diff --git a/hpvm/projects/hpvm-tensor-rt/CMakeLists.txt b/hpvm/projects/hpvm-tensor-rt/CMakeLists.txt
index 0b66eb695097f368b2513055e35662a6a8b95fa1..d28868892f6d45e6905594e143a13aa83b1db9d6 100644
--- a/hpvm/projects/hpvm-tensor-rt/CMakeLists.txt
+++ b/hpvm/projects/hpvm-tensor-rt/CMakeLists.txt
@@ -209,6 +209,12 @@ target_link_libraries(vgg16_cifar100_fp32 tensor_runtime_online ${GPU_PROFILER_
add_executable(mobilenet_cifar10_fp32 dnn_sources/src/fp32/mobilenet.cc)
target_link_libraries(mobilenet_cifar10_fp32 tensor_runtime_online ${GPU_PROFILER_LIB} ${SOC_SIMULATOR_LIB})
+add_executable(alexnet_imagenet_fp32 dnn_sources/src/fp32/alexnet_imagenet.cc)
+target_link_libraries(alexnet_imagenet_fp32 tensor_runtime_online ${GPU_PROFILER_LIB} ${SOC_SIMULATOR_LIB})
+
+add_executable(vgg16_imagenet_fp32 dnn_sources/src/fp32/vgg16_imagenet.cc)
+target_link_libraries(vgg16_imagenet_fp32 tensor_runtime_online ${GPU_PROFILER_LIB} ${SOC_SIMULATOR_LIB})
+
add_executable(resnet50_imagenet_fp32 dnn_sources/src/fp32/resnet50_imagenet.cc)
target_link_libraries(resnet50_imagenet_fp32 tensor_runtime_online ${GPU_PROFILER_LIB} ${SOC_SIMULATOR_LIB})
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/include/utils.h b/hpvm/projects/hpvm-tensor-rt/dnn_sources/include/utils.h
index 500ff63bc86dce6cae0dee3f942639c07bf14ab3..5d1e0e66ad1a3402981682ed97e664ddcc173787 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/include/utils.h
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/include/utils.h
@@ -13,9 +13,11 @@
#include <tensor_runtime.h>
#include <tensor.h>
#include <cmath>
+#include <string.h>
std::vector<float> run_accuracies;
+std::string model_params_path = "../../../build/model_params/";
void printTensorInfo(void* tensor_ptr){
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/alexnet2_cifar10_half.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/alexnet2_cifar10_half.cc
index 161cdd249cc1e94f0a739772e0b9b9ea86993be8..d93110945b1d1a70ec29c7788d9133dc16551ee5 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/alexnet2_cifar10_half.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/alexnet2_cifar10_half.cc
@@ -18,7 +18,7 @@ void testCifarNet(){
printf("********* Alexnet2 CIFAR-10 DNN ********** \n");
- std::string dir_prefix = std::string("../model_params/alexnet2_cifar10/");
+ std::string dir_prefix = model_params_path + std::string("/alexnet2_cifar10/");
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels.bin");
std::string labels32_path = dir_prefix + std::string("labels32.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/alexnet_cifar10_half.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/alexnet_cifar10_half.cc
index 8a429862f34f95793dd9ca7caa619b10dbe568ab..b7695bbd7a24712e335f0cf8bbd25290f3261dea 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/alexnet_cifar10_half.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/alexnet_cifar10_half.cc
@@ -14,7 +14,7 @@ int main(){
llvm_hpvm_initTensorRt(0);
- std::string dir_prefix = std::string("../model_params/alexnet_cifar10/");
+ std::string dir_prefix = model_params_path + std::string("/alexnet_cifar10/");
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/lenet_mnist_half.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/lenet_mnist_half.cc
index f04ec041644394e2258414575162b961f9849667..29f392c630a36a6044c5f804e5d3a7b252591831 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/lenet_mnist_half.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/lenet_mnist_half.cc
@@ -21,7 +21,7 @@ void testLenetTanh(){
int test_batch_size = 5000;
- std::string dir_prefix = std::string("../model_params/lenet_mnist/");
+ std::string dir_prefix = model_params_path + std::string("/lenet_mnist/");
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/mobilenet_half.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/mobilenet_half.cc
index dabafd4345f29d00c7271c796a8497aba8b7772d..d662dc1584c7810d8d3631d5ac16c427c3ff8b02 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/mobilenet_half.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/mobilenet_half.cc
@@ -14,7 +14,7 @@ int main(){
llvm_hpvm_initTensorRt(0);
- std::string dir_prefix = std::string("../model_params/mobilenet/");
+ std::string dir_prefix = model_params_path + std::string("/mobilenet/");
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels.bin");
std::string conv2d_1_w_path = dir_prefix + std::string("conv2d_1_w.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/mobilenet_shallow_half.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/mobilenet_shallow_half.cc
deleted file mode 100644
index 7ce9a90e10697c979adc470345244a2cc326f0cb..0000000000000000000000000000000000000000
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/mobilenet_shallow_half.cc
+++ /dev/null
@@ -1,235 +0,0 @@
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <string.h>
-
-#include "../../../tensor_runtime/include/tensor_runtime.h"
-#include "../../include/utils.h"
-
-
-int main(int argc, char* argv[]){
-
- llvm_hpvm_initTensorRt(0);
-
-
- std::string dir_prefix = std::string("../model_params/mobilenet_shallow/");
-
- std::string input_path = dir_prefix + std::string("input.bin");
- std::string labels_path = dir_prefix + std::string("labels.bin");
- 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(), 0,32,3,3,3);
- std::string batch_normalization_1_gamma_path = dir_prefix + std::string("batch_normalization_1_gamma.bin");
- void* batch_normalization_1_gamma = readTrainedWeights(batch_normalization_1_gamma_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_beta_path = dir_prefix + std::string("batch_normalization_1_beta.bin");
- void* batch_normalization_1_beta = readTrainedWeights(batch_normalization_1_beta_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_mean_path = dir_prefix + std::string("batch_normalization_1_mean.bin");
- void* batch_normalization_1_mean = readTrainedWeights(batch_normalization_1_mean_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_variance_path = dir_prefix + std::string("batch_normalization_1_variance.bin");
- void* batch_normalization_1_variance = readTrainedWeights(batch_normalization_1_variance_path.c_str(), 0,1,32,1,1);
- std::string depthwise_conv2d_1_w_path = dir_prefix + std::string("depthwise_conv2d_1_w.bin");
- void* depthwise_conv2d_1_w = readTrainedWeights(depthwise_conv2d_1_w_path.c_str(), 0,32,1,3,3);
- std::string batch_normalization_2_gamma_path = dir_prefix + std::string("batch_normalization_2_gamma.bin");
- void* batch_normalization_2_gamma = readTrainedWeights(batch_normalization_2_gamma_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_beta_path = dir_prefix + std::string("batch_normalization_2_beta.bin");
- void* batch_normalization_2_beta = readTrainedWeights(batch_normalization_2_beta_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_mean_path = dir_prefix + std::string("batch_normalization_2_mean.bin");
- void* batch_normalization_2_mean = readTrainedWeights(batch_normalization_2_mean_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_variance_path = dir_prefix + std::string("batch_normalization_2_variance.bin");
- void* batch_normalization_2_variance = readTrainedWeights(batch_normalization_2_variance_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(), 0,64,32,1,1);
- std::string batch_normalization_3_gamma_path = dir_prefix + std::string("batch_normalization_3_gamma.bin");
- void* batch_normalization_3_gamma = readTrainedWeights(batch_normalization_3_gamma_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_beta_path = dir_prefix + std::string("batch_normalization_3_beta.bin");
- void* batch_normalization_3_beta = readTrainedWeights(batch_normalization_3_beta_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_mean_path = dir_prefix + std::string("batch_normalization_3_mean.bin");
- void* batch_normalization_3_mean = readTrainedWeights(batch_normalization_3_mean_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_variance_path = dir_prefix + std::string("batch_normalization_3_variance.bin");
- void* batch_normalization_3_variance = readTrainedWeights(batch_normalization_3_variance_path.c_str(), 0,1,64,1,1);
- std::string depthwise_conv2d_2_w_path = dir_prefix + std::string("depthwise_conv2d_2_w.bin");
- void* depthwise_conv2d_2_w = readTrainedWeights(depthwise_conv2d_2_w_path.c_str(), 0,64,1,3,3);
- std::string batch_normalization_4_gamma_path = dir_prefix + std::string("batch_normalization_4_gamma.bin");
- void* batch_normalization_4_gamma = readTrainedWeights(batch_normalization_4_gamma_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_beta_path = dir_prefix + std::string("batch_normalization_4_beta.bin");
- void* batch_normalization_4_beta = readTrainedWeights(batch_normalization_4_beta_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_mean_path = dir_prefix + std::string("batch_normalization_4_mean.bin");
- void* batch_normalization_4_mean = readTrainedWeights(batch_normalization_4_mean_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_variance_path = dir_prefix + std::string("batch_normalization_4_variance.bin");
- void* batch_normalization_4_variance = readTrainedWeights(batch_normalization_4_variance_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(), 0,128,64,1,1);
- std::string batch_normalization_5_gamma_path = dir_prefix + std::string("batch_normalization_5_gamma.bin");
- void* batch_normalization_5_gamma = readTrainedWeights(batch_normalization_5_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_beta_path = dir_prefix + std::string("batch_normalization_5_beta.bin");
- void* batch_normalization_5_beta = readTrainedWeights(batch_normalization_5_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_mean_path = dir_prefix + std::string("batch_normalization_5_mean.bin");
- void* batch_normalization_5_mean = readTrainedWeights(batch_normalization_5_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_variance_path = dir_prefix + std::string("batch_normalization_5_variance.bin");
- void* batch_normalization_5_variance = readTrainedWeights(batch_normalization_5_variance_path.c_str(), 0,1,128,1,1);
- std::string depthwise_conv2d_3_w_path = dir_prefix + std::string("depthwise_conv2d_3_w.bin");
- void* depthwise_conv2d_3_w = readTrainedWeights(depthwise_conv2d_3_w_path.c_str(), 0,128,1,3,3);
- std::string batch_normalization_6_gamma_path = dir_prefix + std::string("batch_normalization_6_gamma.bin");
- void* batch_normalization_6_gamma = readTrainedWeights(batch_normalization_6_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_beta_path = dir_prefix + std::string("batch_normalization_6_beta.bin");
- void* batch_normalization_6_beta = readTrainedWeights(batch_normalization_6_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_mean_path = dir_prefix + std::string("batch_normalization_6_mean.bin");
- void* batch_normalization_6_mean = readTrainedWeights(batch_normalization_6_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_variance_path = dir_prefix + std::string("batch_normalization_6_variance.bin");
- void* batch_normalization_6_variance = readTrainedWeights(batch_normalization_6_variance_path.c_str(), 0,1,128,1,1);
- std::string conv2d_4_w_path = dir_prefix + std::string("conv2d_4_w.bin");
- void* conv2d_4_w = readTrainedWeights(conv2d_4_w_path.c_str(), 0,128,128,1,1);
- std::string batch_normalization_7_gamma_path = dir_prefix + std::string("batch_normalization_7_gamma.bin");
- void* batch_normalization_7_gamma = readTrainedWeights(batch_normalization_7_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_beta_path = dir_prefix + std::string("batch_normalization_7_beta.bin");
- void* batch_normalization_7_beta = readTrainedWeights(batch_normalization_7_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_mean_path = dir_prefix + std::string("batch_normalization_7_mean.bin");
- void* batch_normalization_7_mean = readTrainedWeights(batch_normalization_7_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_variance_path = dir_prefix + std::string("batch_normalization_7_variance.bin");
- void* batch_normalization_7_variance = readTrainedWeights(batch_normalization_7_variance_path.c_str(), 0,1,128,1,1);
- std::string depthwise_conv2d_4_w_path = dir_prefix + std::string("depthwise_conv2d_4_w.bin");
- void* depthwise_conv2d_4_w = readTrainedWeights(depthwise_conv2d_4_w_path.c_str(), 0,128,1,3,3);
- std::string batch_normalization_8_gamma_path = dir_prefix + std::string("batch_normalization_8_gamma.bin");
- void* batch_normalization_8_gamma = readTrainedWeights(batch_normalization_8_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_beta_path = dir_prefix + std::string("batch_normalization_8_beta.bin");
- void* batch_normalization_8_beta = readTrainedWeights(batch_normalization_8_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_mean_path = dir_prefix + std::string("batch_normalization_8_mean.bin");
- void* batch_normalization_8_mean = readTrainedWeights(batch_normalization_8_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_variance_path = dir_prefix + std::string("batch_normalization_8_variance.bin");
- void* batch_normalization_8_variance = readTrainedWeights(batch_normalization_8_variance_path.c_str(), 0,1,128,1,1);
- std::string conv2d_5_w_path = dir_prefix + std::string("conv2d_5_w.bin");
- void* conv2d_5_w = readTrainedWeights(conv2d_5_w_path.c_str(), 0,256,128,1,1);
- std::string batch_normalization_9_gamma_path = dir_prefix + std::string("batch_normalization_9_gamma.bin");
- void* batch_normalization_9_gamma = readTrainedWeights(batch_normalization_9_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_beta_path = dir_prefix + std::string("batch_normalization_9_beta.bin");
- void* batch_normalization_9_beta = readTrainedWeights(batch_normalization_9_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_mean_path = dir_prefix + std::string("batch_normalization_9_mean.bin");
- void* batch_normalization_9_mean = readTrainedWeights(batch_normalization_9_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_variance_path = dir_prefix + std::string("batch_normalization_9_variance.bin");
- void* batch_normalization_9_variance = readTrainedWeights(batch_normalization_9_variance_path.c_str(), 0,1,256,1,1);
- std::string depthwise_conv2d_5_w_path = dir_prefix + std::string("depthwise_conv2d_5_w.bin");
- void* depthwise_conv2d_5_w = readTrainedWeights(depthwise_conv2d_5_w_path.c_str(), 0,256,1,3,3);
- std::string batch_normalization_10_gamma_path = dir_prefix + std::string("batch_normalization_10_gamma.bin");
- void* batch_normalization_10_gamma = readTrainedWeights(batch_normalization_10_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_beta_path = dir_prefix + std::string("batch_normalization_10_beta.bin");
- void* batch_normalization_10_beta = readTrainedWeights(batch_normalization_10_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_mean_path = dir_prefix + std::string("batch_normalization_10_mean.bin");
- void* batch_normalization_10_mean = readTrainedWeights(batch_normalization_10_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_variance_path = dir_prefix + std::string("batch_normalization_10_variance.bin");
- void* batch_normalization_10_variance = readTrainedWeights(batch_normalization_10_variance_path.c_str(), 0,1,256,1,1);
- std::string conv2d_6_w_path = dir_prefix + std::string("conv2d_6_w.bin");
- void* conv2d_6_w = readTrainedWeights(conv2d_6_w_path.c_str(), 0,256,256,1,1);
- std::string batch_normalization_11_gamma_path = dir_prefix + std::string("batch_normalization_11_gamma.bin");
- void* batch_normalization_11_gamma = readTrainedWeights(batch_normalization_11_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_beta_path = dir_prefix + std::string("batch_normalization_11_beta.bin");
- void* batch_normalization_11_beta = readTrainedWeights(batch_normalization_11_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_mean_path = dir_prefix + std::string("batch_normalization_11_mean.bin");
- void* batch_normalization_11_mean = readTrainedWeights(batch_normalization_11_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_variance_path = dir_prefix + std::string("batch_normalization_11_variance.bin");
- void* batch_normalization_11_variance = readTrainedWeights(batch_normalization_11_variance_path.c_str(), 0,1,256,1,1);
- std::string depthwise_conv2d_6_w_path = dir_prefix + std::string("depthwise_conv2d_6_w.bin");
- void* depthwise_conv2d_6_w = readTrainedWeights(depthwise_conv2d_6_w_path.c_str(), 0,256,1,3,3);
- std::string batch_normalization_12_gamma_path = dir_prefix + std::string("batch_normalization_12_gamma.bin");
- void* batch_normalization_12_gamma = readTrainedWeights(batch_normalization_12_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_beta_path = dir_prefix + std::string("batch_normalization_12_beta.bin");
- void* batch_normalization_12_beta = readTrainedWeights(batch_normalization_12_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_mean_path = dir_prefix + std::string("batch_normalization_12_mean.bin");
- void* batch_normalization_12_mean = readTrainedWeights(batch_normalization_12_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_variance_path = dir_prefix + std::string("batch_normalization_12_variance.bin");
- void* batch_normalization_12_variance = readTrainedWeights(batch_normalization_12_variance_path.c_str(), 0,1,256,1,1);
- std::string conv2d_7_w_path = dir_prefix + std::string("conv2d_7_w.bin");
- void* conv2d_7_w = readTrainedWeights(conv2d_7_w_path.c_str(), 0,512,256,1,1);
- std::string batch_normalization_13_gamma_path = dir_prefix + std::string("batch_normalization_13_gamma.bin");
- void* batch_normalization_13_gamma = readTrainedWeights(batch_normalization_13_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_beta_path = dir_prefix + std::string("batch_normalization_13_beta.bin");
- void* batch_normalization_13_beta = readTrainedWeights(batch_normalization_13_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_mean_path = dir_prefix + std::string("batch_normalization_13_mean.bin");
- void* batch_normalization_13_mean = readTrainedWeights(batch_normalization_13_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_variance_path = dir_prefix + std::string("batch_normalization_13_variance.bin");
- void* batch_normalization_13_variance = readTrainedWeights(batch_normalization_13_variance_path.c_str(), 0,1,512,1,1);
- 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(), 0,1,1,2048,10);
- 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(), 0,1,10,1,1);
-
-
- startMemTracking();
-
- int test_input_size = 2000;
- int batch_size = 1000;
- int batch_count = test_input_size / batch_size;
-
-
- float final_accuracy = 0.0;
-
- for(int i = 0; i < batch_count; i++){
-
- int start = i * batch_size;
- int end = (i + 1) * batch_size;
-
- void* input = readInputBatch(input_path.c_str(),0,start,end,3,32,32);
-
- void* var_0 = tensorHalfConvolution(input, conv2d_1_w, 1, 1, 1, 1, 1, 1);
- void* var_1 = tensorHalfBatchNorm(var_0, batch_normalization_1_gamma, batch_normalization_1_beta, batch_normalization_1_mean, batch_normalization_1_variance, 0.001);
- void* var_2 = tensorHalfRelu(var_1);
- void* var_4 = tensorHalfConvCutlass(var_2, depthwise_conv2d_1_w, 1, 1, 1, 1, 1, 32);
- void* var_5 = tensorHalfBatchNorm(var_4, batch_normalization_2_gamma, batch_normalization_2_beta, batch_normalization_2_mean, batch_normalization_2_variance, 0.001);
- void* var_6 = tensorHalfRelu(var_5);
- void* var_7 = tensorHalfConvolution(var_6, conv2d_2_w, 0, 0, 1, 1, 1, 1);
- void* var_8 = tensorHalfBatchNorm(var_7, batch_normalization_3_gamma, batch_normalization_3_beta, batch_normalization_3_mean, batch_normalization_3_variance, 0.001);
- void* var_9 = tensorHalfRelu(var_8);
- void* var_11 = tensorHalfConvCutlass(var_9, depthwise_conv2d_2_w, 1, 1, 2, 2, 1, 64);
- void* var_12 = tensorHalfBatchNorm(var_11, batch_normalization_4_gamma, batch_normalization_4_beta, batch_normalization_4_mean, batch_normalization_4_variance, 0.001);
- void* var_13 = tensorHalfRelu(var_12);
- void* var_14 = tensorHalfConvolution(var_13, conv2d_3_w, 0, 0, 1, 1, 1, 1);
- void* var_15 = tensorHalfBatchNorm(var_14, batch_normalization_5_gamma, batch_normalization_5_beta, batch_normalization_5_mean, batch_normalization_5_variance, 0.001);
- void* var_16 = tensorHalfRelu(var_15);
- void* var_18 = tensorHalfConvCutlass(var_16, depthwise_conv2d_3_w, 1, 1, 1, 1, 1, 128);
- void* var_19 = tensorHalfBatchNorm(var_18, batch_normalization_6_gamma, batch_normalization_6_beta, batch_normalization_6_mean, batch_normalization_6_variance, 0.001);
- void* var_20 = tensorHalfRelu(var_19);
- void* var_21 = tensorHalfConvolution(var_20, conv2d_4_w, 0, 0, 1, 1, 1, 1);
- void* var_22 = tensorHalfBatchNorm(var_21, batch_normalization_7_gamma, batch_normalization_7_beta, batch_normalization_7_mean, batch_normalization_7_variance, 0.001);
- void* var_23 = tensorHalfRelu(var_22);
- void* var_26 = tensorHalfConvCutlass(var_23, depthwise_conv2d_4_w, 1, 1, 2, 2, 1, 128);
- void* var_27 = tensorHalfBatchNorm(var_26, batch_normalization_8_gamma, batch_normalization_8_beta, batch_normalization_8_mean, batch_normalization_8_variance, 0.001);
- void* var_28 = tensorHalfRelu(var_27);
- void* var_29 = tensorHalfConvolution(var_28, conv2d_5_w, 0, 0, 1, 1, 1, 1);
- void* var_30 = tensorHalfBatchNorm(var_29, batch_normalization_9_gamma, batch_normalization_9_beta, batch_normalization_9_mean, batch_normalization_9_variance, 0.001);
- void* var_31 = tensorHalfRelu(var_30);
- void* var_33 = tensorHalfConvCutlass(var_31, depthwise_conv2d_5_w, 1, 1, 1, 1, 1, 256);
- void* var_34 = tensorHalfBatchNorm(var_33, batch_normalization_10_gamma, batch_normalization_10_beta, batch_normalization_10_mean, batch_normalization_10_variance, 0.001);
- void* var_35 = tensorHalfRelu(var_34);
- void* var_36 = tensorHalfConvolution(var_35, conv2d_6_w, 0, 0, 1, 1, 1, 1);
- void* var_37 = tensorHalfBatchNorm(var_36, batch_normalization_11_gamma, batch_normalization_11_beta, batch_normalization_11_mean, batch_normalization_11_variance, 0.001);
- void* var_38 = tensorHalfRelu(var_37);
- void* var_41 = tensorHalfConvCutlass(var_38, depthwise_conv2d_6_w, 1, 1, 2, 2, 1, 256);
- void* var_42 = tensorHalfBatchNorm(var_41, batch_normalization_12_gamma, batch_normalization_12_beta, batch_normalization_12_mean, batch_normalization_12_variance, 0.001);
- void* var_43 = tensorHalfRelu(var_42);
- void* var_44 = tensorHalfConvolution(var_43, conv2d_7_w, 0, 0, 1, 1, 1, 1);
- void* var_45 = tensorHalfBatchNorm(var_44, batch_normalization_13_gamma, batch_normalization_13_beta, batch_normalization_13_mean, batch_normalization_13_variance, 0.001);
- void* var_46 = tensorHalfRelu(var_45);
- void* var_47 = tensorHalfPooling(var_46,1,2,2,0,0,2,2);
- void* var_49 = tensorHalfGemmGPU(var_47, dense_1_w);
- void* var_50 = tensorHalfAdd(var_49, dense_1_b);
- void* var_51 = tensorSoftmax(var_50);
-
- uint8_t* labels = readLabelsBatch(labels_path.c_str(),start,end);
-
- float accuracy = computeAccuracy2(labels, batch_size, var_51);
- final_accuracy += accuracy;
- freeBatchMemory();
-
- }
-
- final_accuracy = final_accuracy / batch_count;
- dumpFinalAccuracy(final_accuracy);
-
- dumpExecutionAccuracies();
-
- llvm_hpvm_cleanupTensorRt();
-
- return 0;
-
-}
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/alexnet2_cifar10_half_profiling.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/alexnet2_cifar10_half_profiling.cc
deleted file mode 100644
index 82fe03247f36dbe6de31205a60344b7f44f85bad..0000000000000000000000000000000000000000
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/alexnet2_cifar10_half_profiling.cc
+++ /dev/null
@@ -1,169 +0,0 @@
-#include "/home/nvidia/Gitlab/hpvm/llvm/projects/gpu_profiler/include/profiler.h"
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <string.h>
-
-#include "../../../../tensor_runtime/include/tensor_runtime.h"
-#include "../../../include/utils.h"
-
-/* NOTE: Reference Architecture to use for profiling */
-void testCifarNet(){
-
- printf("********* Alexnet2 CIFAR-10 DNN ********** \n");
-
- std::string dir_prefix = std::string("../model_params/alexnet2_cifar10/");
- std::string input_path = dir_prefix + std::string("norm_cifar_input.bin");
- std::string labels_path = dir_prefix + std::string("test_labels.bin");
-
- void* conv1_filter = readTrainedWeights("../model_params/alexnet2_cifar10/conv1.bin",
- float_type, 32, 3, 3, 3);
- void* conv1_bias = readTrainedWeights("../model_params/alexnet2_cifar10/conv1_bias.bin",
- float_type, 1, 32, 1, 1);
- void* conv2_filter = readTrainedWeights("../model_params/alexnet2_cifar10/conv2.bin",
- float_type, 32, 32, 3, 3);
- void* conv2_bias = readTrainedWeights("../model_params/alexnet2_cifar10/conv2_bias.bin",
- float_type, 1, 32, 1, 1);
- void* conv3_filter = readTrainedWeights("../model_params/alexnet2_cifar10/conv3.bin",
- float_type, 64, 32, 3, 3);
- void* conv3_bias = readTrainedWeights("../model_params/alexnet2_cifar10/conv3_bias.bin",
- float_type, 1, 64, 1, 1);
- void* conv4_filter = readTrainedWeights("../model_params/alexnet2_cifar10/conv4.bin",
- float_type, 64, 64, 3, 3);
- void* conv4_bias = readTrainedWeights("../model_params/alexnet2_cifar10/conv4_bias.bin",
- float_type, 1, 64, 1, 1);
- void* conv5_filter = readTrainedWeights("../model_params/alexnet2_cifar10/conv5.bin",
- float_type, 128, 64, 3, 3);
- void* conv5_bias = readTrainedWeights("../model_params/alexnet2_cifar10/conv5_bias.bin",
- float_type, 1, 128, 1, 1);
- void* conv6_filter = readTrainedWeights("../model_params/alexnet2_cifar10/conv6.bin",
- float_type, 128, 128, 3, 3);
- void* conv6_bias = readTrainedWeights("../model_params/alexnet2_cifar10/conv6_bias.bin",
- float_type, 1, 128, 1, 1);
-
- void* fc1_weights = readTrainedWeights("../model_params/alexnet2_cifar10/fc1.bin",
- float_type, 1, 1, 2048, 10);
- void* fc1_bias = readTrainedWeights("../model_params/alexnet2_cifar10/fc1_bias.bin",
- float_type, 1, 10, 1, 1);
-
-
- int conv_mode = 1; // NOTE: using CROSS_CORRELATION
- int conv_precision = 0; // NOTE: using Float as compute precision. FIXIT: use enum
-
-
- startMemTracking();
-
- int test_input_size = 5000;
- int batch_size = 1000;
- int batch_count = test_input_size / batch_size;
- float final_accuracy = 0.0;
-
- int total_runs = 10;
-
- // NOTE: Starting time profiling
- startProfiling();
-
- Profiler profiler;
- profiler.start_profiler();
-
- double total_time = 0.0;
- double total_energy = 0.0;
-
- for (int i = 0; i < total_runs; i++){
- for(int i = 0; i < batch_count; i++){
-
- int start = i * batch_size;
- int end = (i + 1) * batch_size;
- void* input = readInputBatch(input_path.c_str(), 0,start,end,3,32,32);
-
- profiler.resume_profiler();
-
- void* conv1out = tensorHalfConvolution(input, conv1_filter, 1, 1, 1, 1,
- conv_mode, conv_precision);
- tensorHalfAdd(conv1out, conv1_bias);
- void* conv1_tanh = tensorHalfTanh(conv1out);
-
- // 2nd Layer
- void* conv2out = tensorHalfConvolution(conv1_tanh, conv2_filter, 1, 1, 1, 1,
- conv_mode, conv_precision);
- tensorHalfAdd(conv2out, conv2_bias);
- void* conv2_tanh = tensorHalfTanh(conv2out);
- void* pool2out = tensorHalfPooling(conv2_tanh, 0, 2, 2, 0, 0, 2, 2);
-
- // 3rd Layer
- void* conv3out = tensorHalfConvolution(pool2out, conv3_filter, 1, 1, 1, 1,
- conv_mode, conv_precision);
- tensorHalfAdd(conv3out, conv3_bias);
- void* conv3_tanh = tensorHalfTanh(conv3out);
-
- // 4th Layer
- void* conv4out = tensorHalfConvolution(conv3_tanh, conv4_filter, 1, 1, 1, 1,
- conv_mode, conv_precision);
- tensorHalfAdd(conv4out, conv4_bias);
- void* conv4_tanh = tensorHalfTanh(conv4out);
- void* pool4out = tensorHalfPooling(conv4_tanh, 0, 2, 2, 0, 0, 2, 2);
-
- // 5th Layer
- void* conv5out = tensorHalfConvolution(pool4out, conv5_filter, 1, 1, 1, 1,
- conv_mode, conv_precision);
- tensorHalfAdd(conv5out, conv5_bias);
- void* conv5_tanh = tensorHalfTanh(conv5out);
-
- // 6th Layer
- void* conv6out = tensorHalfConvolution(conv5_tanh, conv6_filter, 1, 1, 1, 1,
- conv_mode, conv_precision);
- tensorHalfAdd(conv6out, conv6_bias);
-
- void* conv6_tanh = tensorHalfTanh(conv6out);
- void* pool6out = tensorHalfPooling(conv6_tanh, 0, 2, 2, 0, 0, 2, 2);
-
- // final FC Layer
- void* gemm1out = tensorHalfGemmGPU(pool6out, fc1_weights);
- void* gemm1biasout = tensorHalfAdd(gemm1out, fc1_bias);
- void* result = tensorSoftmax(gemm1biasout);
-
- profiler.pause_profiler();
- auto time_energy = profiler.get_time_energy();
