Add software fallback for YUV444 using libswscale

README.md

@@ -52,10 +52,10 @@ Hosting for Moonlight's Debian and L4T package repositories is graciously provid
 ### Linux/Unix Build Requirements
 * Qt 6 is recommended, but Qt 5.9 or later is also supported (replace `qmake6` with `qmake` when using Qt 5).
 * GCC or Clang
-* FFmpeg 4.0 or later
+* FFmpeg 5.0 or later
 * Install the required packages:
   * Debian/Ubuntu:
-    * Base Requirements: `libegl1-mesa-dev libgl1-mesa-dev libopus-dev libsdl2-dev libsdl2-ttf-dev libssl-dev libavcodec-dev libavformat-dev libva-dev libvdpau-dev libxkbcommon-dev wayland-protocols libdrm-dev`
+    * Base Requirements: `libegl1-mesa-dev libgl1-mesa-dev libopus-dev libsdl2-dev libsdl2-ttf-dev libssl-dev libavcodec-dev libavformat-dev libswscale-dev libva-dev libvdpau-dev libxkbcommon-dev wayland-protocols libdrm-dev`
     * Qt 6 (Recommended): `qt6-base-dev qt6-declarative-dev libqt6svg6-dev qml6-module-qtquick-controls qml6-module-qtquick-templates qml6-module-qtquick-layouts qml6-module-qtqml-workerscript qml6-module-qtquick-window qml6-module-qtquick`
     * Qt 5: `qtbase5-dev qt5-qmake qtdeclarative5-dev qtquickcontrols2-5-dev qml-module-qtquick-controls2 qml-module-qtquick-layouts qml-module-qtquick-window2 qml-module-qtquick2 qtwayland5`
   * RedHat/Fedora (RPM Fusion repo required):

app/app.pro

@@ -73,7 +73,7 @@ unix:if(!macx|disable-prebuilts) {
 
     !disable-ffmpeg {
         packagesExist(libavcodec) {
-            PKGCONFIG += libavcodec libavutil
+            PKGCONFIG += libavcodec libavutil libswscale
             CONFIG += ffmpeg
 
             !disable-libva {
@@ -148,7 +148,7 @@ unix:if(!macx|disable-prebuilts) {
     }
 }
 win32 {
-    LIBS += -llibssl -llibcrypto -lSDL2 -lSDL2_ttf -lavcodec -lavutil -lopus -ldxgi -ld3d11
+    LIBS += -llibssl -llibcrypto -lSDL2 -lSDL2_ttf -lavcodec -lavutil -lswscale -lopus -ldxgi -ld3d11
     CONFIG += ffmpeg
     contains(QT_ARCH, x86_64) {
         LIBS += -llibplacebo
@@ -160,7 +160,7 @@ win32:!winrt {
 }
 macx {
     !disable-prebuilts {
-        LIBS += -lssl -lcrypto -lavcodec.61 -lavutil.59 -lopus -framework SDL2 -framework SDL2_ttf
+        LIBS += -lssl -lcrypto -lavcodec.61 -lavutil.59 -lswscale.8 -lopus -framework SDL2 -framework SDL2_ttf
         CONFIG += discord-rpc
     }
 

app/streaming/video/ffmpeg-renderers/sdlvid.cpp

@@ -7,11 +7,19 @@
 
 #include <SDL_syswm.h>
 
+extern "C" {
+#include <libavutil/pixdesc.h>
+#include <libavutil/opt.h>
+}
+
 SdlRenderer::SdlRenderer()
     : m_VideoFormat(0),
       m_Renderer(nullptr),
       m_Texture(nullptr),
       m_ColorSpace(-1),
+      m_NeedsYuvToRgbConversion(false),
+      m_SwsContext(nullptr),
+      m_RgbFrame(av_frame_alloc()),
       m_SwFrameMapper(this)
 {
     SDL_zero(m_OverlayTextures);
@@ -35,6 +43,9 @@ SdlRenderer::~SdlRenderer()
         }
     }
 
