diff --git a/llvm/projects/hpvm-tensor-rt/dnn_sources/include/utils.h b/llvm/projects/hpvm-tensor-rt/dnn_sources/include/utils.h
index c086602506e8ed521c7deb99f7121f9d8159190e..8976f7dc8fa5df24536b051a5d436da6555161cf 100644
--- a/llvm/projects/hpvm-tensor-rt/dnn_sources/include/utils.h
+++ b/llvm/projects/hpvm-tensor-rt/dnn_sources/include/utils.h
@@ -407,7 +407,24 @@ uint8_t* readLabels(const char* labels_file, int num_labels){
fclose(file);
- // printf("--labels bytes_read = %lu \n", bytes_read);
+ return labels;
+}
+
+
+
+uint32_t* readLabels2(const char* labels_file, int num_labels){
+
+ uint32_t* labels = (uint32_t*) malloc(sizeof(uint32_t) * num_labels);
+ FILE* file = fopen(labels_file, "rb");
+ if(file == NULL){
+ printf("Data file %s is not found. Aborting...\n", labels_file);
+ abort();
+ }
+
+ size_t bytes_read = fread(labels, 1, sizeof(uint32_t) * num_labels, file);
+
+ fclose(file);
+
return labels;
}
@@ -436,6 +453,29 @@ uint8_t* readLabelsBatch(const char* labels_file, int start, int end){
}
+uint32_t* readLabelsBatch2(const char* labels_file, int start, int end){
+
+ int num_labels = end - start;
+ int file_header_size = sizeof(uint32_t) * start;
+
+ uint32_t* labels = (uint32_t*) malloc(sizeof(uint32_t) * num_labels);
+ FILE* file = fopen(labels_file, "rb");
+ if(file == NULL){
+ printf("Data file %s is not found. Aborting...\n", labels_file);
+ abort();
+ }
+
+ fseek(file, file_header_size, SEEK_SET); // Skipping the file header
+
+ size_t bytes_read = fread(labels, 1, sizeof(uint32_t) * num_labels, file);
+
+
+ fclose(file);
+
+ return labels;
+}
+
+
void computeAccuracy(const char* labels_file, int num_labels, void* result_ptr){
@@ -478,7 +518,8 @@ void computeAccuracy(const char* labels_file, int num_labels, void* result_ptr){
-float computeAccuracy2(uint8_t* labels, int num_labels, void* result_ptr, unsigned num_classes = 10){
+float computeAccuracy2(uint8_t* labels, int num_labels,
+ void* result_ptr, unsigned num_classes = 10){
unsigned num_zeros = 0;
@@ -494,7 +535,8 @@ float computeAccuracy2(uint8_t* labels, int num_labels, void* result_ptr, unsign
for(int i = 0; i < num_labels; i++){
int chosen = 0;
- for (int id = 1; id < num_classes; ++id){
+ //for (int id = 1; id < num_classes; ++id){
+ for (int id = 1; id < channels; ++id){
if (data[i * channels + chosen] < data[i * channels + id]) chosen = id;
}
@@ -538,7 +580,8 @@ bool descendFloatComp(ClassProb obj1, ClassProb obj2){
}
-float computeTop5Accuracy(uint8_t* labels, int num_labels, void* result_ptr, unsigned num_classes = 10){
+float computeTop5Accuracy(uint8_t* labels, int num_labels,
+ void* result_ptr, unsigned num_classes = 10){
struct Tensor* result = (struct Tensor*) result_ptr;
diff --git a/llvm/projects/hpvm-tensor-rt/tensor_runtime/include/hpvm-rt-controller.h b/llvm/projects/hpvm-tensor-rt/tensor_runtime/include/hpvm-rt-controller.h
index 7558a131a700e73af4624c12a4a6b6ab99a5f8aa..7b3ba18cd85c3b3e56bab6ecd45b1c165febb028 100644
--- a/llvm/projects/hpvm-tensor-rt/tensor_runtime/include/hpvm-rt-controller.h
+++ b/llvm/projects/hpvm-tensor-rt/tensor_runtime/include/hpvm-rt-controller.h
@@ -49,6 +49,7 @@ class RuntimeController {
void init(const char *Cstr, const char *Qstr) {
readQuantizationFile(Qstr);
+ printf("Going to read Conf file \n");
