IntenseWebs/openvino
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[GPU] print datashape of input for benchmark_app (#20943)

Browse Source 
* print datashape of input for benchmark_app
This commit is contained in:
Sungeun Kim 2023-11-09 20:46:50 +09:00 committed by GitHub
parent 09a45bceae
commit 9cc4c25e48
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 26 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
src/plugins/intel_gpu/include/intel_gpu/runtime/debug_configuration.hpp
View File

@ -100,6 +100,7 @@ public:
int verbose_color; // Print verbose color
int list_layers; // Print list layers
int print_multi_kernel_perf; // Print execution time of each kernel in multi-kernel primitimive
int print_input_data_shapes; // Print the input data_shape for benchmark_app.
int disable_usm; // Disable usm usage
int disable_onednn; // Disable onednn for discrete GPU (no effect for integrated GPU)
int disable_onednn_opt_post_ops; // Disable onednn optimize post operators

22
src/plugins/intel_gpu/src/graph/network.cpp
View File

@ -1378,6 +1378,28 @@ void network::execute_impl(const std::vector<event::ptr>& events) {
}
}
// print '-data_shape' option for benchmark_app
GPU_DEBUG_IF(debug_config->print_input_data_shapes == 1) {
std::stringstream data_shape_str;
auto add_string = [&data_shape_str](std::string str) {
data_shape_str << ((data_shape_str.rdbuf()->in_avail() == 0) ? " -data_shape " : ",") << str;
};
for (auto& inst : _exec_order) {
auto name = inst->id();
auto pos = name.find(':');
auto type = name.substr(0, pos);
name.erase(0, pos + 1);
if (inst->is_input() && type == "parameter") {
add_string(name + inst->get_output_layout().get_partial_shape().to_string());
}
}
GPU_DEBUG_COUT << "[program:" << std::setw(2) << ((get_program() != nullptr) ? get_program()->get_id() : 0)
<< "|network:" << std::setw(2) << get_id() << "|iter:" << std::setw(4) << curr_iter << "] benchmark_app cmd: "
<< data_shape_str.str() << std::endl;
}
// Store events only in case of OOO queue or enabled Profiling
auto store_events = is_out_of_order_queue || _enable_profiling;
if (store_events) {

3
src/plugins/intel_gpu/src/runtime/debug_configuration.cpp
View File

@ -108,6 +108,7 @@ static void print_help_messages() {
message_list.emplace_back("OV_GPU_VerboseColor", "Print verbose color");
message_list.emplace_back("OV_GPU_ListLayers", "Print layers names");
message_list.emplace_back("OV_GPU_PrintMultiKernelPerf", "Print execution time of each kernel in multi-kernel primitimive");
message_list.emplace_back("OV_GPU_PrintInputDataShapes", "Print data_shapes of input layers for benchmark_app.");
message_list.emplace_back("OV_GPU_DisableUsm", "Disable usm usage");
message_list.emplace_back("OV_GPU_DisableOnednn", "Disable onednn for discrete GPU (no effect for integrated GPU)");
message_list.emplace_back("OV_GPU_DisableOnednnOptPostOps", "Disable onednn optimize post operators");
@ -173,6 +174,7 @@ debug_configuration::debug_configuration()
, verbose_color(0)
, list_layers(0)
, print_multi_kernel_perf(0)
, print_input_data_shapes(0)
, disable_usm(0)
, disable_onednn(0)
, disable_onednn_opt_post_ops(0)
@ -206,6 +208,7 @@ debug_configuration::debug_configuration()
get_gpu_debug_env_var("VerboseColor", verbose_color);
get_gpu_debug_env_var("ListLayers", list_layers);
get_gpu_debug_env_var("PrintMultiKernelPerf", print_multi_kernel_perf);
get_gpu_debug_env_var("PrintInputDataShapes", print_input_data_shapes);
get_gpu_debug_env_var("DisableUsm", disable_usm);
get_gpu_debug_env_var("DumpGraphs", dump_graphs);
get_gpu_debug_env_var("DumpSources", dump_sources);