvm__tensor_tanh(t1);
__hpvm__return(2, r, (size_t)0);
}
void var_7_node(void* t1, size_t bytes_t1) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(1, t1, 0);
void var_7_node(void *t1, size_t bytes_t1) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(1, t1, 0);
void* r = __hpvm__tensor_pool_max(t1, 2, 2, 0, 0, 2, 2);
__hpvm__return(2, r, (size_t) 0);
void *r = __hpvm__tensor_pool_max(t1, 2, 2, 0, 0, 2, 2);
__hpvm__return(2, r, (size_t)0);
}
void var_8_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void var_8_node(void *t1, size_t bytes_t1, void *t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void *r = __hpvm__tensor_mul(t1, t2);
__hpvm__return(2, r, (size_t) 0);
void *r = __hpvm__tensor_mul(t1, t2);
__hpvm__return(2, r, (size_t)0);
}
void var_9_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void var_9_node(void *t1, size_t bytes_t1, void *t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void *r = __hpvm__tensor_add(t1, t2);
__hpvm__return(2, r, (size_t) 0);
void *r = __hpvm__tensor_add(t1, t2);
__hpvm__return(2, r, (size_t)0);
}
void var_10_node(void* t1, size_t bytes_t1) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(1, t1, 0);
void var_10_node(void *t1, size_t bytes_t1) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(1, t1, 0);
void* r = __hpvm__tensor_tanh(t1);
__hpvm__return(2, r, (size_t) 0);
void *r = __hpvm__tensor_tanh(t1);
__hpvm__return(2, r, (size_t)0);
}
void var_11_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void var_11_node(void *t1, size_t bytes_t1, void *t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void *r = __hpvm__tensor_mul(t1, t2);
__hpvm__return(2, r, (size_t) 0);
void *r = __hpvm__tensor_mul(t1, t2);
__hpvm__return(2, r, (size_t)0);
}
void var_12_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void var_12_node(void *t1, size_t bytes_t1, void *t2, size_t bytes_t2) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(2, t1, t2, 0);
void *r = __hpvm__tensor_add(t1, t2);
__hpvm__return(2, r, (size_t) 0);
void *r = __hpvm__tensor_add(t1, t2);
__hpvm__return(2, r, (size_t)0);
}
void var_13_node(void* t1, size_t bytes_t1) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(1, t1, 0);
void var_13_node(void *t1, size_t bytes_t1) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(1, t1, 0);
void* r = __hpvm__tensor_tanh(t1);
__hpvm__return(2, r, (size_t) 0);
void *r = __hpvm__tensor_tanh(t1);
__hpvm__return(2, r, (size_t)0);
}
void var_14_node(void* t1, size_t bytes_t1) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(1, t1, 0);
void var_14_node(void *t1, size_t bytes_t1) {
__hpvm__hint(hpvm::CUDNN_TARGET);
__hpvm__attributes(1, t1, 0);
void* r = __hpvm__tensor_softmax(t1);
__hpvm__return(2, r, (size_t) 0);
void *r = __hpvm__tensor_softmax(t1);
__hpvm__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* 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){
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 *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) {
__hpvm__hint(hpvm::CPU_TARGET);
__hpvm__attributes(9, input, conv2d_1_w, conv2d_1_b, conv2d_2_w, conv2d_2_b,
dense_1_w, dense_1_b, dense_2_w, dense_2_b, 0);
__hpvm__hint(hpvm::CPU_TARGET);
__hpvm__attributes(9, input, conv2d_1_w, conv2d_1_b, conv2d_2_w, conv2d_2_b, dense_1_w, dense_1_b, dense_2_w, dense_2_b, 0);
void *var_0 = __hpvm__createNodeND(0, var_0_node);
__hpvm__bindIn(var_0, 0, 0, 0);
__hpvm__bindIn(var_0, 1, 1, 0);
__hpvm__bindIn(var_0, 2, 2, 0);
__hpvm__bindIn(var_0, 3, 3, 0);
void* var_0 = __hpvm__createNodeND(0, var_0_node);
void *var_1 = __hpvm__createNodeND(0, var_1_node);
__hpvm__bindIn(var_0, 0, 0, 0);
__hpvm__bindIn(var_0, 1, 1, 0);
__hpvm__bindIn(var_0, 2, 2, 0);
__hpvm__bindIn(var_0, 3, 3, 0);
__hpvm__edge(var_0, var_1, 1, 0, 0, 0);
__hpvm__edge(var_0, var_1, 1, 1, 1, 0);
__hpvm__bindIn(var_1, 4, 2, 0);
__hpvm__bindIn(var_1, 5, 3, 0);
void* var_1 = __hpvm__createNodeND(0, var_1_node);