readConfigurationFile(Cstr);
}
@@ -322,7 +323,12 @@ void llvm_hpvm_clearRuntimeController() {
}
+void llvm_hpvm_invokeRtControl(void* result, uint8_t* labels){
+}
+
+void llvm_hpvm_invokeRtControl2(void* result, uint32_t* labels){
+}
#endif
diff --git a/llvm/projects/hpvm-tensor-rt/tensor_runtime/include/rt-controller-api.h b/llvm/projects/hpvm-tensor-rt/tensor_runtime/include/rt-controller-api.h
index 17441fc756be513915b4125ded958493d2c8bf46..903c9674bdcd679d2252f4ce6469f4979cf7dab8 100644
--- a/llvm/projects/hpvm-tensor-rt/tensor_runtime/include/rt-controller-api.h
+++ b/llvm/projects/hpvm-tensor-rt/tensor_runtime/include/rt-controller-api.h
@@ -5,6 +5,8 @@ extern "C"{
void llvm_hpvm_initializeRuntimeController(const char *, const char *);
void llvm_hpvm_clearRuntimeController();
+ void llvm_hpvm_invokeRtControl(void* result, uint8_t* labels);
+ void llvm_hpvm_invokeRtControl2(void* result, uint32_t* labels);
}
diff --git a/llvm/test/VISC/DNN_Benchmarks/benchmarks/alexnet/src/alexnet_loop.cpp b/llvm/test/VISC/DNN_Benchmarks/benchmarks/alexnet/src/alexnet_loop.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..7b0fb69dd853dbe678c1dee535d0ba0fdbdebcaa
--- /dev/null
+++ b/llvm/test/VISC/DNN_Benchmarks/benchmarks/alexnet/src/alexnet_loop.cpp
@@ -0,0 +1,475 @@
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <sys/stat.h>
+#include <cstring>
+#include <visc.h>
+#include <tensorTypes.h>
+#include <tensorUtils.h>
+
+void var_0_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(2, t1, t2, 0);
+
+ void *r = __visc__tensor_convolution(t1, t2, 5, 5, 1, 1);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_1_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(2, t1, t2, 0);
+
+ void *r = __visc__tensor_add(t1, t2);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_2_node(void* t1, size_t bytes_t1) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(1, t1, 0);
+
+ void* r = __visc__tensor_tanh(t1);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_3_node(void* t1, size_t bytes_t1) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(1, t1, 0);
+
+ void* r = __visc__tensor_pool_max(t1, 2, 2, 0, 0, 2, 2);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_4_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(2, t1, t2, 0);
+
+ void *r = __visc__tensor_convolution(t1, t2, 2, 2, 1, 1);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_5_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(2, t1, t2, 0);
+
+ void *r = __visc__tensor_add(t1, t2);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_6_node(void* t1, size_t bytes_t1) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(1, t1, 0);
+
+ void* r = __visc__tensor_tanh(t1);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_7_node(void* t1, size_t bytes_t1) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(1, t1, 0);
+
+ void* r = __visc__tensor_pool_max(t1, 2, 2, 0, 0, 2, 2);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_8_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(2, t1, t2, 0);
+
+ void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_9_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(2, t1, t2, 0);
+
+ void *r = __visc__tensor_add(t1, t2);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_10_node(void* t1, size_t bytes_t1) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(1, t1, 0);
