Add Windows GPU support again (#702)

umireon · web-flow · commit 43d0bb09a943 · 2025-11-12T04:28:30.000+09:00
* Update DownloadOnnxruntime.cmake

* Add TensorRT execution provider support

This commit adds support for the ONNX Runtime TensorRT execution provider, including detection in CMake, new localization strings, and integration in background and enhance filters. It also updates session creation logic to handle TensorRT-specific options and cache management.

* Update DownloadOnnxruntime.cmake

* Update DownloadOnnxruntime.cmake

* Update CMakePresets.json

* Update background-filter.cpp

* Update ort-session-utils.cpp

* Update ort-session-utils.cpp
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -183,6 +183,17 @@ if(HAVE_ONNXRUNTIME_ROCM_EP)
   target_compile_definitions(${CMAKE_PROJECT_NAME} PRIVATE HAVE_ONNXRUNTIME_ROCM_EP)
 endif()
 
+check_library_exists(
+  onnxruntime::onnxruntime
+  OrtSessionOptionsAppendExecutionProvider_Tensorrt
+  onnxruntime_c_api.h
+  HAVE_ONNXRUNTIME_TENSORRT_EP
+)
+if(HAVE_ONNXRUNTIME_TENSORRT_EP)
+  message(STATUS "ONNX Runtime TensorRT Execution Provider found")
+  target_compile_definitions(${CMAKE_PROJECT_NAME} PRIVATE HAVE_ONNXRUNTIME_TENSORRT_EP)
+endif()
+
 target_sources(
   ${CMAKE_PROJECT_NAME}
   PRIVATE
diff --git a/CMakePresets.json b/CMakePresets.json
@@ -76,7 +76,7 @@
       "displayName": "Windows x64 CI build",
       "description": "Build for Windows x64 on CI",
       "cacheVariables": {
-        "CMAKE_COMPILE_WARNING_AS_ERROR": false
+        "CMAKE_COMPILE_WARNING_AS_ERROR": true
       }
     },
     {
diff --git a/cmake/DownloadOnnxruntime.cmake b/cmake/DownloadOnnxruntime.cmake
@@ -26,14 +26,14 @@ if(PLATFORM STREQUAL "macos")
 elseif(PLATFORM STREQUAL "windows")
   file(
     DOWNLOAD
-      https://github.com/microsoft/onnxruntime/releases/download/v1.23.2/onnxruntime-win-x64-1.23.2.zip
-      onnxruntime-win-x64-1.23.2.zip
-    EXPECTED_HASH SHA256=0b38df9af21834e41e73d602d90db5cb06dbd1ca618948b8f1d66d607ac9f3cd
+      https://github.com/microsoft/onnxruntime/releases/download/v1.23.2/onnxruntime-win-x64-gpu-1.23.2.zip
+      onnxruntime-win-x64-gpu-1.23.2.zip
+    EXPECTED_HASH SHA256=e77afdbbc2b8cb6da4e5a50d89841b48c44f3e47dce4fb87b15a2743786d0bb9
   )
   execute_process(
-    COMMAND ${CMAKE_COMMAND} -E tar xf onnxruntime-win-x64-1.23.2.zip
+    COMMAND ${CMAKE_COMMAND} -E tar xf onnxruntime-win-x64-gpu-1.23.2.zip
   )
-  file(RENAME onnxruntime-win-x64-1.23.2 onnxruntime)
+  file(RENAME onnxruntime-win-x64-gpu-1.23.2 onnxruntime)
 elseif(PLATFORM STREQUAL "linux")
   file(
     DOWNLOAD
diff --git a/data/locale/ar-EG.ini b/data/locale/ar-EG.ini
@@ -9,6 +9,7 @@ InferenceDevice="جهاز الإستدلال"
 CPU="وحدة المعالجة المركزية"
 GPUCUDA="الوحدة المركزية - CUDA"
 GPUROCM="GPU - ROCM"
+TENSORRT="TensorRT"
 CoreML="CoreML"
 SegmentationModel="نموذج التقسيم"
 SINet="SINet"
diff --git a/data/locale/bn-IN.ini b/data/locale/bn-IN.ini
@@ -9,6 +9,7 @@ InferenceDevice="নিখরচনা ডিভাইস"
 CPU="সিপিইউ"
 GPUCUDA="জিপিইউ - কুড়া"
 GPUROCM="GPU - ROCM"
+TENSORRT="TensorRT"
 CoreML="কোরএমএল"
 SegmentationModel="সেগমেন্টেশন মডেল"
 SINet="এসআইনেট"
diff --git a/data/locale/en-US.ini b/data/locale/en-US.ini