void *var_2 = __hpvm__createNodeND(0, var_2_node);
__hpvm__edge(var_0, var_1, 1, 0, 0, 0);
__hpvm__edge(var_0, var_1, 1, 1, 1, 0);
__hpvm__bindIn(var_1, 4, 2, 0);
__hpvm__bindIn(var_1, 5, 3, 0);
__hpvm__edge(var_1, var_2, 1, 0, 0, 0);
__hpvm__edge(var_1, var_2, 1, 1, 1, 0);
void* var_2 = __hpvm__createNodeND(0, var_2_node);
void *var_3 = __hpvm__createNodeND(0, var_3_node);
__hpvm__edge(var_1, var_2, 1, 0, 0, 0);
__hpvm__edge(var_1, var_2, 1, 1, 1, 0);
__hpvm__edge(var_2, var_3, 1, 0, 0, 0);
__hpvm__edge(var_2, var_3, 1, 1, 1, 0);
void* var_3 = __hpvm__createNodeND(0, var_3_node);
void *var_4 = __hpvm__createNodeND(0, var_4_node);
__hpvm__edge(var_2, var_3, 1, 0, 0, 0);
__hpvm__edge(var_2, var_3, 1, 1, 1, 0);
__hpvm__edge(var_3, var_4, 1, 0, 0, 0);
__hpvm__edge(var_3, var_4, 1, 1, 1, 0);
__hpvm__bindIn(var_4, 6, 2, 0);
__hpvm__bindIn(var_4, 7, 3, 0);
void* var_4 = __hpvm__createNodeND(0, var_4_node);
void *var_5 = __hpvm__createNodeND(0, var_5_node);
__hpvm__edge(var_3, var_4, 1, 0, 0, 0);
__hpvm__edge(var_3, var_4, 1, 1, 1, 0);
__hpvm__bindIn(var_4, 6, 2, 0);
__hpvm__bindIn(var_4, 7, 3, 0);
__hpvm__edge(var_4, var_5, 1, 0, 0, 0);
__hpvm__edge(var_4, var_5, 1, 1, 1, 0);
__hpvm__bindIn(var_5, 8, 2, 0);
__hpvm__bindIn(var_5, 9, 3, 0);
void* var_5 = __hpvm__createNodeND(0, var_5_node);
void *var_6 = __hpvm__createNodeND(0, var_6_node);
__hpvm__edge(var_4, var_5, 1, 0, 0, 0);
__hpvm__edge(var_4, var_5, 1, 1, 1, 0);
__hpvm__bindIn(var_5, 8, 2, 0);
__hpvm__bindIn(var_5, 9, 3, 0);
__hpvm__edge(var_5, var_6, 1, 0, 0, 0);
__hpvm__edge(var_5, var_6, 1, 1, 1, 0);
void* var_6 = __hpvm__createNodeND(0, var_6_node);
void *var_7 = __hpvm__createNodeND(0, var_7_node);
__hpvm__edge(var_5, var_6, 1, 0, 0, 0);
__hpvm__edge(var_5, var_6, 1, 1, 1, 0);
__hpvm__edge(var_6, var_7, 1, 0, 0, 0);
__hpvm__edge(var_6, var_7, 1, 1, 1, 0);
void* var_7 = __hpvm__createNodeND(0, var_7_node);
void *var_8 = __hpvm__createNodeND(0, var_8_node);
__hpvm__edge(var_6, var_7, 1, 0, 0, 0);
__hpvm__edge(var_6, var_7, 1, 1, 1, 0);
__hpvm__edge(var_7, var_8, 1, 0, 0, 0);
__hpvm__edge(var_7, var_8, 1, 1, 1, 0);
__hpvm__bindIn(var_8, 10, 2, 0);
__hpvm__bindIn(var_8, 11, 3, 0);
void* var_8 = __hpvm__createNodeND(0, var_8_node);
void *var_9 = __hpvm__createNodeND(0, var_9_node);
__hpvm__edge(var_7, var_8, 1, 0, 0, 0);
__hpvm__edge(var_7, var_8, 1, 1, 1, 0);
__hpvm__bindIn(var_8, 10, 2, 0);
__hpvm__bindIn(var_8, 11, 3, 0);
__hpvm__edge(var_8, var_9, 1, 0, 0, 0);
__hpvm__edge(var_8, var_9, 1, 1, 1, 0);
__hpvm__bindIn(var_9, 12, 2, 0);
__hpvm__bindIn(var_9, 13, 3, 0);
void* var_9 = __hpvm__createNodeND(0, var_9_node);
void *var_10 = __hpvm__createNodeND(0, var_10_node);
__hpvm__edge(var_8, var_9, 1, 0, 0, 0);
__hpvm__edge(var_8, var_9, 1, 1, 1, 0);
__hpvm__bindIn(var_9, 12, 2, 0);
__hpvm__bindIn(var_9, 13, 3, 0);
__hpvm__edge(var_9, var_10, 1, 0, 0, 0);
__hpvm__edge(var_9, var_10, 1, 1, 1, 0);
void* var_10 = __hpvm__createNodeND(0, var_10_node);
void *var_11 = __hpvm__createNodeND(0, var_11_node);
__hpvm__edge(var_9, var_10, 1, 0, 0, 0);
__hpvm__edge(var_9, var_10, 1, 1, 1, 0);
__hpvm__edge(var_10, var_11, 1, 0, 0, 0);
__hpvm__edge(var_10, var_11, 1, 1, 1, 0);
__hpvm__bindIn(var_11, 14, 2, 0);
__hpvm__bindIn(var_11, 15, 3, 0);
void* var_11 = __hpvm__createNodeND(0, var_11_node);
void *var_12 = __hpvm__createNodeND(0, var_12_node);
__hpvm__edge(var_10, var_11, 1, 0, 0, 0);
__hpvm__edge(var_10, var_11, 1, 1, 1, 0);
__hpvm__bindIn(var_11, 14, 2, 0);
__hpvm__bindIn(var_11, 15, 3, 0);
__hpvm__edge(var_11, var_12, 1, 0, 0, 0);
__hpvm__edge(var_11, var_12, 1, 1, 1, 0);
__hpvm__bindIn(var_12, 16, 2, 0);
__hpvm__bindIn(var_12, 17, 3, 0);
void* var_12 = __hpvm__createNodeND(0, var_12_node);
void *var_13 = __hpvm__createNodeND(0, var_13_node);
