update

microsoft · Feb 24, 2024 · df7b525 · df7b525
1 parent 3b4de7a
commit df7b525
Showing 1 changed file with 0 additions and 11 deletions.
diff --git a/onnxruntime/core/providers/cuda/cuda_graph.cc b/onnxruntime/core/providers/cuda/cuda_graph.cc
@@ -22,14 +22,11 @@ void CUDAGraph::SetGraphAnnotation(GraphAnnotationOptional_t cuda_graph_annotati
 
 void CUDAGraph::CaptureBegin() {
   if (!cuda_graph_annotation_id_.has_value()) {
-    //std::cout << "CaptureBegin: cuda_graph_annotation_id is empty" << std::endl;
     ORT_ENFORCE(!has_graph_exec_,
                 "This cuda graph has already captured a graph. "
                 "Create a new instance to capture a new graph.");
   } else {
-    //std::cout << "CaptureBegin: cuda_graph_annotation_id is " << *cuda_graph_annotation_id_ << std::endl;
     if (!IsGraphCaptureAllowedOnRun()) {
-      //std::cout << "CaptureBegin: Graph capture is not allowed on this run" << std::endl;
       return;
     }
   }
@@ -39,18 +36,15 @@ void CUDAGraph::CaptureBegin() {
   // will support multiple threads. For multiple threads with multiple graphs
   // and streams, `cudaStreamCaptureModeGlobal` needs to be changed to
   // `cudaStreamCaptureModeThreadLocal`
-  //std::cout << "REAL cuda graph capture begins" << std::endl;
   CUDA_CALL_THROW(cudaStreamBeginCapture(stream_, cudaStreamCaptureModeGlobal));
 }
 
 void CUDAGraph::CaptureEnd() {
-  //std::cout << "CUDAGraph::CaptureEnd()" << std::endl;
   if (!IsGraphCaptureAllowedOnRun()) {
     return;
   }
 
   if (cuda_graph_annotation_id_.has_value()) {
-    //std::cout << "CaptureEnd: cuda_graph_annotation_id is " << *cuda_graph_annotation_id_ << std::endl;
     CUDA_CALL_THROW(cudaStreamEndCapture(stream_, &additional_graph_));
     if (additional_graph_ == NULL) {
       ORT_THROW("CUDAGraph::CaptureEnd: additional_graph_ is NULL");
@@ -59,7 +53,6 @@ void CUDAGraph::CaptureEnd() {
     cudaGraphExec_t graph_exec = NULL;
 
     has_additional_graph_ = true;
-    //std::cout << "REAL cuda graph capture ends" << std::endl;
     CUDA_CALL_THROW(cudaGraphInstantiate(&graph_exec, additional_graph_, NULL, NULL, 0));
     CUDA_CALL_THROW(cudaGraphDestroy(additional_graph_));
     has_additional_graph_ = false;
@@ -70,14 +63,12 @@ void CUDAGraph::CaptureEnd() {
     return;
   }
 
-  //std::cout << "CaptureEnd: cuda_graph_annotation_id is empty" << std::endl;
   CUDA_CALL_THROW(cudaStreamEndCapture(stream_, &graph_));
   if (graph_ == NULL) {
     ORT_THROW("CUDAGraph::CaptureEnd: graph_ is NULL");
   }
 
   has_graph_ = true;
-  //std::cout << "REAL cuda graph capture ends" << std::endl;
   CUDA_CALL_THROW(cudaGraphInstantiate(&graph_exec_, graph_, NULL, NULL, 0));
   has_graph_exec_ = true;
   CUDA_CALL_THROW(cudaGraphDestroy(graph_));
@@ -91,7 +82,6 @@ Status CUDAGraph::Replay(GraphAnnotationOptional_t cuda_graph_annotation_id) {
   // Although this function is not thread safe, the lock is not needed here because
   // CUDA EP maintains a separate cuda graph per thread
   if (cuda_graph_annotation_id_.has_value()) {
-    //std::cout << "Replaying CUDA graph on stream " << stream_ << " with cuda_graph_annotation_id " << *cuda_graph_annotation_id << std::endl;
     LOGS_DEFAULT(INFO) << "Replaying CUDA graph on stream " << stream_ << " with cuda_graph_annotation_id " << *cuda_graph_annotation_id;
     auto it = graph_exec_map_.find(*cuda_graph_annotation_id);
     if (it == graph_exec_map_.end()) {
@@ -101,7 +91,6 @@ Status CUDAGraph::Replay(GraphAnnotationOptional_t cuda_graph_annotation_id) {
     }
     CUDA_RETURN_IF_ERROR(cudaGraphLaunch(it->second, stream_));
   } else {
-    //std::cout << "Replaying CUDA graph on stream " << stream_ << std::endl;
     LOGS_DEFAULT(INFO) << "Replaying CUDA graph on stream " << stream_;
     CUDA_RETURN_IF_ERROR(cudaGraphLaunch(graph_exec_, stream_));
   }