@@ -9,6 +9,7 @@ InferenceDevice="Inference device"
 CPU="CPU"
 GPUCUDA="GPU - CUDA"
 GPUROCM="GPU - ROCM"
+TENSORRT="TensorRT"
 CoreML="CoreML"
 SegmentationModel="Segmentation model"
 SINet="SINet"
diff --git a/data/locale/es-SP.ini b/data/locale/es-SP.ini
@@ -9,6 +9,7 @@ InferenceDevice="Dispositivo de inferencia"
 CPU="CPU"
 GPUCUDA="GPU - CUDA"
 GPUROCM="GPU - ROCM"
+TENSORRT="TensorRT"
 CoreML="CoreML"
 SegmentationModel="Modelo de segmentación"
 SINet="SINet"
diff --git a/data/locale/fr-FR.ini b/data/locale/fr-FR.ini
@@ -9,6 +9,7 @@ InferenceDevice="Dispositif d'inférence"
 CPU="CPU"
 GPUCUDA="GPU - CUDA"
 GPUROCM="GPU - ROCM"
+TENSORRT="TensorRT"
 CoreML="CoreML"
 SegmentationModel="Modèle de segmentation"
 SINet="SINet"
diff --git a/data/locale/hi-IN.ini b/data/locale/hi-IN.ini
@@ -9,6 +9,7 @@ InferenceDevice="संदर्भ डिवाइस"
 CPU="सीपीयू"
 GPUCUDA="जीपीयू - क्यूडा"
 GPUROCM="जीपीयू - ROCM"
+TENSORRT="TensorRT"
 CoreML="कोरएमएल"
 SegmentationModel="सेगमेंटेशन मॉडल"
 SINet="ऐसआईनेट"
diff --git a/data/locale/it-IT.ini b/data/locale/it-IT.ini
@@ -9,6 +9,7 @@ InferenceDevice="Dispositivo di inferenza"
 CPU="CPU"
 GPUCUDA="GPU - CUDA"
 GPUROCM="GPU - ROCM"
+TENSORRT="TensorRT"
 CoreML="CoreML"
 SegmentationModel="Modello di segmentazione".
 SINet="SINet"
diff --git a/data/locale/ja-JP.ini b/data/locale/ja-JP.ini
@@ -9,6 +9,7 @@ InferenceDevice="推論デバイス"
 CPU="CPU"
 GPUCUDA="GPU - CUDA"
 GPUROCM="GPU - ROCM"
+TENSORRT="TensorRT"
 CoreML="CoreML"
 SegmentationModel="セグメンテーションモデル"
 SINet="SINet"
diff --git a/data/locale/ko-KR.ini b/data/locale/ko-KR.ini
@@ -9,6 +9,7 @@ InferenceDevice="추론 장치"
 CPU="CPU"
 GPUCUDA="GPU - CUDA"
 GPUROCM="GPU - ROCM"
+TENSORRT="TensorRT"
 CoreML="CoreML"
 SegmentationModel="세분화 모델"
 SINet="SINet"
diff --git a/data/locale/pt-BR.ini b/data/locale/pt-BR.ini
@@ -9,6 +9,7 @@ InferenceDevice="Dispositivo de Inferência"
 CPU="CPU"
 GPUCUDA="GPU - CUDA"
 GPUROCM="GPU - ROCM"
+TENSORRT="TensorRT"
 CoreML="CoreML"
 SegmentationModel="Modelo de segmentação"
 SINet="SINet"
diff --git a/data/locale/ru-RU.ini b/data/locale/ru-RU.ini
@@ -9,6 +9,7 @@ InferenceDevice="Устройство вывода"
 CPU="ЦПУ"
 GPUCUDA="ГПУ - CUDA"
 GPUROCM="ГПУ - ROCM"
+TENSORRT="TensorRT"
 CoreML="CoreML"
 SegmentationModel="Модель сегментации"
 SINet="SINet"
diff --git a/data/locale/ta-IN.ini b/data/locale/ta-IN.ini
@@ -9,6 +9,7 @@ InferenceDevice="பரிமாற்ற சாதனம்"
 CPU="சிபியூ"
 GPUCUDA="GPU-CUDA"
 GPUROCM="GPU - ROCM"
+TENSORRT="TensorRT"
 CoreML="CoreML"
 SegmentationModel="பிரிவு மாதிரி"
 SINet="SINet"
diff --git a/data/locale/tr-TR.ini b/data/locale/tr-TR.ini
@@ -9,6 +9,7 @@ InferenceDevice="Tahmin Cihazı"
 CPU="CPU"
 GPUCUDA="GPU - CUDA"
 GPUROCM="GPU - ROCM"
+TENSORRT="TensorRT"
 CoreML="CoreML"
 SegmentationModel="Segmentasyon Modeli"
 SINet="SINet"
diff --git a/data/locale/zh-CN.ini b/data/locale/zh-CN.ini
@@ -9,6 +9,7 @@ InferenceDevice="推断设备"
 CPU="CPU"
 GPUCUDA="GPU-CUDA"
 GPUROCM="GPU - ROCM"
+TENSORRT="TensorRT"
 CoreML="CoreML"
 SegmentationModel="分割模型"
 SINet="SINet"
diff --git a/src/background-filter.cpp b/src/background-filter.cpp
@@ -158,10 +158,13 @@ obs_properties_t *background_filter_properties(void *data)
 
