Merge branch 'master' of https://github.com/openvinotoolkit/openvino …

…into as/weights_bank_dq_update

src/plugins/intel_gpu/include/intel_gpu/graph/network.hpp

@@ -4,17 +4,15 @@
 
 #pragma once
 
-#include "openvino/runtime/threading/cpu_streams_executor.hpp"
+#include "openvino/runtime/threading/istreams_executor.hpp"
 
 #include "intel_gpu/graph/topology.hpp"
 #include "intel_gpu/graph/program.hpp"
 #include "intel_gpu/graph/serialization/binary_buffer.hpp"
-#include "intel_gpu/runtime/compounds.hpp"
 #include "intel_gpu/runtime/memory.hpp"
 #include "intel_gpu/runtime/engine.hpp"
 #include "intel_gpu/runtime/event.hpp"
 #include "intel_gpu/runtime/stream.hpp"
-#include "intel_gpu/runtime/lru_cache.hpp"
 #include "intel_gpu/runtime/shape_predictor.hpp"
 #include "intel_gpu/plugin/variable_state.hpp"
 
@@ -211,7 +209,7 @@ struct network {
     bool is_dynamic() const { return _is_dynamic; }
     size_t get_weights_cache_capacity() const { return _weights_cache_capacity; }
 
-    memory_pool& get_memory_pool() {
+    memory_pool& get_memory_pool() const {
         return *_memory_pool;
     }
 
@@ -284,7 +282,9 @@ struct network {
     void dump_memory_pool(std::string dump_path, int64_t curr_iter);
 
 #ifdef GPU_DEBUG_CONFIG
-    int64_t iteration = 0;
+    mutable int64_t iteration = 0;
+    friend class NetworkDebugHelper;
+    friend class NodeDebugHelper;
 #endif
 };
 }  // namespace cldnn

src/plugins/intel_gpu/src/graph/debug_helper.hpp

@@ -0,0 +1,69 @@
+// Copyright (C) 2018-2024 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#pragma once
+
+#include "intel_gpu/graph/network.hpp"
+#include "intel_gpu/graph/program.hpp"
+#include "intel_gpu/runtime/stream.hpp"
+#include "intel_gpu/runtime/debug_configuration.hpp"
+#include "primitive_inst.h"
+
+namespace cldnn {
+
+#ifdef GPU_DEBUG_CONFIG
+
+class NodeDebugHelper {
+public:
+    NodeDebugHelper(const primitive_inst& inst);
+    ~NodeDebugHelper();
+
+private:
+    std::string get_iteration_prefix() {
+        if (m_iter < 0)
+            return std::string("");
+        return std::to_string(m_iter) + "_";
+    }
+
+    std::string get_file_prefix() {
+        auto prog_id = ((m_program != nullptr) ? m_program->get_id() : 0);
+        auto net_id = m_network.get_id();
+
+        return "program" + std::to_string(prog_id) + "_network" + std::to_string(net_id) + "_" + get_iteration_prefix() + m_inst.id();
+    }
+
+
+    const primitive_inst& m_inst;
+    stream& m_stream;
+    const network& m_network;
+    const program* m_program;
+    const size_t m_iter;
+
+    const debug_configuration* debug_config = cldnn ::debug_configuration ::get_instance();
+};
+
+class NetworkDebugHelper {
+public:
+    NetworkDebugHelper(const network& net);
+    ~NetworkDebugHelper();
+
+private:
+    void dump_memory_pool(std::string dump_path, int64_t curr_iter) const;
+    const network& m_network;
+    const size_t m_iter;
+
+    const debug_configuration* debug_config = cldnn ::debug_configuration ::get_instance();
+};
+
+#define NETWORK_DEBUG(net) NetworkDebugHelper __network_debug_helper(net)
+#define NODE_DEBUG(inst) NodeDebugHelper __node_debug_helper(inst)
+
+#else
+
+#define NETWORK_DEBUG(...)
+#define NODE_DEBUG(...)
+
+#endif  // GPU_DEBUG_CONFIG
+
+}  // namespace cldnn

src/plugins/intel_gpu/src/graph/include/program_dump_graph.h

@@ -14,6 +14,6 @@ std::string get_dir_path(const ExecutionConfig& config);
 void dump_graph_optimized(std::ofstream&, const program&);
 void dump_graph_processing_order(std::ofstream&, const program&);
 void dump_graph_init(std::ofstream&, const program&,
-                     std::function<std::shared_ptr<primitive_inst>(const primitive_id&)> get_primitive_inst = nullptr);
+                     std::function<std::shared_ptr<const primitive_inst>(const primitive_id&)> get_primitive_inst = nullptr);
 void dump_graph_info(std::ofstream&, const program&);
 }  // namespace cldnn

src/plugins/intel_gpu/src/graph/layout_optimizer.cpp

@@ -1089,11 +1089,13 @@ format layout_optimizer::get_expected_format(quantize_node const& node) {
     auto use_onednn_impls = _optimization_attributes.use_onednn_impls;
 
     if (use_onednn_impls) {
-        auto& user = node.get_users().front();
-        if (user != nullptr && user->get_preferred_input_fmt(user->get_dependency_index(node)) != format::any) {
-            expected = user->get_preferred_input_fmt(user->get_dependency_index(node));
-        } else {
-            expected = format::any;
+        expected = format::any;
+        auto& users = node.get_users();
+        if (users.size() != 0) {
+            auto& user = users.front();
+            if (user != nullptr && user->get_preferred_input_fmt(user->get_dependency_index(node)) != format::any) {
+                expected = user->get_preferred_input_fmt(user->get_dependency_index(node));
+            }
         }
     } else if (only_gemm_users(node)) {
         // TODO: Gemm is not supporting fsv layouts