+
+ void* r = __visc__tensor_tanh(t1);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_11_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(2, t1, t2, 0);
+
+ void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_12_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(2, t1, t2, 0);
+
+ void *r = __visc__tensor_add(t1, t2);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_13_node(void* t1, size_t bytes_t1) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(1, t1, 0);
+
+ void* r = __visc__tensor_tanh(t1);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_14_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(2, t1, t2, 0);
+
+ void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_15_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(2, t1, t2, 0);
+
+ void *r = __visc__tensor_add(t1, t2);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_16_node(void* t1, size_t bytes_t1) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(1, t1, 0);
+
+ void* r = __visc__tensor_tanh(t1);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_17_node(void* t1, size_t bytes_t1) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(1, t1, 0);
+
+ void* r = __visc__tensor_pool_max(t1, 2, 2, 0, 0, 2, 2);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_18_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(2, t1, t2, 0);
+
+ void *r = __visc__tensor_mul(t1, t2);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_19_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(2, t1, t2, 0);
+
+ void *r = __visc__tensor_add(t1, t2);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void var_20_node(void* t1, size_t bytes_t1) {
+ __visc__hint(visc::CUDNN_TARGET);
+ __visc__attributes(1, t1, 0);
+
+ void* r = __visc__tensor_softmax(t1);
+ __visc__return(2, r, (size_t) 0);
+}
+
+void root(void* input, size_t input_bytes,
+ void* conv2d_1_w, size_t conv2d_1_w_bytes,
+ void* conv2d_1_b, size_t conv2d_1_b_bytes,
+ void* conv2d_2_w, size_t conv2d_2_w_bytes,
+ void* conv2d_2_b, size_t conv2d_2_b_bytes,
+ void* conv2d_3_w, size_t conv2d_3_w_bytes,
+ void* conv2d_3_b, size_t conv2d_3_b_bytes,
+ void* conv2d_4_w, size_t conv2d_4_w_bytes,
+ void* conv2d_4_b, size_t conv2d_4_b_bytes,
+ void* conv2d_5_w, size_t conv2d_5_w_bytes,
+ void* conv2d_5_b, size_t conv2d_5_b_bytes,
+ void* dense_1_w, size_t dense_1_w_bytes,
+ void* dense_1_b, size_t dense_1_b_bytes){
+
+
+ __visc__hint(visc::CPU_TARGET);
+ __visc__attributes(13, input, conv2d_1_w, conv2d_1_b, conv2d_2_w, conv2d_2_b, conv2d_3_w, conv2d_3_b, conv2d_4_w, conv2d_4_b, conv2d_5_w, conv2d_5_b, dense_1_w, dense_1_b, 0);
+
+
+ void* var_0 = __visc__createNodeND(0, var_0_node);
+
+ __visc__bindIn(var_0, 0, 0, 0);
+ __visc__bindIn(var_0, 1, 1, 0);
+ __visc__bindIn(var_0, 2, 2, 0);
+ __visc__bindIn(var_0, 3, 3, 0);
+
+ void* var_1 = __visc__createNodeND(0, var_1_node);
+
+ __visc__edge(var_0, var_1, 1, 0, 0, 0);
+ __visc__edge(var_0, var_1, 1, 1, 1, 0);
+ __visc__bindIn(var_1, 4, 2, 0);
+ __visc__bindIn(var_1, 5, 3, 0);
+
+ void* var_2 = __visc__createNodeND(0, var_2_node);
+
+ __visc__edge(var_1, var_2, 1, 0, 0, 0);
+ __visc__edge(var_1, var_2, 1, 1, 1, 0);
+
+ void* var_3 = __visc__createNodeND(0, var_3_node);
+
