From 187249d59cdb5c6733983adc19fc80b95fd3bcde Mon Sep 17 00:00:00 2001
From: WolframRhodium <WolframRhodium@users.noreply.github.com>
Date: Sat, 20 Apr 2024 11:21:42 +0800
Subject: [PATCH] vsort/vs_onnxruntime.cpp: disable ort cuda stream sync

---
 vsort/vs_onnxruntime.cpp | 25 ++++++++++++++++++++++---
 1 file changed, 22 insertions(+), 3 deletions(-)

diff --git a/vsort/vs_onnxruntime.cpp b/vsort/vs_onnxruntime.cpp
index ecd5725..c775c06 100644
--- a/vsort/vs_onnxruntime.cpp
+++ b/vsort/vs_onnxruntime.cpp
@@ -25,6 +25,7 @@ using namespace std::chrono_literals;
 #define NOMINMAX
 
 #include <onnxruntime_c_api.h>
+#include <onnxruntime_run_options_config_keys.h>
 
 #ifdef ENABLE_CUDA
 #include <cuda_runtime.h>
@@ -571,9 +572,23 @@ static const VSFrameRef *VS_CC vsOrtGetFrame(
             return nullptr;
         };
 
+        OrtRunOptions * run_options {};
+
 #ifdef ENABLE_CUDA
         if (d->backend == Backend::CUDA) {
             checkCUDAError(cudaSetDevice(d->device_id));
+
+#if ORT_API_VERSION >= 16
+            checkError(ortapi->CreateRunOptions(&run_options));
+            if (run_options == nullptr) {
+                return set_error("create run_options failed");
+            }
+            checkError(ortapi->AddRunConfigEntry(
+                run_options,
+                kOrtRunOptionsConfigDisableSynchronizeExecutionProviders,
+                "1"
+            ));
+#endif // ORT_API_VERSION >= 16
         }
 #endif // ENABLE_CUDA
 
@@ -650,17 +665,17 @@ static const VSFrameRef *VS_CC vsOrtGetFrame(
                     // note that this applies only to stream capture from the ort library
                     // this fails when another plugin also uses global-mode stream capture
                     std::lock_guard _ { capture_lock };
-                    checkError(ortapi->RunWithBinding(resource.session, nullptr, resource.binding));
+                    checkError(ortapi->RunWithBinding(resource.session, run_options, resource.binding));
 
                     // onnxruntime replays the graph itself in CUDAExecutionProvider::OnRunEnd
                 } else
 #endif // ENABLE_CUDA
                 if (d->backend == Backend::CPU || d->backend == Backend::CUDA) {
-                    checkError(ortapi->RunWithBinding(resource.session, nullptr, resource.binding));
+                    checkError(ortapi->RunWithBinding(resource.session, run_options, resource.binding));
                 } else {
                     checkError(ortapi->Run(
                         resource.session,
-                        nullptr,
+                        run_options,
                         &resource.input_name,
                         &resource.input_tensor,
                         1,
@@ -741,6 +756,10 @@ static const VSFrameRef *VS_CC vsOrtGetFrame(
             y = std::min(y + step_h, src_height - src_tile_h);
         }
 
+        if (run_options) {
+            ortapi->ReleaseRunOptions(run_options);
+        }
+
         d->release(ticket);
 
         for (const auto & frame : src_frames) {