- total_time += time_energy.first;
- total_energy += time_energy.second;
-
- profiler.reset();
-
- uint8_t* labels = readLabelsBatch(labels_path.c_str(), start, end);
-
- float accuracy = computeAccuracy2(labels, batch_size, result);
- final_accuracy += accuracy;
-
- freeBatchMemory();
- }
- }
- profiler.stop_profiler();
-
- std::cout<<"---------------------------------------\n";
- std::cout<<"Average time: " << total_time / total_runs << '\n';
- std::cout<<"Average energy: " << total_energy / total_runs << '\n';
- std::cout<<"---------------------------------------\n";
-
- stopProfiling();
-
- final_accuracy = final_accuracy / batch_count / total_runs;
- dumpFinalAccuracy(final_accuracy);
-
-}
-
-
-int main(int argc, char* argv[]){
-
- llvm_hpvm_initTensorRt(0);
-
- testCifarNet();
-
- llvm_hpvm_cleanupTensorRt();
-
- return 0;
-}
-
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/alexnet_cifar10_half_profiling.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/alexnet_cifar10_half_profiling.cc
deleted file mode 100644
index 965e3170ea5c9df7dec1abe13d06581fe56f3b21..0000000000000000000000000000000000000000
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/alexnet_cifar10_half_profiling.cc
+++ /dev/null
@@ -1,126 +0,0 @@
-#include "/home/nvidia/Gitlab/hpvm/llvm/projects/gpu_profiler/include/profiler.h"
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <string.h>
-#include "../../../../tensor_runtime/include/tensor_runtime.h"
-#include "../../../include/utils.h"
-
-int main(){
-
- llvm_hpvm_initTensorRt(0);
-
-
- std::string dir_prefix = std::string("../model_params/alexnet_cifar10_front/");
- std::string input_path = dir_prefix + std::string("input.bin");
- //void* input = readTrainedWeights(input_path.c_str(), 0,10000,3,32,32);
- 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("conv0.bin");
- void* conv2d_1_w = readTrainedWeights(conv2d_1_w_path.c_str(), 0,64,3,11,11);
- std::string conv2d_1_b_path = dir_prefix + std::string("conv_bias0.bin");
- void* conv2d_1_b = readTrainedWeights(conv2d_1_b_path.c_str(), 0,1,64,1,1);
- std::string conv2d_2_w_path = dir_prefix + std::string("conv3.bin");
- void* conv2d_2_w = readTrainedWeights(conv2d_2_w_path.c_str(), 0,192,64,5,5);
- std::string conv2d_2_b_path = dir_prefix + std::string("conv_bias3.bin");
- void* conv2d_2_b = readTrainedWeights(conv2d_2_b_path.c_str(), 0,1,192,1,1);
- std::string conv2d_3_w_path = dir_prefix + std::string("conv6.bin");
- void* conv2d_3_w = readTrainedWeights(conv2d_3_w_path.c_str(), 0,384,192,3,3);
- std::string conv2d_3_b_path = dir_prefix + std::string("conv_bias6.bin");
- void* conv2d_3_b = readTrainedWeights(conv2d_3_b_path.c_str(), 0,1,384,1,1);
- std::string conv2d_4_w_path = dir_prefix + std::string("conv7.bin");
- void* conv2d_4_w = readTrainedWeights(conv2d_4_w_path.c_str(), 0,256,384,3,3);
- std::string conv2d_4_b_path = dir_prefix + std::string("conv_bias7.bin");
- void* conv2d_4_b = readTrainedWeights(conv2d_4_b_path.c_str(), 0,1,256,1,1);
- std::string conv2d_5_w_path = dir_prefix + std::string("conv8.bin");
- void* conv2d_5_w = readTrainedWeights(conv2d_5_w_path.c_str(), 0,256,256,3,3);
- std::string conv2d_5_b_path = dir_prefix + std::string("conv_bias8.bin");
- void* conv2d_5_b = readTrainedWeights(conv2d_5_b_path.c_str(), 0,1,256,1,1);
- std::string dense_1_w_path = dir_prefix + std::string("fc12.bin");
- void* dense_1_w = readTrainedWeights(dense_1_w_path.c_str(), 0,1,1,4096,10);
- std::string dense_1_b_path = dir_prefix + std::string("fc_bias12.bin");
- void* dense_1_b = readTrainedWeights(dense_1_b_path.c_str(), 0,1,10,1,1);
-
-
- startMemTracking();
-
- int test_input_size = 5000;
- int batch_size = 1000;
- int batch_count = test_input_size / batch_size;
- float final_accuracy = 0.0;
-
- int total_runs = 10;
- Profiler profiler;
- profiler.start_profiler();
-
- double total_time = 0.0;
- double total_energy = 0.0;
-
- // NOTE: Starting time profiling
- startProfiling();
- for (int i = 0; i < total_runs; i++){
- for(int i = 0; i < batch_count; i++){
-
- int start = i * batch_size;
- int end = (i + 1) * batch_size;
- void* input = readInputBatch(input_path.c_str(), 0,start,end,3,32,32);
-
- profiler.resume_profiler();
- void* var_0 = tensorHalfConvolution(input, conv2d_1_w, 5, 5, 1, 1, 1, 0);
- void* var_1 = tensorHalfAdd(var_0, conv2d_1_b);
- void* var_2 = tensorHalfTanh(var_1);
- void* var_3 = tensorHalfPooling(var_2,0,2,2,0,0,2,2);
- void* var_5 = tensorHalfConvolution(var_3, conv2d_2_w, 2, 2, 1, 1, 1, 0);
- void* var_6 = tensorHalfAdd(var_5, conv2d_2_b);
- void* var_7 = tensorHalfTanh(var_6);
- void* var_8 = tensorHalfPooling(var_7,0,2,2,0,0,2,2);
- void* var_10 = tensorHalfConvolution(var_8, conv2d_3_w, 1, 1, 1, 1, 1, 0);
- void* var_11 = tensorHalfAdd(var_10, conv2d_3_b);
- void* var_12 = tensorHalfTanh(var_11);
- void* var_13 = tensorHalfConvolution(var_12, conv2d_4_w, 1, 1, 1, 1, 1, 0);
- void* var_14 = tensorHalfAdd(var_13, conv2d_4_b);
- void* var_15 = tensorHalfTanh(var_14);
- void* var_16 = tensorHalfConvolution(var_15, conv2d_5_w, 1, 1, 1, 1, 1, 0);
- void* var_17 = tensorHalfAdd(var_16, conv2d_5_b);
- void* var_18 = tensorHalfTanh(var_17);
- void* var_19 = tensorHalfPooling(var_18,0,2,2,0,0,2,2);
- void* var_22 = tensorHalfGemmGPU(var_19, dense_1_w);
- void* var_23 = tensorHalfAdd(var_22, dense_1_b);
- void* var_24 = tensorSoftmax(var_23);
-
- profiler.pause_profiler();
- auto time_energy = profiler.get_time_energy();
- total_time += time_energy.first;
- total_energy += time_energy.second;
- profiler.reset();
-
- uint8_t* labels = readLabelsBatch(labels_path.c_str(), start, end);
-
- float accuracy = computeAccuracy2(labels,batch_size,var_24);
- final_accuracy += accuracy;
-
- freeBatchMemory();
- }
- }
- profiler.stop_profiler();
-
- std::cout<<"---------------------------------------\n";
- std::cout<<"Average time: " << total_time / total_runs << '\n';
- std::cout<<"Average energy: " << total_energy / total_runs << '\n';
- std::cout<<"---------------------------------------\n";
-
- stopProfiling();
-
- final_accuracy = final_accuracy / batch_count / total_runs;
- dumpFinalAccuracy(final_accuracy);
-
-
- llvm_hpvm_cleanupTensorRt();
-
- return 0;
-
-}
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/lenet_keras_half_profiling.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/lenet_keras_half_profiling.cc
deleted file mode 100644
index e6ffd6b03de4901780511e56afdb5faac85bb807..0000000000000000000000000000000000000000
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/lenet_keras_half_profiling.cc
+++ /dev/null
@@ -1,186 +0,0 @@
-#include "/home/nvidia/Gitlab/hpvm/llvm/projects/gpu_profiler/include/profiler.h"
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <string.h>
-
-
-#include "../../../../tensor_runtime/include/tensor_runtime.h"
-#include "../../../include/utils.h"
-
-
-bool Opentuner_run = false;
-
-
-/* NOTE: Reference Architecture to use for profiling */
-void testLenetTanh(){
-
- int total_runs = 10;
-
-
- printf("********* Lenet-2 Architecture ********** \n");
- // FIXIT: Extend this to batch of images - currently 5 images
-
- int test_batch_size = 5000;
-
- uint8_t* labels = readLabels("../model_params/lenet_params/datasets/t10k-labels-idx1-ubyte", test_batch_size);
-
- void* input = readInputTensor("../model_params/lenet_params/datasets/t10k-images-idx3-ubyte",
- CUDNN_DATA_FLOAT,
- test_batch_size, 1, 28, 28);
-
- // NOTE: Filter descriptors do NOT have batch size
- // NOTE: First two dims are output channels (configurable), input channels (MUST match input channels)
- // IMP: The output channels matches the trained model - not the Lenet arch proposed in Andrew Ng's class
- void* conv1_filter = readTrainedWeights("../model_params/lenet_keras/conv1.bin",
- float_type, 32, 1, 5, 5);
- void* conv1_bias = readTrainedWeights("../model_params/lenet_keras/conv1_bias.bin",
- float_type, 1, 32, 1, 1);
- void* conv2_filter = readTrainedWeights("../model_params/lenet_keras/conv2.bin",
- float_type, 64, 32, 5, 5);
- void* conv2_bias = readTrainedWeights("../model_params/lenet_keras/conv2_bias.bin",
- float_type, 1, 64, 1, 1);
- void* fc1_weights = readTrainedWeights("../model_params/lenet_keras/fc1.bin",
- float_type, 1, 1, 7*7*64, 1024);
- void* fc1_bias = readTrainedWeights("../model_params/lenet_keras/fc1_bias.bin",
- float_type, 1, 1024, 1, 1);
- void* fc2_weights = readTrainedWeights("../model_params/lenet_keras/fc2.bin",
- float_type, 1, 1, 1024, 10);
- void* fc2_bias = readTrainedWeights("../model_params/lenet_keras/fc2_bias.bin",
- float_type, 1, 10, 1, 1);
-
-
-
- clearTensorMap();
-
- Profiler profiler;
- profiler.start_profiler();
-
- double total_time = 0.0;
- float final_accuracy = 0.0;
-
- for(int i = 0; i < total_runs; i++){
-
- if(Opentuner_run){
-
- char* myfifo = "/tmp/myfifo";
- int fd = open(myfifo, O_RDONLY);
-
- int ret_val = fcntl(fd, F_GETFD);
- if(ret_val == -1){
- printf("Invalid descriptor \n");
- abort();
- }
-
- char str[100];
- read(fd, str, 80);
- if(strcmp(str, "stop_run") == 0){
- abort();
- }
-
- close(fd);
- }
-
-
- readOpenTunerFlags("opentuner_flags"); // Resets the OpenTuner counters
-
- // Start power and performnce profiling
- profiler.resume_profiler();
- startProfiling();
-
- int conv_mode = 1; // NOTE: using CROSS_CORRELATION
- int conv_precision = 0; // NOTE: using Float as compute precision. FIXIT: use enum
-
- // NOTE: 'SAME' convolution
- void* conv1out = tensorHalfConvolution(input, conv1_filter, 2, 2, 1, 1,
- conv_mode, conv_precision);
-
- // NOTE: For tensorAdd, the only dimension that MUST match is channels
- tensorHalfAdd(conv1out, conv1_bias); // NOTE: In place operation
-
- void* pool1out = tensorHalfPooling(conv1out, 0, 2, 2, 0, 0, 2, 2);
-
- void* conv1_tanh = tensorHalfTanh(pool1out);
-
- // NOTE: input channels have to match between tensor op inputs and outputs
- void* conv2out = tensorHalfConvolution(conv1_tanh, conv2_filter, 2, 2, 1, 1,
- conv_mode, conv_precision);
- tensorHalfAdd(conv2out, conv2_bias); // NOTE: In place operation
-
- void* pool2out = tensorHalfPooling(conv2out, 0, 2, 2, 0, 0, 2, 2);
-
- void* conv2_tanh = tensorHalfTanh(pool2out);
-
- void* gemm1out = tensorHalfGemm(conv2_tanh, fc1_weights);
-
- void* gemm1biasout = tensorHalfAdd(gemm1out, fc1_bias);
-
- void* tanh1out = tensorHalfTanh(gemm1biasout);
-
- void* gemm2out = tensorHalfGemm(tanh1out, fc2_weights);
-
- void* gemm2_biasout = tensorHalfAdd(gemm2out, fc2_bias);
-
- void* tanh2out = tensorHalfTanh(gemm2_biasout);
-
- void* result = tensorSoftmax(tanh2out);
-
- profiler.pause_profiler();
- auto time_energy = profiler.get_time_energy();
- total_time += time_energy.first;
- profiler.reset();
-
- // End profiling and dump output to profile.txt
- stopProfiling();
-
- float accuracy = computeAccuracy2(labels, test_batch_size, result);
- final_accuracy += accuracy;
- dumpAccuracyNorms();
- freeOutputTensors();
-
- if(Opentuner_run){
-
- char* myfifo = "/tmp/myfifo";
- int fd_out = open(myfifo, O_WRONLY);
- int ret_val = fcntl(fd_out, F_GETFD);
- if(ret_val == -1){
- printf("Invalid descriptor \n");
- abort();
- }
-
- const char* str = "completed***!\n\0";
- write(fd_out, str, 80);
- close(fd_out);
- }
-
- }
-
- profiler.stop_profiler();
-
- std::cout<<"---------------------------------------\n";
- std::cout<<"Average time: " << total_time / total_runs << '\n';
- std::cout<<"---------------------------------------\n";
-
- final_accuracy = final_accuracy / total_runs;
- dumpFinalAccuracy(final_accuracy);
-}
-
-
-int main(int argc, char* argv[]){
-
- if(argc > 1)
- Opentuner_run = true;
-
- llvm_hpvm_initTensorRt(0);
-
- testLenetTanh();
-
- llvm_hpvm_cleanupTensorRt();
-
- return 0;
-}
-
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/mobilenet_depthwise_half_profiling.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/mobilenet_depthwise_half_profiling.cc
deleted file mode 100644
index 641047b50dc1219f1d02bbfb75e2014840c90d96..0000000000000000000000000000000000000000
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/mobilenet_depthwise_half_profiling.cc
+++ /dev/null
@@ -1,416 +0,0 @@
-#include "/home/nvidia/Gitlab/hpvm/llvm/projects/gpu_profiler/include/profiler.h"
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <string.h>
-
-#include "../../../../tensor_runtime/include/tensor_runtime.h"
-#include "../../../include/utils.h"
-
-int main(){
-
- llvm_hpvm_initTensorRt(0);
-
-
- std::string dir_prefix = std::string("../model_params/mobilenet/");
- std::string input_path = dir_prefix + std::string("input.bin");
- std::string labels_path = dir_prefix + std::string("labels.bin");
- 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(), 0,32,3,3,3);
- std::string batch_normalization_1_gamma_path = dir_prefix + std::string("batch_normalization_1_gamma.bin");
- void* batch_normalization_1_gamma = readTrainedWeights(batch_normalization_1_gamma_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_beta_path = dir_prefix + std::string("batch_normalization_1_beta.bin");
- void* batch_normalization_1_beta = readTrainedWeights(batch_normalization_1_beta_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_mean_path = dir_prefix + std::string("batch_normalization_1_mean.bin");
- void* batch_normalization_1_mean = readTrainedWeights(batch_normalization_1_mean_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_variance_path = dir_prefix + std::string("batch_normalization_1_variance.bin");
- void* batch_normalization_1_variance = readTrainedWeights(batch_normalization_1_variance_path.c_str(), 0,1,32,1,1);
- std::string depthwise_conv2d_1_w_path = dir_prefix + std::string("depthwise_conv2d_1_w.bin");
- void* depthwise_conv2d_1_w = readTrainedWeights(depthwise_conv2d_1_w_path.c_str(), 0,32,1,3,3);
- std::string batch_normalization_2_gamma_path = dir_prefix + std::string("batch_normalization_2_gamma.bin");
- void* batch_normalization_2_gamma = readTrainedWeights(batch_normalization_2_gamma_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_beta_path = dir_prefix + std::string("batch_normalization_2_beta.bin");
- void* batch_normalization_2_beta = readTrainedWeights(batch_normalization_2_beta_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_mean_path = dir_prefix + std::string("batch_normalization_2_mean.bin");
- void* batch_normalization_2_mean = readTrainedWeights(batch_normalization_2_mean_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_variance_path = dir_prefix + std::string("batch_normalization_2_variance.bin");
- void* batch_normalization_2_variance = readTrainedWeights(batch_normalization_2_variance_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(), 0,64,32,1,1);
- std::string batch_normalization_3_gamma_path = dir_prefix + std::string("batch_normalization_3_gamma.bin");
- void* batch_normalization_3_gamma = readTrainedWeights(batch_normalization_3_gamma_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_beta_path = dir_prefix + std::string("batch_normalization_3_beta.bin");
- void* batch_normalization_3_beta = readTrainedWeights(batch_normalization_3_beta_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_mean_path = dir_prefix + std::string("batch_normalization_3_mean.bin");
- void* batch_normalization_3_mean = readTrainedWeights(batch_normalization_3_mean_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_variance_path = dir_prefix + std::string("batch_normalization_3_variance.bin");
- void* batch_normalization_3_variance = readTrainedWeights(batch_normalization_3_variance_path.c_str(), 0,1,64,1,1);
- std::string depthwise_conv2d_2_w_path = dir_prefix + std::string("depthwise_conv2d_2_w.bin");
- void* depthwise_conv2d_2_w = readTrainedWeights(depthwise_conv2d_2_w_path.c_str(), 0,64,1,3,3);
- std::string batch_normalization_4_gamma_path = dir_prefix + std::string("batch_normalization_4_gamma.bin");
- void* batch_normalization_4_gamma = readTrainedWeights(batch_normalization_4_gamma_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_beta_path = dir_prefix + std::string("batch_normalization_4_beta.bin");
- void* batch_normalization_4_beta = readTrainedWeights(batch_normalization_4_beta_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_mean_path = dir_prefix + std::string("batch_normalization_4_mean.bin");
- void* batch_normalization_4_mean = readTrainedWeights(batch_normalization_4_mean_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_variance_path = dir_prefix + std::string("batch_normalization_4_variance.bin");
- void* batch_normalization_4_variance = readTrainedWeights(batch_normalization_4_variance_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(), 0,128,64,1,1);
- std::string batch_normalization_5_gamma_path = dir_prefix + std::string("batch_normalization_5_gamma.bin");
- void* batch_normalization_5_gamma = readTrainedWeights(batch_normalization_5_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_beta_path = dir_prefix + std::string("batch_normalization_5_beta.bin");
- void* batch_normalization_5_beta = readTrainedWeights(batch_normalization_5_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_mean_path = dir_prefix + std::string("batch_normalization_5_mean.bin");
- void* batch_normalization_5_mean = readTrainedWeights(batch_normalization_5_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_variance_path = dir_prefix + std::string("batch_normalization_5_variance.bin");
- void* batch_normalization_5_variance = readTrainedWeights(batch_normalization_5_variance_path.c_str(), 0,1,128,1,1);
- std::string depthwise_conv2d_3_w_path = dir_prefix + std::string("depthwise_conv2d_3_w.bin");
- void* depthwise_conv2d_3_w = readTrainedWeights(depthwise_conv2d_3_w_path.c_str(), 0,128,1,3,3);
- std::string batch_normalization_6_gamma_path = dir_prefix + std::string("batch_normalization_6_gamma.bin");
- void* batch_normalization_6_gamma = readTrainedWeights(batch_normalization_6_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_beta_path = dir_prefix + std::string("batch_normalization_6_beta.bin");
- void* batch_normalization_6_beta = readTrainedWeights(batch_normalization_6_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_mean_path = dir_prefix + std::string("batch_normalization_6_mean.bin");
- void* batch_normalization_6_mean = readTrainedWeights(batch_normalization_6_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_variance_path = dir_prefix + std::string("batch_normalization_6_variance.bin");
- void* batch_normalization_6_variance = readTrainedWeights(batch_normalization_6_variance_path.c_str(), 0,1,128,1,1);
- std::string conv2d_4_w_path = dir_prefix + std::string("conv2d_4_w.bin");
- void* conv2d_4_w = readTrainedWeights(conv2d_4_w_path.c_str(), 0,128,128,1,1);
- std::string batch_normalization_7_gamma_path = dir_prefix + std::string("batch_normalization_7_gamma.bin");
- void* batch_normalization_7_gamma = readTrainedWeights(batch_normalization_7_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_beta_path = dir_prefix + std::string("batch_normalization_7_beta.bin");
- void* batch_normalization_7_beta = readTrainedWeights(batch_normalization_7_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_mean_path = dir_prefix + std::string("batch_normalization_7_mean.bin");
- void* batch_normalization_7_mean = readTrainedWeights(batch_normalization_7_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_variance_path = dir_prefix + std::string("batch_normalization_7_variance.bin");
- void* batch_normalization_7_variance = readTrainedWeights(batch_normalization_7_variance_path.c_str(), 0,1,128,1,1);
- std::string depthwise_conv2d_4_w_path = dir_prefix + std::string("depthwise_conv2d_4_w.bin");
- void* depthwise_conv2d_4_w = readTrainedWeights(depthwise_conv2d_4_w_path.c_str(), 0,128,1,3,3);
- std::string batch_normalization_8_gamma_path = dir_prefix + std::string("batch_normalization_8_gamma.bin");
- void* batch_normalization_8_gamma = readTrainedWeights(batch_normalization_8_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_beta_path = dir_prefix + std::string("batch_normalization_8_beta.bin");
- void* batch_normalization_8_beta = readTrainedWeights(batch_normalization_8_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_mean_path = dir_prefix + std::string("batch_normalization_8_mean.bin");
- void* batch_normalization_8_mean = readTrainedWeights(batch_normalization_8_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_variance_path = dir_prefix + std::string("batch_normalization_8_variance.bin");
- void* batch_normalization_8_variance = readTrainedWeights(batch_normalization_8_variance_path.c_str(), 0,1,128,1,1);
- std::string conv2d_5_w_path = dir_prefix + std::string("conv2d_5_w.bin");
- void* conv2d_5_w = readTrainedWeights(conv2d_5_w_path.c_str(), 0,256,128,1,1);
- std::string batch_normalization_9_gamma_path = dir_prefix + std::string("batch_normalization_9_gamma.bin");
- void* batch_normalization_9_gamma = readTrainedWeights(batch_normalization_9_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_beta_path = dir_prefix + std::string("batch_normalization_9_beta.bin");
- void* batch_normalization_9_beta = readTrainedWeights(batch_normalization_9_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_mean_path = dir_prefix + std::string("batch_normalization_9_mean.bin");
- void* batch_normalization_9_mean = readTrainedWeights(batch_normalization_9_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_variance_path = dir_prefix + std::string("batch_normalization_9_variance.bin");
- void* batch_normalization_9_variance = readTrainedWeights(batch_normalization_9_variance_path.c_str(), 0,1,256,1,1);
- std::string depthwise_conv2d_5_w_path = dir_prefix + std::string("depthwise_conv2d_5_w.bin");
- void* depthwise_conv2d_5_w = readTrainedWeights(depthwise_conv2d_5_w_path.c_str(), 0,256,1,3,3);
- std::string batch_normalization_10_gamma_path = dir_prefix + std::string("batch_normalization_10_gamma.bin");
- void* batch_normalization_10_gamma = readTrainedWeights(batch_normalization_10_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_beta_path = dir_prefix + std::string("batch_normalization_10_beta.bin");
- void* batch_normalization_10_beta = readTrainedWeights(batch_normalization_10_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_mean_path = dir_prefix + std::string("batch_normalization_10_mean.bin");
- void* batch_normalization_10_mean = readTrainedWeights(batch_normalization_10_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_variance_path = dir_prefix + std::string("batch_normalization_10_variance.bin");
- void* batch_normalization_10_variance = readTrainedWeights(batch_normalization_10_variance_path.c_str(), 0,1,256,1,1);
- std::string conv2d_6_w_path = dir_prefix + std::string("conv2d_6_w.bin");
- void* conv2d_6_w = readTrainedWeights(conv2d_6_w_path.c_str(), 0,256,256,1,1);
- std::string batch_normalization_11_gamma_path = dir_prefix + std::string("batch_normalization_11_gamma.bin");
- void* batch_normalization_11_gamma = readTrainedWeights(batch_normalization_11_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_beta_path = dir_prefix + std::string("batch_normalization_11_beta.bin");
- void* batch_normalization_11_beta = readTrainedWeights(batch_normalization_11_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_mean_path = dir_prefix + std::string("batch_normalization_11_mean.bin");
- void* batch_normalization_11_mean = readTrainedWeights(batch_normalization_11_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_variance_path = dir_prefix + std::string("batch_normalization_11_variance.bin");
- void* batch_normalization_11_variance = readTrainedWeights(batch_normalization_11_variance_path.c_str(), 0,1,256,1,1);
- std::string depthwise_conv2d_6_w_path = dir_prefix + std::string("depthwise_conv2d_6_w.bin");
- void* depthwise_conv2d_6_w = readTrainedWeights(depthwise_conv2d_6_w_path.c_str(), 0,256,1,3,3);
- std::string batch_normalization_12_gamma_path = dir_prefix + std::string("batch_normalization_12_gamma.bin");
- void* batch_normalization_12_gamma = readTrainedWeights(batch_normalization_12_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_beta_path = dir_prefix + std::string("batch_normalization_12_beta.bin");
- void* batch_normalization_12_beta = readTrainedWeights(batch_normalization_12_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_mean_path = dir_prefix + std::string("batch_normalization_12_mean.bin");
- void* batch_normalization_12_mean = readTrainedWeights(batch_normalization_12_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_variance_path = dir_prefix + std::string("batch_normalization_12_variance.bin");
- void* batch_normalization_12_variance = readTrainedWeights(batch_normalization_12_variance_path.c_str(), 0,1,256,1,1);
- std::string conv2d_7_w_path = dir_prefix + std::string("conv2d_7_w.bin");
- void* conv2d_7_w = readTrainedWeights(conv2d_7_w_path.c_str(), 0,512,256,1,1);
- std::string batch_normalization_13_gamma_path = dir_prefix + std::string("batch_normalization_13_gamma.bin");
- void* batch_normalization_13_gamma = readTrainedWeights(batch_normalization_13_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_beta_path = dir_prefix + std::string("batch_normalization_13_beta.bin");
- void* batch_normalization_13_beta = readTrainedWeights(batch_normalization_13_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_mean_path = dir_prefix + std::string("batch_normalization_13_mean.bin");
- void* batch_normalization_13_mean = readTrainedWeights(batch_normalization_13_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_variance_path = dir_prefix + std::string("batch_normalization_13_variance.bin");
- void* batch_normalization_13_variance = readTrainedWeights(batch_normalization_13_variance_path.c_str(), 0,1,512,1,1);
- std::string depthwise_conv2d_7_w_path = dir_prefix + std::string("depthwise_conv2d_7_w.bin");
- void* depthwise_conv2d_7_w = readTrainedWeights(depthwise_conv2d_7_w_path.c_str(), 0,512,1,3,3);
- std::string batch_normalization_14_gamma_path = dir_prefix + std::string("batch_normalization_14_gamma.bin");
- void* batch_normalization_14_gamma = readTrainedWeights(batch_normalization_14_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_14_beta_path = dir_prefix + std::string("batch_normalization_14_beta.bin");
- void* batch_normalization_14_beta = readTrainedWeights(batch_normalization_14_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_14_mean_path = dir_prefix + std::string("batch_normalization_14_mean.bin");
- void* batch_normalization_14_mean = readTrainedWeights(batch_normalization_14_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_14_variance_path = dir_prefix + std::string("batch_normalization_14_variance.bin");
- void* batch_normalization_14_variance = readTrainedWeights(batch_normalization_14_variance_path.c_str(), 0,1,512,1,1);
- std::string conv2d_8_w_path = dir_prefix + std::string("conv2d_8_w.bin");
- void* conv2d_8_w = readTrainedWeights(conv2d_8_w_path.c_str(), 0,512,512,1,1);
- std::string batch_normalization_15_gamma_path = dir_prefix + std::string("batch_normalization_15_gamma.bin");
- void* batch_normalization_15_gamma = readTrainedWeights(batch_normalization_15_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_15_beta_path = dir_prefix + std::string("batch_normalization_15_beta.bin");
- void* batch_normalization_15_beta = readTrainedWeights(batch_normalization_15_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_15_mean_path = dir_prefix + std::string("batch_normalization_15_mean.bin");
- void* batch_normalization_15_mean = readTrainedWeights(batch_normalization_15_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_15_variance_path = dir_prefix + std::string("batch_normalization_15_variance.bin");
- void* batch_normalization_15_variance = readTrainedWeights(batch_normalization_15_variance_path.c_str(), 0,1,512,1,1);
- std::string depthwise_conv2d_8_w_path = dir_prefix + std::string("depthwise_conv2d_8_w.bin");
- void* depthwise_conv2d_8_w = readTrainedWeights(depthwise_conv2d_8_w_path.c_str(), 0,512,1,3,3);
- std::string batch_normalization_16_gamma_path = dir_prefix + std::string("batch_normalization_16_gamma.bin");
- void* batch_normalization_16_gamma = readTrainedWeights(batch_normalization_16_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_16_beta_path = dir_prefix + std::string("batch_normalization_16_beta.bin");
- void* batch_normalization_16_beta = readTrainedWeights(batch_normalization_16_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_16_mean_path = dir_prefix + std::string("batch_normalization_16_mean.bin");