__hpvm__edge(var_11, var_12, 1, 0, 0, 0);
__hpvm__edge(var_11, var_12, 1, 1, 1, 0);
__hpvm__bindIn(var_12, 16, 2, 0);
__hpvm__bindIn(var_12, 17, 3, 0);
__hpvm__edge(var_12, var_13, 1, 0, 0, 0);
__hpvm__edge(var_12, var_13, 1, 1, 1, 0);
void* var_13 = __hpvm__createNodeND(0, var_13_node);
void *var_14 = __hpvm__createNodeND(0, var_14_node);
__hpvm__edge(var_12, var_13, 1, 0, 0, 0);
__hpvm__edge(var_12, var_13, 1, 1, 1, 0);
void* var_14 = __hpvm__createNodeND(0, var_14_node);
__hpvm__edge(var_13, var_14, 1, 0, 0, 0);
__hpvm__edge(var_13, var_14, 1, 1, 1, 0);
__hpvm__bindOut(var_14, 0, 0, 0);
__hpvm__bindOut(var_14, 1, 1, 0);
__hpvm__edge(var_13, var_14, 1, 0, 0, 0);
__hpvm__edge(var_13, var_14, 1, 1, 1, 0);
__hpvm__bindOut(var_14, 0, 0, 0);
__hpvm__bindOut(var_14, 1, 1, 0);
}
struct ret_t {
void* tensor;
size_t bytes;
};
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* 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_mnist/");
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels32.bin");
std::string conv2d_1_w_path = dir_prefix + std::string("conv1.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("conv1_bias.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("conv2.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("conv2_bias.bin");
void* conv2d_2_b = readTrainedWeights(conv2d_2_b_path.c_str(), 0,1,64,1,1);
std::string dense_1_w_path = dir_prefix + std::string("fc1.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("fc1_bias.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("fc2.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("fc2_bias.bin");
void* dense_2_b = readTrainedWeights(dense_2_b_path.c_str(), 0,1,10,1,1);
void* input = readTrainedWeights(input_path.c_str(), 0, 5000,1,28,28);
uint32_t* labels = readLabels3(labels_path.c_str(), 5000);
__hpvm__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->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 = __hpvm__launch(0, root, (void*) args);
__hpvm__wait(dfg);
void *result = static_cast<RootIn*>(args)->input;
hpvm_request_tensor(result, 0);
__hpvm__cleanup();
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 *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(MODEL_PARAMS_DIR) + "/lenet_mnist/";
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels32.bin");
std::string conv2d_1_w_path = dir_prefix + std::string("conv1.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("conv1_bias.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("conv2.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("conv2_bias.bin");
void *conv2d_2_b =
readTrainedWeights(conv2d_2_b_path.c_str(), 0, 1, 64, 1, 1);
std::string dense_1_w_path = dir_prefix + std::string("fc1.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("fc1_bias.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("fc2.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("fc2_bias.bin");
void *dense_2_b = readTrainedWeights(dense_2_b_path.c_str(), 0, 1, 10, 1, 1);
void *input = readTrainedWeights(input_path.c_str(), 0, 5000, 1, 28, 28);
uint32_t *labels = readLabels3(labels_path.c_str(), 5000);
__hpvm__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->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 = __hpvm__launch(0, root, (void *)args);
__hpvm__wait(dfg);
void *result = static_cast<RootIn *>(args)->input;
hpvm_request_tensor(result, 0);
__hpvm__cleanup();
computeAccuracy3(labels, result);
return 0;
}
return 0;
}
hpvm/test/dnn_benchmarks/benchmarks/mobilenet/mobilenet.cpp
+ 2519
− 2124
View file @ 93538c03
This diff is collapsed.
hpvm/test/dnn_benchmarks/benchmarks/resnet18/resnet18.cpp
+ 1088
− 1058
View file @ 93538c03
This diff is collapsed.
hpvm/test/dnn_benchmarks/benchmarks/resnet18/resnet18_loop.cpp
+ 1176
− 1221
View file @ 93538c03
This diff is collapsed.
hpvm/test/dnn_benchmarks/include/config.h.in 0 → 100644
+ 1
− 0
View file @ 93538c03
#define MODEL_PARAMS_DIR "@MODEL_PARAMS_DIR@"
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