[GPU] Enable qdq_stripping path for GPU (#694)

* update the statement so that we run CreateModelWithStrippedQDQNodes on GPU

* ensure the capability checks are also updated

* update the comment

Signed-off-by: bfilipek <[email protected]>

---------

Signed-off-by: bfilipek <[email protected]>

Author: intbf

onnxruntime/core/providers/openvino/backend_manager.cc

@@ -380,8 +380,9 @@ BackendManager::GetModelProtoFromFusedNode(const onnxruntime::Node& fused_node,
 #endif
 
   const auto& onnx_model_path_name = subgraph.ModelPath();
-  // QDQ stripping enabled only for the NPU
-  if (session_context_.device_type.find("NPU") != std::string::npos &&
+  // QDQ stripping enabled only for the NPU and experimentally on the GPU
+  if ((session_context_.device_type.find("NPU") != std::string::npos ||
+      session_context_.device_type.find("GPU") != std::string::npos) &&
       (enable_ovep_qdq_optimizer || session_context_.so_share_ep_contexts)) {
     std::unique_ptr<onnxruntime::Model> model;
     Status status = CreateModelWithStrippedQDQNodes(subgraph, logger, session_context_.so_share_ep_contexts, enable_ovep_qdq_optimizer, model, shared_context_.shared_weights);

onnxruntime/core/providers/openvino/ov_versions/capability.cc

@@ -34,7 +34,7 @@ GetCapability::GetCapability(const EPCtxHandler& ep_ctx_handler,
                                                                 graph_viewer_(graph_viewer_param),
                                                                 device_type_(std::move(device_type_param)) {
   bool npu_qdq_optimizer_enabled = false;
-  if (device_type_.find("NPU") != std::string::npos) {
+  if (device_type_.find("NPU") != std::string::npos || device_type_.find("GPU") != std::string::npos) {
     device_type_ = "CPU";
     if (enable_qdq_optimizer) npu_qdq_optimizer_enabled = true;
   }