+    av_frame_free(&m_RgbFrame);
+    sws_freeContext(m_SwsContext);
+
     if (m_Texture != nullptr) {
         SDL_DestroyTexture(m_Texture);
     }
@@ -82,15 +93,27 @@ bool SdlRenderer::isRenderThreadSupported()
 
 bool SdlRenderer::isPixelFormatSupported(int videoFormat, AVPixelFormat pixelFormat)
 {
-    if (videoFormat & VIDEO_FORMAT_MASK_10BIT) {
+    if (videoFormat & (VIDEO_FORMAT_MASK_10BIT | VIDEO_FORMAT_MASK_YUV444)) {
-        // SDL2 doesn't support 10-bit pixel formats
+        // SDL2 can't natively handle textures with these formats, but we can perform
+        // conversion on the CPU using swscale then upload them as an RGB texture.
+        const AVPixFmtDescriptor* formatDesc = av_pix_fmt_desc_get(pixelFormat);
+        if (!formatDesc) {
+            SDL_assert(formatDesc);
             return false;
         }
-    else if (videoFormat & VIDEO_FORMAT_MASK_YUV444) {
+
-        // SDL2 doesn't support YUV444 pixel formats
+        const int expectedPixelDepth = (videoFormat & VIDEO_FORMAT_MASK_10BIT) ? 10 : 8;
-        return false;
+        const int expectedLog2ChromaW = (videoFormat & VIDEO_FORMAT_MASK_YUV444) ? 0 : 1;
+        const int expectedLog2ChromaH = (videoFormat & VIDEO_FORMAT_MASK_YUV444) ? 0 : 1;
+
+        return formatDesc->comp[0].depth == expectedPixelDepth &&
+               formatDesc->log2_chroma_w == expectedLog2ChromaW &&
+               formatDesc->log2_chroma_h == expectedLog2ChromaH;
     }
     else {
+        // The formats listed below are natively supported by SDL, so it can handle
+        // YUV to RGB conversion on the GPU using pixel shaders.
+        //
         // Remember to keep this in sync with SdlRenderer::renderFrame()!
         switch (pixelFormat) {
         case AV_PIX_FMT_YUV420P:
@@ -112,8 +135,8 @@ bool SdlRenderer::initialize(PDECODER_PARAMETERS params)
     m_VideoFormat = params->videoFormat;
     m_SwFrameMapper.setVideoFormat(m_VideoFormat);
 
-    if (params->videoFormat & (VIDEO_FORMAT_MASK_10BIT | VIDEO_FORMAT_MASK_YUV444)) {
+    if (params->videoFormat & VIDEO_FORMAT_MASK_10BIT) {
-        // SDL doesn't support rendering YUV444 or 10-bit textures yet
+        // SDL doesn't support rendering HDR yet
         return false;
     }
 
@@ -233,6 +256,11 @@ void SdlRenderer::renderOverlay(Overlay::OverlayType type)
     }
 }
 
+void SdlRenderer::ffNoopFree(void*, uint8_t*)
+{
+    // Nothing
+}
+
 void SdlRenderer::renderFrame(AVFrame* frame)
 {
     int err;
@@ -286,6 +314,7 @@ ReadbackRetry:
         Uint32 sdlFormat;
 
         // Remember to keep this in sync with SdlRenderer::isPixelFormatSupported()!
+        m_NeedsYuvToRgbConversion = false;
         switch (frame->format)
         {
         case AV_PIX_FMT_YUV420P:
@@ -300,10 +329,57 @@ ReadbackRetry:
             sdlFormat = SDL_PIXELFORMAT_NV21;
             break;
         default:
-            SDL_assert(false);
+            SDL_LogWarn(SDL_LOG_CATEGORY_APPLICATION,
+                        "Performing color conversion on CPU due to lack of SDL support for format: %u",
+                        frame->format);
+            sdlFormat = SDL_PIXELFORMAT_XRGB8888;
+            m_NeedsYuvToRgbConversion = true;
+            break;
+        }
+
+        if (m_NeedsYuvToRgbConversion) {
+            m_RgbFrame->width = frame->width;
+            m_RgbFrame->height = frame->height;
+            m_RgbFrame->format = AV_PIX_FMT_BGR0;
+
+            sws_freeContext(m_SwsContext);
+            m_SwsContext = sws_alloc_context();
+            if (!m_SwsContext) {
+                SDL_LogError(SDL_LOG_CATEGORY_APPLICATION,
+                             "sws_alloc_context() failed");
                 goto Exit;
             }
 
