[GPU] Disable using port indices for query mode

Lyamin-Roman · Mar 12, 2024 · bbda257 · bbda257
1 parent 9086a07
commit bbda257
Show file tree

Hide file tree

Showing 3 changed files with 26 additions and 9 deletions.
diff --git a/src/plugins/intel_gpu/include/intel_gpu/plugin/program_builder.hpp b/src/plugins/intel_gpu/include/intel_gpu/plugin/program_builder.hpp
@@ -142,6 +142,7 @@ class ProgramBuilder final {
     bool use_new_shape_infer() const { return allow_new_shape_infer; }
     bool requires_new_shape_infer(const std::shared_ptr<ov::Node>& op) const;
     bool is_inner_program() const { return m_is_inner_program; }
+    bool is_query_mode() { return queryMode; }
 
     std::shared_ptr<ov::threading::IStreamsExecutor> get_task_executor() const { return m_task_executor; }
     std::shared_ptr<cldnn::ICompilationContext> get_compilation_context() const { return m_compilation_context; }

diff --git a/src/plugins/intel_gpu/src/plugin/ops/parameter.cpp b/src/plugins/intel_gpu/src/plugin/ops/parameter.cpp
@@ -34,8 +34,13 @@ static void CreateParameterOp(ProgramBuilder& p, const std::shared_ptr<ov::op::v
     // look at the expected color format of this input
     auto input_name = layer_type_name_ID(op);
     cldnn::layout input_layout(input_pshape, element_type, input_format);
-    int64_t port_index = p.get_parameter_index(op);
-    OPENVINO_ASSERT(port_index != -1, "[GPU] Parameter port index for ", input_name, " not found");
+
+    bool query_mode = p.is_query_mode();
+    int64_t port_index = -1;
+    if (!query_mode) {
+        port_index = p.get_parameter_index(op);
+        OPENVINO_ASSERT(port_index != -1, "[GPU] Parameter port index for ", input_name, " not found");
+    }
 
     auto is_convert_color_type = [](const std::shared_ptr<ov::Node> &node) {
         return ov::is_type<ov::op::v8::NV12toRGB>(node) ||
@@ -85,7 +90,10 @@ static void CreateParameterOp(ProgramBuilder& p, const std::shared_ptr<ov::op::v
         size_t batch = input_pshape[0].get_length();
         input_layout.format = cldnn::format::nv12;
         input_layout.set_partial_shape({ 1, input_pshape[1], input_pshape[2], input_pshape[3] });
-        p.inputLayouts.insert({ port_index, input_layout });
+
+        if (!query_mode) {
+            p.inputLayouts.insert({ port_index, input_layout });
+        }
 
         std::string suffix = "";
         std::vector<cldnn::input_info> surfaces_inputs;
@@ -94,7 +102,10 @@ static void CreateParameterOp(ProgramBuilder& p, const std::shared_ptr<ov::op::v
                 suffix = "_" + std::to_string(i);
             std::string batched_name = input_name + suffix;
             p.add_primitive(*op, cldnn::input_layout(batched_name, input_layout));
-            p.inputPrimitiveIDs[port_index].emplace_back(batched_name);
+
+            if (!query_mode) {
+                p.inputPrimitiveIDs[port_index].emplace_back(batched_name);
+            }
 
             auto reorder_layout = input_layout;
             reorder_layout.format = cldnn::format::bfyx;
@@ -116,8 +127,11 @@ static void CreateParameterOp(ProgramBuilder& p, const std::shared_ptr<ov::op::v
         auto reorder_name = "reorder:" + input_name + ProgramBuilder::m_preProcessTag;
 
         p.add_primitive(*op, cldnn::input_layout(input_name, input_layout));
-        p.inputPrimitiveIDs[port_index] = { input_name };
-        p.inputLayouts.insert({ port_index, input_layout });
+
+        if (!query_mode) {
+            p.inputPrimitiveIDs[port_index] = { input_name };
+            p.inputLayouts.insert({ port_index, input_layout });
+        }
 
         if (connected_to_quantize(op)) {
             // Techically this reorder is not needed, but for some reason it impacts layout propagation logic

diff --git a/src/plugins/intel_gpu/src/plugin/ops/result.cpp b/src/plugins/intel_gpu/src/plugin/ops/result.cpp
@@ -38,9 +38,11 @@ static void CreateResultOp(ProgramBuilder& p, const std::shared_ptr<ov::op::v0::
                                             out_data_type);
     p.add_primitive(*op, reorder_primitive, { input_id, op->get_friendly_name() });
 
-    int64_t port_index = p.get_result_index(op);
-    OPENVINO_ASSERT(port_index != -1, "[GPU] Result port index for ", input_id, " not found");
-    p.prevPrimitiveIDs[port_index] = input_id;
+    if (!p.is_query_mode()) {
+        int64_t port_index = p.get_result_index(op);
+        OPENVINO_ASSERT(port_index != -1, "[GPU] Result port index for ", input_id, " not found");
+        p.prevPrimitiveIDs[port_index] = input_id;
+    }
 }
 
 REGISTER_FACTORY_IMPL(v0, Result);