+ __visc__edge(var_2, var_3, 1, 0, 0, 0);
+ __visc__edge(var_2, var_3, 1, 1, 1, 0);
+
+ void* var_4 = __visc__createNodeND(0, var_4_node);
+
+ __visc__edge(var_3, var_4, 1, 0, 0, 0);
+ __visc__edge(var_3, var_4, 1, 1, 1, 0);
+ __visc__bindIn(var_4, 6, 2, 0);
+ __visc__bindIn(var_4, 7, 3, 0);
+
+ void* var_5 = __visc__createNodeND(0, var_5_node);
+
+ __visc__edge(var_4, var_5, 1, 0, 0, 0);
+ __visc__edge(var_4, var_5, 1, 1, 1, 0);
+ __visc__bindIn(var_5, 8, 2, 0);
+ __visc__bindIn(var_5, 9, 3, 0);
+
+ void* var_6 = __visc__createNodeND(0, var_6_node);
+
+ __visc__edge(var_5, var_6, 1, 0, 0, 0);
+ __visc__edge(var_5, var_6, 1, 1, 1, 0);
+
+ void* var_7 = __visc__createNodeND(0, var_7_node);
+
+ __visc__edge(var_6, var_7, 1, 0, 0, 0);
+ __visc__edge(var_6, var_7, 1, 1, 1, 0);
+
+ void* var_8 = __visc__createNodeND(0, var_8_node);
+
+ __visc__edge(var_7, var_8, 1, 0, 0, 0);
+ __visc__edge(var_7, var_8, 1, 1, 1, 0);
+ __visc__bindIn(var_8, 10, 2, 0);
+ __visc__bindIn(var_8, 11, 3, 0);
+
+ void* var_9 = __visc__createNodeND(0, var_9_node);
+
+ __visc__edge(var_8, var_9, 1, 0, 0, 0);
+ __visc__edge(var_8, var_9, 1, 1, 1, 0);
+ __visc__bindIn(var_9, 12, 2, 0);
+ __visc__bindIn(var_9, 13, 3, 0);
+
+ void* var_10 = __visc__createNodeND(0, var_10_node);
+
+ __visc__edge(var_9, var_10, 1, 0, 0, 0);
+ __visc__edge(var_9, var_10, 1, 1, 1, 0);
+
+ void* var_11 = __visc__createNodeND(0, var_11_node);
+
+ __visc__edge(var_10, var_11, 1, 0, 0, 0);
+ __visc__edge(var_10, var_11, 1, 1, 1, 0);
+ __visc__bindIn(var_11, 14, 2, 0);
+ __visc__bindIn(var_11, 15, 3, 0);
+
+ void* var_12 = __visc__createNodeND(0, var_12_node);
+
+ __visc__edge(var_11, var_12, 1, 0, 0, 0);
+ __visc__edge(var_11, var_12, 1, 1, 1, 0);
+ __visc__bindIn(var_12, 16, 2, 0);
+ __visc__bindIn(var_12, 17, 3, 0);
+
+ void* var_13 = __visc__createNodeND(0, var_13_node);
+
+ __visc__edge(var_12, var_13, 1, 0, 0, 0);
+ __visc__edge(var_12, var_13, 1, 1, 1, 0);
+
+ void* var_14 = __visc__createNodeND(0, var_14_node);
+
+ __visc__edge(var_13, var_14, 1, 0, 0, 0);
+ __visc__edge(var_13, var_14, 1, 1, 1, 0);
+ __visc__bindIn(var_14, 18, 2, 0);
+ __visc__bindIn(var_14, 19, 3, 0);
+
+ void* var_15 = __visc__createNodeND(0, var_15_node);
+
+ __visc__edge(var_14, var_15, 1, 0, 0, 0);
+ __visc__edge(var_14, var_15, 1, 1, 1, 0);
+ __visc__bindIn(var_15, 20, 2, 0);
+ __visc__bindIn(var_15, 21, 3, 0);
+
+ void* var_16 = __visc__createNodeND(0, var_16_node);
+
+ __visc__edge(var_15, var_16, 1, 0, 0, 0);
+ __visc__edge(var_15, var_16, 1, 1, 1, 0);
+
+ void* var_17 = __visc__createNodeND(0, var_17_node);
+
+ __visc__edge(var_16, var_17, 1, 0, 0, 0);
+ __visc__edge(var_16, var_17, 1, 1, 1, 0);
+
+ void* var_18 = __visc__createNodeND(0, var_18_node);
+
+ __visc__edge(var_17, var_18, 1, 0, 0, 0);
+ __visc__edge(var_17, var_18, 1, 1, 1, 0);
+ __visc__bindIn(var_18, 22, 2, 0);
+ __visc__bindIn(var_18, 23, 3, 0);
+
+ void* var_19 = __visc__createNodeND(0, var_19_node);
+
+ __visc__edge(var_18, var_19, 1, 0, 0, 0);
+ __visc__edge(var_18, var_19, 1, 1, 1, 0);
+ __visc__bindIn(var_19, 24, 2, 0);
+ __visc__bindIn(var_19, 25, 3, 0);
+
+ void* var_20 = __visc__createNodeND(0, var_20_node);
+
+ __visc__edge(var_19, var_20, 1, 0, 0, 0);
+ __visc__edge(var_19, var_20, 1, 1, 1, 0);
+
+ __visc__bindOut(var_20, 0, 0, 0);