- void* batch_normalization_16_mean = readTrainedWeights(batch_normalization_16_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_16_variance_path = dir_prefix + std::string("batch_normalization_16_variance.bin");
- void* batch_normalization_16_variance = readTrainedWeights(batch_normalization_16_variance_path.c_str(), 0,1,512,1,1);
- std::string conv2d_9_w_path = dir_prefix + std::string("conv2d_9_w.bin");
- void* conv2d_9_w = readTrainedWeights(conv2d_9_w_path.c_str(), 0,512,512,1,1);
- std::string batch_normalization_17_gamma_path = dir_prefix + std::string("batch_normalization_17_gamma.bin");
- void* batch_normalization_17_gamma = readTrainedWeights(batch_normalization_17_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_17_beta_path = dir_prefix + std::string("batch_normalization_17_beta.bin");
- void* batch_normalization_17_beta = readTrainedWeights(batch_normalization_17_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_17_mean_path = dir_prefix + std::string("batch_normalization_17_mean.bin");
- void* batch_normalization_17_mean = readTrainedWeights(batch_normalization_17_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_17_variance_path = dir_prefix + std::string("batch_normalization_17_variance.bin");
- void* batch_normalization_17_variance = readTrainedWeights(batch_normalization_17_variance_path.c_str(), 0,1,512,1,1);
- std::string depthwise_conv2d_9_w_path = dir_prefix + std::string("depthwise_conv2d_9_w.bin");
- void* depthwise_conv2d_9_w = readTrainedWeights(depthwise_conv2d_9_w_path.c_str(), 0,512,1,3,3);
- std::string batch_normalization_18_gamma_path = dir_prefix + std::string("batch_normalization_18_gamma.bin");
- void* batch_normalization_18_gamma = readTrainedWeights(batch_normalization_18_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_18_beta_path = dir_prefix + std::string("batch_normalization_18_beta.bin");
- void* batch_normalization_18_beta = readTrainedWeights(batch_normalization_18_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_18_mean_path = dir_prefix + std::string("batch_normalization_18_mean.bin");
- void* batch_normalization_18_mean = readTrainedWeights(batch_normalization_18_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_18_variance_path = dir_prefix + std::string("batch_normalization_18_variance.bin");
- void* batch_normalization_18_variance = readTrainedWeights(batch_normalization_18_variance_path.c_str(), 0,1,512,1,1);
- std::string conv2d_10_w_path = dir_prefix + std::string("conv2d_10_w.bin");
- void* conv2d_10_w = readTrainedWeights(conv2d_10_w_path.c_str(), 0,512,512,1,1);
- std::string batch_normalization_19_gamma_path = dir_prefix + std::string("batch_normalization_19_gamma.bin");
- void* batch_normalization_19_gamma = readTrainedWeights(batch_normalization_19_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_19_beta_path = dir_prefix + std::string("batch_normalization_19_beta.bin");
- void* batch_normalization_19_beta = readTrainedWeights(batch_normalization_19_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_19_mean_path = dir_prefix + std::string("batch_normalization_19_mean.bin");
- void* batch_normalization_19_mean = readTrainedWeights(batch_normalization_19_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_19_variance_path = dir_prefix + std::string("batch_normalization_19_variance.bin");
- void* batch_normalization_19_variance = readTrainedWeights(batch_normalization_19_variance_path.c_str(), 0,1,512,1,1);
- std::string depthwise_conv2d_10_w_path = dir_prefix + std::string("depthwise_conv2d_10_w.bin");
- void* depthwise_conv2d_10_w = readTrainedWeights(depthwise_conv2d_10_w_path.c_str(), 0,512,1,3,3);
- std::string batch_normalization_20_gamma_path = dir_prefix + std::string("batch_normalization_20_gamma.bin");
- void* batch_normalization_20_gamma = readTrainedWeights(batch_normalization_20_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_20_beta_path = dir_prefix + std::string("batch_normalization_20_beta.bin");
- void* batch_normalization_20_beta = readTrainedWeights(batch_normalization_20_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_20_mean_path = dir_prefix + std::string("batch_normalization_20_mean.bin");
- void* batch_normalization_20_mean = readTrainedWeights(batch_normalization_20_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_20_variance_path = dir_prefix + std::string("batch_normalization_20_variance.bin");
- void* batch_normalization_20_variance = readTrainedWeights(batch_normalization_20_variance_path.c_str(), 0,1,512,1,1);
- std::string conv2d_11_w_path = dir_prefix + std::string("conv2d_11_w.bin");
- void* conv2d_11_w = readTrainedWeights(conv2d_11_w_path.c_str(), 0,512,512,1,1);
- std::string batch_normalization_21_gamma_path = dir_prefix + std::string("batch_normalization_21_gamma.bin");
- void* batch_normalization_21_gamma = readTrainedWeights(batch_normalization_21_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_21_beta_path = dir_prefix + std::string("batch_normalization_21_beta.bin");
- void* batch_normalization_21_beta = readTrainedWeights(batch_normalization_21_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_21_mean_path = dir_prefix + std::string("batch_normalization_21_mean.bin");
- void* batch_normalization_21_mean = readTrainedWeights(batch_normalization_21_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_21_variance_path = dir_prefix + std::string("batch_normalization_21_variance.bin");
- void* batch_normalization_21_variance = readTrainedWeights(batch_normalization_21_variance_path.c_str(), 0,1,512,1,1);
- std::string depthwise_conv2d_11_w_path = dir_prefix + std::string("depthwise_conv2d_11_w.bin");
- void* depthwise_conv2d_11_w = readTrainedWeights(depthwise_conv2d_11_w_path.c_str(), 0,512,1,3,3);
- std::string batch_normalization_22_gamma_path = dir_prefix + std::string("batch_normalization_22_gamma.bin");
- void* batch_normalization_22_gamma = readTrainedWeights(batch_normalization_22_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_22_beta_path = dir_prefix + std::string("batch_normalization_22_beta.bin");
- void* batch_normalization_22_beta = readTrainedWeights(batch_normalization_22_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_22_mean_path = dir_prefix + std::string("batch_normalization_22_mean.bin");
- void* batch_normalization_22_mean = readTrainedWeights(batch_normalization_22_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_22_variance_path = dir_prefix + std::string("batch_normalization_22_variance.bin");
- void* batch_normalization_22_variance = readTrainedWeights(batch_normalization_22_variance_path.c_str(), 0,1,512,1,1);
- std::string conv2d_12_w_path = dir_prefix + std::string("conv2d_12_w.bin");
- void* conv2d_12_w = readTrainedWeights(conv2d_12_w_path.c_str(), 0,512,512,1,1);
- std::string batch_normalization_23_gamma_path = dir_prefix + std::string("batch_normalization_23_gamma.bin");
- void* batch_normalization_23_gamma = readTrainedWeights(batch_normalization_23_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_23_beta_path = dir_prefix + std::string("batch_normalization_23_beta.bin");
- void* batch_normalization_23_beta = readTrainedWeights(batch_normalization_23_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_23_mean_path = dir_prefix + std::string("batch_normalization_23_mean.bin");
- void* batch_normalization_23_mean = readTrainedWeights(batch_normalization_23_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_23_variance_path = dir_prefix + std::string("batch_normalization_23_variance.bin");
- void* batch_normalization_23_variance = readTrainedWeights(batch_normalization_23_variance_path.c_str(), 0,1,512,1,1);
- std::string depthwise_conv2d_12_w_path = dir_prefix + std::string("depthwise_conv2d_12_w.bin");
- void* depthwise_conv2d_12_w = readTrainedWeights(depthwise_conv2d_12_w_path.c_str(), 0,512,1,3,3);
- std::string batch_normalization_24_gamma_path = dir_prefix + std::string("batch_normalization_24_gamma.bin");
- void* batch_normalization_24_gamma = readTrainedWeights(batch_normalization_24_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_24_beta_path = dir_prefix + std::string("batch_normalization_24_beta.bin");
- void* batch_normalization_24_beta = readTrainedWeights(batch_normalization_24_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_24_mean_path = dir_prefix + std::string("batch_normalization_24_mean.bin");
- void* batch_normalization_24_mean = readTrainedWeights(batch_normalization_24_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_24_variance_path = dir_prefix + std::string("batch_normalization_24_variance.bin");
- void* batch_normalization_24_variance = readTrainedWeights(batch_normalization_24_variance_path.c_str(), 0,1,512,1,1);
- std::string conv2d_13_w_path = dir_prefix + std::string("conv2d_13_w.bin");
- void* conv2d_13_w = readTrainedWeights(conv2d_13_w_path.c_str(), 0,1024,512,1,1);
- std::string batch_normalization_25_gamma_path = dir_prefix + std::string("batch_normalization_25_gamma.bin");
- void* batch_normalization_25_gamma = readTrainedWeights(batch_normalization_25_gamma_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_25_beta_path = dir_prefix + std::string("batch_normalization_25_beta.bin");
- void* batch_normalization_25_beta = readTrainedWeights(batch_normalization_25_beta_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_25_mean_path = dir_prefix + std::string("batch_normalization_25_mean.bin");
- void* batch_normalization_25_mean = readTrainedWeights(batch_normalization_25_mean_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_25_variance_path = dir_prefix + std::string("batch_normalization_25_variance.bin");
- void* batch_normalization_25_variance = readTrainedWeights(batch_normalization_25_variance_path.c_str(), 0,1,1024,1,1);
- std::string depthwise_conv2d_13_w_path = dir_prefix + std::string("depthwise_conv2d_13_w.bin");
- void* depthwise_conv2d_13_w = readTrainedWeights(depthwise_conv2d_13_w_path.c_str(), 0,1024,1,3,3);
- std::string batch_normalization_26_gamma_path = dir_prefix + std::string("batch_normalization_26_gamma.bin");
- void* batch_normalization_26_gamma = readTrainedWeights(batch_normalization_26_gamma_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_26_beta_path = dir_prefix + std::string("batch_normalization_26_beta.bin");
- void* batch_normalization_26_beta = readTrainedWeights(batch_normalization_26_beta_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_26_mean_path = dir_prefix + std::string("batch_normalization_26_mean.bin");
- void* batch_normalization_26_mean = readTrainedWeights(batch_normalization_26_mean_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_26_variance_path = dir_prefix + std::string("batch_normalization_26_variance.bin");
- void* batch_normalization_26_variance = readTrainedWeights(batch_normalization_26_variance_path.c_str(), 0,1,1024,1,1);
- std::string conv2d_14_w_path = dir_prefix + std::string("conv2d_14_w.bin");
- void* conv2d_14_w = readTrainedWeights(conv2d_14_w_path.c_str(), 0,1024,1024,1,1);
- std::string batch_normalization_27_gamma_path = dir_prefix + std::string("batch_normalization_27_gamma.bin");
- void* batch_normalization_27_gamma = readTrainedWeights(batch_normalization_27_gamma_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_27_beta_path = dir_prefix + std::string("batch_normalization_27_beta.bin");
- void* batch_normalization_27_beta = readTrainedWeights(batch_normalization_27_beta_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_27_mean_path = dir_prefix + std::string("batch_normalization_27_mean.bin");
- void* batch_normalization_27_mean = readTrainedWeights(batch_normalization_27_mean_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_27_variance_path = dir_prefix + std::string("batch_normalization_27_variance.bin");
- void* batch_normalization_27_variance = readTrainedWeights(batch_normalization_27_variance_path.c_str(), 0,1,1024,1,1);
- 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(), 0,1,1,1024,10);
- 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(), 0,1,10,1,1);
-
-
-
- startMemTracking();
-
- int test_input_size = 5000;
- int batch_size = 1000;
- int batch_count = test_input_size / batch_size;
-
- int total_runs = 10;
- float final_accuracy = 0.0;
-
- for (int run_num = 0; run_num < total_runs; run_num++){
- for(int i = 0; i < batch_count; i++){
-
- int start = i * batch_size;
- int end = (i + 1) * batch_size;
-
- void* input = readInputBatch(input_path.c_str(),0,start,end,3,32,32);
-
- void* var_0 = tensorHalfConvolution(input, conv2d_1_w, 1, 1, 1, 1, 1, 1);
- void* var_1 = tensorHalfBatchNorm(var_0, batch_normalization_1_gamma, batch_normalization_1_beta, batch_normalization_1_mean, batch_normalization_1_variance, 0.001);
- void* var_2 = tensorHalfRelu(var_1);
- void* var_4 = tensorHalfConvCutlass(var_2, depthwise_conv2d_1_w, 1, 1, 1, 1, 1, 32);
- void* var_5 = tensorHalfBatchNorm(var_4, batch_normalization_2_gamma, batch_normalization_2_beta, batch_normalization_2_mean, batch_normalization_2_variance, 0.001);
- void* var_6 = tensorHalfRelu(var_5);
- void* var_7 = tensorHalfConvolution(var_6, conv2d_2_w, 0, 0, 1, 1, 1, 1);
- void* var_8 = tensorHalfBatchNorm(var_7, batch_normalization_3_gamma, batch_normalization_3_beta, batch_normalization_3_mean, batch_normalization_3_variance, 0.001);
- void* var_9 = tensorHalfRelu(var_8);
- void* var_11 = tensorHalfConvCutlass(var_9, depthwise_conv2d_2_w, 1, 1, 2, 2, 1, 64);
- void* var_12 = tensorHalfBatchNorm(var_11, batch_normalization_4_gamma, batch_normalization_4_beta, batch_normalization_4_mean, batch_normalization_4_variance, 0.001);
- void* var_13 = tensorHalfRelu(var_12);
- void* var_14 = tensorHalfConvolution(var_13, conv2d_3_w, 0, 0, 1, 1, 1, 1);
- void* var_15 = tensorHalfBatchNorm(var_14, batch_normalization_5_gamma, batch_normalization_5_beta, batch_normalization_5_mean, batch_normalization_5_variance, 0.001);
- void* var_16 = tensorHalfRelu(var_15);
- void* var_18 = tensorHalfConvCutlass(var_16, depthwise_conv2d_3_w, 1, 1, 1, 1, 1, 128);
- void* var_19 = tensorHalfBatchNorm(var_18, batch_normalization_6_gamma, batch_normalization_6_beta, batch_normalization_6_mean, batch_normalization_6_variance, 0.001);
- void* var_20 = tensorHalfRelu(var_19);
- void* var_21 = tensorHalfConvolution(var_20, conv2d_4_w, 0, 0, 1, 1, 1, 1);
- void* var_22 = tensorHalfBatchNorm(var_21, batch_normalization_7_gamma, batch_normalization_7_beta, batch_normalization_7_mean, batch_normalization_7_variance, 0.001);
- void* var_23 = tensorHalfRelu(var_22);
- void* var_26 = tensorHalfConvCutlass(var_23, depthwise_conv2d_4_w, 1, 1, 2, 2, 1, 128);
- void* var_27 = tensorHalfBatchNorm(var_26, batch_normalization_8_gamma, batch_normalization_8_beta, batch_normalization_8_mean, batch_normalization_8_variance, 0.001);
- void* var_28 = tensorHalfRelu(var_27);
- void* var_29 = tensorHalfConvolution(var_28, conv2d_5_w, 0, 0, 1, 1, 1, 1);
- void* var_30 = tensorHalfBatchNorm(var_29, batch_normalization_9_gamma, batch_normalization_9_beta, batch_normalization_9_mean, batch_normalization_9_variance, 0.001);
- void* var_31 = tensorHalfRelu(var_30);
- void* var_33 = tensorHalfConvCutlass(var_31, depthwise_conv2d_5_w, 1, 1, 1, 1, 1, 256);
- void* var_34 = tensorHalfBatchNorm(var_33, batch_normalization_10_gamma, batch_normalization_10_beta, batch_normalization_10_mean, batch_normalization_10_variance, 0.001);
- void* var_35 = tensorHalfRelu(var_34);
- void* var_36 = tensorHalfConvolution(var_35, conv2d_6_w, 0, 0, 1, 1, 1, 1);
- void* var_37 = tensorHalfBatchNorm(var_36, batch_normalization_11_gamma, batch_normalization_11_beta, batch_normalization_11_mean, batch_normalization_11_variance, 0.001);
- void* var_38 = tensorHalfRelu(var_37);
- void* var_41 = tensorHalfConvCutlass(var_38, depthwise_conv2d_6_w, 1, 1, 2, 2, 1, 256);
- void* var_42 = tensorHalfBatchNorm(var_41, batch_normalization_12_gamma, batch_normalization_12_beta, batch_normalization_12_mean, batch_normalization_12_variance, 0.001);
- void* var_43 = tensorHalfRelu(var_42);
- void* var_44 = tensorHalfConvolution(var_43, conv2d_7_w, 0, 0, 1, 1, 1, 1);
- void* var_45 = tensorHalfBatchNorm(var_44, batch_normalization_13_gamma, batch_normalization_13_beta, batch_normalization_13_mean, batch_normalization_13_variance, 0.001);
- void* var_46 = tensorHalfRelu(var_45);
- void* var_48 = tensorHalfConvCutlass(var_46, depthwise_conv2d_7_w, 1, 1, 1, 1, 1, 512);
- void* var_49 = tensorHalfBatchNorm(var_48, batch_normalization_14_gamma, batch_normalization_14_beta, batch_normalization_14_mean, batch_normalization_14_variance, 0.001);
- void* var_50 = tensorHalfRelu(var_49);
- void* var_51 = tensorHalfConvolution(var_50, conv2d_8_w, 0, 0, 1, 1, 1, 1);
- void* var_52 = tensorHalfBatchNorm(var_51, batch_normalization_15_gamma, batch_normalization_15_beta, batch_normalization_15_mean, batch_normalization_15_variance, 0.001);
- void* var_53 = tensorHalfRelu(var_52);
- void* var_55 = tensorHalfConvCutlass(var_53, depthwise_conv2d_8_w, 1, 1, 1, 1, 1, 512);
- void* var_56 = tensorHalfBatchNorm(var_55, batch_normalization_16_gamma, batch_normalization_16_beta, batch_normalization_16_mean, batch_normalization_16_variance, 0.001);
- void* var_57 = tensorHalfRelu(var_56);
- void* var_58 = tensorHalfConvolution(var_57, conv2d_9_w, 0, 0, 1, 1, 1, 1);
- void* var_59 = tensorHalfBatchNorm(var_58, batch_normalization_17_gamma, batch_normalization_17_beta, batch_normalization_17_mean, batch_normalization_17_variance, 0.001);
- void* var_60 = tensorHalfRelu(var_59);
- void* var_63 = tensorHalfConvCutlass(var_60, depthwise_conv2d_9_w, 1, 1, 1, 1, 1, 512);
- void* var_64 = tensorHalfBatchNorm(var_63, batch_normalization_18_gamma, batch_normalization_18_beta, batch_normalization_18_mean, batch_normalization_18_variance, 0.001);
- void* var_65 = tensorHalfRelu(var_64);
- void* var_66 = tensorHalfConvolution(var_65, conv2d_10_w, 0, 0, 1, 1, 1, 1);
- void* var_67 = tensorHalfBatchNorm(var_66, batch_normalization_19_gamma, batch_normalization_19_beta, batch_normalization_19_mean, batch_normalization_19_variance, 0.001);
- void* var_68 = tensorHalfRelu(var_67);
- void* var_70 = tensorHalfConvCutlass(var_68, depthwise_conv2d_10_w, 1, 1, 1, 1, 1, 512);
- void* var_71 = tensorHalfBatchNorm(var_70, batch_normalization_20_gamma, batch_normalization_20_beta, batch_normalization_20_mean, batch_normalization_20_variance, 0.001);
- void* var_72 = tensorHalfRelu(var_71);
- void* var_73 = tensorHalfConvolution(var_72, conv2d_11_w, 0, 0, 1, 1, 1, 1);
- void* var_74 = tensorHalfBatchNorm(var_73, batch_normalization_21_gamma, batch_normalization_21_beta, batch_normalization_21_mean, batch_normalization_21_variance, 0.001);
- void* var_75 = tensorHalfRelu(var_74);
- void* var_77 = tensorHalfConvCutlass(var_75, depthwise_conv2d_11_w, 1, 1, 1, 1, 1, 512);
- void* var_78 = tensorHalfBatchNorm(var_77, batch_normalization_22_gamma, batch_normalization_22_beta, batch_normalization_22_mean, batch_normalization_22_variance, 0.001);
- void* var_79 = tensorHalfRelu(var_78);
- void* var_80 = tensorHalfConvolution(var_79, conv2d_12_w, 0, 0, 1, 1, 1, 1);
- void* var_81 = tensorHalfBatchNorm(var_80, batch_normalization_23_gamma, batch_normalization_23_beta, batch_normalization_23_mean, batch_normalization_23_variance, 0.001);
- void* var_82 = tensorHalfRelu(var_81);
- void* var_85 = tensorHalfConvCutlass(var_82, depthwise_conv2d_12_w, 1, 1, 2, 2, 1, 512);
- void* var_86 = tensorHalfBatchNorm(var_85, batch_normalization_24_gamma, batch_normalization_24_beta, batch_normalization_24_mean, batch_normalization_24_variance, 0.001);
- void* var_87 = tensorHalfRelu(var_86);
- void* var_88 = tensorHalfConvolution(var_87, conv2d_13_w, 0, 0, 1, 1, 1, 1);
- void* var_89 = tensorHalfBatchNorm(var_88, batch_normalization_25_gamma, batch_normalization_25_beta, batch_normalization_25_mean, batch_normalization_25_variance, 0.001);
- void* var_90 = tensorHalfRelu(var_89);
- void* var_92 = tensorHalfConvCutlass(var_90, depthwise_conv2d_13_w, 1, 1, 1, 1, 1, 1024);
- void* var_93 = tensorHalfBatchNorm(var_92, batch_normalization_26_gamma, batch_normalization_26_beta, batch_normalization_26_mean, batch_normalization_26_variance, 0.001);
- void* var_94 = tensorHalfRelu(var_93);
- void* var_95 = tensorHalfConvolution(var_94, conv2d_14_w, 0, 0, 1, 1, 1, 1);
- void* var_96 = tensorHalfBatchNorm(var_95, batch_normalization_27_gamma, batch_normalization_27_beta, batch_normalization_27_mean, batch_normalization_27_variance, 0.001);
- void* var_97 = tensorHalfRelu(var_96);
- void* var_99 = tensorHalfPooling(var_97,1,2,2,0,0,2,2);
- void* var_101 = tensorHalfGemmGPU(var_99, dense_1_w);
- void* var_102 = tensorHalfAdd(var_101, dense_1_b);
- void* var_103 = tensorSoftmax(var_102);
-
- uint8_t* labels = readLabelsBatch(labels_path.c_str(),start,end);
-
- float accuracy = computeAccuracy2(labels, batch_size, var_103);
- final_accuracy += accuracy;
- freeBatchMemory();
- }
- }
- final_accuracy = final_accuracy / batch_count;
- dumpFinalAccuracy(final_accuracy);
-
- llvm_hpvm_cleanupTensorRt();
-
- return 0;
-
-}
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/mobilenet_half_cifar10_profiling.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/mobilenet_half_cifar10_profiling.cc
deleted file mode 100644
index 1c6a3955b1ad644363947106bb0f77d6b9a77050..0000000000000000000000000000000000000000
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/mobilenet_half_cifar10_profiling.cc
+++ /dev/null
@@ -1,438 +0,0 @@
-#include "/home/nvidia/Gitlab/hpvm/llvm/projects/gpu_profiler/include/profiler.h"
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <string.h>
-
-#include "../../../../tensor_runtime/include/tensor_runtime.h"
-#include "../../../include/utils.h"
-
-int main(){
-
- llvm_hpvm_initTensorRt(0);
-
-
- std::string dir_prefix = std::string("../model_params/mobilenet_quant/");
- std::string input_path = dir_prefix + std::string("input.bin");
- std::string labels_path = dir_prefix + std::string("labels.bin");
- 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(), 0,32,3,3,3);
- std::string batch_normalization_1_gamma_path = dir_prefix + std::string("batch_normalization_1_gamma.bin");
- void* batch_normalization_1_gamma = readTrainedWeights(batch_normalization_1_gamma_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_beta_path = dir_prefix + std::string("batch_normalization_1_beta.bin");
- void* batch_normalization_1_beta = readTrainedWeights(batch_normalization_1_beta_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_mean_path = dir_prefix + std::string("batch_normalization_1_mean.bin");
- void* batch_normalization_1_mean = readTrainedWeights(batch_normalization_1_mean_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_variance_path = dir_prefix + std::string("batch_normalization_1_variance.bin");
- void* batch_normalization_1_variance = readTrainedWeights(batch_normalization_1_variance_path.c_str(), 0,1,32,1,1);
- std::string depthwise_conv2d_1_w_path = dir_prefix + std::string("depthwise_conv2d_1_w.bin");
- void* depthwise_conv2d_1_w = readTrainedWeights(depthwise_conv2d_1_w_path.c_str(), 0,32,1,3,3);
- std::string batch_normalization_2_gamma_path = dir_prefix + std::string("batch_normalization_2_gamma.bin");
- void* batch_normalization_2_gamma = readTrainedWeights(batch_normalization_2_gamma_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_beta_path = dir_prefix + std::string("batch_normalization_2_beta.bin");
- void* batch_normalization_2_beta = readTrainedWeights(batch_normalization_2_beta_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_mean_path = dir_prefix + std::string("batch_normalization_2_mean.bin");
- void* batch_normalization_2_mean = readTrainedWeights(batch_normalization_2_mean_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_variance_path = dir_prefix + std::string("batch_normalization_2_variance.bin");
- void* batch_normalization_2_variance = readTrainedWeights(batch_normalization_2_variance_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(), 0,64,32,1,1);
- std::string batch_normalization_3_gamma_path = dir_prefix + std::string("batch_normalization_3_gamma.bin");
- void* batch_normalization_3_gamma = readTrainedWeights(batch_normalization_3_gamma_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_beta_path = dir_prefix + std::string("batch_normalization_3_beta.bin");
- void* batch_normalization_3_beta = readTrainedWeights(batch_normalization_3_beta_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_mean_path = dir_prefix + std::string("batch_normalization_3_mean.bin");
- void* batch_normalization_3_mean = readTrainedWeights(batch_normalization_3_mean_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_variance_path = dir_prefix + std::string("batch_normalization_3_variance.bin");
- void* batch_normalization_3_variance = readTrainedWeights(batch_normalization_3_variance_path.c_str(), 0,1,64,1,1);
- std::string depthwise_conv2d_2_w_path = dir_prefix + std::string("depthwise_conv2d_2_w.bin");
- void* depthwise_conv2d_2_w = readTrainedWeights(depthwise_conv2d_2_w_path.c_str(), 0,64,1,3,3);
- std::string batch_normalization_4_gamma_path = dir_prefix + std::string("batch_normalization_4_gamma.bin");
- void* batch_normalization_4_gamma = readTrainedWeights(batch_normalization_4_gamma_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_beta_path = dir_prefix + std::string("batch_normalization_4_beta.bin");
- void* batch_normalization_4_beta = readTrainedWeights(batch_normalization_4_beta_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_mean_path = dir_prefix + std::string("batch_normalization_4_mean.bin");
- void* batch_normalization_4_mean = readTrainedWeights(batch_normalization_4_mean_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_variance_path = dir_prefix + std::string("batch_normalization_4_variance.bin");
- void* batch_normalization_4_variance = readTrainedWeights(batch_normalization_4_variance_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(), 0,128,64,1,1);
- std::string batch_normalization_5_gamma_path = dir_prefix + std::string("batch_normalization_5_gamma.bin");
- void* batch_normalization_5_gamma = readTrainedWeights(batch_normalization_5_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_beta_path = dir_prefix + std::string("batch_normalization_5_beta.bin");
- void* batch_normalization_5_beta = readTrainedWeights(batch_normalization_5_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_mean_path = dir_prefix + std::string("batch_normalization_5_mean.bin");
- void* batch_normalization_5_mean = readTrainedWeights(batch_normalization_5_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_variance_path = dir_prefix + std::string("batch_normalization_5_variance.bin");
- void* batch_normalization_5_variance = readTrainedWeights(batch_normalization_5_variance_path.c_str(), 0,1,128,1,1);
- std::string depthwise_conv2d_3_w_path = dir_prefix + std::string("depthwise_conv2d_3_w.bin");
- void* depthwise_conv2d_3_w = readTrainedWeights(depthwise_conv2d_3_w_path.c_str(), 0,128,1,3,3);
- std::string batch_normalization_6_gamma_path = dir_prefix + std::string("batch_normalization_6_gamma.bin");
- void* batch_normalization_6_gamma = readTrainedWeights(batch_normalization_6_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_beta_path = dir_prefix + std::string("batch_normalization_6_beta.bin");
- void* batch_normalization_6_beta = readTrainedWeights(batch_normalization_6_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_mean_path = dir_prefix + std::string("batch_normalization_6_mean.bin");
- void* batch_normalization_6_mean = readTrainedWeights(batch_normalization_6_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_variance_path = dir_prefix + std::string("batch_normalization_6_variance.bin");
- void* batch_normalization_6_variance = readTrainedWeights(batch_normalization_6_variance_path.c_str(), 0,1,128,1,1);
- std::string conv2d_4_w_path = dir_prefix + std::string("conv2d_4_w.bin");
- void* conv2d_4_w = readTrainedWeights(conv2d_4_w_path.c_str(), 0,128,128,1,1);
- std::string batch_normalization_7_gamma_path = dir_prefix + std::string("batch_normalization_7_gamma.bin");
- void* batch_normalization_7_gamma = readTrainedWeights(batch_normalization_7_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_beta_path = dir_prefix + std::string("batch_normalization_7_beta.bin");
