[webgpu] Use pushErrorScope()/popErrorScope() once for an inference run

onnxruntime/core/providers/webgpu/compute_context.cc

@@ -13,29 +13,5 @@ ComputeContext::ComputeContext(OpKernelContext& kernel_context)
       kernel_context_{kernel_context} {
 }
 
-void ComputeContext::PushErrorScope() {
-  if (webgpu_context_.ValidationMode() >= ValidationMode::Basic) {
-    webgpu_context_.Device().PushErrorScope(wgpu::ErrorFilter::Validation);
-  }
-}
-
-Status ComputeContext::PopErrorScope() {
-  Status status{};
-
-  if (webgpu_context_.ValidationMode() >= ValidationMode::Basic) {
-    ORT_RETURN_IF_ERROR(webgpu_context_.Wait(
-        webgpu_context_.Device().PopErrorScope(
-            wgpu::CallbackMode::WaitAnyOnly, [](wgpu::PopErrorScopeStatus pop_status, wgpu::ErrorType error_type, char const* message, Status* status) {
-              ORT_ENFORCE(pop_status == wgpu::PopErrorScopeStatus::Success, "Instance dropped.");
-              if (error_type == wgpu::ErrorType::NoError) {
-                return;
-              }
-              *status = ORT_MAKE_STATUS(ONNXRUNTIME, FAIL, "WebGPU validation failed. ", message);
-            },
-            &status)));
-  }
-  return status;
-}
-
 }  // namespace webgpu
 }  // namespace onnxruntime

onnxruntime/core/providers/webgpu/compute_context.h

@@ -112,20 +112,6 @@ class ComputeContext {
     return webgpu_context_.Run(*this, program);
   }
 
-  //
-  // Push error scope.
-  //
-  // This is useful only when "skip_validation" is not set.
-  //
-  void PushErrorScope();
-
-  //
-  // Pop error scope.
-  //
-  // This is useful only when "skip_validation" is not set.
-  //
-  Status PopErrorScope();
-
  protected:
   WebGpuContext& webgpu_context_;
   OpKernelContext& kernel_context_;

onnxruntime/core/providers/webgpu/webgpu_execution_provider.cc

@@ -832,6 +832,10 @@ std::unique_ptr<profiling::EpProfiler> WebGpuExecutionProvider::GetProfiler() {
 }
 
 Status WebGpuExecutionProvider::OnRunStart(const onnxruntime::RunOptions& /*run_options*/) {
+  if (context_.ValidationMode() >= ValidationMode::Basic) {
+    context_.Device().PushErrorScope(wgpu::ErrorFilter::Validation);
+  }
+
   if (profiler_->Enabled()) {
     context_.StartProfiling();
   }
@@ -858,7 +862,20 @@ Status WebGpuExecutionProvider::OnRunEnd(bool /* sync_stream */, const onnxrunti
     context_.CollectProfilingData(profiler_->Events());
   }
 
-  return Status::OK();
+  Status status{};
+  if (context_.ValidationMode() >= ValidationMode::Basic) {
+    ORT_RETURN_IF_ERROR(context_.Wait(
+        context_.Device().PopErrorScope(
+            wgpu::CallbackMode::WaitAnyOnly, [](wgpu::PopErrorScopeStatus pop_status, wgpu::ErrorType error_type, char const* message, Status* status) {
+              ORT_ENFORCE(pop_status == wgpu::PopErrorScopeStatus::Success, "Instance dropped.");
+              if (error_type == wgpu::ErrorType::NoError) {
+                return;
+              }
+              *status = ORT_MAKE_STATUS(ONNXRUNTIME, FAIL, "WebGPU validation failed. ", message);
+            },
+            &status)));
+  }
+  return status;
 }
 
 bool WebGpuExecutionProvider::IsGraphCaptureEnabled() const {

onnxruntime/core/providers/webgpu/webgpu_kernel.h

@@ -23,9 +23,7 @@ class WebGpuKernel : public OpKernel {
   Status Compute(OpKernelContext* p_op_kernel_context) const override {
     ComputeContext context{*p_op_kernel_context};
 
-    context.PushErrorScope();
     Status s = ComputeInternal(context);
-    ORT_RETURN_IF_ERROR(context.PopErrorScope());
 
     return s;
   }