+ __visc__bindOut(var_20, 1, 1, 0);
+
+}
+
+struct ret_t {
+ void* tensor;
+ size_t bytes;
+};
+
+typedef struct __attribute__((__packed__)) {
+ void* input;
+ size_t input_bytes;
+ void* conv2d_1_w;
+ size_t conv2d_1_w_bytes;
+ void* conv2d_1_b;
+ size_t conv2d_1_b_bytes;
+ void* conv2d_2_w;
+ size_t conv2d_2_w_bytes;
+ void* conv2d_2_b;
+ size_t conv2d_2_b_bytes;
+ void* conv2d_3_w;
+ size_t conv2d_3_w_bytes;
+ void* conv2d_3_b;
+ size_t conv2d_3_b_bytes;
+ void* conv2d_4_w;
+ size_t conv2d_4_w_bytes;
+ void* conv2d_4_b;
+ size_t conv2d_4_b_bytes;
+ void* conv2d_5_w;
+ size_t conv2d_5_w_bytes;
+ void* conv2d_5_b;
+ size_t conv2d_5_b_bytes;
+ void* dense_1_w;
+ size_t dense_1_w_bytes;
+ void* dense_1_b;
+ size_t dense_1_b_bytes;
+
+ struct ret_t r;
+}
+RootIn;
+
+int main(){
+
+ std::string dir_prefix = std::string("../../../../../../projects/hpvm-tensor-rt/model_params/alexnet_cifar10_test/");
+
+
+ std::string labels_path = dir_prefix + std::string("labels.bin");
+ //uint8_t* labels = readLabels(labels_path.c_str(),10000);
+ std::string conv2d_1_w_path = dir_prefix + std::string("conv2d_1_w.bin");
+ void* conv2d_1_w = readTrainedWeights(conv2d_1_w_path.c_str(), 0,64,3,11,11);
+ 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(), 0,192,64,5,5);
+ std::string conv2d_2_b_path = dir_prefix + std::string("conv2d_2_b.bin");
+ void* conv2d_2_b = readTrainedWeights(conv2d_2_b_path.c_str(), 0,1,192,1,1);
+ 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(), 0,384,192,3,3);
+ 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(), 0,1,384,1,1);
+ 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(), 0,256,384,3,3);
+ 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(), 0,1,256,1,1);
+ 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(), 0,256,256,3,3);
+ 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 dense_1_w_path = dir_prefix + std::string("dense_1_w.bin");
+ void* dense_1_w = readTrainedWeights(dense_1_w_path.c_str(), 0,1,1,4096,10);
+ 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);
+
+
+
+ __visc__init();
+ RootIn* args = static_cast<RootIn*>(malloc(sizeof(RootIn)));
+
+ args->conv2d_1_w = conv2d_1_w;
+ args->conv2d_1_w_bytes = 0;
+ args->conv2d_1_b = conv2d_1_b;
+ args->conv2d_1_b_bytes = 0;
+ args->conv2d_2_w = conv2d_2_w;
+ args->conv2d_2_w_bytes = 0;
+ args->conv2d_2_b = conv2d_2_b;
+ args->conv2d_2_b_bytes = 0;
+ args->conv2d_3_w = conv2d_3_w;
+ args->conv2d_3_w_bytes = 0;
+ args->conv2d_3_b = conv2d_3_b;
+ args->conv2d_3_b_bytes = 0;
+ args->conv2d_4_w = conv2d_4_w;
+ args->conv2d_4_w_bytes = 0;
+ args->conv2d_4_b = conv2d_4_b;
+ args->conv2d_4_b_bytes = 0;
+ args->conv2d_5_w = conv2d_5_w;
+ args->conv2d_5_w_bytes = 0;
+ args->conv2d_5_b = conv2d_5_b;
+ args->conv2d_5_b_bytes = 0;
+ args->dense_1_w = dense_1_w;
+ args->dense_1_w_bytes = 0;
+ args->dense_1_b = dense_1_b;
+ args->dense_1_b_bytes = 0;
+
+ int batch_size = 500;
+ int test_input_size = 10000;
+ int batch_count = test_input_size / batch_size;
+
+ std::string input_path = dir_prefix + std::string("input.bin");
+ void* input = create4DTensor(0,nchw,batch_size,3,32,32);
+
+ startMemTracking();
+ for (int i = 0; i < batch_count; i++){
+
+ int start = i * batch_size;
+ int end = (i + 1) * batch_size;
+