- void* batch_normalization_7_beta = readTrainedWeights(batch_normalization_7_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_mean_path = dir_prefix + std::string("batch_normalization_7_mean.bin");
- void* batch_normalization_7_mean = readTrainedWeights(batch_normalization_7_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_variance_path = dir_prefix + std::string("batch_normalization_7_variance.bin");
- void* batch_normalization_7_variance = readTrainedWeights(batch_normalization_7_variance_path.c_str(), 0,1,128,1,1);
- std::string depthwise_conv2d_4_w_path = dir_prefix + std::string("depthwise_conv2d_4_w.bin");
- void* depthwise_conv2d_4_w = readTrainedWeights(depthwise_conv2d_4_w_path.c_str(), 0,128,1,3,3);
- std::string batch_normalization_8_gamma_path = dir_prefix + std::string("batch_normalization_8_gamma.bin");
- void* batch_normalization_8_gamma = readTrainedWeights(batch_normalization_8_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_beta_path = dir_prefix + std::string("batch_normalization_8_beta.bin");
- void* batch_normalization_8_beta = readTrainedWeights(batch_normalization_8_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_mean_path = dir_prefix + std::string("batch_normalization_8_mean.bin");
- void* batch_normalization_8_mean = readTrainedWeights(batch_normalization_8_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_variance_path = dir_prefix + std::string("batch_normalization_8_variance.bin");
- void* batch_normalization_8_variance = readTrainedWeights(batch_normalization_8_variance_path.c_str(), 0,1,128,1,1);
- std::string conv2d_5_w_path = dir_prefix + std::string("conv2d_5_w.bin");
- void* conv2d_5_w = readTrainedWeights(conv2d_5_w_path.c_str(), 0,256,128,1,1);
- std::string batch_normalization_9_gamma_path = dir_prefix + std::string("batch_normalization_9_gamma.bin");
- void* batch_normalization_9_gamma = readTrainedWeights(batch_normalization_9_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_beta_path = dir_prefix + std::string("batch_normalization_9_beta.bin");
- void* batch_normalization_9_beta = readTrainedWeights(batch_normalization_9_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_mean_path = dir_prefix + std::string("batch_normalization_9_mean.bin");
- void* batch_normalization_9_mean = readTrainedWeights(batch_normalization_9_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_variance_path = dir_prefix + std::string("batch_normalization_9_variance.bin");
- void* batch_normalization_9_variance = readTrainedWeights(batch_normalization_9_variance_path.c_str(), 0,1,256,1,1);
- std::string depthwise_conv2d_5_w_path = dir_prefix + std::string("depthwise_conv2d_5_w.bin");
- void* depthwise_conv2d_5_w = readTrainedWeights(depthwise_conv2d_5_w_path.c_str(), 0,256,1,3,3);
- std::string batch_normalization_10_gamma_path = dir_prefix + std::string("batch_normalization_10_gamma.bin");
- void* batch_normalization_10_gamma = readTrainedWeights(batch_normalization_10_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_beta_path = dir_prefix + std::string("batch_normalization_10_beta.bin");
- void* batch_normalization_10_beta = readTrainedWeights(batch_normalization_10_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_mean_path = dir_prefix + std::string("batch_normalization_10_mean.bin");
- void* batch_normalization_10_mean = readTrainedWeights(batch_normalization_10_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_variance_path = dir_prefix + std::string("batch_normalization_10_variance.bin");
- void* batch_normalization_10_variance = readTrainedWeights(batch_normalization_10_variance_path.c_str(), 0,1,256,1,1);
- std::string conv2d_6_w_path = dir_prefix + std::string("conv2d_6_w.bin");
- void* conv2d_6_w = readTrainedWeights(conv2d_6_w_path.c_str(), 0,256,256,1,1);
- std::string batch_normalization_11_gamma_path = dir_prefix + std::string("batch_normalization_11_gamma.bin");
- void* batch_normalization_11_gamma = readTrainedWeights(batch_normalization_11_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_beta_path = dir_prefix + std::string("batch_normalization_11_beta.bin");
- void* batch_normalization_11_beta = readTrainedWeights(batch_normalization_11_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_mean_path = dir_prefix + std::string("batch_normalization_11_mean.bin");
- void* batch_normalization_11_mean = readTrainedWeights(batch_normalization_11_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_variance_path = dir_prefix + std::string("batch_normalization_11_variance.bin");
- void* batch_normalization_11_variance = readTrainedWeights(batch_normalization_11_variance_path.c_str(), 0,1,256,1,1);
- std::string depthwise_conv2d_6_w_path = dir_prefix + std::string("depthwise_conv2d_6_w.bin");
- void* depthwise_conv2d_6_w = readTrainedWeights(depthwise_conv2d_6_w_path.c_str(), 0,256,1,3,3);
- std::string batch_normalization_12_gamma_path = dir_prefix + std::string("batch_normalization_12_gamma.bin");
- void* batch_normalization_12_gamma = readTrainedWeights(batch_normalization_12_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_beta_path = dir_prefix + std::string("batch_normalization_12_beta.bin");
- void* batch_normalization_12_beta = readTrainedWeights(batch_normalization_12_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_mean_path = dir_prefix + std::string("batch_normalization_12_mean.bin");
- void* batch_normalization_12_mean = readTrainedWeights(batch_normalization_12_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_variance_path = dir_prefix + std::string("batch_normalization_12_variance.bin");
- void* batch_normalization_12_variance = readTrainedWeights(batch_normalization_12_variance_path.c_str(), 0,1,256,1,1);
- std::string conv2d_7_w_path = dir_prefix + std::string("conv2d_7_w.bin");
- void* conv2d_7_w = readTrainedWeights(conv2d_7_w_path.c_str(), 0,512,256,1,1);
- std::string batch_normalization_13_gamma_path = dir_prefix + std::string("batch_normalization_13_gamma.bin");
- void* batch_normalization_13_gamma = readTrainedWeights(batch_normalization_13_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_beta_path = dir_prefix + std::string("batch_normalization_13_beta.bin");
- void* batch_normalization_13_beta = readTrainedWeights(batch_normalization_13_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_mean_path = dir_prefix + std::string("batch_normalization_13_mean.bin");
- void* batch_normalization_13_mean = readTrainedWeights(batch_normalization_13_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_variance_path = dir_prefix + std::string("batch_normalization_13_variance.bin");
- void* batch_normalization_13_variance = readTrainedWeights(batch_normalization_13_variance_path.c_str(), 0,1,512,1,1);
- std::string depthwise_conv2d_7_w_path = dir_prefix + std::string("depthwise_conv2d_7_w.bin");
- void* depthwise_conv2d_7_w = readTrainedWeights(depthwise_conv2d_7_w_path.c_str(), 0,512,1,3,3);
- std::string batch_normalization_14_gamma_path = dir_prefix + std::string("batch_normalization_14_gamma.bin");
- void* batch_normalization_14_gamma = readTrainedWeights(batch_normalization_14_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_14_beta_path = dir_prefix + std::string("batch_normalization_14_beta.bin");
- void* batch_normalization_14_beta = readTrainedWeights(batch_normalization_14_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_14_mean_path = dir_prefix + std::string("batch_normalization_14_mean.bin");
- void* batch_normalization_14_mean = readTrainedWeights(batch_normalization_14_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_14_variance_path = dir_prefix + std::string("batch_normalization_14_variance.bin");
- void* batch_normalization_14_variance = readTrainedWeights(batch_normalization_14_variance_path.c_str(), 0,1,512,1,1);
- std::string conv2d_8_w_path = dir_prefix + std::string("conv2d_8_w.bin");
- void* conv2d_8_w = readTrainedWeights(conv2d_8_w_path.c_str(), 0,512,512,1,1);
- std::string batch_normalization_15_gamma_path = dir_prefix + std::string("batch_normalization_15_gamma.bin");
- void* batch_normalization_15_gamma = readTrainedWeights(batch_normalization_15_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_15_beta_path = dir_prefix + std::string("batch_normalization_15_beta.bin");
- void* batch_normalization_15_beta = readTrainedWeights(batch_normalization_15_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_15_mean_path = dir_prefix + std::string("batch_normalization_15_mean.bin");
- void* batch_normalization_15_mean = readTrainedWeights(batch_normalization_15_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_15_variance_path = dir_prefix + std::string("batch_normalization_15_variance.bin");
- void* batch_normalization_15_variance = readTrainedWeights(batch_normalization_15_variance_path.c_str(), 0,1,512,1,1);
- std::string depthwise_conv2d_8_w_path = dir_prefix + std::string("depthwise_conv2d_8_w.bin");
- void* depthwise_conv2d_8_w = readTrainedWeights(depthwise_conv2d_8_w_path.c_str(), 0,512,1,3,3);
- std::string batch_normalization_16_gamma_path = dir_prefix + std::string("batch_normalization_16_gamma.bin");
- void* batch_normalization_16_gamma = readTrainedWeights(batch_normalization_16_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_16_beta_path = dir_prefix + std::string("batch_normalization_16_beta.bin");
- void* batch_normalization_16_beta = readTrainedWeights(batch_normalization_16_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_16_mean_path = dir_prefix + std::string("batch_normalization_16_mean.bin");
- void* batch_normalization_16_mean = readTrainedWeights(batch_normalization_16_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_16_variance_path = dir_prefix + std::string("batch_normalization_16_variance.bin");
- void* batch_normalization_16_variance = readTrainedWeights(batch_normalization_16_variance_path.c_str(), 0,1,512,1,1);
- std::string conv2d_9_w_path = dir_prefix + std::string("conv2d_9_w.bin");
- void* conv2d_9_w = readTrainedWeights(conv2d_9_w_path.c_str(), 0,512,512,1,1);
- std::string batch_normalization_17_gamma_path = dir_prefix + std::string("batch_normalization_17_gamma.bin");
- void* batch_normalization_17_gamma = readTrainedWeights(batch_normalization_17_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_17_beta_path = dir_prefix + std::string("batch_normalization_17_beta.bin");
- void* batch_normalization_17_beta = readTrainedWeights(batch_normalization_17_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_17_mean_path = dir_prefix + std::string("batch_normalization_17_mean.bin");
- void* batch_normalization_17_mean = readTrainedWeights(batch_normalization_17_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_17_variance_path = dir_prefix + std::string("batch_normalization_17_variance.bin");
- void* batch_normalization_17_variance = readTrainedWeights(batch_normalization_17_variance_path.c_str(), 0,1,512,1,1);
- std::string depthwise_conv2d_9_w_path = dir_prefix + std::string("depthwise_conv2d_9_w.bin");
- void* depthwise_conv2d_9_w = readTrainedWeights(depthwise_conv2d_9_w_path.c_str(), 0,512,1,3,3);
- std::string batch_normalization_18_gamma_path = dir_prefix + std::string("batch_normalization_18_gamma.bin");
- void* batch_normalization_18_gamma = readTrainedWeights(batch_normalization_18_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_18_beta_path = dir_prefix + std::string("batch_normalization_18_beta.bin");
- void* batch_normalization_18_beta = readTrainedWeights(batch_normalization_18_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_18_mean_path = dir_prefix + std::string("batch_normalization_18_mean.bin");
- void* batch_normalization_18_mean = readTrainedWeights(batch_normalization_18_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_18_variance_path = dir_prefix + std::string("batch_normalization_18_variance.bin");
- void* batch_normalization_18_variance = readTrainedWeights(batch_normalization_18_variance_path.c_str(), 0,1,512,1,1);
- std::string conv2d_10_w_path = dir_prefix + std::string("conv2d_10_w.bin");
- void* conv2d_10_w = readTrainedWeights(conv2d_10_w_path.c_str(), 0,512,512,1,1);
- std::string batch_normalization_19_gamma_path = dir_prefix + std::string("batch_normalization_19_gamma.bin");
- void* batch_normalization_19_gamma = readTrainedWeights(batch_normalization_19_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_19_beta_path = dir_prefix + std::string("batch_normalization_19_beta.bin");
- void* batch_normalization_19_beta = readTrainedWeights(batch_normalization_19_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_19_mean_path = dir_prefix + std::string("batch_normalization_19_mean.bin");
- void* batch_normalization_19_mean = readTrainedWeights(batch_normalization_19_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_19_variance_path = dir_prefix + std::string("batch_normalization_19_variance.bin");
- void* batch_normalization_19_variance = readTrainedWeights(batch_normalization_19_variance_path.c_str(), 0,1,512,1,1);
- std::string depthwise_conv2d_10_w_path = dir_prefix + std::string("depthwise_conv2d_10_w.bin");
- void* depthwise_conv2d_10_w = readTrainedWeights(depthwise_conv2d_10_w_path.c_str(), 0,512,1,3,3);
- std::string batch_normalization_20_gamma_path = dir_prefix + std::string("batch_normalization_20_gamma.bin");
- void* batch_normalization_20_gamma = readTrainedWeights(batch_normalization_20_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_20_beta_path = dir_prefix + std::string("batch_normalization_20_beta.bin");
- void* batch_normalization_20_beta = readTrainedWeights(batch_normalization_20_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_20_mean_path = dir_prefix + std::string("batch_normalization_20_mean.bin");
- void* batch_normalization_20_mean = readTrainedWeights(batch_normalization_20_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_20_variance_path = dir_prefix + std::string("batch_normalization_20_variance.bin");
- void* batch_normalization_20_variance = readTrainedWeights(batch_normalization_20_variance_path.c_str(), 0,1,512,1,1);
- std::string conv2d_11_w_path = dir_prefix + std::string("conv2d_11_w.bin");
- void* conv2d_11_w = readTrainedWeights(conv2d_11_w_path.c_str(), 0,512,512,1,1);
- std::string batch_normalization_21_gamma_path = dir_prefix + std::string("batch_normalization_21_gamma.bin");
- void* batch_normalization_21_gamma = readTrainedWeights(batch_normalization_21_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_21_beta_path = dir_prefix + std::string("batch_normalization_21_beta.bin");
- void* batch_normalization_21_beta = readTrainedWeights(batch_normalization_21_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_21_mean_path = dir_prefix + std::string("batch_normalization_21_mean.bin");
- void* batch_normalization_21_mean = readTrainedWeights(batch_normalization_21_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_21_variance_path = dir_prefix + std::string("batch_normalization_21_variance.bin");
- void* batch_normalization_21_variance = readTrainedWeights(batch_normalization_21_variance_path.c_str(), 0,1,512,1,1);
- std::string depthwise_conv2d_11_w_path = dir_prefix + std::string("depthwise_conv2d_11_w.bin");
- void* depthwise_conv2d_11_w = readTrainedWeights(depthwise_conv2d_11_w_path.c_str(), 0,512,1,3,3);
- std::string batch_normalization_22_gamma_path = dir_prefix + std::string("batch_normalization_22_gamma.bin");
- void* batch_normalization_22_gamma = readTrainedWeights(batch_normalization_22_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_22_beta_path = dir_prefix + std::string("batch_normalization_22_beta.bin");
- void* batch_normalization_22_beta = readTrainedWeights(batch_normalization_22_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_22_mean_path = dir_prefix + std::string("batch_normalization_22_mean.bin");
- void* batch_normalization_22_mean = readTrainedWeights(batch_normalization_22_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_22_variance_path = dir_prefix + std::string("batch_normalization_22_variance.bin");
- void* batch_normalization_22_variance = readTrainedWeights(batch_normalization_22_variance_path.c_str(), 0,1,512,1,1);
- std::string conv2d_12_w_path = dir_prefix + std::string("conv2d_12_w.bin");
- void* conv2d_12_w = readTrainedWeights(conv2d_12_w_path.c_str(), 0,512,512,1,1);
- std::string batch_normalization_23_gamma_path = dir_prefix + std::string("batch_normalization_23_gamma.bin");
- void* batch_normalization_23_gamma = readTrainedWeights(batch_normalization_23_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_23_beta_path = dir_prefix + std::string("batch_normalization_23_beta.bin");
- void* batch_normalization_23_beta = readTrainedWeights(batch_normalization_23_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_23_mean_path = dir_prefix + std::string("batch_normalization_23_mean.bin");
- void* batch_normalization_23_mean = readTrainedWeights(batch_normalization_23_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_23_variance_path = dir_prefix + std::string("batch_normalization_23_variance.bin");
- void* batch_normalization_23_variance = readTrainedWeights(batch_normalization_23_variance_path.c_str(), 0,1,512,1,1);
- std::string depthwise_conv2d_12_w_path = dir_prefix + std::string("depthwise_conv2d_12_w.bin");
- void* depthwise_conv2d_12_w = readTrainedWeights(depthwise_conv2d_12_w_path.c_str(), 0,512,1,3,3);
- std::string batch_normalization_24_gamma_path = dir_prefix + std::string("batch_normalization_24_gamma.bin");
- void* batch_normalization_24_gamma = readTrainedWeights(batch_normalization_24_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_24_beta_path = dir_prefix + std::string("batch_normalization_24_beta.bin");
- void* batch_normalization_24_beta = readTrainedWeights(batch_normalization_24_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_24_mean_path = dir_prefix + std::string("batch_normalization_24_mean.bin");
- void* batch_normalization_24_mean = readTrainedWeights(batch_normalization_24_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_24_variance_path = dir_prefix + std::string("batch_normalization_24_variance.bin");
- void* batch_normalization_24_variance = readTrainedWeights(batch_normalization_24_variance_path.c_str(), 0,1,512,1,1);
- std::string conv2d_13_w_path = dir_prefix + std::string("conv2d_13_w.bin");
- void* conv2d_13_w = readTrainedWeights(conv2d_13_w_path.c_str(), 0,1024,512,1,1);
- std::string batch_normalization_25_gamma_path = dir_prefix + std::string("batch_normalization_25_gamma.bin");
- void* batch_normalization_25_gamma = readTrainedWeights(batch_normalization_25_gamma_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_25_beta_path = dir_prefix + std::string("batch_normalization_25_beta.bin");
- void* batch_normalization_25_beta = readTrainedWeights(batch_normalization_25_beta_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_25_mean_path = dir_prefix + std::string("batch_normalization_25_mean.bin");
- void* batch_normalization_25_mean = readTrainedWeights(batch_normalization_25_mean_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_25_variance_path = dir_prefix + std::string("batch_normalization_25_variance.bin");
- void* batch_normalization_25_variance = readTrainedWeights(batch_normalization_25_variance_path.c_str(), 0,1,1024,1,1);
- std::string depthwise_conv2d_13_w_path = dir_prefix + std::string("depthwise_conv2d_13_w.bin");
- void* depthwise_conv2d_13_w = readTrainedWeights(depthwise_conv2d_13_w_path.c_str(), 0,1024,1,3,3);
- std::string batch_normalization_26_gamma_path = dir_prefix + std::string("batch_normalization_26_gamma.bin");
- void* batch_normalization_26_gamma = readTrainedWeights(batch_normalization_26_gamma_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_26_beta_path = dir_prefix + std::string("batch_normalization_26_beta.bin");
- void* batch_normalization_26_beta = readTrainedWeights(batch_normalization_26_beta_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_26_mean_path = dir_prefix + std::string("batch_normalization_26_mean.bin");
- void* batch_normalization_26_mean = readTrainedWeights(batch_normalization_26_mean_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_26_variance_path = dir_prefix + std::string("batch_normalization_26_variance.bin");
- void* batch_normalization_26_variance = readTrainedWeights(batch_normalization_26_variance_path.c_str(), 0,1,1024,1,1);
- std::string conv2d_14_w_path = dir_prefix + std::string("conv2d_14_w.bin");
- void* conv2d_14_w = readTrainedWeights(conv2d_14_w_path.c_str(), 0,1024,1024,1,1);
- std::string batch_normalization_27_gamma_path = dir_prefix + std::string("batch_normalization_27_gamma.bin");
- void* batch_normalization_27_gamma = readTrainedWeights(batch_normalization_27_gamma_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_27_beta_path = dir_prefix + std::string("batch_normalization_27_beta.bin");
- void* batch_normalization_27_beta = readTrainedWeights(batch_normalization_27_beta_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_27_mean_path = dir_prefix + std::string("batch_normalization_27_mean.bin");
- void* batch_normalization_27_mean = readTrainedWeights(batch_normalization_27_mean_path.c_str(), 0,1,1024,1,1);
- std::string batch_normalization_27_variance_path = dir_prefix + std::string("batch_normalization_27_variance.bin");
- void* batch_normalization_27_variance = readTrainedWeights(batch_normalization_27_variance_path.c_str(), 0,1,1024,1,1);
- 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(), 0,1,1,1024,10);
- 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(), 0,1,10,1,1);
-
-
- startMemTracking();
-
- startProfiling();
-
- int test_input_size = 5000;
- int batch_size = 1000;
- int batch_count = test_input_size / batch_size;
- float final_accuracy = 0.0;
-
- int total_runs = 10;
-
- Profiler profiler;
- profiler.start_profiler();
-
- double total_time = 0.0;
-
- for(int i = 0; i < total_runs; i++){
- for(int i = 0; i < batch_count; i++){
-
- int start = i * batch_size;
- int end = (i + 1) * batch_size;
-
- void* input = readInputBatch(input_path.c_str(),0,start,end,3,32,32);
-
- profiler.resume_profiler();
- void* var_0 = tensorHalfConvolution(input, conv2d_1_w, 1, 1, 1, 1, 1, 1);
- void* var_1 = tensorHalfBatchNorm(var_0, batch_normalization_1_gamma, batch_normalization_1_beta, batch_normalization_1_mean, batch_normalization_1_variance, 0.001);
- void* var_2 = tensorHalfRelu(var_1);
- void* var_4 = tensorHalfConvolution(var_2, depthwise_conv2d_1_w, 1, 1, 1, 1, 1, 32);
- void* var_5 = tensorHalfBatchNorm(var_4, batch_normalization_2_gamma, batch_normalization_2_beta, batch_normalization_2_mean, batch_normalization_2_variance, 0.001);
- void* var_6 = tensorHalfRelu(var_5);
- void* var_7 = tensorHalfConvolution(var_6, conv2d_2_w, 0, 0, 1, 1, 1, 1);
- void* var_8 = tensorHalfBatchNorm(var_7, batch_normalization_3_gamma, batch_normalization_3_beta, batch_normalization_3_mean, batch_normalization_3_variance, 0.001);
- void* var_9 = tensorHalfRelu(var_8);
- void* var_11 = tensorHalfConvolution(var_9, depthwise_conv2d_2_w, 1, 1, 2, 2, 1, 64);
- void* var_12 = tensorHalfBatchNorm(var_11, batch_normalization_4_gamma, batch_normalization_4_beta, batch_normalization_4_mean, batch_normalization_4_variance, 0.001);
- void* var_13 = tensorHalfRelu(var_12);
- void* var_14 = tensorHalfConvolution(var_13, conv2d_3_w, 0, 0, 1, 1, 1, 1);
- void* var_15 = tensorHalfBatchNorm(var_14, batch_normalization_5_gamma, batch_normalization_5_beta, batch_normalization_5_mean, batch_normalization_5_variance, 0.001);
- void* var_16 = tensorHalfRelu(var_15);
- void* var_18 = tensorHalfConvolution(var_16, depthwise_conv2d_3_w, 1, 1, 1, 1, 1, 128);
- void* var_19 = tensorHalfBatchNorm(var_18, batch_normalization_6_gamma, batch_normalization_6_beta, batch_normalization_6_mean, batch_normalization_6_variance, 0.001);
- void* var_20 = tensorHalfRelu(var_19);
- void* var_21 = tensorHalfConvolution(var_20, conv2d_4_w, 0, 0, 1, 1, 1, 1);
- void* var_22 = tensorHalfBatchNorm(var_21, batch_normalization_7_gamma, batch_normalization_7_beta, batch_normalization_7_mean, batch_normalization_7_variance, 0.001);
- void* var_23 = tensorHalfRelu(var_22);
- void* var_26 = tensorHalfConvolution(var_23, depthwise_conv2d_4_w, 1, 1, 2, 2, 1, 128);
- void* var_27 = tensorHalfBatchNorm(var_26, batch_normalization_8_gamma, batch_normalization_8_beta, batch_normalization_8_mean, batch_normalization_8_variance, 0.001);
- void* var_28 = tensorHalfRelu(var_27);
- void* var_29 = tensorHalfConvolution(var_28, conv2d_5_w, 0, 0, 1, 1, 1, 1);
- void* var_30 = tensorHalfBatchNorm(var_29, batch_normalization_9_gamma, batch_normalization_9_beta, batch_normalization_9_mean, batch_normalization_9_variance, 0.001);
- void* var_31 = tensorHalfRelu(var_30);
- void* var_33 = tensorHalfConvolution(var_31, depthwise_conv2d_5_w, 1, 1, 1, 1, 1, 256);
- void* var_34 = tensorHalfBatchNorm(var_33, batch_normalization_10_gamma, batch_normalization_10_beta, batch_normalization_10_mean, batch_normalization_10_variance, 0.001);
- void* var_35 = tensorHalfRelu(var_34);
- void* var_36 = tensorHalfConvolution(var_35, conv2d_6_w, 0, 0, 1, 1, 1, 1);
- void* var_37 = tensorHalfBatchNorm(var_36, batch_normalization_11_gamma, batch_normalization_11_beta, batch_normalization_11_mean, batch_normalization_11_variance, 0.001);
- void* var_38 = tensorHalfRelu(var_37);
- void* var_41 = tensorHalfConvolution(var_38, depthwise_conv2d_6_w, 1, 1, 2, 2, 1, 256);
- void* var_42 = tensorHalfBatchNorm(var_41, batch_normalization_12_gamma, batch_normalization_12_beta, batch_normalization_12_mean, batch_normalization_12_variance, 0.001);
- void* var_43 = tensorHalfRelu(var_42);
- void* var_44 = tensorHalfConvolution(var_43, conv2d_7_w, 0, 0, 1, 1, 1, 1);
- void* var_45 = tensorHalfBatchNorm(var_44, batch_normalization_13_gamma, batch_normalization_13_beta, batch_normalization_13_mean, batch_normalization_13_variance, 0.001);
- void* var_46 = tensorHalfRelu(var_45);
- void* var_48 = tensorHalfConvolution(var_46, depthwise_conv2d_7_w, 1, 1, 1, 1, 1, 512);
- void* var_49 = tensorHalfBatchNorm(var_48, batch_normalization_14_gamma, batch_normalization_14_beta, batch_normalization_14_mean, batch_normalization_14_variance, 0.001);
- void* var_50 = tensorHalfRelu(var_49);
- void* var_51 = tensorHalfConvolution(var_50, conv2d_8_w, 0, 0, 1, 1, 1, 1);
- void* var_52 = tensorHalfBatchNorm(var_51, batch_normalization_15_gamma, batch_normalization_15_beta, batch_normalization_15_mean, batch_normalization_15_variance, 0.001);
- void* var_53 = tensorHalfRelu(var_52);
- void* var_55 = tensorHalfConvolution(var_53, depthwise_conv2d_8_w, 1, 1, 1, 1, 1, 512);
- void* var_56 = tensorHalfBatchNorm(var_55, batch_normalization_16_gamma, batch_normalization_16_beta, batch_normalization_16_mean, batch_normalization_16_variance, 0.001);
- void* var_57 = tensorHalfRelu(var_56);
- void* var_58 = tensorHalfConvolution(var_57, conv2d_9_w, 0, 0, 1, 1, 1, 1);
- void* var_59 = tensorHalfBatchNorm(var_58, batch_normalization_17_gamma, batch_normalization_17_beta, batch_normalization_17_mean, batch_normalization_17_variance, 0.001);
- void* var_60 = tensorHalfRelu(var_59);
- void* var_63 = tensorHalfConvolution(var_60, depthwise_conv2d_9_w, 1, 1, 1, 1, 1, 512);
- void* var_64 = tensorHalfBatchNorm(var_63, batch_normalization_18_gamma, batch_normalization_18_beta, batch_normalization_18_mean, batch_normalization_18_variance, 0.001);
- void* var_65 = tensorHalfRelu(var_64);
- void* var_66 = tensorHalfConvolution(var_65, conv2d_10_w, 0, 0, 1, 1, 1, 1);
- void* var_67 = tensorHalfBatchNorm(var_66, batch_normalization_19_gamma, batch_normalization_19_beta, batch_normalization_19_mean, batch_normalization_19_variance, 0.001);
- void* var_68 = tensorHalfRelu(var_67);
- void* var_70 = tensorHalfConvolution(var_68, depthwise_conv2d_10_w, 1, 1, 1, 1, 1, 512);
- void* var_71 = tensorHalfBatchNorm(var_70, batch_normalization_20_gamma, batch_normalization_20_beta, batch_normalization_20_mean, batch_normalization_20_variance, 0.001);
- void* var_72 = tensorHalfRelu(var_71);
- void* var_73 = tensorHalfConvolution(var_72, conv2d_11_w, 0, 0, 1, 1, 1, 1);
- void* var_74 = tensorHalfBatchNorm(var_73, batch_normalization_21_gamma, batch_normalization_21_beta, batch_normalization_21_mean, batch_normalization_21_variance, 0.001);
- void* var_75 = tensorHalfRelu(var_74);
