From 9cc4c25e481e96760eda7ee553220f2492122303 Mon Sep 17 00:00:00 2001
From: Sungeun Kim <sungeun.kim@intel.com>
Date: Thu, 9 Nov 2023 20:46:50 +0900
Subject: [PATCH] [GPU] print datashape of input for benchmark_app (#20943)

* print datashape of input for benchmark_app
---
 .../intel_gpu/runtime/debug_configuration.hpp |  1 +
 src/plugins/intel_gpu/src/graph/network.cpp   | 22 +++++++++++++++++++
 .../src/runtime/debug_configuration.cpp       |  3 +++
 3 files changed, 26 insertions(+)

diff --git a/src/plugins/intel_gpu/include/intel_gpu/runtime/debug_configuration.hpp b/src/plugins/intel_gpu/include/intel_gpu/runtime/debug_configuration.hpp
index 0e6269641e0..0a5fb7513ee 100644
--- a/src/plugins/intel_gpu/include/intel_gpu/runtime/debug_configuration.hpp
+++ b/src/plugins/intel_gpu/include/intel_gpu/runtime/debug_configuration.hpp
@@ -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
diff --git a/src/plugins/intel_gpu/src/graph/network.cpp b/src/plugins/intel_gpu/src/graph/network.cpp
index 2aaca0c73b2..cd5d0124cd1 100644
--- a/src/plugins/intel_gpu/src/graph/network.cpp
+++ b/src/plugins/intel_gpu/src/graph/network.cpp
@@ -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) {
diff --git a/src/plugins/intel_gpu/src/runtime/debug_configuration.cpp b/src/plugins/intel_gpu/src/runtime/debug_configuration.cpp
index eb7324cb7b5..55f166c4880 100644
--- a/src/plugins/intel_gpu/src/runtime/debug_configuration.cpp
+++ b/src/plugins/intel_gpu/src/runtime/debug_configuration.cpp
@@ -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);