 	obs_property_list_add_string(p_use_gpu, obs_module_text("CPU"), USEGPU_CPU);
 #ifdef HAVE_ONNXRUNTIME_CUDA_CP
-	obs_property_list_add_string(p_use_gpu, obs_module_text("GPUCuda"), USEGPU_CUDA);
+	obs_property_list_add_string(p_use_gpu, obs_module_text("GPUCUDA"), USEGPU_CUDA);
 #endif
 #ifdef HAVE_ONNXRUNTIME_ROCM_EP
-	obs_property_list_add_string(p_use_gpu, obs_module_text("GPURocm"), USEGPU_ROCM);
+	obs_property_list_add_string(p_use_gpu, obs_module_text("GPUROCM"), USEGPU_ROCM);
+#endif
+#ifdef HAVE_ONNXRUNTIME_TENSORRT_EP
+	obs_property_list_add_string(p_use_gpu, obs_module_text("TENSORRT"), USEGPU_TENSORRT);
 #endif
 #if defined(__APPLE__)
 	obs_property_list_add_string(p_use_gpu, obs_module_text("CoreML"), USEGPU_COREML);
@@ -265,7 +268,7 @@ void background_filter_update(void *data, obs_data_t *settings)
 
 	tf->contourFilter = (float)obs_data_get_double(settings, "contour_filter");
 	tf->smoothContour = (float)obs_data_get_double(settings, "smooth_contour");
-	tf->maskExpansion = (float)obs_data_get_double(settings, "mask_expansion");
+	tf->maskExpansion = (int)obs_data_get_double(settings, "mask_expansion");
 	tf->feather = (float)obs_data_get_double(settings, "feather");
 	tf->maskEveryXFrames = (int)obs_data_get_int(settings, "mask_every_x_frames");
 	tf->maskEveryXFramesCount = (int)(0);
diff --git a/src/consts.h b/src/consts.h
@@ -16,6 +16,7 @@ const char *const MODEL_RMBG = "models/bria_rmbg_1_4_qint8.onnx";
 const char *const USEGPU_CPU = "cpu";
 const char *const USEGPU_CUDA = "cuda";
 const char *const USEGPU_ROCM = "rocm";
+const char *const USEGPU_TENSORRT = "tensorrt";
 const char *const USEGPU_COREML = "coreml";
 
 const char *const EFFECT_PATH = "effects/mask_alpha_filter.effect";
diff --git a/src/enhance-filter.cpp b/src/enhance-filter.cpp
@@ -59,6 +59,9 @@ obs_properties_t *enhance_filter_properties(void *data)
 #ifdef HAVE_ONNXRUNTIME_ROCM_EP
 	obs_property_list_add_string(p_use_gpu, obs_module_text("GPUROCM"), USEGPU_ROCM);
 #endif
+#ifdef HAVE_ONNXRUNTIME_TENSORRT_EP
+	obs_property_list_add_string(p_use_gpu, obs_module_text("TENSORRT"), USEGPU_TENSORRT);
+#endif
 #if defined(__APPLE__)
 	obs_property_list_add_string(p_use_gpu, obs_module_text("CoreML"), USEGPU_COREML);
 #endif
diff --git a/src/ort-utils/ort-session-utils.cpp b/src/ort-utils/ort-session-utils.cpp
@@ -65,6 +65,11 @@ int createOrtSession(filter_data *tf)
 			Ort::ThrowOnError(OrtSessionOptionsAppendExecutionProvider_ROCM(sessionOptions, 0));
 		}
 #endif
+#ifdef HAVE_ONNXRUNTIME_TENSORRT_EP
+		if (tf->useGPU == USEGPU_TENSORRT) {
+			Ort::ThrowOnError(OrtSessionOptionsAppendExecutionProvider_Tensorrt(sessionOptions, 0));
+		}
+#endif
 #if defined(__APPLE__)
 		if (tf->useGPU == USEGPU_COREML) {
 			uint32_t coreml_flags = 0;

Original file line number	Diff line number	Diff line change
`@@ -76,7 +76,7 @@`
`76`	`76`	`"displayName": "Windows x64 CI build",`
`77`	`77`	`"description": "Build for Windows x64 on CI",`
`78`	`78`	`"cacheVariables": {`
`79`		`- "CMAKE_COMPILE_WARNING_AS_ERROR": false`
	`79`	`+ "CMAKE_COMPILE_WARNING_AS_ERROR": true`
`80`	`80`	`}`
`81`	`81`	`},`
`82`	`82`	`{`