- void* var_77 = tensorHalfConvolution(var_75, depthwise_conv2d_11_w, 1, 1, 1, 1, 1, 512);
- void* var_78 = tensorHalfBatchNorm(var_77, batch_normalization_22_gamma, batch_normalization_22_beta, batch_normalization_22_mean, batch_normalization_22_variance, 0.001);
- void* var_79 = tensorHalfRelu(var_78);
- void* var_80 = tensorHalfConvolution(var_79, conv2d_12_w, 0, 0, 1, 1, 1, 1);
- void* var_81 = tensorHalfBatchNorm(var_80, batch_normalization_23_gamma, batch_normalization_23_beta, batch_normalization_23_mean, batch_normalization_23_variance, 0.001);
- void* var_82 = tensorHalfRelu(var_81);
- void* var_85 = tensorHalfConvolution(var_82, depthwise_conv2d_12_w, 1, 1, 2, 2, 1, 512);
- void* var_86 = tensorHalfBatchNorm(var_85, batch_normalization_24_gamma, batch_normalization_24_beta, batch_normalization_24_mean, batch_normalization_24_variance, 0.001);
- void* var_87 = tensorHalfRelu(var_86);
- void* var_88 = tensorHalfConvolution(var_87, conv2d_13_w, 0, 0, 1, 1, 1, 1);
- void* var_89 = tensorHalfBatchNorm(var_88, batch_normalization_25_gamma, batch_normalization_25_beta, batch_normalization_25_mean, batch_normalization_25_variance, 0.001);
- void* var_90 = tensorHalfRelu(var_89);
- void* var_92 = tensorHalfConvolution(var_90, depthwise_conv2d_13_w, 1, 1, 1, 1, 1, 1024);
- void* var_93 = tensorHalfBatchNorm(var_92, batch_normalization_26_gamma, batch_normalization_26_beta, batch_normalization_26_mean, batch_normalization_26_variance, 0.001);
- void* var_94 = tensorHalfRelu(var_93);
- void* var_95 = tensorHalfConvolution(var_94, conv2d_14_w, 0, 0, 1, 1, 1, 1);
- void* var_96 = tensorHalfBatchNorm(var_95, batch_normalization_27_gamma, batch_normalization_27_beta, batch_normalization_27_mean, batch_normalization_27_variance, 0.001);
- void* var_97 = tensorHalfRelu(var_96);
- void* var_99 = tensorHalfPooling(var_97,1,2,2,0,0,2,2);
- void* var_101 = tensorHalfGemmGPU(var_99, dense_1_w);
- void* var_102 = tensorHalfAdd(var_101, dense_1_b);
- void* var_103 = tensorSoftmax(var_102);
-
- profiler.pause_profiler();
- auto time_energy = profiler.get_time_energy();
- total_time += time_energy.first;
- profiler.reset();
-
- uint8_t* labels = readLabelsBatch(labels_path.c_str(),start,end);
-
- float accuracy = computeAccuracy2(labels, batch_size, var_103);
- final_accuracy += accuracy;
- freeBatchMemory();
-
- }
- }
- profiler.stop_profiler();
-
- std::cout<<"---------------------------------------\n";
- std::cout<<"Average time: " << total_time / total_runs << '\n';
- std::cout<<"---------------------------------------\n";
-
- stopProfiling();
-
- final_accuracy = final_accuracy / batch_count / total_runs;
- dumpFinalAccuracy(final_accuracy);
-
-
- llvm_hpvm_cleanupTensorRt();
-
- return 0;
-
-}
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/mobilenet_shallow_depthwise_half_profiling.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/mobilenet_shallow_depthwise_half_profiling.cc
deleted file mode 100644
index f68eb1793b66b0579f2ed6dbff26a56677f2aa95..0000000000000000000000000000000000000000
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/mobilenet_shallow_depthwise_half_profiling.cc
+++ /dev/null
@@ -1,249 +0,0 @@
-#include "/home/nvidia/Gitlab/hpvm/llvm/projects/gpu_profiler/include/profiler.h"
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <string.h>
-
-#include "../../../../tensor_runtime/include/tensor_runtime.h"
-#include "../../../include/utils.h"
-
-
-int main(int argc, char* argv[]){
-
- int total_runs = 10;
- if (argc > 1){
- total_runs = atoi(argv[1]);
- }
-
-
- llvm_hpvm_initTensorRt(0);
-
- //std::string dir_prefix = std::string("../../keras/data/mobilenet_shallow_nathan/");
-
- std::string dir_prefix = std::string("../model_params/mobilenet_shallow/");
-
- std::string input_path = dir_prefix + std::string("input.bin");
- std::string labels_path = dir_prefix + std::string("labels.bin");
- 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(), 0,32,3,3,3);
- std::string batch_normalization_1_gamma_path = dir_prefix + std::string("batch_normalization_1_gamma.bin");
- void* batch_normalization_1_gamma = readTrainedWeights(batch_normalization_1_gamma_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_beta_path = dir_prefix + std::string("batch_normalization_1_beta.bin");
- void* batch_normalization_1_beta = readTrainedWeights(batch_normalization_1_beta_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_mean_path = dir_prefix + std::string("batch_normalization_1_mean.bin");
- void* batch_normalization_1_mean = readTrainedWeights(batch_normalization_1_mean_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_variance_path = dir_prefix + std::string("batch_normalization_1_variance.bin");
- void* batch_normalization_1_variance = readTrainedWeights(batch_normalization_1_variance_path.c_str(), 0,1,32,1,1);
- std::string depthwise_conv2d_1_w_path = dir_prefix + std::string("depthwise_conv2d_1_w.bin");
- void* depthwise_conv2d_1_w = readTrainedWeights(depthwise_conv2d_1_w_path.c_str(), 0,32,1,3,3);
- std::string batch_normalization_2_gamma_path = dir_prefix + std::string("batch_normalization_2_gamma.bin");
- void* batch_normalization_2_gamma = readTrainedWeights(batch_normalization_2_gamma_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_beta_path = dir_prefix + std::string("batch_normalization_2_beta.bin");
- void* batch_normalization_2_beta = readTrainedWeights(batch_normalization_2_beta_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_mean_path = dir_prefix + std::string("batch_normalization_2_mean.bin");
- void* batch_normalization_2_mean = readTrainedWeights(batch_normalization_2_mean_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_variance_path = dir_prefix + std::string("batch_normalization_2_variance.bin");
- void* batch_normalization_2_variance = readTrainedWeights(batch_normalization_2_variance_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(), 0,64,32,1,1);
- std::string batch_normalization_3_gamma_path = dir_prefix + std::string("batch_normalization_3_gamma.bin");
- void* batch_normalization_3_gamma = readTrainedWeights(batch_normalization_3_gamma_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_beta_path = dir_prefix + std::string("batch_normalization_3_beta.bin");
- void* batch_normalization_3_beta = readTrainedWeights(batch_normalization_3_beta_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_mean_path = dir_prefix + std::string("batch_normalization_3_mean.bin");
- void* batch_normalization_3_mean = readTrainedWeights(batch_normalization_3_mean_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_variance_path = dir_prefix + std::string("batch_normalization_3_variance.bin");
- void* batch_normalization_3_variance = readTrainedWeights(batch_normalization_3_variance_path.c_str(), 0,1,64,1,1);
- std::string depthwise_conv2d_2_w_path = dir_prefix + std::string("depthwise_conv2d_2_w.bin");
- void* depthwise_conv2d_2_w = readTrainedWeights(depthwise_conv2d_2_w_path.c_str(), 0,64,1,3,3);
- std::string batch_normalization_4_gamma_path = dir_prefix + std::string("batch_normalization_4_gamma.bin");
- void* batch_normalization_4_gamma = readTrainedWeights(batch_normalization_4_gamma_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_beta_path = dir_prefix + std::string("batch_normalization_4_beta.bin");
- void* batch_normalization_4_beta = readTrainedWeights(batch_normalization_4_beta_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_mean_path = dir_prefix + std::string("batch_normalization_4_mean.bin");
- void* batch_normalization_4_mean = readTrainedWeights(batch_normalization_4_mean_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_variance_path = dir_prefix + std::string("batch_normalization_4_variance.bin");
- void* batch_normalization_4_variance = readTrainedWeights(batch_normalization_4_variance_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(), 0,128,64,1,1);
- std::string batch_normalization_5_gamma_path = dir_prefix + std::string("batch_normalization_5_gamma.bin");
- void* batch_normalization_5_gamma = readTrainedWeights(batch_normalization_5_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_beta_path = dir_prefix + std::string("batch_normalization_5_beta.bin");
- void* batch_normalization_5_beta = readTrainedWeights(batch_normalization_5_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_mean_path = dir_prefix + std::string("batch_normalization_5_mean.bin");
- void* batch_normalization_5_mean = readTrainedWeights(batch_normalization_5_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_variance_path = dir_prefix + std::string("batch_normalization_5_variance.bin");
- void* batch_normalization_5_variance = readTrainedWeights(batch_normalization_5_variance_path.c_str(), 0,1,128,1,1);
- std::string depthwise_conv2d_3_w_path = dir_prefix + std::string("depthwise_conv2d_3_w.bin");
- void* depthwise_conv2d_3_w = readTrainedWeights(depthwise_conv2d_3_w_path.c_str(), 0,128,1,3,3);
- std::string batch_normalization_6_gamma_path = dir_prefix + std::string("batch_normalization_6_gamma.bin");
- void* batch_normalization_6_gamma = readTrainedWeights(batch_normalization_6_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_beta_path = dir_prefix + std::string("batch_normalization_6_beta.bin");
- void* batch_normalization_6_beta = readTrainedWeights(batch_normalization_6_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_mean_path = dir_prefix + std::string("batch_normalization_6_mean.bin");
- void* batch_normalization_6_mean = readTrainedWeights(batch_normalization_6_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_variance_path = dir_prefix + std::string("batch_normalization_6_variance.bin");
- void* batch_normalization_6_variance = readTrainedWeights(batch_normalization_6_variance_path.c_str(), 0,1,128,1,1);
- std::string conv2d_4_w_path = dir_prefix + std::string("conv2d_4_w.bin");
- void* conv2d_4_w = readTrainedWeights(conv2d_4_w_path.c_str(), 0,128,128,1,1);
- std::string batch_normalization_7_gamma_path = dir_prefix + std::string("batch_normalization_7_gamma.bin");
- void* batch_normalization_7_gamma = readTrainedWeights(batch_normalization_7_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_beta_path = dir_prefix + std::string("batch_normalization_7_beta.bin");
- void* batch_normalization_7_beta = readTrainedWeights(batch_normalization_7_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_mean_path = dir_prefix + std::string("batch_normalization_7_mean.bin");
- void* batch_normalization_7_mean = readTrainedWeights(batch_normalization_7_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_variance_path = dir_prefix + std::string("batch_normalization_7_variance.bin");
- void* batch_normalization_7_variance = readTrainedWeights(batch_normalization_7_variance_path.c_str(), 0,1,128,1,1);
- std::string depthwise_conv2d_4_w_path = dir_prefix + std::string("depthwise_conv2d_4_w.bin");
- void* depthwise_conv2d_4_w = readTrainedWeights(depthwise_conv2d_4_w_path.c_str(), 0,128,1,3,3);
- std::string batch_normalization_8_gamma_path = dir_prefix + std::string("batch_normalization_8_gamma.bin");
- void* batch_normalization_8_gamma = readTrainedWeights(batch_normalization_8_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_beta_path = dir_prefix + std::string("batch_normalization_8_beta.bin");
- void* batch_normalization_8_beta = readTrainedWeights(batch_normalization_8_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_mean_path = dir_prefix + std::string("batch_normalization_8_mean.bin");
- void* batch_normalization_8_mean = readTrainedWeights(batch_normalization_8_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_variance_path = dir_prefix + std::string("batch_normalization_8_variance.bin");
- void* batch_normalization_8_variance = readTrainedWeights(batch_normalization_8_variance_path.c_str(), 0,1,128,1,1);
- std::string conv2d_5_w_path = dir_prefix + std::string("conv2d_5_w.bin");
- void* conv2d_5_w = readTrainedWeights(conv2d_5_w_path.c_str(), 0,256,128,1,1);
- std::string batch_normalization_9_gamma_path = dir_prefix + std::string("batch_normalization_9_gamma.bin");
- void* batch_normalization_9_gamma = readTrainedWeights(batch_normalization_9_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_beta_path = dir_prefix + std::string("batch_normalization_9_beta.bin");
- void* batch_normalization_9_beta = readTrainedWeights(batch_normalization_9_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_mean_path = dir_prefix + std::string("batch_normalization_9_mean.bin");
- void* batch_normalization_9_mean = readTrainedWeights(batch_normalization_9_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_variance_path = dir_prefix + std::string("batch_normalization_9_variance.bin");
- void* batch_normalization_9_variance = readTrainedWeights(batch_normalization_9_variance_path.c_str(), 0,1,256,1,1);
- std::string depthwise_conv2d_5_w_path = dir_prefix + std::string("depthwise_conv2d_5_w.bin");
- void* depthwise_conv2d_5_w = readTrainedWeights(depthwise_conv2d_5_w_path.c_str(), 0,256,1,3,3);
- std::string batch_normalization_10_gamma_path = dir_prefix + std::string("batch_normalization_10_gamma.bin");
- void* batch_normalization_10_gamma = readTrainedWeights(batch_normalization_10_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_beta_path = dir_prefix + std::string("batch_normalization_10_beta.bin");
- void* batch_normalization_10_beta = readTrainedWeights(batch_normalization_10_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_mean_path = dir_prefix + std::string("batch_normalization_10_mean.bin");
- void* batch_normalization_10_mean = readTrainedWeights(batch_normalization_10_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_variance_path = dir_prefix + std::string("batch_normalization_10_variance.bin");
- void* batch_normalization_10_variance = readTrainedWeights(batch_normalization_10_variance_path.c_str(), 0,1,256,1,1);
- std::string conv2d_6_w_path = dir_prefix + std::string("conv2d_6_w.bin");
- void* conv2d_6_w = readTrainedWeights(conv2d_6_w_path.c_str(), 0,256,256,1,1);
- std::string batch_normalization_11_gamma_path = dir_prefix + std::string("batch_normalization_11_gamma.bin");
- void* batch_normalization_11_gamma = readTrainedWeights(batch_normalization_11_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_beta_path = dir_prefix + std::string("batch_normalization_11_beta.bin");
- void* batch_normalization_11_beta = readTrainedWeights(batch_normalization_11_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_mean_path = dir_prefix + std::string("batch_normalization_11_mean.bin");
- void* batch_normalization_11_mean = readTrainedWeights(batch_normalization_11_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_variance_path = dir_prefix + std::string("batch_normalization_11_variance.bin");
- void* batch_normalization_11_variance = readTrainedWeights(batch_normalization_11_variance_path.c_str(), 0,1,256,1,1);
- std::string depthwise_conv2d_6_w_path = dir_prefix + std::string("depthwise_conv2d_6_w.bin");
- void* depthwise_conv2d_6_w = readTrainedWeights(depthwise_conv2d_6_w_path.c_str(), 0,256,1,3,3);
- std::string batch_normalization_12_gamma_path = dir_prefix + std::string("batch_normalization_12_gamma.bin");
- void* batch_normalization_12_gamma = readTrainedWeights(batch_normalization_12_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_beta_path = dir_prefix + std::string("batch_normalization_12_beta.bin");
- void* batch_normalization_12_beta = readTrainedWeights(batch_normalization_12_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_mean_path = dir_prefix + std::string("batch_normalization_12_mean.bin");
- void* batch_normalization_12_mean = readTrainedWeights(batch_normalization_12_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_variance_path = dir_prefix + std::string("batch_normalization_12_variance.bin");
- void* batch_normalization_12_variance = readTrainedWeights(batch_normalization_12_variance_path.c_str(), 0,1,256,1,1);
- std::string conv2d_7_w_path = dir_prefix + std::string("conv2d_7_w.bin");
- void* conv2d_7_w = readTrainedWeights(conv2d_7_w_path.c_str(), 0,512,256,1,1);
- std::string batch_normalization_13_gamma_path = dir_prefix + std::string("batch_normalization_13_gamma.bin");
- void* batch_normalization_13_gamma = readTrainedWeights(batch_normalization_13_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_beta_path = dir_prefix + std::string("batch_normalization_13_beta.bin");
- void* batch_normalization_13_beta = readTrainedWeights(batch_normalization_13_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_mean_path = dir_prefix + std::string("batch_normalization_13_mean.bin");
- void* batch_normalization_13_mean = readTrainedWeights(batch_normalization_13_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_variance_path = dir_prefix + std::string("batch_normalization_13_variance.bin");
- void* batch_normalization_13_variance = readTrainedWeights(batch_normalization_13_variance_path.c_str(), 0,1,512,1,1);
- 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(), 0,1,1,2048,10);
- 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(), 0,1,10,1,1);
-
-
- startMemTracking();
-
- int test_input_size = 5000;
- int batch_size = 1000;
- int batch_count = test_input_size / batch_size;
-
-
- float final_accuracy = 0.0;
-
- for(int j = 0; j < total_runs; j++){
- for(int i = 0; i < batch_count; i++){
-
- int start = i * batch_size;
- int end = (i + 1) * batch_size;
-
- void* input = readInputBatch(input_path.c_str(),0,start,end,3,32,32);
-
- void* var_0 = tensorHalfConvolution(input, conv2d_1_w, 1, 1, 1, 1, 1, 1);
- void* var_1 = tensorHalfBatchNorm(var_0, batch_normalization_1_gamma, batch_normalization_1_beta, batch_normalization_1_mean, batch_normalization_1_variance, 0.001);
- void* var_2 = tensorHalfRelu(var_1);
- void* var_4 = tensorHalfConvCutlass(var_2, depthwise_conv2d_1_w, 1, 1, 1, 1, 1, 32);
- void* var_5 = tensorHalfBatchNorm(var_4, batch_normalization_2_gamma, batch_normalization_2_beta, batch_normalization_2_mean, batch_normalization_2_variance, 0.001);
- void* var_6 = tensorHalfRelu(var_5);
- void* var_7 = tensorHalfConvolution(var_6, conv2d_2_w, 0, 0, 1, 1, 1, 1);
- void* var_8 = tensorHalfBatchNorm(var_7, batch_normalization_3_gamma, batch_normalization_3_beta, batch_normalization_3_mean, batch_normalization_3_variance, 0.001);
- void* var_9 = tensorHalfRelu(var_8);
- void* var_11 = tensorHalfConvCutlass(var_9, depthwise_conv2d_2_w, 1, 1, 2, 2, 1, 64);
- void* var_12 = tensorHalfBatchNorm(var_11, batch_normalization_4_gamma, batch_normalization_4_beta, batch_normalization_4_mean, batch_normalization_4_variance, 0.001);
- void* var_13 = tensorHalfRelu(var_12);
- void* var_14 = tensorHalfConvolution(var_13, conv2d_3_w, 0, 0, 1, 1, 1, 1);
- void* var_15 = tensorHalfBatchNorm(var_14, batch_normalization_5_gamma, batch_normalization_5_beta, batch_normalization_5_mean, batch_normalization_5_variance, 0.001);
- void* var_16 = tensorHalfRelu(var_15);
- void* var_18 = tensorHalfConvCutlass(var_16, depthwise_conv2d_3_w, 1, 1, 1, 1, 1, 128);
- void* var_19 = tensorHalfBatchNorm(var_18, batch_normalization_6_gamma, batch_normalization_6_beta, batch_normalization_6_mean, batch_normalization_6_variance, 0.001);
- void* var_20 = tensorHalfRelu(var_19);
- void* var_21 = tensorHalfConvolution(var_20, conv2d_4_w, 0, 0, 1, 1, 1, 1);
- void* var_22 = tensorHalfBatchNorm(var_21, batch_normalization_7_gamma, batch_normalization_7_beta, batch_normalization_7_mean, batch_normalization_7_variance, 0.001);
- void* var_23 = tensorHalfRelu(var_22);
- void* var_26 = tensorHalfConvCutlass(var_23, depthwise_conv2d_4_w, 1, 1, 2, 2, 1, 128);
- void* var_27 = tensorHalfBatchNorm(var_26, batch_normalization_8_gamma, batch_normalization_8_beta, batch_normalization_8_mean, batch_normalization_8_variance, 0.001);
- void* var_28 = tensorHalfRelu(var_27);
- void* var_29 = tensorHalfConvolution(var_28, conv2d_5_w, 0, 0, 1, 1, 1, 1);
- void* var_30 = tensorHalfBatchNorm(var_29, batch_normalization_9_gamma, batch_normalization_9_beta, batch_normalization_9_mean, batch_normalization_9_variance, 0.001);
- void* var_31 = tensorHalfRelu(var_30);
- void* var_33 = tensorHalfConvCutlass(var_31, depthwise_conv2d_5_w, 1, 1, 1, 1, 1, 256);
- void* var_34 = tensorHalfBatchNorm(var_33, batch_normalization_10_gamma, batch_normalization_10_beta, batch_normalization_10_mean, batch_normalization_10_variance, 0.001);
- void* var_35 = tensorHalfRelu(var_34);
- void* var_36 = tensorHalfConvolution(var_35, conv2d_6_w, 0, 0, 1, 1, 1, 1);
- void* var_37 = tensorHalfBatchNorm(var_36, batch_normalization_11_gamma, batch_normalization_11_beta, batch_normalization_11_mean, batch_normalization_11_variance, 0.001);
- void* var_38 = tensorHalfRelu(var_37);
- void* var_41 = tensorHalfConvCutlass(var_38, depthwise_conv2d_6_w, 1, 1, 2, 2, 1, 256);
- void* var_42 = tensorHalfBatchNorm(var_41, batch_normalization_12_gamma, batch_normalization_12_beta, batch_normalization_12_mean, batch_normalization_12_variance, 0.001);
- void* var_43 = tensorHalfRelu(var_42);
- void* var_44 = tensorHalfConvolution(var_43, conv2d_7_w, 0, 0, 1, 1, 1, 1);
- void* var_45 = tensorHalfBatchNorm(var_44, batch_normalization_13_gamma, batch_normalization_13_beta, batch_normalization_13_mean, batch_normalization_13_variance, 0.001);
- void* var_46 = tensorHalfRelu(var_45);
- void* var_47 = tensorHalfPooling(var_46,1,2,2,0,0,2,2);
- void* var_49 = tensorHalfGemmGPU(var_47, dense_1_w);
- void* var_50 = tensorHalfAdd(var_49, dense_1_b);
- void* var_51 = tensorSoftmax(var_50);
-
- uint8_t* labels = readLabelsBatch(labels_path.c_str(),start,end);
-
- float accuracy = computeAccuracy2(labels, batch_size, var_51);
- final_accuracy += accuracy;
- freeBatchMemory();
-
- }
-
- //final_accuracy = final_accuracy / batch_count;
- dumpFinalAccuracy(final_accuracy / batch_count);
- }
-
- final_accuracy = final_accuracy / batch_count / total_runs;
- dumpFinalAccuracy(final_accuracy);
-
- dumpExecutionAccuracies();
-
- llvm_hpvm_cleanupTensorRt();
-
- return 0;
-
-}
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/mobilenet_shallow_half_profiling.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/mobilenet_shallow_half_profiling.cc
deleted file mode 100644
index c641db1a05efe44d4801da1ebdcaf2ae8945e7f2..0000000000000000000000000000000000000000
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/mobilenet_shallow_half_profiling.cc
+++ /dev/null
@@ -1,225 +0,0 @@
-#include "/home/nvidia/Gitlab/hpvm/llvm/projects/gpu_profiler/include/profiler.h"
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <string.h>
-
-#include "../../../../tensor_runtime/include/tensor_runtime.h"
-#include "../../../include/utils.h"
-
-int main(){
-
- llvm_hpvm_initTensorRt(0);
-
- std::string dir_prefix = std::string("../model_params/mobilenet_shallow/");
- std::string input_path = dir_prefix + std::string("input.bin");
- std::string labels_path = dir_prefix + std::string("labels.bin");
- 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(), 0,32,3,3,3);
- std::string batch_normalization_1_gamma_path = dir_prefix + std::string("batch_normalization_1_gamma.bin");
- void* batch_normalization_1_gamma = readTrainedWeights(batch_normalization_1_gamma_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_beta_path = dir_prefix + std::string("batch_normalization_1_beta.bin");
- void* batch_normalization_1_beta = readTrainedWeights(batch_normalization_1_beta_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_mean_path = dir_prefix + std::string("batch_normalization_1_mean.bin");
- void* batch_normalization_1_mean = readTrainedWeights(batch_normalization_1_mean_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_variance_path = dir_prefix + std::string("batch_normalization_1_variance.bin");
- void* batch_normalization_1_variance = readTrainedWeights(batch_normalization_1_variance_path.c_str(), 0,1,32,1,1);
- std::string depthwise_conv2d_1_w_path = dir_prefix + std::string("depthwise_conv2d_1_w.bin");
- void* depthwise_conv2d_1_w = readTrainedWeights(depthwise_conv2d_1_w_path.c_str(), 0,32,1,3,3);
- std::string batch_normalization_2_gamma_path = dir_prefix + std::string("batch_normalization_2_gamma.bin");
- void* batch_normalization_2_gamma = readTrainedWeights(batch_normalization_2_gamma_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_beta_path = dir_prefix + std::string("batch_normalization_2_beta.bin");
- void* batch_normalization_2_beta = readTrainedWeights(batch_normalization_2_beta_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_mean_path = dir_prefix + std::string("batch_normalization_2_mean.bin");
- void* batch_normalization_2_mean = readTrainedWeights(batch_normalization_2_mean_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_variance_path = dir_prefix + std::string("batch_normalization_2_variance.bin");
- void* batch_normalization_2_variance = readTrainedWeights(batch_normalization_2_variance_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(), 0,64,32,1,1);
- std::string batch_normalization_3_gamma_path = dir_prefix + std::string("batch_normalization_3_gamma.bin");
- void* batch_normalization_3_gamma = readTrainedWeights(batch_normalization_3_gamma_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_beta_path = dir_prefix + std::string("batch_normalization_3_beta.bin");
- void* batch_normalization_3_beta = readTrainedWeights(batch_normalization_3_beta_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_mean_path = dir_prefix + std::string("batch_normalization_3_mean.bin");
- void* batch_normalization_3_mean = readTrainedWeights(batch_normalization_3_mean_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_variance_path = dir_prefix + std::string("batch_normalization_3_variance.bin");
- void* batch_normalization_3_variance = readTrainedWeights(batch_normalization_3_variance_path.c_str(), 0,1,64,1,1);
- std::string depthwise_conv2d_2_w_path = dir_prefix + std::string("depthwise_conv2d_2_w.bin");
- void* depthwise_conv2d_2_w = readTrainedWeights(depthwise_conv2d_2_w_path.c_str(), 0,64,1,3,3);
- std::string batch_normalization_4_gamma_path = dir_prefix + std::string("batch_normalization_4_gamma.bin");
- void* batch_normalization_4_gamma = readTrainedWeights(batch_normalization_4_gamma_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_beta_path = dir_prefix + std::string("batch_normalization_4_beta.bin");
- void* batch_normalization_4_beta = readTrainedWeights(batch_normalization_4_beta_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_mean_path = dir_prefix + std::string("batch_normalization_4_mean.bin");
- void* batch_normalization_4_mean = readTrainedWeights(batch_normalization_4_mean_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_variance_path = dir_prefix + std::string("batch_normalization_4_variance.bin");
- void* batch_normalization_4_variance = readTrainedWeights(batch_normalization_4_variance_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(), 0,64,64,1,1);
- std::string batch_normalization_5_gamma_path = dir_prefix + std::string("batch_normalization_5_gamma.bin");
- void* batch_normalization_5_gamma = readTrainedWeights(batch_normalization_5_gamma_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_5_beta_path = dir_prefix + std::string("batch_normalization_5_beta.bin");
- void* batch_normalization_5_beta = readTrainedWeights(batch_normalization_5_beta_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_5_mean_path = dir_prefix + std::string("batch_normalization_5_mean.bin");
- void* batch_normalization_5_mean = readTrainedWeights(batch_normalization_5_mean_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_5_variance_path = dir_prefix + std::string("batch_normalization_5_variance.bin");
- void* batch_normalization_5_variance = readTrainedWeights(batch_normalization_5_variance_path.c_str(), 0,1,64,1,1);
- std::string depthwise_conv2d_3_w_path = dir_prefix + std::string("depthwise_conv2d_3_w.bin");
- void* depthwise_conv2d_3_w = readTrainedWeights(depthwise_conv2d_3_w_path.c_str(), 0,64,1,3,3);
- std::string batch_normalization_6_gamma_path = dir_prefix + std::string("batch_normalization_6_gamma.bin");
- void* batch_normalization_6_gamma = readTrainedWeights(batch_normalization_6_gamma_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_6_beta_path = dir_prefix + std::string("batch_normalization_6_beta.bin");
- void* batch_normalization_6_beta = readTrainedWeights(batch_normalization_6_beta_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_6_mean_path = dir_prefix + std::string("batch_normalization_6_mean.bin");