+            AVDictionary *options { nullptr };
+            av_dict_set_int(&options, "srcw", frame->width, 0);
+            av_dict_set_int(&options, "srch", frame->height, 0);
+            av_dict_set_int(&options, "src_format", frame->format, 0);
+            av_dict_set_int(&options, "dstw", m_RgbFrame->width, 0);
+            av_dict_set_int(&options, "dsth", m_RgbFrame->height, 0);
+            av_dict_set_int(&options, "dst_format", m_RgbFrame->format, 0);
+            av_dict_set_int(&options, "threads", std::min(SDL_GetCPUCount(), 4), 0); // Up to 4 threads
+
+            err = av_opt_set_dict(m_SwsContext, &options);
+            av_dict_free(&options);
+            if (err < 0) {
+                char string[AV_ERROR_MAX_STRING_SIZE];
+                SDL_LogError(SDL_LOG_CATEGORY_APPLICATION,
+                             "av_opt_set_dict() failed: %s",
+                             av_make_error_string(string, sizeof(string), err));
+                goto Exit;
+            }
+
+            err = sws_init_context(m_SwsContext, nullptr, nullptr);
+            if (err < 0) {
+                char string[AV_ERROR_MAX_STRING_SIZE];
+                SDL_LogError(SDL_LOG_CATEGORY_APPLICATION,
+                             "sws_init_context() failed: %s",
+                             av_make_error_string(string, sizeof(string), err));
+                goto Exit;
+            }
+        }
+        else {
+            // SDL will perform YUV conversion on the GPU
             switch (colorspace)
             {
             case COLORSPACE_REC_709:
@@ -322,6 +398,7 @@ ReadbackRetry:
             default:
                 break;
             }
+        }
 
         m_Texture = SDL_CreateTexture(m_Renderer,
                                       sdlFormat,
@@ -371,7 +448,7 @@ ReadbackRetry:
                              frame->data[2],
                              frame->linesize[2]);
     }
-    else {
+    else if (!m_NeedsYuvToRgbConversion) {
 #if SDL_VERSION_ATLEAST(2, 0, 15)
         // SDL_UpdateNVTexture is not supported on all renderer backends,
         // (notably not DX9), so we must have a fallback in case it's not
@@ -430,6 +507,37 @@ ReadbackRetry:
             SDL_UnlockTexture(m_Texture);
         }
     }
+    else {
+        // We have a pixel format that SDL doesn't natively support, so we must use
+        // swscale to convert the YUV frame into an RGB frame to upload to the GPU.
+        uint8_t* pixels;
+        int texturePitch;
+
+        err = SDL_LockTexture(m_Texture, nullptr, (void**)&pixels, &texturePitch);
+        if (err < 0) {
+            SDL_LogError(SDL_LOG_CATEGORY_APPLICATION,
+                         "SDL_LockTexture() failed: %s",
+                         SDL_GetError());
+            goto Exit;
+        }
+
+        // Create a buffer to wrap our locked texture buffer
+        m_RgbFrame->buf[0] = av_buffer_create(pixels, m_RgbFrame->height * texturePitch, ffNoopFree, nullptr, 0);
+        m_RgbFrame->data[0] = pixels;
+        m_RgbFrame->linesize[0] = texturePitch;
+
+        // Perform multi-threaded color conversion into the locked texture buffer
+        err = sws_scale_frame(m_SwsContext, m_RgbFrame, frame);
+        av_buffer_unref(&m_RgbFrame->buf[0]);
+        SDL_UnlockTexture(m_Texture);
+        if (err < 0) {
+            char string[AV_ERROR_MAX_STRING_SIZE];
+            SDL_LogError(SDL_LOG_CATEGORY_APPLICATION,
+                         "sws_scale_frame() failed: %s",
+                         av_make_error_string(string, AV_ERROR_MAX_STRING_SIZE, err));
+            goto Exit;
+        }
+    }
 
     SDL_RenderClear(m_Renderer);
 

app/streaming/video/ffmpeg-renderers/sdlvid.h

@@ -7,6 +7,10 @@
 #include "cuda.h"
 #endif
 
+extern "C" {
+#include <libswscale/swscale.h>
+}
+
 class SdlRenderer : public IFFmpegRenderer {
 public:
     SdlRenderer();
@@ -23,6 +27,8 @@ public:
 private:
     void renderOverlay(Overlay::OverlayType type);
 
+    static void ffNoopFree(void *opaque, uint8_t *data);
+
     int m_VideoFormat;
     SDL_Renderer* m_Renderer;
     SDL_Texture* m_Texture;
@@ -30,6 +36,11 @@ private:
     SDL_Texture* m_OverlayTextures[Overlay::OverlayMax];
     SDL_Rect m_OverlayRects[Overlay::OverlayMax];
 