+ copyInputBatch(input_path.c_str(),start,end,3,32,32, input);
+
+ args->input = input;
+ args->input_bytes = 0;
+
+ //void* input = readInputBatch(input_path.c_str(),0,start,end,3,32,32);
+
+ void* dfg = __visc__launch(0, root, (void*) args);
+
+ __visc__wait(dfg);
+
+ void *result = static_cast<RootIn*>(args)->input;
+ hpvm_request_tensor(result, 0);
+
+
+ uint8_t* labels = readLabelsBatch(labels_path.c_str(),start,end);
+
+ computeAccuracy2(labels, batch_size, result);
+
+ llvm_hpvm_invokeRtControl(result, labels);
+
+ freeBatchMemory();
+ }
+
+
+ __visc__cleanup();
+
+
+ return 0;
+
+}
diff --git a/llvm/test/VISC/DNN_Benchmarks/common/include/tensorUtils.h b/llvm/test/VISC/DNN_Benchmarks/common/include/tensorUtils.h
index c086602506e8ed521c7deb99f7121f9d8159190e..8976f7dc8fa5df24536b051a5d436da6555161cf 100644
--- a/llvm/test/VISC/DNN_Benchmarks/common/include/tensorUtils.h
+++ b/llvm/test/VISC/DNN_Benchmarks/common/include/tensorUtils.h
@@ -407,7 +407,24 @@ uint8_t* readLabels(const char* labels_file, int num_labels){
fclose(file);
- // printf("--labels bytes_read = %lu \n", bytes_read);
+ return labels;
+}
+
+
+
+uint32_t* readLabels2(const char* labels_file, int num_labels){
+
+ uint32_t* labels = (uint32_t*) malloc(sizeof(uint32_t) * num_labels);
+ FILE* file = fopen(labels_file, "rb");
+ if(file == NULL){
+ printf("Data file %s is not found. Aborting...\n", labels_file);
+ abort();
+ }
+
+ size_t bytes_read = fread(labels, 1, sizeof(uint32_t) * num_labels, file);
+
+ fclose(file);
+
return labels;
}
@@ -436,6 +453,29 @@ uint8_t* readLabelsBatch(const char* labels_file, int start, int end){
}
+uint32_t* readLabelsBatch2(const char* labels_file, int start, int end){
+
+ int num_labels = end - start;
+ int file_header_size = sizeof(uint32_t) * start;
+
+ uint32_t* labels = (uint32_t*) malloc(sizeof(uint32_t) * num_labels);
+ FILE* file = fopen(labels_file, "rb");
+ if(file == NULL){
+ printf("Data file %s is not found. Aborting...\n", labels_file);
+ abort();
+ }
+
+ fseek(file, file_header_size, SEEK_SET); // Skipping the file header
+
+ size_t bytes_read = fread(labels, 1, sizeof(uint32_t) * num_labels, file);
+
+
+ fclose(file);
+
+ return labels;
+}
+
+
void computeAccuracy(const char* labels_file, int num_labels, void* result_ptr){
@@ -478,7 +518,8 @@ void computeAccuracy(const char* labels_file, int num_labels, void* result_ptr){
-float computeAccuracy2(uint8_t* labels, int num_labels, void* result_ptr, unsigned num_classes = 10){
+float computeAccuracy2(uint8_t* labels, int num_labels,
+ void* result_ptr, unsigned num_classes = 10){
unsigned num_zeros = 0;
@@ -494,7 +535,8 @@ float computeAccuracy2(uint8_t* labels, int num_labels, void* result_ptr, unsign
for(int i = 0; i < num_labels; i++){
int chosen = 0;
- for (int id = 1; id < num_classes; ++id){
+ //for (int id = 1; id < num_classes; ++id){
+ for (int id = 1; id < channels; ++id){
if (data[i * channels + chosen] < data[i * channels + id]) chosen = id;
}
@@ -538,7 +580,8 @@ bool descendFloatComp(ClassProb obj1, ClassProb obj2){
}
-float computeTop5Accuracy(uint8_t* labels, int num_labels, void* result_ptr, unsigned num_classes = 10){
+float computeTop5Accuracy(uint8_t* labels, int num_labels,
+ void* result_ptr, unsigned num_classes = 10){
struct Tensor* result = (struct Tensor*) result_ptr;