- void* batch_normalization_6_mean = readTrainedWeights(batch_normalization_6_mean_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_6_variance_path = dir_prefix + std::string("batch_normalization_6_variance.bin");
- void* batch_normalization_6_variance = readTrainedWeights(batch_normalization_6_variance_path.c_str(), 0,1,64,1,1);
- std::string conv2d_4_w_path = dir_prefix + std::string("conv2d_4_w.bin");
- void* conv2d_4_w = readTrainedWeights(conv2d_4_w_path.c_str(), 0,128,64,1,1);
- std::string batch_normalization_7_gamma_path = dir_prefix + std::string("batch_normalization_7_gamma.bin");
- void* batch_normalization_7_gamma = readTrainedWeights(batch_normalization_7_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_beta_path = dir_prefix + std::string("batch_normalization_7_beta.bin");
- void* batch_normalization_7_beta = readTrainedWeights(batch_normalization_7_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_mean_path = dir_prefix + std::string("batch_normalization_7_mean.bin");
- void* batch_normalization_7_mean = readTrainedWeights(batch_normalization_7_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_variance_path = dir_prefix + std::string("batch_normalization_7_variance.bin");
- void* batch_normalization_7_variance = readTrainedWeights(batch_normalization_7_variance_path.c_str(), 0,1,128,1,1);
- std::string depthwise_conv2d_4_w_path = dir_prefix + std::string("depthwise_conv2d_4_w.bin");
- void* depthwise_conv2d_4_w = readTrainedWeights(depthwise_conv2d_4_w_path.c_str(), 0,128,1,3,3);
- std::string batch_normalization_8_gamma_path = dir_prefix + std::string("batch_normalization_8_gamma.bin");
- void* batch_normalization_8_gamma = readTrainedWeights(batch_normalization_8_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_beta_path = dir_prefix + std::string("batch_normalization_8_beta.bin");
- void* batch_normalization_8_beta = readTrainedWeights(batch_normalization_8_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_mean_path = dir_prefix + std::string("batch_normalization_8_mean.bin");
- void* batch_normalization_8_mean = readTrainedWeights(batch_normalization_8_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_variance_path = dir_prefix + std::string("batch_normalization_8_variance.bin");
- void* batch_normalization_8_variance = readTrainedWeights(batch_normalization_8_variance_path.c_str(), 0,1,128,1,1);
- std::string conv2d_5_w_path = dir_prefix + std::string("conv2d_5_w.bin");
- void* conv2d_5_w = readTrainedWeights(conv2d_5_w_path.c_str(), 0,256,128,1,1);
- std::string batch_normalization_9_gamma_path = dir_prefix + std::string("batch_normalization_9_gamma.bin");
- void* batch_normalization_9_gamma = readTrainedWeights(batch_normalization_9_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_beta_path = dir_prefix + std::string("batch_normalization_9_beta.bin");
- void* batch_normalization_9_beta = readTrainedWeights(batch_normalization_9_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_mean_path = dir_prefix + std::string("batch_normalization_9_mean.bin");
- void* batch_normalization_9_mean = readTrainedWeights(batch_normalization_9_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_variance_path = dir_prefix + std::string("batch_normalization_9_variance.bin");
- void* batch_normalization_9_variance = readTrainedWeights(batch_normalization_9_variance_path.c_str(), 0,1,256,1,1);
- std::string depthwise_conv2d_5_w_path = dir_prefix + std::string("depthwise_conv2d_5_w.bin");
- void* depthwise_conv2d_5_w = readTrainedWeights(depthwise_conv2d_5_w_path.c_str(), 0,256,1,3,3);
- std::string batch_normalization_10_gamma_path = dir_prefix + std::string("batch_normalization_10_gamma.bin");
- void* batch_normalization_10_gamma = readTrainedWeights(batch_normalization_10_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_beta_path = dir_prefix + std::string("batch_normalization_10_beta.bin");
- void* batch_normalization_10_beta = readTrainedWeights(batch_normalization_10_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_mean_path = dir_prefix + std::string("batch_normalization_10_mean.bin");
- void* batch_normalization_10_mean = readTrainedWeights(batch_normalization_10_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_variance_path = dir_prefix + std::string("batch_normalization_10_variance.bin");
- void* batch_normalization_10_variance = readTrainedWeights(batch_normalization_10_variance_path.c_str(), 0,1,256,1,1);
- std::string conv2d_6_w_path = dir_prefix + std::string("conv2d_6_w.bin");
- void* conv2d_6_w = readTrainedWeights(conv2d_6_w_path.c_str(), 0,256,256,1,1);
- std::string batch_normalization_11_gamma_path = dir_prefix + std::string("batch_normalization_11_gamma.bin");
- void* batch_normalization_11_gamma = readTrainedWeights(batch_normalization_11_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_beta_path = dir_prefix + std::string("batch_normalization_11_beta.bin");
- void* batch_normalization_11_beta = readTrainedWeights(batch_normalization_11_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_mean_path = dir_prefix + std::string("batch_normalization_11_mean.bin");
- void* batch_normalization_11_mean = readTrainedWeights(batch_normalization_11_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_variance_path = dir_prefix + std::string("batch_normalization_11_variance.bin");
- void* batch_normalization_11_variance = readTrainedWeights(batch_normalization_11_variance_path.c_str(), 0,1,256,1,1);
- 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(), 0,1,1,1024,10);
- 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(), 0,1,10,1,1);
-
-
- startMemTracking();
-
- int test_input_size = 5000;
- int batch_size = 1000;
- int batch_count = test_input_size / batch_size;
- float final_accuracy = 0.0;
-
- int total_runs = 10;
- Profiler profiler;
- profiler.start_profiler();
-
- double total_time = 0.0;
-
- for(int i = 0; i < total_runs; i++){
- for(int i = 0; i < batch_count; i++){
-
- int start = i * batch_size;
- int end = (i + 1) * batch_size;
-
- void* input = readInputBatch(input_path.c_str(),0,start,end,3,32,32);
-
- profiler.resume_profiler();
-
- void* var_0 = tensorHalfConvolution(input, conv2d_1_w, 1, 1, 1, 1, 1, 1);
- void* var_1 = tensorBatchNorm(var_0, batch_normalization_1_gamma, batch_normalization_1_beta, batch_normalization_1_mean, batch_normalization_1_variance, 0.001);
- void* var_2 = tensorHalfRelu(var_1);
- void* var_4 = tensorHalfConvolution(var_2, depthwise_conv2d_1_w, 1, 1, 1, 1, 1, 32);
- void* var_5 = tensorBatchNorm(var_4, batch_normalization_2_gamma, batch_normalization_2_beta, batch_normalization_2_mean, batch_normalization_2_variance, 0.001);
- void* var_6 = tensorHalfRelu(var_5);
- void* var_7 = tensorHalfConvolution(var_6, conv2d_2_w, 0, 0, 1, 1, 1, 1);
- void* var_8 = tensorBatchNorm(var_7, batch_normalization_3_gamma, batch_normalization_3_beta, batch_normalization_3_mean, batch_normalization_3_variance, 0.001);
- void* var_9 = tensorHalfRelu(var_8);
- void* var_11 = tensorHalfConvolution(var_9, depthwise_conv2d_2_w, 1, 1, 2, 2, 1, 64);
- void* var_12 = tensorBatchNorm(var_11, batch_normalization_4_gamma, batch_normalization_4_beta, batch_normalization_4_mean, batch_normalization_4_variance, 0.001);
- void* var_13 = tensorHalfRelu(var_12);
- void* var_14 = tensorHalfConvolution(var_13, conv2d_3_w, 0, 0, 1, 1, 1, 1);
- void* var_15 = tensorBatchNorm(var_14, batch_normalization_5_gamma, batch_normalization_5_beta, batch_normalization_5_mean, batch_normalization_5_variance, 0.001);
- void* var_16 = tensorHalfRelu(var_15);
- void* var_18 = tensorHalfConvolution(var_16, depthwise_conv2d_3_w, 1, 1, 2, 2, 1, 64);
- void* var_19 = tensorBatchNorm(var_18, batch_normalization_6_gamma, batch_normalization_6_beta, batch_normalization_6_mean, batch_normalization_6_variance, 0.001);
- void* var_20 = tensorHalfRelu(var_19);
- void* var_21 = tensorHalfConvolution(var_20, conv2d_4_w, 0, 0, 1, 1, 1, 1);
- void* var_22 = tensorBatchNorm(var_21, batch_normalization_7_gamma, batch_normalization_7_beta, batch_normalization_7_mean, batch_normalization_7_variance, 0.001);
- void* var_23 = tensorHalfRelu(var_22);
- void* var_26 = tensorHalfConvolution(var_23, depthwise_conv2d_4_w, 1, 1, 2, 2, 1, 128);
- void* var_27 = tensorBatchNorm(var_26, batch_normalization_8_gamma, batch_normalization_8_beta, batch_normalization_8_mean, batch_normalization_8_variance, 0.001);
- void* var_28 = tensorHalfRelu(var_27);
- void* var_29 = tensorHalfConvolution(var_28, conv2d_5_w, 0, 0, 1, 1, 1, 1);
- void* var_30 = tensorBatchNorm(var_29, batch_normalization_9_gamma, batch_normalization_9_beta, batch_normalization_9_mean, batch_normalization_9_variance, 0.001);
- void* var_31 = tensorHalfRelu(var_30);
- void* var_33 = tensorHalfConvolution(var_31, depthwise_conv2d_5_w, 1, 1, 1, 1, 1, 256);
- void* var_34 = tensorBatchNorm(var_33, batch_normalization_10_gamma, batch_normalization_10_beta, batch_normalization_10_mean, batch_normalization_10_variance, 0.001);
- void* var_35 = tensorHalfRelu(var_34);
- void* var_36 = tensorHalfConvolution(var_35, conv2d_6_w, 0, 0, 1, 1, 1, 1);
- void* var_37 = tensorBatchNorm(var_36, batch_normalization_11_gamma, batch_normalization_11_beta, batch_normalization_11_mean, batch_normalization_11_variance, 0.001);
- void* var_38 = tensorHalfRelu(var_37);
- void* var_40 = tensorHalfPooling(var_38,1,2,2,0,0,2,2);
- void* var_42 = tensorHalfGemmGPU(var_40, dense_1_w);
- void* var_43 = tensorHalfAdd(var_42, dense_1_b);
- void* var_44 = tensorSoftmax(var_43);
-
- profiler.pause_profiler();
- auto time_energy = profiler.get_time_energy();
- total_time += time_energy.first;
- profiler.reset();
-
- uint8_t* labels = readLabelsBatch(labels_path.c_str(),start,end);
-
- float accuracy = computeAccuracy2(labels, batch_size, var_44);
- final_accuracy += accuracy;
- freeBatchMemory();
- }
- }
-
- profiler.stop_profiler();
-
- std::cout<<"---------------------------------------\n";
- std::cout<<"Average time: " << total_time / total_runs << '\n';
- std::cout<<"---------------------------------------\n";
-
- final_accuracy = final_accuracy / batch_count / total_runs;
- dumpFinalAccuracy(final_accuracy);
-
-
- llvm_hpvm_cleanupTensorRt();
-
- return 0;
-
-}
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/resnet18_cifar10_half_profiling.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/resnet18_cifar10_half_profiling.cc
deleted file mode 100644
index f91814e8390a400159467298a3702147cbf2f4b3..0000000000000000000000000000000000000000
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/resnet18_cifar10_half_profiling.cc
+++ /dev/null
@@ -1,242 +0,0 @@
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <string.h>
-
-#include "../../../../tensor_runtime/include/tensor_runtime.h"
-#include "../../../include/utils.h"
-
-#include "/home/nvidia/Gitlab/hpvm/llvm/projects/gpu_profiler/include/profiler.h"
-
-int main(){
-
- llvm_hpvm_initTensorRt(0);
-
- std::string dir_prefix = std::string("../model_params/resnet18_cifar10_3/");
- std::string input_path = dir_prefix + std::string("input.bin");
- //void* input = readTrainedWeights(input_path.c_str(), 0, batch_size,3,32,32);
- std::string labels_path = dir_prefix + std::string("labels.bin");
- //uint8_t* labels = readLabels(labels_path.c_str(), batch_size);
- 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(), 0,16,3,3,3);
- 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(), 0,1,16,1,1);
- 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(), 0,16,16,3,3);
- 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(), 0,1,16,1,1);
- 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(), 0,16,16,3,3);
- 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(), 0,1,16,1,1);
- std::string conv2d_4_w_path = dir_prefix + std::string("conv2d_4_w.bin");
- void* conv2d_4_w = readTrainedWeights(conv2d_4_w_path.c_str(), 0,16,16,3,3);
- std::string conv2d_4_b_path = dir_prefix + std::string("conv2d_4_b.bin");
- void* conv2d_4_b = readTrainedWeights(conv2d_4_b_path.c_str(), 0,1,16,1,1);
- std::string conv2d_5_w_path = dir_prefix + std::string("conv2d_5_w.bin");
- void* conv2d_5_w = readTrainedWeights(conv2d_5_w_path.c_str(), 0,16,16,3,3);
- std::string conv2d_5_b_path = dir_prefix + std::string("conv2d_5_b.bin");
- void* conv2d_5_b = readTrainedWeights(conv2d_5_b_path.c_str(), 0,1,16,1,1);
- std::string conv2d_6_w_path = dir_prefix + std::string("conv2d_6_w.bin");
- void* conv2d_6_w = readTrainedWeights(conv2d_6_w_path.c_str(), 0,16,16,3,3);
- std::string conv2d_6_b_path = dir_prefix + std::string("conv2d_6_b.bin");
- void* conv2d_6_b = readTrainedWeights(conv2d_6_b_path.c_str(), 0,1,16,1,1);
- std::string conv2d_7_w_path = dir_prefix + std::string("conv2d_7_w.bin");
- void* conv2d_7_w = readTrainedWeights(conv2d_7_w_path.c_str(), 0,16,16,3,3);
- std::string conv2d_7_b_path = dir_prefix + std::string("conv2d_7_b.bin");
- void* conv2d_7_b = readTrainedWeights(conv2d_7_b_path.c_str(), 0,1,16,1,1);
- std::string conv2d_8_w_path = dir_prefix + std::string("conv2d_8_w.bin");
- void* conv2d_8_w = readTrainedWeights(conv2d_8_w_path.c_str(), 0,32,16,3,3);
- std::string conv2d_8_b_path = dir_prefix + std::string("conv2d_8_b.bin");
- void* conv2d_8_b = readTrainedWeights(conv2d_8_b_path.c_str(), 0,1,32,1,1);
- std::string conv2d_10_w_path = dir_prefix + std::string("conv2d_10_w.bin");
- void* conv2d_10_w = readTrainedWeights(conv2d_10_w_path.c_str(), 0,32,16,1,1);
- std::string conv2d_10_b_path = dir_prefix + std::string("conv2d_10_b.bin");
- void* conv2d_10_b = readTrainedWeights(conv2d_10_b_path.c_str(), 0,1,32,1,1);
- std::string conv2d_9_w_path = dir_prefix + std::string("conv2d_9_w.bin");
- void* conv2d_9_w = readTrainedWeights(conv2d_9_w_path.c_str(), 0,32,32,3,3);
- std::string conv2d_9_b_path = dir_prefix + std::string("conv2d_9_b.bin");
- void* conv2d_9_b = readTrainedWeights(conv2d_9_b_path.c_str(), 0,1,32,1,1);
- std::string conv2d_11_w_path = dir_prefix + std::string("conv2d_11_w.bin");
- void* conv2d_11_w = readTrainedWeights(conv2d_11_w_path.c_str(), 0,32,32,3,3);
- std::string conv2d_11_b_path = dir_prefix + std::string("conv2d_11_b.bin");
- void* conv2d_11_b = readTrainedWeights(conv2d_11_b_path.c_str(), 0,1,32,1,1);
- std::string conv2d_12_w_path = dir_prefix + std::string("conv2d_12_w.bin");
- void* conv2d_12_w = readTrainedWeights(conv2d_12_w_path.c_str(), 0,32,32,3,3);
- std::string conv2d_12_b_path = dir_prefix + std::string("conv2d_12_b.bin");
- void* conv2d_12_b = readTrainedWeights(conv2d_12_b_path.c_str(), 0,1,32,1,1);
- std::string conv2d_13_w_path = dir_prefix + std::string("conv2d_13_w.bin");
- void* conv2d_13_w = readTrainedWeights(conv2d_13_w_path.c_str(), 0,32,32,3,3);
- std::string conv2d_13_b_path = dir_prefix + std::string("conv2d_13_b.bin");
- void* conv2d_13_b = readTrainedWeights(conv2d_13_b_path.c_str(), 0,1,32,1,1);
- std::string conv2d_14_w_path = dir_prefix + std::string("conv2d_14_w.bin");
- void* conv2d_14_w = readTrainedWeights(conv2d_14_w_path.c_str(), 0,32,32,3,3);
- std::string conv2d_14_b_path = dir_prefix + std::string("conv2d_14_b.bin");
- void* conv2d_14_b = readTrainedWeights(conv2d_14_b_path.c_str(), 0,1,32,1,1);
- std::string conv2d_15_w_path = dir_prefix + std::string("conv2d_15_w.bin");
- void* conv2d_15_w = readTrainedWeights(conv2d_15_w_path.c_str(), 0,64,32,3,3);
- std::string conv2d_15_b_path = dir_prefix + std::string("conv2d_15_b.bin");
- void* conv2d_15_b = readTrainedWeights(conv2d_15_b_path.c_str(), 0,1,64,1,1);
- std::string conv2d_17_w_path = dir_prefix + std::string("conv2d_17_w.bin");
- void* conv2d_17_w = readTrainedWeights(conv2d_17_w_path.c_str(), 0,64,32,1,1);
- std::string conv2d_17_b_path = dir_prefix + std::string("conv2d_17_b.bin");
- void* conv2d_17_b = readTrainedWeights(conv2d_17_b_path.c_str(), 0,1,64,1,1);
- std::string conv2d_16_w_path = dir_prefix + std::string("conv2d_16_w.bin");
- void* conv2d_16_w = readTrainedWeights(conv2d_16_w_path.c_str(), 0,64,64,3,3);
- std::string conv2d_16_b_path = dir_prefix + std::string("conv2d_16_b.bin");
- void* conv2d_16_b = readTrainedWeights(conv2d_16_b_path.c_str(), 0,1,64,1,1);
- std::string conv2d_18_w_path = dir_prefix + std::string("conv2d_18_w.bin");
- void* conv2d_18_w = readTrainedWeights(conv2d_18_w_path.c_str(), 0,64,64,3,3);
- std::string conv2d_18_b_path = dir_prefix + std::string("conv2d_18_b.bin");
- void* conv2d_18_b = readTrainedWeights(conv2d_18_b_path.c_str(), 0,1,64,1,1);
- std::string conv2d_19_w_path = dir_prefix + std::string("conv2d_19_w.bin");
- void* conv2d_19_w = readTrainedWeights(conv2d_19_w_path.c_str(), 0,64,64,3,3);
- std::string conv2d_19_b_path = dir_prefix + std::string("conv2d_19_b.bin");
- void* conv2d_19_b = readTrainedWeights(conv2d_19_b_path.c_str(), 0,1,64,1,1);
- std::string conv2d_20_w_path = dir_prefix + std::string("conv2d_20_w.bin");
- void* conv2d_20_w = readTrainedWeights(conv2d_20_w_path.c_str(), 0,64,64,3,3);
- std::string conv2d_20_b_path = dir_prefix + std::string("conv2d_20_b.bin");
- void* conv2d_20_b = readTrainedWeights(conv2d_20_b_path.c_str(), 0,1,64,1,1);
- std::string conv2d_21_w_path = dir_prefix + std::string("conv2d_21_w.bin");
- void* conv2d_21_w = readTrainedWeights(conv2d_21_w_path.c_str(), 0,64,64,3,3);
- std::string conv2d_21_b_path = dir_prefix + std::string("conv2d_21_b.bin");
- void* conv2d_21_b = readTrainedWeights(conv2d_21_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(), 0,1,1,64,10);
- 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(), 0,1,10,1,1);
-
-
- startMemTracking();
-
- int test_input_size = 5000;
- int batch_size = 1000;
- int batch_count = test_input_size / batch_size;
- float final_accuracy = 0.0;
-
- int total_runs = 10; //100;
-
- // NOTE: Starting time profiling
- startProfiling();
-
- Profiler profiler;
- profiler.start_profiler();
- double total_time = 0.0;
-
- for (int itrs = 0; itrs < total_runs; itrs++){
- for(int i = 0; i < batch_count; i++){
-
- int start = i * batch_size;
- int end = (i + 1) * batch_size;
-
- void* input = readInputBatch(input_path.c_str(), 0,start,end,3,32,32);
-
- profiler.resume_profiler();
-
- void* var_2 = tensorHalfConvolution(input, conv2d_1_w, 1, 1, 1, 1, 1, 0);
- void* var_3 = tensorHalfAdd(var_2, conv2d_1_b);
- void* var_4 = tensorHalfRelu(var_3);
- void* var_6 = tensorHalfConvolution(var_4, conv2d_2_w, 1, 1, 1, 1, 1, 0);
- void* var_7 = tensorHalfAdd(var_6, conv2d_2_b);
- void* var_8 = tensorHalfRelu(var_7);
- void* var_10 = tensorHalfConvolution(var_8, conv2d_3_w, 1, 1, 1, 1, 1, 0);
- void* var_11 = tensorHalfAdd(var_10, conv2d_3_b);
- void* var_12 = tensorHalfAdd(var_4, var_11);
- void* var_13 = tensorHalfRelu(var_12);
- void* var_15 = tensorHalfConvolution(var_13, conv2d_4_w, 1, 1, 1, 1, 1, 0);
- void* var_16 = tensorHalfAdd(var_15, conv2d_4_b);
- void* var_17 = tensorHalfRelu(var_16);
- void* var_19 = tensorHalfConvolution(var_17, conv2d_5_w, 1, 1, 1, 1, 1, 0);
- void* var_20 = tensorHalfAdd(var_19, conv2d_5_b);
- void* var_21 = tensorHalfAdd(var_13, var_20);
- void* var_22 = tensorHalfRelu(var_21);
- void* var_24 = tensorHalfConvolution(var_22, conv2d_6_w, 1, 1, 1, 1, 1, 0);
- void* var_25 = tensorHalfAdd(var_24, conv2d_6_b);
- void* var_26 = tensorHalfRelu(var_25);
- void* var_28 = tensorHalfConvolution(var_26, conv2d_7_w, 1, 1, 1, 1, 1, 0);
- void* var_29 = tensorHalfAdd(var_28, conv2d_7_b);
- void* var_30 = tensorHalfAdd(var_22, var_29);
- void* var_31 = tensorHalfRelu(var_30);
- void* var_33 = tensorHalfConvolution(var_31, conv2d_8_w, 1, 1, 2, 2, 1, 0);
- void* var_34 = tensorHalfAdd(var_33, conv2d_8_b);
- void* var_35 = tensorHalfRelu(var_34);
- void* var_37 = tensorHalfConvolution(var_35, conv2d_9_w, 1, 1, 1, 1, 1, 0);
- void* var_38 = tensorHalfAdd(var_37, conv2d_9_b);
- void* var_40 = tensorHalfConvolution(var_31, conv2d_10_w, 0, 0, 2, 2, 1, 0);
- void* var_41 = tensorHalfAdd(var_40, conv2d_10_b);
- void* var_42 = tensorHalfAdd(var_41, var_38);
- void* var_43 = tensorHalfRelu(var_42);
- void* var_45 = tensorHalfConvolution(var_43, conv2d_11_w, 1, 1, 1, 1, 1, 0);
- void* var_46 = tensorHalfAdd(var_45, conv2d_11_b);
- void* var_47 = tensorHalfRelu(var_46);
- void* var_49 = tensorHalfConvolution(var_47, conv2d_12_w, 1, 1, 1, 1, 1, 0);
- void* var_50 = tensorHalfAdd(var_49, conv2d_12_b);
- void* var_51 = tensorHalfAdd(var_43, var_50);
- void* var_52 = tensorHalfRelu(var_51);
- void* var_54 = tensorHalfConvolution(var_52, conv2d_13_w, 1, 1, 1, 1, 1, 0);
- void* var_55 = tensorHalfAdd(var_54, conv2d_13_b);
- void* var_56 = tensorHalfRelu(var_55);
- void* var_58 = tensorHalfConvolution(var_56, conv2d_14_w, 1, 1, 1, 1, 1, 0);
- void* var_59 = tensorHalfAdd(var_58, conv2d_14_b);
- void* var_60 = tensorHalfAdd(var_52, var_59);
- void* var_61 = tensorHalfRelu(var_60);
- void* var_63 = tensorHalfConvolution(var_61, conv2d_15_w, 1, 1, 2, 2, 1, 0);
- void* var_64 = tensorHalfAdd(var_63, conv2d_15_b);
- void* var_65 = tensorHalfRelu(var_64);
- void* var_67 = tensorHalfConvolution(var_65, conv2d_16_w, 1, 1, 1, 1, 1, 0);
- void* var_68 = tensorHalfAdd(var_67, conv2d_16_b);
- void* var_70 = tensorHalfConvolution(var_61, conv2d_17_w, 0, 0, 2, 2, 1, 0);
- void* var_71 = tensorHalfAdd(var_70, conv2d_17_b);
- void* var_72 = tensorHalfAdd(var_71, var_68);
- void* var_73 = tensorHalfRelu(var_72);
- void* var_75 = tensorHalfConvolution(var_73, conv2d_18_w, 1, 1, 1, 1, 1, 0);
- void* var_76 = tensorHalfAdd(var_75, conv2d_18_b);
- void* var_77 = tensorHalfRelu(var_76);
- void* var_79 = tensorHalfConvolution(var_77, conv2d_19_w, 1, 1, 1, 1, 1, 0);
- void* var_80 = tensorHalfAdd(var_79, conv2d_19_b);
- void* var_81 = tensorHalfAdd(var_73, var_80);
- void* var_82 = tensorHalfRelu(var_81);
- void* var_84 = tensorHalfConvolution(var_82, conv2d_20_w, 1, 1, 1, 1, 1, 0);
- void* var_85 = tensorHalfAdd(var_84, conv2d_20_b);
- void* var_86 = tensorHalfRelu(var_85);
- void* var_88 = tensorHalfConvolution(var_86, conv2d_21_w, 1, 1, 1, 1, 1, 0);
- void* var_89 = tensorHalfAdd(var_88, conv2d_21_b);
- void* var_90 = tensorHalfAdd(var_82, var_89);
- void* var_91 = tensorHalfRelu(var_90);
- void* var_92 = tensorHalfPooling(var_91,1,8,8,0,0,8,8);
- void* var_94 = tensorHalfGemmGPU(var_92, dense_1_w);
- void* var_95 = tensorHalfAdd(var_94, dense_1_b);
- void* var_96 = tensorSoftmax(var_95);
-
- profiler.pause_profiler();
- auto time_energy = profiler.get_time_energy();
- total_time += time_energy.first;
- profiler.reset();
-
- uint8_t* labels = readLabelsBatch(labels_path.c_str(), start, end);
-
- float accuracy = computeAccuracy2(labels,batch_size,var_96);
- final_accuracy += accuracy;
-
- freeBatchMemory();
- }
- }
- stopProfiling();
-
- profiler.stop_profiler();
-
- final_accuracy = final_accuracy / batch_count / total_runs;
- dumpFinalAccuracy(final_accuracy);
-
- std::cout<<"---------------------------------------\n";
- std::cout<<"Average time: " << total_time / total_runs << '\n';
- std::cout<<"---------------------------------------\n";
-
- llvm_hpvm_cleanupTensorRt();
-
- return 0;
-
-}
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/vgg16_cifar100_half_profiling.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/vgg16_cifar100_half_profiling.cc
deleted file mode 100644
index b778b1720c8a2db2f90230c3e57d0e0928f8665b..0000000000000000000000000000000000000000
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/vgg16_cifar100_half_profiling.cc
+++ /dev/null
@@ -1,182 +0,0 @@
-#include "/home/nvidia/Gitlab/hpvm/llvm/projects/gpu_profiler/include/profiler.h"
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <string.h>
-
-#include "../../../../tensor_runtime/include/tensor_runtime.h"
-#include "../../../include/utils.h"
-
-int main(){
-
- llvm_hpvm_initTensorRt(0);
-
- std::string dir_prefix = std::string("../model_params/vgg16_cifar100_front/");
- std::string input_path = dir_prefix + std::string("input.bin");
- std::string labels_path = dir_prefix + std::string("labels.bin");
- 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(), 0,64,3,3,3);
- 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(), 0,1,64,1,1);
- 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(), 0,64,64,3,3);
- 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(), 0,128,64,3,3);
- 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(), 0,1,128,1,1);
- std::string conv2d_4_w_path = dir_prefix + std::string("conv2d_4_w.bin");
- void* conv2d_4_w = readTrainedWeights(conv2d_4_w_path.c_str(), 0,128,128,3,3);
- std::string conv2d_4_b_path = dir_prefix + std::string("conv2d_4_b.bin");
- void* conv2d_4_b = readTrainedWeights(conv2d_4_b_path.c_str(), 0,1,128,1,1);
- std::string conv2d_5_w_path = dir_prefix + std::string("conv2d_5_w.bin");
- void* conv2d_5_w = readTrainedWeights(conv2d_5_w_path.c_str(), 0,256,128,3,3);
- std::string conv2d_5_b_path = dir_prefix + std::string("conv2d_5_b.bin");
- void* conv2d_5_b = readTrainedWeights(conv2d_5_b_path.c_str(), 0,1,256,1,1);
- std::string conv2d_6_w_path = dir_prefix + std::string("conv2d_6_w.bin");
- void* conv2d_6_w = readTrainedWeights(conv2d_6_w_path.c_str(), 0,256,256,3,3);
- std::string conv2d_6_b_path = dir_prefix + std::string("conv2d_6_b.bin");
- void* conv2d_6_b = readTrainedWeights(conv2d_6_b_path.c_str(), 0,1,256,1,1);
- std::string conv2d_7_w_path = dir_prefix + std::string("conv2d_7_w.bin");
- void* conv2d_7_w = readTrainedWeights(conv2d_7_w_path.c_str(), 0,256,256,3,3);
- std::string conv2d_7_b_path = dir_prefix + std::string("conv2d_7_b.bin");
- void* conv2d_7_b = readTrainedWeights(conv2d_7_b_path.c_str(), 0,1,256,1,1);
- std::string conv2d_8_w_path = dir_prefix + std::string("conv2d_8_w.bin");
- void* conv2d_8_w = readTrainedWeights(conv2d_8_w_path.c_str(), 0,512,256,3,3);
- std::string conv2d_8_b_path = dir_prefix + std::string("conv2d_8_b.bin");
- void* conv2d_8_b = readTrainedWeights(conv2d_8_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_9_w_path = dir_prefix + std::string("conv2d_9_w.bin");
- void* conv2d_9_w = readTrainedWeights(conv2d_9_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_9_b_path = dir_prefix + std::string("conv2d_9_b.bin");
- void* conv2d_9_b = readTrainedWeights(conv2d_9_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_10_w_path = dir_prefix + std::string("conv2d_10_w.bin");
- void* conv2d_10_w = readTrainedWeights(conv2d_10_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_10_b_path = dir_prefix + std::string("conv2d_10_b.bin");
- void* conv2d_10_b = readTrainedWeights(conv2d_10_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_11_w_path = dir_prefix + std::string("conv2d_11_w.bin");
- void* conv2d_11_w = readTrainedWeights(conv2d_11_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_11_b_path = dir_prefix + std::string("conv2d_11_b.bin");
- void* conv2d_11_b = readTrainedWeights(conv2d_11_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_12_w_path = dir_prefix + std::string("conv2d_12_w.bin");
- void* conv2d_12_w = readTrainedWeights(conv2d_12_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_12_b_path = dir_prefix + std::string("conv2d_12_b.bin");
- void* conv2d_12_b = readTrainedWeights(conv2d_12_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_13_w_path = dir_prefix + std::string("conv2d_13_w.bin");
- void* conv2d_13_w = readTrainedWeights(conv2d_13_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_13_b_path = dir_prefix + std::string("conv2d_13_b.bin");
- void* conv2d_13_b = readTrainedWeights(conv2d_13_b_path.c_str(), 0,1,512,1,1);
- 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(), 0,1,1,512,512);
- 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(), 0,1,512,1,1);
- 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(), 0,1,1,512,100);
- 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(), 0,1,100,1,1);
-
-
- startMemTracking();
-
- int test_input_size = 5000;
- int batch_size = 1000;
- int batch_count = test_input_size / batch_size;
- float final_accuracy = 0.0;
-
- int total_runs = 10;
- Profiler profiler;
- profiler.start_profiler();
- double total_time = 0.0;
-
- for (int i = 0; i < total_runs; i++){
- for(int i = 0; i < batch_count; i++){
-
- int start = i * batch_size;
- int end = (i + 1) * batch_size;
-
- void* input = readInputBatch(input_path.c_str(),0,start,end,3,32,32);
-
- profiler.resume_profiler();
-
- void* var_0 = tensorHalfConvolution(input, conv2d_1_w, 1, 1, 1, 1, 1, 0);
- void* var_1 = tensorHalfAdd(var_0, conv2d_1_b);