+    // Used for CPU conversion of YUV to RGB if needed
+    bool m_NeedsYuvToRgbConversion;
+    SwsContext* m_SwsContext;
+    AVFrame* m_RgbFrame;
+
     SwFrameMapper m_SwFrameMapper;
 
 #ifdef HAVE_CUDA

appveyor.yml

@@ -13,7 +13,7 @@ environment:
       BUILD_TARGET: steamlink
     - APPVEYOR_BUILD_WORKER_IMAGE: Ubuntu2004
       BUILD_TARGET: linux
-      FFMPEG_CONFIGURE_ARGS: --enable-pic --disable-static --enable-shared --disable-all --enable-avcodec --enable-avformat --enable-decoder=h264 --enable-decoder=hevc --enable-decoder=av1 --enable-hwaccel=h264_vaapi --enable-hwaccel=hevc_vaapi --enable-hwaccel=av1_vaapi --enable-hwaccel=h264_vdpau --enable-hwaccel=hevc_vdpau --enable-hwaccel=av1_vdpau --enable-libdrm --enable-hwaccel=h264_vulkan --enable-hwaccel=hevc_vulkan --enable-hwaccel=av1_vulkan --enable-libdav1d --enable-decoder=libdav1d
+      FFMPEG_CONFIGURE_ARGS: --enable-pic --disable-static --enable-shared --disable-all --enable-avcodec --enable-avformat --enable-swscale --enable-decoder=h264 --enable-decoder=hevc --enable-decoder=av1 --enable-hwaccel=h264_vaapi --enable-hwaccel=hevc_vaapi --enable-hwaccel=av1_vaapi --enable-hwaccel=h264_vdpau --enable-hwaccel=hevc_vdpau --enable-hwaccel=av1_vdpau --enable-libdrm --enable-hwaccel=h264_vulkan --enable-hwaccel=hevc_vulkan --enable-hwaccel=av1_vulkan --enable-libdav1d --enable-decoder=libdav1d
 
 install:
   - cmd: 'copy /y scripts\appveyor\qmake.bat %QTDIR%\msvc2019_arm64\bin\'

libs

@@ -1 +1 @@
-Subproject commit f04ef02a95815a6a679f0d82ca88115edb413f26
+Subproject commit f9b4803fdfcfe12cb94076702bdf1ae70a7a4545

scripts/clean-libs.sh

@@ -46,8 +46,8 @@ while [[ "$#" -gt 0 ]]; do
       shift
       ;;
     --ffmpeg_win)
-      rm -r $LIB_PATH/windows/include/*/libavcodec $LIB_PATH/windows/include/*/libavutil $LIB_PATH/windows/include/*/libavformat
+      rm -r $LIB_PATH/windows/include/*/libavcodec $LIB_PATH/windows/include/*/libavutil $LIB_PATH/windows/include/*/libavformat $LIB_PATH/windows/include/*/libswscale
-      rm $LIB_PATH/windows/lib/*/avcodec* $LIB_PATH/windows/lib/*/avutil* $LIB_PATH/windows/lib/*/avformat*
+      rm $LIB_PATH/windows/lib/*/avcodec* $LIB_PATH/windows/lib/*/avutil* $LIB_PATH/windows/lib/*/avformat* $LIB_PATH/windows/lib/*/swscale*
       shift
       ;;
     --dav1d_win)
@@ -55,8 +55,8 @@ while [[ "$#" -gt 0 ]]; do
       shift
       ;;
     --ffmpeg_mac)
-      rm -r $LIB_PATH/mac/include/libavcodec $LIB_PATH/mac/include/libavutil $LIB_PATH/mac/include/libavformat
+      rm -r $LIB_PATH/mac/include/libavcodec $LIB_PATH/mac/include/libavutil $LIB_PATH/mac/include/libavformat $LIB_PATH/mac/include/libswscale
-      rm $LIB_PATH/mac/lib/libavcodec* $LIB_PATH/mac/lib/libavutil* $LIB_PATH/mac/lib/libavformat*
+      rm $LIB_PATH/mac/lib/libavcodec* $LIB_PATH/mac/lib/libavutil* $LIB_PATH/mac/lib/libavformat* $LIB_PATH/mac/lib/libswscale*
       shift
       ;;
     --libplacebo_win)