Fix error in creation of gemm impl

(fused inputs should not be added to "input" of kernel arg)
yeonbok · Aug 9, 2022 · 88b1e16 · 88b1e16
1 parent 3d8b166
commit 88b1e16
Showing 1 changed file with 7 additions and 1 deletion.
diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/gemm.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/gemm.cpp
@@ -115,12 +115,18 @@ struct gemm_impl : typed_primitive_impl_ocl<gemm> {
         };
         const auto input_layouts = get_gemm_input_layouts(impl_param->input_layouts, impl_param->output_layout);
         const auto output_layout = get_gemm_output_layout(input_layouts, impl_param->output_layout);
+
+        auto first_fused_input_idx = input_layouts.size();
+        const auto fused_descs = impl_param->fused_desc;
+        if (fused_descs.size() > 0) {
+            first_fused_input_idx = fused_descs[0].dep_start_idx;
+        }
         auto gemm_params = get_default_params<kernel_selector::gemm_params>(*impl_param, 1);
         auto gemm_optional_params =
             get_default_optional_params<kernel_selector::gemm_optional_params>(arg.get_program());
 
         gemm_params.inputs.clear();
-        for (size_t i = 0; i < input_layouts.size(); i++) {
+        for (size_t i = 0; i < std::min(input_layouts.size(), first_fused_input_idx); i++) {
             gemm_params.inputs.push_back(convert_data_tensor(input_layouts[i]));
         }
         gemm_params.outputs[0] = convert_data_tensor(output_layout);