- void* var_2 = tensorHalfRelu(var_1);
- void* var_4 = tensorHalfConvolution(var_2, conv2d_2_w, 1, 1, 1, 1, 1, 0);
- void* var_5 = tensorHalfAdd(var_4, conv2d_2_b);
- void* var_6 = tensorHalfRelu(var_5);
- void* var_7 = tensorHalfPooling(var_6,0,2,2,0,0,2,2);
- void* var_8 = tensorHalfConvolution(var_7, conv2d_3_w, 1, 1, 1, 1, 1, 0);
- void* var_9 = tensorHalfAdd(var_8, conv2d_3_b);
- void* var_10 = tensorHalfRelu(var_9);
- void* var_12 = tensorHalfConvolution(var_10, conv2d_4_w, 1, 1, 1, 1, 1, 0);
- void* var_13 = tensorHalfAdd(var_12, conv2d_4_b);
- void* var_14 = tensorHalfRelu(var_13);
- void* var_15 = tensorHalfPooling(var_14,0,2,2,0,0,2,2);
- void* var_16 = tensorHalfConvolution(var_15, conv2d_5_w, 1, 1, 1, 1, 1, 0);
- void* var_17 = tensorHalfAdd(var_16, conv2d_5_b);
- void* var_18 = tensorHalfRelu(var_17);
- void* var_20 = tensorHalfConvolution(var_18, conv2d_6_w, 1, 1, 1, 1, 1, 0);
- void* var_21 = tensorHalfAdd(var_20, conv2d_6_b);
- void* var_22 = tensorHalfRelu(var_21);
- void* var_24 = tensorHalfConvolution(var_22, conv2d_7_w, 1, 1, 1, 1, 1, 0);
- void* var_25 = tensorHalfAdd(var_24, conv2d_7_b);
- void* var_26 = tensorHalfRelu(var_25);
- void* var_27 = tensorHalfPooling(var_26,0,2,2,0,0,2,2);
- void* var_28 = tensorHalfConvolution(var_27, conv2d_8_w, 1, 1, 1, 1, 1, 0);
- void* var_29 = tensorHalfAdd(var_28, conv2d_8_b);
- void* var_30 = tensorHalfRelu(var_29);
- void* var_32 = tensorHalfConvolution(var_30, conv2d_9_w, 1, 1, 1, 1, 1, 0);
- void* var_33 = tensorHalfAdd(var_32, conv2d_9_b);
- void* var_34 = tensorHalfRelu(var_33);
- void* var_36 = tensorHalfConvolution(var_34, conv2d_10_w, 1, 1, 1, 1, 1, 0);
- void* var_37 = tensorHalfAdd(var_36, conv2d_10_b);
- void* var_38 = tensorHalfRelu(var_37);
- void* var_39 = tensorHalfPooling(var_38,0,2,2,0,0,2,2);
- void* var_40 = tensorHalfConvolution(var_39, conv2d_11_w, 1, 1, 1, 1, 1, 0);
- void* var_41 = tensorHalfAdd(var_40, conv2d_11_b);
- void* var_42 = tensorHalfRelu(var_41);
- void* var_44 = tensorHalfConvolution(var_42, conv2d_12_w, 1, 1, 1, 1, 1, 0);
- void* var_45 = tensorHalfAdd(var_44, conv2d_12_b);
- void* var_46 = tensorHalfRelu(var_45);
- void* var_48 = tensorHalfConvolution(var_46, conv2d_13_w, 1, 1, 1, 1, 1, 0);
- void* var_49 = tensorHalfAdd(var_48, conv2d_13_b);
- void* var_50 = tensorHalfRelu(var_49);
- void* var_51 = tensorHalfPooling(var_50,0,2,2,0,0,2,2);
- void* var_54 = tensorHalfGemmGPU(var_51, dense_1_w);
- void* var_55 = tensorHalfAdd(var_54, dense_1_b);
- void* var_56 = tensorHalfRelu(var_55);
- void* var_58 = tensorHalfGemmGPU(var_56, dense_2_w);
- void* var_59 = tensorHalfAdd(var_58, dense_2_b);
- void* var_60 = tensorSoftmax(var_59);
-
- profiler.pause_profiler();
- auto time_energy = profiler.get_time_energy();
- total_time += time_energy.first;
- profiler.reset();
-
- uint8_t* labels = readLabelsBatch(labels_path.c_str(),start,end);
-
- float accuracy = computeAccuracy2(labels, batch_size, var_60, 100);
- final_accuracy += accuracy;
- freeBatchMemory();
-
- }
- }
-
- profiler.stop_profiler();
-
- std::cout<<"---------------------------------------\n";
- std::cout<<"Average time: " << total_time / total_runs << '\n';
- std::cout<<"---------------------------------------\n";
-
- final_accuracy = final_accuracy / batch_count / total_runs;
- dumpFinalAccuracy(final_accuracy);
-
- llvm_hpvm_cleanupTensorRt();
-
- return 0;
-}
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/vgg16_cifar10_half_profiling.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/vgg16_cifar10_half_profiling.cc
deleted file mode 100644
index 3f97e5dbde3b6d124888a8c74d435880097a394c..0000000000000000000000000000000000000000
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/profiling/vgg16_cifar10_half_profiling.cc
+++ /dev/null
@@ -1,189 +0,0 @@
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <string.h>
-
-#include "../../../../tensor_runtime/include/tensor_runtime.h"
-#include "../../../include/utils.h"
-
-#include "/home/nvidia/Gitlab/hpvm/llvm/projects/gpu_profiler/include/profiler.h"
-
-int main(){
-
- llvm_hpvm_initTensorRt(0);
-
- std::string dir_prefix = std::string("../model_params/vgg16_cifar10_2/");
- std::string input_path = dir_prefix + std::string("input.bin");
- std::string labels_path = dir_prefix + std::string("labels.bin");
- 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(), 0,64,3,3,3);
- 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(), 0,1,64,1,1);
- 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(), 0,64,64,3,3);
- 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(), 0,128,64,3,3);
- 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(), 0,1,128,1,1);
- std::string conv2d_4_w_path = dir_prefix + std::string("conv2d_4_w.bin");
- void* conv2d_4_w = readTrainedWeights(conv2d_4_w_path.c_str(), 0,128,128,3,3);
- std::string conv2d_4_b_path = dir_prefix + std::string("conv2d_4_b.bin");
- void* conv2d_4_b = readTrainedWeights(conv2d_4_b_path.c_str(), 0,1,128,1,1);
- std::string conv2d_5_w_path = dir_prefix + std::string("conv2d_5_w.bin");
- void* conv2d_5_w = readTrainedWeights(conv2d_5_w_path.c_str(), 0,256,128,3,3);
- std::string conv2d_5_b_path = dir_prefix + std::string("conv2d_5_b.bin");
- void* conv2d_5_b = readTrainedWeights(conv2d_5_b_path.c_str(), 0,1,256,1,1);
- std::string conv2d_6_w_path = dir_prefix + std::string("conv2d_6_w.bin");
- void* conv2d_6_w = readTrainedWeights(conv2d_6_w_path.c_str(), 0,256,256,3,3);
- std::string conv2d_6_b_path = dir_prefix + std::string("conv2d_6_b.bin");
- void* conv2d_6_b = readTrainedWeights(conv2d_6_b_path.c_str(), 0,1,256,1,1);
- std::string conv2d_7_w_path = dir_prefix + std::string("conv2d_7_w.bin");
- void* conv2d_7_w = readTrainedWeights(conv2d_7_w_path.c_str(), 0,256,256,3,3);
- std::string conv2d_7_b_path = dir_prefix + std::string("conv2d_7_b.bin");
- void* conv2d_7_b = readTrainedWeights(conv2d_7_b_path.c_str(), 0,1,256,1,1);
- std::string conv2d_8_w_path = dir_prefix + std::string("conv2d_8_w.bin");
- void* conv2d_8_w = readTrainedWeights(conv2d_8_w_path.c_str(), 0,512,256,3,3);
- std::string conv2d_8_b_path = dir_prefix + std::string("conv2d_8_b.bin");
- void* conv2d_8_b = readTrainedWeights(conv2d_8_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_9_w_path = dir_prefix + std::string("conv2d_9_w.bin");
- void* conv2d_9_w = readTrainedWeights(conv2d_9_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_9_b_path = dir_prefix + std::string("conv2d_9_b.bin");
- void* conv2d_9_b = readTrainedWeights(conv2d_9_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_10_w_path = dir_prefix + std::string("conv2d_10_w.bin");
- void* conv2d_10_w = readTrainedWeights(conv2d_10_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_10_b_path = dir_prefix + std::string("conv2d_10_b.bin");
- void* conv2d_10_b = readTrainedWeights(conv2d_10_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_11_w_path = dir_prefix + std::string("conv2d_11_w.bin");
- void* conv2d_11_w = readTrainedWeights(conv2d_11_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_11_b_path = dir_prefix + std::string("conv2d_11_b.bin");
- void* conv2d_11_b = readTrainedWeights(conv2d_11_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_12_w_path = dir_prefix + std::string("conv2d_12_w.bin");
- void* conv2d_12_w = readTrainedWeights(conv2d_12_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_12_b_path = dir_prefix + std::string("conv2d_12_b.bin");
- void* conv2d_12_b = readTrainedWeights(conv2d_12_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_13_w_path = dir_prefix + std::string("conv2d_13_w.bin");
- void* conv2d_13_w = readTrainedWeights(conv2d_13_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_13_b_path = dir_prefix + std::string("conv2d_13_b.bin");
- void* conv2d_13_b = readTrainedWeights(conv2d_13_b_path.c_str(), 0,1,512,1,1);
- 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(), 0,1,1,512,512);
- 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(), 0,1,512,1,1);
- 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(), 0,1,1,512,10);
- 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);
-
-
- startMemTracking();
-
- int test_input_size = 5000;
- int batch_size = 1000;
- int batch_count = test_input_size / batch_size;
- float final_accuracy = 0.0;
-
- int total_runs = 10;
-
- // NOTE: Starting time profiling
- startProfiling();
-
- Profiler profiler;
- profiler.start_profiler();
-
- double total_time = 0.0;
- for (int itrs = 0; itrs < total_runs; itrs++){
- for(int i = 0; i < batch_count; i++){
-
- int start = i * batch_size;
- int end = (i + 1) * batch_size;
-
- void* input = readInputBatch(input_path.c_str(), 0,start,end,3,32,32);
-
- profiler.resume_profiler();
-
- void* var_0 = tensorHalfConvolution(input, conv2d_1_w, 1, 1, 1, 1, 1, 0);
- void* var_1 = tensorHalfAdd(var_0, conv2d_1_b);
- void* var_2 = tensorHalfRelu(var_1);
- void* var_4 = tensorHalfConvolution(var_2, conv2d_2_w, 1, 1, 1, 1, 1, 0);
- void* var_5 = tensorHalfAdd(var_4, conv2d_2_b);
- void* var_6 = tensorHalfRelu(var_5);
- void* var_7 = tensorHalfPooling(var_6,0,2,2,0,0,2,2);
- void* var_8 = tensorHalfConvolution(var_7, conv2d_3_w, 1, 1, 1, 1, 1, 0);
- void* var_9 = tensorHalfAdd(var_8, conv2d_3_b);
- void* var_10 = tensorHalfRelu(var_9);
- void* var_12 = tensorHalfConvolution(var_10, conv2d_4_w, 1, 1, 1, 1, 1, 0);
- void* var_13 = tensorHalfAdd(var_12, conv2d_4_b);
- void* var_14 = tensorHalfRelu(var_13);
- void* var_15 = tensorHalfPooling(var_14,0,2,2,0,0,2,2);
- void* var_16 = tensorHalfConvolution(var_15, conv2d_5_w, 1, 1, 1, 1, 1, 0);
- void* var_17 = tensorHalfAdd(var_16, conv2d_5_b);
- void* var_18 = tensorHalfRelu(var_17);
- void* var_20 = tensorHalfConvolution(var_18, conv2d_6_w, 1, 1, 1, 1, 1, 0);
- void* var_21 = tensorHalfAdd(var_20, conv2d_6_b);
- void* var_22 = tensorHalfRelu(var_21);
- void* var_24 = tensorHalfConvolution(var_22, conv2d_7_w, 1, 1, 1, 1, 1, 0);
- void* var_25 = tensorHalfAdd(var_24, conv2d_7_b);
- void* var_26 = tensorHalfRelu(var_25);
- void* var_27 = tensorHalfPooling(var_26,0,2,2,0,0,2,2);
- void* var_28 = tensorHalfConvolution(var_27, conv2d_8_w, 1, 1, 1, 1, 1, 0);
- void* var_29 = tensorHalfAdd(var_28, conv2d_8_b);
- void* var_30 = tensorHalfRelu(var_29);
- void* var_32 = tensorHalfConvolution(var_30, conv2d_9_w, 1, 1, 1, 1, 1, 0);
- void* var_33 = tensorHalfAdd(var_32, conv2d_9_b);
- void* var_34 = tensorHalfRelu(var_33);
- void* var_36 = tensorHalfConvolution(var_34, conv2d_10_w, 1, 1, 1, 1, 1, 0);
- void* var_37 = tensorHalfAdd(var_36, conv2d_10_b);
- void* var_38 = tensorHalfRelu(var_37);
- void* var_39 = tensorHalfPooling(var_38,0,2,2,0,0,2,2);
- void* var_40 = tensorHalfConvolution(var_39, conv2d_11_w, 1, 1, 1, 1, 1, 0);
- void* var_41 = tensorHalfAdd(var_40, conv2d_11_b);
- void* var_42 = tensorHalfRelu(var_41);
- void* var_44 = tensorHalfConvolution(var_42, conv2d_12_w, 1, 1, 1, 1, 1, 0);
- void* var_45 = tensorHalfAdd(var_44, conv2d_12_b);
- void* var_46 = tensorHalfRelu(var_45);
- void* var_48 = tensorHalfConvolution(var_46, conv2d_13_w, 1, 1, 1, 1, 1, 0);
- void* var_49 = tensorHalfAdd(var_48, conv2d_13_b);
- void* var_50 = tensorHalfRelu(var_49);
- void* var_51 = tensorHalfPooling(var_50,0,2,2,0,0,2,2);
- void* var_54 = tensorHalfGemmGPU(var_51, dense_1_w);
- void* var_55 = tensorHalfAdd(var_54, dense_1_b);
- void* var_56 = tensorHalfRelu(var_55);
- void* var_58 = tensorHalfGemmGPU(var_56, dense_2_w);
- void* var_59 = tensorHalfAdd(var_58, dense_2_b);
- void* var_60 = tensorSoftmax(var_59);
-
- profiler.pause_profiler();
- auto time_energy = profiler.get_time_energy();
- total_time += time_energy.first;
- profiler.reset();
-
- uint8_t* labels = readLabelsBatch(labels_path.c_str(), start, end);
-
- float accuracy = computeAccuracy2(labels,batch_size,var_60);
- final_accuracy += accuracy;
-
- freeBatchMemory();
- }
- }
- std::cout<<"---------------------------------------\n";
- std::cout<<"Average time: " << total_time / total_runs << '\n';
- std::cout<<"---------------------------------------\n";
-
- profiler.stop_profiler();
- // Start power and performance profiling
- stopProfiling();
-
- final_accuracy = final_accuracy / batch_count / total_runs;
- dumpFinalAccuracy(final_accuracy);
-
- llvm_hpvm_cleanupTensorRt();
-
- return 0;
-
-}
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/resnet18_cifar10_half.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/resnet18_cifar10_half.cc
index 9779b95d865d1939244f50c3910d7ed770b0729d..741c4a443cc9a56c443ec5858aaed5a7d5705268 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/resnet18_cifar10_half.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/resnet18_cifar10_half.cc
@@ -13,7 +13,7 @@ int main(){
llvm_hpvm_initTensorRt(0);
- std::string dir_prefix = std::string("../model_params/resnet18_cifar10/");
+ std::string dir_prefix = model_params_path + std::string("/resnet18_cifar10/");
std::string input_path = dir_prefix + std::string("input.bin");
//void* input = readTrainedWeights(input_path.c_str(), 0, batch_size,3,32,32);
std::string labels_path = dir_prefix + std::string("labels.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/vgg16_cifar100_half.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/vgg16_cifar100_half.cc
index 7107defe9d154731a46efaf5c8ad244ceb69bad7..9ac1deea68c693f8baf2df2d9f2b626b3597ad7f 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/vgg16_cifar100_half.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/vgg16_cifar100_half.cc
@@ -13,7 +13,7 @@ int main(){
llvm_hpvm_initTensorRt(0);
- std::string dir_prefix = std::string("../model_params/vgg16_cifar100/");
+ std::string dir_prefix = model_params_path + std::string("/vgg16_cifar100/");
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels.bin");
std::string conv2d_1_w_path = dir_prefix + std::string("conv2d_1_w.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/vgg16_cifar10_half.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/vgg16_cifar10_half.cc
index 45e74fbe32e053e2d43c1dde0f90460c21ab0118..f92bac10e27162fe0bc59c07aa4f9ede542ccd6e 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/vgg16_cifar10_half.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp16/vgg16_cifar10_half.cc
@@ -13,7 +13,7 @@ int main(){
llvm_hpvm_initTensorRt(0);
- std::string dir_prefix = std::string("../model_params/vgg16_cifar10/");
+ std::string dir_prefix = model_params_path + std::string("/vgg16_cifar10/");
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels.bin");
std::string conv2d_1_w_path = dir_prefix + std::string("conv2d_1_w.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/alexnet2_canny.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/alexnet2_canny.cc
deleted file mode 100644
index 628ce6616cde37a5eddde5ab6049001525203580..0000000000000000000000000000000000000000
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/alexnet2_canny.cc
+++ /dev/null
@@ -1,255 +0,0 @@
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <vector>
-#include <string.h>
-#include "tensor_runtime.h"
-#include "utils.h"
-
-#include "tensor_custom_ops_cpu.h"
-
-
-
-
-Tensor* gaussianFilter(float sigma, size_t w, size_t h, size_t n_chan) {
- int64_t m = (w - 1) / 2, n = (h - 1) / 2;
- auto *data = new float[w * h];
- float sum = 0.0f;
- for (int64_t i = -m; i <= m; i++)
- for (int64_t j = -n; j <= n; j++) {
- size_t idx = (i + m) * h + (j + n);
- float exponent = -(i * i + j * j) / (2.0 * sigma * sigma);
- data[idx] = exp(exponent);
- sum += data[idx];
- }
- if (sum != 0.0f)
- for (size_t i = 0; i < w * h; i++)
- data[i] /= sum;
- return (Tensor *)createFilterFromData(CUDNN_DATA_FLOAT, data, w, h, n_chan);
-}
-
-std::pair<Tensor*, Tensor*> getSobelKernels() {
- std::vector<float> k1({-1, 0, 1, -2, 0, 2, -1, 0, 1});
- std::vector<float> k2({1, 2, 1, 0, 0, 0, -1, -2, -1});
- auto *t1 =
- (Tensor *)createFilterFromData(CUDNN_DATA_FLOAT, k1.data(), 3, 3, 1);
- auto *t2 =
- (Tensor *)createFilterFromData(CUDNN_DATA_FLOAT, k2.data(), 3, 3, 1);
- return std::make_pair(t1, t2);
-}
-
-/***
-
-TODOs:
-
-* Precision calculation?
-* tensorArgMax?
-* tensorSelect?
-* tensorContract
-* autotuning support for these functions
-* FP32 vs F16 versions of sampling perforation?
-* Need tensorRT version and a PROMISE API version
-* How to Profile? are profileEvent calls added
-* Pytorch version
-
-
-****/
-
-void* canny_filter(void* dataset) {
-
- Tensor* gaussian = gaussianFilter(1.4, 5, 5, 1);
- Tensor* kernel_x, *kernel_y;
- std::tie(kernel_x, kernel_y) = getSobelKernels();
-
- // 0. Grayscale
- auto* summed_image = tensorReduce(dataset, 1, MathOp::Add);
- auto* grayscale_image = tensorMap1(MathOp::Avg3, summed_image);
- // 1. Denoise
-
- auto* image2 = tensorConvolution(grayscale_image, gaussian,
- 2, 2, // padding
- 1, 1, // strides
- 1, 0); // conv_mode, conv_groups
-
- // 2. Get edge gradient / direction
- auto *grad_x = tensorConvolution(image2, kernel_x,
- 1, 1,
- 1, 1,
- 1, 0);
-
- auto *grad_y = tensorConvolution(image2, kernel_y,
- 1, 1,
- 1, 1,
- 1, 0);
-
- auto *grad_mag = tensorMap2(MathOp::Hypot, grad_x, grad_y);
- // 2.5. Normalize grad magnitude
- auto *grad_max_1D = tensorReduce(grad_mag, 2, MathOp::Max);
- auto *grad_max = tensorReduce(grad_max_1D, 3, MathOp::Max);
- auto *grad_mag_norm = tensorMap2(MathOp::Div, grad_mag, grad_max);
- return grad_mag_norm;
-}
-
-
-
-
-void* invoke_canny(void* input) {
-
- auto* result = canny_filter(input);
-
- printf("Done with Canny \n");
-
- return result;
-}
-
-
-
-
-
-
-
-
-int main(){
-
- llvm_hpvm_initTensorRt(0);
-
- std::string dir_prefix = std::string("../model_params/alexnet2_cifar10/");
- std::string input_path = dir_prefix + std::string("norm_cifar_input.bin");
- std::string canny_input_path = dir_prefix + std::string("canny_input.bin");
- std::string labels_path = dir_prefix + std::string("test_labels.bin");
-
- void* conv1_filter = readTrainedWeights("../model_params/alexnet2_cifar10/conv1.bin",
- float_type, 32, 3, 3, 3);
- void* conv1_bias = readTrainedWeights("../model_params/alexnet2_cifar10/conv1_bias.bin",
- float_type, 1, 32, 1, 1);
- void* conv2_filter = readTrainedWeights("../model_params/alexnet2_cifar10/conv2.bin",
- float_type, 32, 32, 3, 3);
- void* conv2_bias = readTrainedWeights("../model_params/alexnet2_cifar10/conv2_bias.bin",
- float_type, 1, 32, 1, 1);
- void* conv3_filter = readTrainedWeights("../model_params/alexnet2_cifar10/conv3.bin",
- float_type, 64, 32, 3, 3);
- void* conv3_bias = readTrainedWeights("../model_params/alexnet2_cifar10/conv3_bias.bin",
- float_type, 1, 64, 1, 1);
- void* conv4_filter = readTrainedWeights("../model_params/alexnet2_cifar10/conv4.bin",
- float_type, 64, 64, 3, 3);
- void* conv4_bias = readTrainedWeights("../model_params/alexnet2_cifar10/conv4_bias.bin",
- float_type, 1, 64, 1, 1);
- void* conv5_filter = readTrainedWeights("../model_params/alexnet2_cifar10/conv5.bin",
- float_type, 128, 64, 3, 3);
- void* conv5_bias = readTrainedWeights("../model_params/alexnet2_cifar10/conv5_bias.bin",
- float_type, 1, 128, 1, 1);
- void* conv6_filter = readTrainedWeights("../model_params/alexnet2_cifar10/conv6.bin",
- float_type, 128, 128, 3, 3);
- void* conv6_bias = readTrainedWeights("../model_params/alexnet2_cifar10/conv6_bias.bin",
- float_type, 1, 128, 1, 1);
-
- void* fc1_weights = readTrainedWeights("../model_params/alexnet2_cifar10/fc1.bin",
- float_type, 1, 1, 2048, 10);
- void* fc1_bias = readTrainedWeights("../model_params/alexnet2_cifar10/fc1_bias.bin",
- float_type, 1, 10, 1, 1);
-
-
- int test_input_size = 5000;
- int batch_size = 500;
- int batch_count = test_input_size / batch_size;
- float final_accuracy = 0.0;
-
- // NOTE: Starting time profiling
- startProfiling();
- startMemTracking();
-
- int conv_mode = 1; // NOTE: using CROSS_CORRELATION
- int conv_precision = 0; // NOTE: using Float as compute precision. FIXIT: use enum
-
- for(int i = 0; i < batch_count; i++){
-
- int start = i * batch_size;
- int end = (i + 1) * batch_size;
-
-
- void* input = readInputBatch(input_path.c_str(), 0,start,end,3,32,32);
- void* canny_input = readInputBatch(canny_input_path.c_str(), 0,start,end, 3, 128, 128);
-
- void* conv1out = tensorConvolution(input, conv1_filter, 1, 1, 1, 1,
- conv_mode, conv_precision);
- tensorAdd(conv1out, conv1_bias);
- void* conv1_tanh = tensorTanh(conv1out);
-
- // 2nd Layer
- void* conv2out = tensorConvolution(conv1_tanh, conv2_filter, 1, 1, 1, 1,
- conv_mode, conv_precision);
- tensorAdd(conv2out, conv2_bias);
- void* conv2_tanh = tensorTanh(conv2out);
- void* pool2out = tensorPooling(conv2_tanh, 0, 2, 2, 0, 0, 2, 2);
-
- // 3rd Layer
- void* conv3out = tensorConvolution(pool2out, conv3_filter, 1, 1, 1, 1,
- conv_mode, conv_precision);
- tensorAdd(conv3out, conv3_bias);
- void* conv3_tanh = tensorTanh(conv3out);
-
- // 4th Layer
- void* conv4out = tensorConvolution(conv3_tanh, conv4_filter, 1, 1, 1, 1,
- conv_mode, conv_precision);
- tensorAdd(conv4out, conv4_bias);
- void* conv4_tanh = tensorTanh(conv4out);
- void* pool4out = tensorPooling(conv4_tanh, 0, 2, 2, 0, 0, 2, 2);
-
- // 5th Layer
- void* conv5out = tensorConvolution(pool4out, conv5_filter, 1, 1, 1, 1,
- conv_mode, conv_precision);
- tensorAdd(conv5out, conv5_bias);
- void* conv5_tanh = tensorTanh(conv5out);
-
- // 6th Layer
- void* conv6out = tensorConvolution(conv5_tanh, conv6_filter, 1, 1, 1, 1,
- conv_mode, conv_precision);
- tensorAdd(conv6out, conv6_bias);
-
- void* conv6_tanh = tensorTanh(conv6out);
- void* pool6out = tensorPooling(conv6_tanh, 0, 2, 2, 0, 0, 2, 2);
-
- // final FC Layer
- void* gemm1out = tensorGemmGPU(pool6out, fc1_weights);
- void* gemm1biasout = tensorAdd(gemm1out, fc1_bias);
- void* result = tensorSoftmax(gemm1biasout);
-
- uint8_t* labels = readLabelsBatch(labels_path.c_str(), start, end);
-
- float accuracy = computeAccuracy2(labels, batch_size, result);
- final_accuracy += accuracy;
-
-
- std::vector<int> index_vector;
- index_vector.push_back(1);
- index_vector.push_back(2);
- index_vector.push_back(3);
- index_vector.push_back(4);
- index_vector.push_back(5);
-
-
- void* argmax_out = tensorArgMax(result);
- void* select_out = tensorSelect2(argmax_out, index_vector);
- void* reduced_input = tensorContract(canny_input, select_out);
-
- invoke_canny(reduced_input);
-
-
- freeBatchMemory();
- }
-
- stopProfiling();
-
- final_accuracy = final_accuracy / batch_count;
- dumpFinalAccuracy(final_accuracy);
-
-
- llvm_hpvm_cleanupTensorRt();
-
- return 0;
-
-}
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/alexnet2_cifar10.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/alexnet2_cifar10.cc
index 5918f4f18ebdb7d4f2fa3e37c0982b8ed8d10932..50d9747f990d486c4543607d16d4a4ccb88b0517 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/alexnet2_cifar10.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/alexnet2_cifar10.cc
@@ -19,7 +19,7 @@ void testCifarNet(){
printf("********* Alexnet2 CIFAR-10 DNN ********** \n");
- std::string dir_prefix = std::string("../model_params/alexnet2_cifar10/");
+ std::string dir_prefix = model_params_path + std::string("/alexnet2_cifar10/");
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels.bin");
std::string labels32_path = dir_prefix + std::string("labels32.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/alexnet_cifar10.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/alexnet_cifar10.cc
index 8129fbfafcdd3e991e67d33fd3013e1700da45c5..1a76f1ae8ba6059124117b82cd72e8ccd6cdeba6 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/alexnet_cifar10.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/alexnet_cifar10.cc
@@ -13,7 +13,7 @@ int main(){
llvm_hpvm_initTensorRt(0);
- std::string dir_prefix = std::string("../model_params/alexnet_cifar10/");
+ std::string dir_prefix = model_params_path + std::string("/alexnet_cifar10/");
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/lenet_mnist.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/lenet_mnist.cc
index c047ffe090a93711cb66973ef6622d46fccdcee3..7508f3119eeb469a164fad9741000308e3e8c031 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/lenet_mnist.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/lenet_mnist.cc
@@ -22,7 +22,7 @@ void testLenetTanh(){
int test_batch_size = 5000;
- std::string dir_prefix = std::string("../model_params/lenet_mnist/");
+ std::string dir_prefix = model_params_path + std::string("/lenet_mnist/");
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/mobilenet.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/mobilenet.cc
index 78ca2dac98435ea146da44a78bb2f7405af8c5ef..7c311a568647caa107112bed4982fb57254dc7b3 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/mobilenet.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/mobilenet.cc
@@ -15,7 +15,7 @@ int main(){
llvm_hpvm_initTensorRt(0);
- std::string dir_prefix = std::string("../model_params/mobilenet/");
+ std::string dir_prefix = model_params_path + std::string("/mobilenet/");
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels.bin");
std::string conv2d_1_w_path = dir_prefix + std::string("conv2d_1_w.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/mobilenet_shallow.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/mobilenet_shallow.cc
deleted file mode 100644
index d30518216f76160e183a915a6e6da2018239ab60..0000000000000000000000000000000000000000
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/mobilenet_shallow.cc
+++ /dev/null
@@ -1,240 +0,0 @@
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <string.h>
-#include "../../tensor_runtime/include/tensor_runtime.h"
-#include "../include/utils.h"
-
-int main(int argc, char* argv[]){
-
- int total_runs = 1;
- if (argc > 1){
- total_runs = atoi(argv[1]);
- }
-
-
- llvm_hpvm_initTensorRt(0);
-
- std::string dir_prefix = std::string("../model_params/mobilenet_shallow/");
-
- std::string input_path = dir_prefix + std::string("input.bin");
- std::string labels_path = dir_prefix + std::string("labels.bin");
- 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(), 0,32,3,3,3);
- std::string batch_normalization_1_gamma_path = dir_prefix + std::string("batch_normalization_1_gamma.bin");
- void* batch_normalization_1_gamma = readTrainedWeights(batch_normalization_1_gamma_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_beta_path = dir_prefix + std::string("batch_normalization_1_beta.bin");
- void* batch_normalization_1_beta = readTrainedWeights(batch_normalization_1_beta_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_mean_path = dir_prefix + std::string("batch_normalization_1_mean.bin");
- void* batch_normalization_1_mean = readTrainedWeights(batch_normalization_1_mean_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_1_variance_path = dir_prefix + std::string("batch_normalization_1_variance.bin");
- void* batch_normalization_1_variance = readTrainedWeights(batch_normalization_1_variance_path.c_str(), 0,1,32,1,1);
- std::string depthwise_conv2d_1_w_path = dir_prefix + std::string("depthwise_conv2d_1_w.bin");
- void* depthwise_conv2d_1_w = readTrainedWeights(depthwise_conv2d_1_w_path.c_str(), 0,32,1,3,3);
- std::string batch_normalization_2_gamma_path = dir_prefix + std::string("batch_normalization_2_gamma.bin");
- void* batch_normalization_2_gamma = readTrainedWeights(batch_normalization_2_gamma_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_beta_path = dir_prefix + std::string("batch_normalization_2_beta.bin");
- void* batch_normalization_2_beta = readTrainedWeights(batch_normalization_2_beta_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_mean_path = dir_prefix + std::string("batch_normalization_2_mean.bin");
- void* batch_normalization_2_mean = readTrainedWeights(batch_normalization_2_mean_path.c_str(), 0,1,32,1,1);
- std::string batch_normalization_2_variance_path = dir_prefix + std::string("batch_normalization_2_variance.bin");
- void* batch_normalization_2_variance = readTrainedWeights(batch_normalization_2_variance_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(), 0,64,32,1,1);
- std::string batch_normalization_3_gamma_path = dir_prefix + std::string("batch_normalization_3_gamma.bin");
- void* batch_normalization_3_gamma = readTrainedWeights(batch_normalization_3_gamma_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_beta_path = dir_prefix + std::string("batch_normalization_3_beta.bin");
- void* batch_normalization_3_beta = readTrainedWeights(batch_normalization_3_beta_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_mean_path = dir_prefix + std::string("batch_normalization_3_mean.bin");
- void* batch_normalization_3_mean = readTrainedWeights(batch_normalization_3_mean_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_3_variance_path = dir_prefix + std::string("batch_normalization_3_variance.bin");
- void* batch_normalization_3_variance = readTrainedWeights(batch_normalization_3_variance_path.c_str(), 0,1,64,1,1);
- std::string depthwise_conv2d_2_w_path = dir_prefix + std::string("depthwise_conv2d_2_w.bin");
- void* depthwise_conv2d_2_w = readTrainedWeights(depthwise_conv2d_2_w_path.c_str(), 0,64,1,3,3);
- std::string batch_normalization_4_gamma_path = dir_prefix + std::string("batch_normalization_4_gamma.bin");
- void* batch_normalization_4_gamma = readTrainedWeights(batch_normalization_4_gamma_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_beta_path = dir_prefix + std::string("batch_normalization_4_beta.bin");
- void* batch_normalization_4_beta = readTrainedWeights(batch_normalization_4_beta_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_mean_path = dir_prefix + std::string("batch_normalization_4_mean.bin");
- void* batch_normalization_4_mean = readTrainedWeights(batch_normalization_4_mean_path.c_str(), 0,1,64,1,1);
- std::string batch_normalization_4_variance_path = dir_prefix + std::string("batch_normalization_4_variance.bin");
- void* batch_normalization_4_variance = readTrainedWeights(batch_normalization_4_variance_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(), 0,128,64,1,1);
- std::string batch_normalization_5_gamma_path = dir_prefix + std::string("batch_normalization_5_gamma.bin");
- void* batch_normalization_5_gamma = readTrainedWeights(batch_normalization_5_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_beta_path = dir_prefix + std::string("batch_normalization_5_beta.bin");
- void* batch_normalization_5_beta = readTrainedWeights(batch_normalization_5_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_mean_path = dir_prefix + std::string("batch_normalization_5_mean.bin");
- void* batch_normalization_5_mean = readTrainedWeights(batch_normalization_5_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_5_variance_path = dir_prefix + std::string("batch_normalization_5_variance.bin");
- void* batch_normalization_5_variance = readTrainedWeights(batch_normalization_5_variance_path.c_str(), 0,1,128,1,1);
- std::string depthwise_conv2d_3_w_path = dir_prefix + std::string("depthwise_conv2d_3_w.bin");
- void* depthwise_conv2d_3_w = readTrainedWeights(depthwise_conv2d_3_w_path.c_str(), 0,128,1,3,3);
- std::string batch_normalization_6_gamma_path = dir_prefix + std::string("batch_normalization_6_gamma.bin");
- void* batch_normalization_6_gamma = readTrainedWeights(batch_normalization_6_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_beta_path = dir_prefix + std::string("batch_normalization_6_beta.bin");
- void* batch_normalization_6_beta = readTrainedWeights(batch_normalization_6_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_mean_path = dir_prefix + std::string("batch_normalization_6_mean.bin");
- void* batch_normalization_6_mean = readTrainedWeights(batch_normalization_6_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_6_variance_path = dir_prefix + std::string("batch_normalization_6_variance.bin");
- void* batch_normalization_6_variance = readTrainedWeights(batch_normalization_6_variance_path.c_str(), 0,1,128,1,1);
- std::string conv2d_4_w_path = dir_prefix + std::string("conv2d_4_w.bin");
- void* conv2d_4_w = readTrainedWeights(conv2d_4_w_path.c_str(), 0,128,128,1,1);
- std::string batch_normalization_7_gamma_path = dir_prefix + std::string("batch_normalization_7_gamma.bin");
- void* batch_normalization_7_gamma = readTrainedWeights(batch_normalization_7_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_beta_path = dir_prefix + std::string("batch_normalization_7_beta.bin");
- void* batch_normalization_7_beta = readTrainedWeights(batch_normalization_7_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_mean_path = dir_prefix + std::string("batch_normalization_7_mean.bin");
- void* batch_normalization_7_mean = readTrainedWeights(batch_normalization_7_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_7_variance_path = dir_prefix + std::string("batch_normalization_7_variance.bin");
- void* batch_normalization_7_variance = readTrainedWeights(batch_normalization_7_variance_path.c_str(), 0,1,128,1,1);
- std::string depthwise_conv2d_4_w_path = dir_prefix + std::string("depthwise_conv2d_4_w.bin");
- void* depthwise_conv2d_4_w = readTrainedWeights(depthwise_conv2d_4_w_path.c_str(), 0,128,1,3,3);
- std::string batch_normalization_8_gamma_path = dir_prefix + std::string("batch_normalization_8_gamma.bin");
- void* batch_normalization_8_gamma = readTrainedWeights(batch_normalization_8_gamma_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_beta_path = dir_prefix + std::string("batch_normalization_8_beta.bin");
- void* batch_normalization_8_beta = readTrainedWeights(batch_normalization_8_beta_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_mean_path = dir_prefix + std::string("batch_normalization_8_mean.bin");
- void* batch_normalization_8_mean = readTrainedWeights(batch_normalization_8_mean_path.c_str(), 0,1,128,1,1);
- std::string batch_normalization_8_variance_path = dir_prefix + std::string("batch_normalization_8_variance.bin");
- void* batch_normalization_8_variance = readTrainedWeights(batch_normalization_8_variance_path.c_str(), 0,1,128,1,1);
- std::string conv2d_5_w_path = dir_prefix + std::string("conv2d_5_w.bin");
- void* conv2d_5_w = readTrainedWeights(conv2d_5_w_path.c_str(), 0,256,128,1,1);
- std::string batch_normalization_9_gamma_path = dir_prefix + std::string("batch_normalization_9_gamma.bin");
- void* batch_normalization_9_gamma = readTrainedWeights(batch_normalization_9_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_beta_path = dir_prefix + std::string("batch_normalization_9_beta.bin");
- void* batch_normalization_9_beta = readTrainedWeights(batch_normalization_9_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_mean_path = dir_prefix + std::string("batch_normalization_9_mean.bin");
- void* batch_normalization_9_mean = readTrainedWeights(batch_normalization_9_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_9_variance_path = dir_prefix + std::string("batch_normalization_9_variance.bin");
- void* batch_normalization_9_variance = readTrainedWeights(batch_normalization_9_variance_path.c_str(), 0,1,256,1,1);
- std::string depthwise_conv2d_5_w_path = dir_prefix + std::string("depthwise_conv2d_5_w.bin");
- void* depthwise_conv2d_5_w = readTrainedWeights(depthwise_conv2d_5_w_path.c_str(), 0,256,1,3,3);
- std::string batch_normalization_10_gamma_path = dir_prefix + std::string("batch_normalization_10_gamma.bin");
- void* batch_normalization_10_gamma = readTrainedWeights(batch_normalization_10_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_beta_path = dir_prefix + std::string("batch_normalization_10_beta.bin");
- void* batch_normalization_10_beta = readTrainedWeights(batch_normalization_10_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_mean_path = dir_prefix + std::string("batch_normalization_10_mean.bin");
- void* batch_normalization_10_mean = readTrainedWeights(batch_normalization_10_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_10_variance_path = dir_prefix + std::string("batch_normalization_10_variance.bin");
- void* batch_normalization_10_variance = readTrainedWeights(batch_normalization_10_variance_path.c_str(), 0,1,256,1,1);
- std::string conv2d_6_w_path = dir_prefix + std::string("conv2d_6_w.bin");
- void* conv2d_6_w = readTrainedWeights(conv2d_6_w_path.c_str(), 0,256,256,1,1);
- std::string batch_normalization_11_gamma_path = dir_prefix + std::string("batch_normalization_11_gamma.bin");
- void* batch_normalization_11_gamma = readTrainedWeights(batch_normalization_11_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_beta_path = dir_prefix + std::string("batch_normalization_11_beta.bin");
- void* batch_normalization_11_beta = readTrainedWeights(batch_normalization_11_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_mean_path = dir_prefix + std::string("batch_normalization_11_mean.bin");
- void* batch_normalization_11_mean = readTrainedWeights(batch_normalization_11_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_11_variance_path = dir_prefix + std::string("batch_normalization_11_variance.bin");
- void* batch_normalization_11_variance = readTrainedWeights(batch_normalization_11_variance_path.c_str(), 0,1,256,1,1);
- std::string depthwise_conv2d_6_w_path = dir_prefix + std::string("depthwise_conv2d_6_w.bin");
- void* depthwise_conv2d_6_w = readTrainedWeights(depthwise_conv2d_6_w_path.c_str(), 0,256,1,3,3);
- std::string batch_normalization_12_gamma_path = dir_prefix + std::string("batch_normalization_12_gamma.bin");
- void* batch_normalization_12_gamma = readTrainedWeights(batch_normalization_12_gamma_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_beta_path = dir_prefix + std::string("batch_normalization_12_beta.bin");
- void* batch_normalization_12_beta = readTrainedWeights(batch_normalization_12_beta_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_mean_path = dir_prefix + std::string("batch_normalization_12_mean.bin");
- void* batch_normalization_12_mean = readTrainedWeights(batch_normalization_12_mean_path.c_str(), 0,1,256,1,1);
- std::string batch_normalization_12_variance_path = dir_prefix + std::string("batch_normalization_12_variance.bin");
- void* batch_normalization_12_variance = readTrainedWeights(batch_normalization_12_variance_path.c_str(), 0,1,256,1,1);
- std::string conv2d_7_w_path = dir_prefix + std::string("conv2d_7_w.bin");
- void* conv2d_7_w = readTrainedWeights(conv2d_7_w_path.c_str(), 0,512,256,1,1);
- std::string batch_normalization_13_gamma_path = dir_prefix + std::string("batch_normalization_13_gamma.bin");
- void* batch_normalization_13_gamma = readTrainedWeights(batch_normalization_13_gamma_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_beta_path = dir_prefix + std::string("batch_normalization_13_beta.bin");
- void* batch_normalization_13_beta = readTrainedWeights(batch_normalization_13_beta_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_mean_path = dir_prefix + std::string("batch_normalization_13_mean.bin");
- void* batch_normalization_13_mean = readTrainedWeights(batch_normalization_13_mean_path.c_str(), 0,1,512,1,1);
- std::string batch_normalization_13_variance_path = dir_prefix + std::string("batch_normalization_13_variance.bin");
- void* batch_normalization_13_variance = readTrainedWeights(batch_normalization_13_variance_path.c_str(), 0,1,512,1,1);
- 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(), 0,1,1,2048,10);
- 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(), 0,1,10,1,1);
-
-
- startMemTracking();
-
- int test_input_size = 5000;
- int batch_size = 2500;
- int batch_count = test_input_size / batch_size;
-
-
- for(int j = 0; j < total_runs; j++){
- float final_accuracy = 0.0;
- for(int i = 0; i < batch_count; i++){
-
- int start = i * batch_size;
- int end = (i + 1) * batch_size;
-
- void* input = readInputBatch(input_path.c_str(),0,start,end,3,32,32);
-
- void* var_0 = tensorConvolution(input, conv2d_1_w, 1, 1, 1, 1, 1, 1);
- void* var_1 = tensorBatchNorm(var_0, batch_normalization_1_gamma, batch_normalization_1_beta, batch_normalization_1_mean, batch_normalization_1_variance, 0.001);
- void* var_2 = tensorRelu(var_1);
- void* var_4 = tensorConvCutlass(var_2, depthwise_conv2d_1_w, 1, 1, 1, 1, 1, 32);
- void* var_5 = tensorBatchNorm(var_4, batch_normalization_2_gamma, batch_normalization_2_beta, batch_normalization_2_mean, batch_normalization_2_variance, 0.001);
- void* var_6 = tensorRelu(var_5);
- void* var_7 = tensorConvolution(var_6, conv2d_2_w, 0, 0, 1, 1, 1, 1);
- void* var_8 = tensorBatchNorm(var_7, batch_normalization_3_gamma, batch_normalization_3_beta, batch_normalization_3_mean, batch_normalization_3_variance, 0.001);
- void* var_9 = tensorRelu(var_8);
- void* var_11 = tensorConvCutlass(var_9, depthwise_conv2d_2_w, 1, 1, 2, 2, 1, 64);
- void* var_12 = tensorBatchNorm(var_11, batch_normalization_4_gamma, batch_normalization_4_beta, batch_normalization_4_mean, batch_normalization_4_variance, 0.001);
- void* var_13 = tensorRelu(var_12);
- void* var_14 = tensorConvolution(var_13, conv2d_3_w, 0, 0, 1, 1, 1, 1);
- void* var_15 = tensorBatchNorm(var_14, batch_normalization_5_gamma, batch_normalization_5_beta, batch_normalization_5_mean, batch_normalization_5_variance, 0.001);
- void* var_16 = tensorRelu(var_15);
- void* var_18 = tensorConvCutlass(var_16, depthwise_conv2d_3_w, 1, 1, 1, 1, 1, 128);
- void* var_19 = tensorBatchNorm(var_18, batch_normalization_6_gamma, batch_normalization_6_beta, batch_normalization_6_mean, batch_normalization_6_variance, 0.001);
- void* var_20 = tensorRelu(var_19);
- void* var_21 = tensorConvolution(var_20, conv2d_4_w, 0, 0, 1, 1, 1, 1);
- void* var_22 = tensorBatchNorm(var_21, batch_normalization_7_gamma, batch_normalization_7_beta, batch_normalization_7_mean, batch_normalization_7_variance, 0.001);
- void* var_23 = tensorRelu(var_22);
- void* var_26 = tensorConvCutlass(var_23, depthwise_conv2d_4_w, 1, 1, 2, 2, 1, 128);
- void* var_27 = tensorBatchNorm(var_26, batch_normalization_8_gamma, batch_normalization_8_beta, batch_normalization_8_mean, batch_normalization_8_variance, 0.001);
- void* var_28 = tensorRelu(var_27);
- void* var_29 = tensorConvolution(var_28, conv2d_5_w, 0, 0, 1, 1, 1, 1);
- void* var_30 = tensorBatchNorm(var_29, batch_normalization_9_gamma, batch_normalization_9_beta, batch_normalization_9_mean, batch_normalization_9_variance, 0.001);
- void* var_31 = tensorRelu(var_30);
- void* var_33 = tensorConvCutlass(var_31, depthwise_conv2d_5_w, 1, 1, 1, 1, 1, 256);
- void* var_34 = tensorBatchNorm(var_33, batch_normalization_10_gamma, batch_normalization_10_beta, batch_normalization_10_mean, batch_normalization_10_variance, 0.001);
- void* var_35 = tensorRelu(var_34);
- void* var_36 = tensorConvolution(var_35, conv2d_6_w, 0, 0, 1, 1, 1, 1);
- void* var_37 = tensorBatchNorm(var_36, batch_normalization_11_gamma, batch_normalization_11_beta, batch_normalization_11_mean, batch_normalization_11_variance, 0.001);
- void* var_38 = tensorRelu(var_37);
- void* var_41 = tensorConvCutlass(var_38, depthwise_conv2d_6_w, 1, 1, 2, 2, 1, 256);
- void* var_42 = tensorBatchNorm(var_41, batch_normalization_12_gamma, batch_normalization_12_beta, batch_normalization_12_mean, batch_normalization_12_variance, 0.001);
- void* var_43 = tensorRelu(var_42);
- void* var_44 = tensorConvolution(var_43, conv2d_7_w, 0, 0, 1, 1, 1, 1);
- void* var_45 = tensorBatchNorm(var_44, batch_normalization_13_gamma, batch_normalization_13_beta, batch_normalization_13_mean, batch_normalization_13_variance, 0.001);
- void* var_46 = tensorRelu(var_45);
- void* var_47 = tensorPooling(var_46,1,2,2,0,0,2,2);
- void* var_49 = tensorGemmGPU(var_47, dense_1_w);
- void* var_50 = tensorAdd(var_49, dense_1_b);
- void* var_51 = tensorSoftmax(var_50);
-
- uint8_t* labels = readLabelsBatch(labels_path.c_str(),start,end);
-
- float accuracy = computeAccuracy2(labels, batch_size, var_51);
- final_accuracy += accuracy;
- freeBatchMemory();
-
- }
-
- final_accuracy = final_accuracy / batch_count;
- dumpFinalAccuracy(final_accuracy);
- }
-
- dumpExecutionAccuracies();
-
- llvm_hpvm_cleanupTensorRt();
-
- return 0;
-
-}
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/resnet18_cifar10.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/resnet18_cifar10.cc
index b0c868085bae1abc2025364609114cc21c7d213a..87b8cd4156ed8d7f882ff7642420c995cd7c3a0f 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/resnet18_cifar10.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/resnet18_cifar10.cc
@@ -13,7 +13,7 @@ int main(){
llvm_hpvm_initTensorRt(1);
- std::string dir_prefix = std::string("../model_params/resnet18_cifar10/");
+ std::string dir_prefix = model_params_path + std::string("/resnet18_cifar10/");
std::string input_path = dir_prefix + std::string("input.bin");
//void* input = readTrainedWeights(input_path.c_str(), 0, batch_size,3,32,32);
std::string labels_path = dir_prefix + std::string("labels.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/resnet50_imagenet.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/resnet50_imagenet.cc
index 1192d04de200c8e8183c35861da2d04aa705e955..0914b3f70c353ee7e56c39ccf52f21914618301e 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/resnet50_imagenet.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/resnet50_imagenet.cc
@@ -15,7 +15,7 @@ int main(){
llvm_hpvm_initTensorRt(0);
- std::string dir_prefix = std::string("/shared/hsharif3/resnet50_imagenet_tune/");
+ std::string dir_prefix = model_params_path + std::string("/shared/hsharif3/resnet50_imagenet/");
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels.bin");
std::string conv2d_1_w_path = dir_prefix + std::string("conv2d_1_w.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/vgg16_cifar10.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/vgg16_cifar10.cc
index e8469e8a4892f51337118e4699f09ae98c13bf71..a6dc7cbc11cf77357a749bff117489fc4b292941 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/vgg16_cifar10.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/vgg16_cifar10.cc
@@ -13,7 +13,7 @@ int main(){
llvm_hpvm_initTensorRt(0);
- std::string dir_prefix = std::string("../model_params/vgg16_cifar10/");
+ std::string dir_prefix = model_params_path + std::string("/vgg16_cifar10/");
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels.bin");
std::string conv2d_1_w_path = dir_prefix + std::string("conv2d_1_w.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/vgg16_cifar100.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/vgg16_cifar100.cc
index 0290a2782880c1aa8c1ea33f5564926665d968d6..2539f8d8722909724a9dc2890e82f4f98853f5cd 100644
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/vgg16_cifar100.cc
+++ b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/vgg16_cifar100.cc
@@ -13,7 +13,7 @@ int main(){
llvm_hpvm_initTensorRt(0);
- std::string dir_prefix = std::string("../model_params/vgg16_cifar100/");
+ std::string dir_prefix = model_params_path + std::string("/vgg16_cifar100/");
std::string input_path = dir_prefix + std::string("input.bin");
std::string labels_path = dir_prefix + std::string("labels.bin");
diff --git a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/vgg16_cifar100_5.cc b/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/vgg16_cifar100_5.cc
deleted file mode 100644
index 3ee273d70aea6d74cfa55f250e999b05506f9b21..0000000000000000000000000000000000000000
--- a/hpvm/projects/hpvm-tensor-rt/dnn_sources/src/fp32/vgg16_cifar100_5.cc
+++ /dev/null
@@ -1,167 +0,0 @@
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <string.h>
-#include "../../tensor_runtime/include/tensor_runtime.h"
-#include "../include/utils.h"
-
-int main(){
-
- llvm_hpvm_initTensorRt(0);
-
- std::string dir_prefix = std::string("../model_params/vgg16_cifar100_front/");
- //std::string input_path = dir_prefix + std::string("vgg16_cifar100_calib.bin");
- //std::string labels_path = dir_prefix + std::string("vgg16_cifar100_train_labels.bin");
-
- std::string input_path = dir_prefix + std::string("input.bin");
- std::string labels_path = dir_prefix + std::string("labels.bin");
-
- 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(), 0,64,3,3,3);
- 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(), 0,1,64,1,1);
- 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(), 0,64,64,3,3);
- 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(), 0,128,64,3,3);
- 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(), 0,1,128,1,1);
- std::string conv2d_4_w_path = dir_prefix + std::string("conv2d_4_w.bin");
- void* conv2d_4_w = readTrainedWeights(conv2d_4_w_path.c_str(), 0,128,128,3,3);
- std::string conv2d_4_b_path = dir_prefix + std::string("conv2d_4_b.bin");
- void* conv2d_4_b = readTrainedWeights(conv2d_4_b_path.c_str(), 0,1,128,1,1);
- std::string conv2d_5_w_path = dir_prefix + std::string("conv2d_5_w.bin");
- void* conv2d_5_w = readTrainedWeights(conv2d_5_w_path.c_str(), 0,256,128,3,3);
- std::string conv2d_5_b_path = dir_prefix + std::string("conv2d_5_b.bin");
- void* conv2d_5_b = readTrainedWeights(conv2d_5_b_path.c_str(), 0,1,256,1,1);
- std::string conv2d_6_w_path = dir_prefix + std::string("conv2d_6_w.bin");
- void* conv2d_6_w = readTrainedWeights(conv2d_6_w_path.c_str(), 0,256,256,3,3);
- std::string conv2d_6_b_path = dir_prefix + std::string("conv2d_6_b.bin");
- void* conv2d_6_b = readTrainedWeights(conv2d_6_b_path.c_str(), 0,1,256,1,1);
- std::string conv2d_7_w_path = dir_prefix + std::string("conv2d_7_w.bin");
- void* conv2d_7_w = readTrainedWeights(conv2d_7_w_path.c_str(), 0,256,256,3,3);
- std::string conv2d_7_b_path = dir_prefix + std::string("conv2d_7_b.bin");
- void* conv2d_7_b = readTrainedWeights(conv2d_7_b_path.c_str(), 0,1,256,1,1);
- std::string conv2d_8_w_path = dir_prefix + std::string("conv2d_8_w.bin");
- void* conv2d_8_w = readTrainedWeights(conv2d_8_w_path.c_str(), 0,512,256,3,3);
- std::string conv2d_8_b_path = dir_prefix + std::string("conv2d_8_b.bin");
- void* conv2d_8_b = readTrainedWeights(conv2d_8_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_9_w_path = dir_prefix + std::string("conv2d_9_w.bin");
- void* conv2d_9_w = readTrainedWeights(conv2d_9_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_9_b_path = dir_prefix + std::string("conv2d_9_b.bin");
- void* conv2d_9_b = readTrainedWeights(conv2d_9_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_10_w_path = dir_prefix + std::string("conv2d_10_w.bin");
- void* conv2d_10_w = readTrainedWeights(conv2d_10_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_10_b_path = dir_prefix + std::string("conv2d_10_b.bin");
- void* conv2d_10_b = readTrainedWeights(conv2d_10_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_11_w_path = dir_prefix + std::string("conv2d_11_w.bin");
- void* conv2d_11_w = readTrainedWeights(conv2d_11_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_11_b_path = dir_prefix + std::string("conv2d_11_b.bin");
- void* conv2d_11_b = readTrainedWeights(conv2d_11_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_12_w_path = dir_prefix + std::string("conv2d_12_w.bin");
- void* conv2d_12_w = readTrainedWeights(conv2d_12_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_12_b_path = dir_prefix + std::string("conv2d_12_b.bin");
- void* conv2d_12_b = readTrainedWeights(conv2d_12_b_path.c_str(), 0,1,512,1,1);
- std::string conv2d_13_w_path = dir_prefix + std::string("conv2d_13_w.bin");
- void* conv2d_13_w = readTrainedWeights(conv2d_13_w_path.c_str(), 0,512,512,3,3);
- std::string conv2d_13_b_path = dir_prefix + std::string("conv2d_13_b.bin");
- void* conv2d_13_b = readTrainedWeights(conv2d_13_b_path.c_str(), 0,1,512,1,1);
- 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(), 0,1,1,512,512);
- 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(), 0,1,512,1,1);
- 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(), 0,1,1,512,100);
- 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(), 0,1,100,1,1);
-
-
- startMemTracking();
-
- int test_input_size = 5000;
- int batch_size = 2500;
- int offset = 5000;
-
- int batch_count = test_input_size / batch_size;
- float final_accuracy = 0.0;
-
- for(int i = 0; i < batch_count; i++){
-
- int start = i * batch_size + offset;
- int end = (i + 1) * batch_size + offset;
-
- void* input = readInputBatch(input_path.c_str(),0,start,end,3,32,32);
-
- void* var_0 = tensorConvolution(input, conv2d_1_w, 1, 1, 1, 1, 1, 0);
- void* var_1 = tensorAdd(var_0, conv2d_1_b);
- void* var_2 = tensorRelu(var_1);
- void* var_4 = tensorConvolution(var_2, conv2d_2_w, 1, 1, 1, 1, 1, 0);
- void* var_5 = tensorAdd(var_4, conv2d_2_b);
- void* var_6 = tensorRelu(var_5);
- void* var_7 = tensorPooling(var_6,0,2,2,0,0,2,2);
- void* var_8 = tensorConvolution(var_7, conv2d_3_w, 1, 1, 1, 1, 1, 0);
- void* var_9 = tensorAdd(var_8, conv2d_3_b);
- void* var_10 = tensorRelu(var_9);
- void* var_12 = tensorConvolution(var_10, conv2d_4_w, 1, 1, 1, 1, 1, 0);
- void* var_13 = tensorAdd(var_12, conv2d_4_b);
- void* var_14 = tensorRelu(var_13);
- void* var_15 = tensorPooling(var_14,0,2,2,0,0,2,2);
- void* var_16 = tensorConvolution(var_15, conv2d_5_w, 1, 1, 1, 1, 1, 0);
- void* var_17 = tensorAdd(var_16, conv2d_5_b);
- void* var_18 = tensorRelu(var_17);
- void* var_20 = tensorConvolution(var_18, conv2d_6_w, 1, 1, 1, 1, 1, 0);
- void* var_21 = tensorAdd(var_20, conv2d_6_b);
- void* var_22 = tensorRelu(var_21);
- void* var_24 = tensorConvolution(var_22, conv2d_7_w, 1, 1, 1, 1, 1, 0);
- void* var_25 = tensorAdd(var_24, conv2d_7_b);
- void* var_26 = tensorRelu(var_25);
- void* var_27 = tensorPooling(var_26,0,2,2,0,0,2,2);
- void* var_28 = tensorConvolution(var_27, conv2d_8_w, 1, 1, 1, 1, 1, 0);
- void* var_29 = tensorAdd(var_28, conv2d_8_b);
- void* var_30 = tensorRelu(var_29);
- void* var_32 = tensorConvolution(var_30, conv2d_9_w, 1, 1, 1, 1, 1, 0);
- void* var_33 = tensorAdd(var_32, conv2d_9_b);
- void* var_34 = tensorRelu(var_33);
- void* var_36 = tensorConvolution(var_34, conv2d_10_w, 1, 1, 1, 1, 1, 0);
- void* var_37 = tensorAdd(var_36, conv2d_10_b);
- void* var_38 = tensorRelu(var_37);
- void* var_39 = tensorPooling(var_38,0,2,2,0,0,2,2);
- void* var_40 = tensorConvolution(var_39, conv2d_11_w, 1, 1, 1, 1, 1, 0);
- void* var_41 = tensorAdd(var_40, conv2d_11_b);
- void* var_42 = tensorRelu(var_41);
- void* var_44 = tensorConvolution(var_42, conv2d_12_w, 1, 1, 1, 1, 1, 0);
- void* var_45 = tensorAdd(var_44, conv2d_12_b);
- void* var_46 = tensorRelu(var_45);
- void* var_48 = tensorConvolution(var_46, conv2d_13_w, 1, 1, 1, 1, 1, 0);
- void* var_49 = tensorAdd(var_48, conv2d_13_b);
- void* var_50 = tensorRelu(var_49);
- void* var_51 = tensorPooling(var_50,0,2,2,0,0,2,2);
- void* var_54 = tensorGemmGPU(var_51, dense_1_w);
- void* var_55 = tensorAdd(var_54, dense_1_b);
- void* var_56 = tensorRelu(var_55);
- void* var_58 = tensorGemmGPU(var_56, dense_2_w);
- void* var_59 = tensorAdd(var_58, dense_2_b);
- void* var_60 = tensorSoftmax(var_59);
-
- uint8_t* labels = readLabelsBatch(labels_path.c_str(),start,end);
-
- //float accuracy = computeAccuracy2(labels, batch_size, var_60, 100);
- float accuracy = computeTop5Accuracy(labels, batch_size, var_60, 100);
- final_accuracy += accuracy;
- freeBatchMemory();
-
- }
-
- final_accuracy = final_accuracy / batch_count;
- dumpFinalAccuracy(final_accuracy);
-
- llvm_hpvm_cleanupTensorRt();
-
- return 0;
-}
diff --git a/hpvm/projects/hpvm-tensor-rt/scripts/test_dnns.py b/hpvm/projects/hpvm-tensor-rt/scripts/test_dnns.py
index 8d13a292372d81d491aedf21341c0e51859be723..11c3584a41e272527bc8141d9e9a9ed2d22ab51b 100644
--- a/hpvm/projects/hpvm-tensor-rt/scripts/test_dnns.py
+++ b/hpvm/projects/hpvm-tensor-rt/scripts/test_dnns.py
@@ -28,9 +28,8 @@ def createBaselineConfig(f_path, base_flag, num_layers):
if __name__ == "__main__":
- FP32_binary_paths = ["alexnet_cifar10", "alexnet2_cifar10", "resnet18_cifar10", "vgg16_cifar10", "vgg16_cifar100", "lenet_mnist", "mobilenet", "mobilenet_shallow"]
- FP16_binary_paths = ["alexnet_half", "alexnet2_half", "resnet18_half", "vgg16_cifar10_half", "vgg16_cifar100_half", "lenet_half", "mobilenet_half", "mobilenet_shallow_half"]
- PROMISE_binary_paths = ["alexnet_promise", "alexnet2_promise", "resnet18_promise", "vgg16_cifar10_promise", "vgg16_cifar100_promise", "mobilenet_promise", "mobilenet_shallow_promise"]
+ FP32_binary_paths = ["alexnet_cifar10_fp32", "alexnet2_cifar10_fp32", "resnet18_cifar10_fp32", "vgg16_cifar10_fp32", "vgg16_cifar100_fp32", "lenet_mnist_fp32", "mobilenet_cifar10_fp32"]
+ FP16_binary_paths = ["alexnet_cifar10_fp16", "alexnet2_cifar10_fp16", "resnet18_cifar10_fp16", "vgg16_cifar10_fp16", "vgg16_cifar100_fp16", "lenet_mnist_fp16", "mobilenet_cifar10_fp16"]
fp32_results = {}
for binary_path in FP32_binary_paths:
@@ -46,15 +45,5 @@ if __name__ == "__main__":
fp16_results[binary_path] = accuracy
- createBaselineConfig("promise_flags", 11, 1000)
- promise_results = {}
- for binary_path in PROMISE_binary_paths:
- subprocess.call("./" + binary_path)
- accuracy = readAccuracy("final_accuracy")
- promise_results[binary_path] = accuracy
-
-
printResults(fp32_results)
printResults(fp16_results)
- printResults(promise_results)
-