Add X11 VAAPI decoder

CMakeLists.txt

@@ -21,20 +21,26 @@ pkg_check_modules(SDL sdl2>=2.0.4)
 pkg_check_modules(AVCODEC libavcodec)
 pkg_check_modules(AVUTIL libavutil)
 pkg_check_modules(XLIB x11)
-pkg_check_modules(LIBVA vdpau)
+pkg_check_modules(VDPAU vdpau)
+pkg_check_modules(LIBVA libva)
+pkg_check_modules(LIBVA_X11 libva-x11)
 pkg_check_modules(PULSE libpulse-simple)
 pkg_check_modules(CEC libcec>=3.0.0)
 pkg_check_modules(EGL egl)
 pkg_check_modules(GLES glesv2)
 
-set(VDPAU_FOUND FALSE)
+set(VDPAU_ACCEL_FOUND FALSE)
+set(VA_ACCEL_FOUND FALSE)
 set(SOFTWARE_FOUND FALSE)
 
 if(AVCODEC_FOUND AND AVUTIL_FOUND)
   if(EGL_FOUND AND GLES_FOUND AND XLIB_FOUND)
     set(X11_FOUND TRUE)
-    if(XLIB_FOUND AND LIBVA_FOUND)
-      set(VDPAU_FOUND TRUE)
+    if(VDPAU_FOUND)
+      set(VDPAU_ACCEL_FOUND TRUE)
+    endif()
+    if (LIBVA_FOUND AND LIBVA_X11_FOUND)
+      set(VA_ACCEL_FOUND TRUE)
     endif()
   endif()
   if(SDL_FOUND OR X11_FOUND)
@@ -61,11 +67,16 @@ if (SOFTWARE_FOUND)
     list(APPEND MOONLIGHT_DEFINITIONS HAVE_X11)
     list(APPEND MOONLIGHT_OPTIONS X11)
   endif()
-  if(VDPAU_FOUND)
+  if(VDPAU_ACCEL_FOUND)
     list(APPEND SRC_LIST ./src/video/ffmpeg_vdpau.c)
     list(APPEND MOONLIGHT_DEFINITIONS HAVE_VDPAU)
     list(APPEND MOONLIGHT_OPTIONS VDPAU)
   endif()
+  if(VA_ACCEL_FOUND)
+    list(APPEND SRC_LIST ./src/video/ffmpeg_vaapi.c)
+    list(APPEND MOONLIGHT_DEFINITIONS HAVE_VAAPI)
+    list(APPEND MOONLIGHT_OPTIONS VAAPI)
+  endif()
 endif()
 
 if (AMLOGIC_FOUND OR BROADCOM_FOUND OR FREESCALE_FOUND OR X11_FOUND)
@@ -147,9 +158,13 @@ endif()
 if (SOFTWARE_FOUND)
   target_include_directories(moonlight PRIVATE ${AVCODEC_INCLUDE_DIRS} ${AVUTIL_INCLUDE_DIRS})
   target_link_libraries(moonlight ${AVCODEC_LIBRARIES} ${AVUTIL_LIBRARIES})
-  if(VDPAU_FOUND)
-    target_include_directories(moonlight PRIVATE ${XLIB_INCLUDE_DIRS} ${LIBVA_INCLUDE_DIRS})
-    target_link_libraries(moonlight ${XLIB_LIBRARIES} ${LIBVA_LIBRARIES})
+  if(VDPAU_ACCEL_FOUND)
+    target_include_directories(moonlight PRIVATE ${VDPAU_INCLUDE_DIRS})
+    target_link_libraries(moonlight ${VDPAU_LIBRARIES})
+  endif()
+  if(VA_ACCEL_FOUND)
+    target_include_directories(moonlight PRIVATE ${LIBVA_INCLUDE_DIRS} ${LIBVA_X11_INCLUDE_DIRS})
+    target_link_libraries(moonlight ${LIBVA_LIBRARIES} ${LIBVA_X11_LIBRARIES})
   endif()
 endif()
 

src/platform.c

@@ -60,10 +60,17 @@ enum platform platform_check(char* name) {
   }
   #endif
   #ifdef HAVE_X11
-  if (std || strcmp(name, "x11") == 0 || strcmp(name, "x11_vdpau") == 0) {
-    int x11 = x11_init(strcmp(name, "x11") != 0);
+  bool x11 = strcmp(name, "x11") == 0;
+  bool vdpau = strcmp(name, "x11_vdpau") == 0;
+  bool vaapi = strcmp(name, "x11_vaapi") == 0;
+  if (std || x11 || vdpau || vaapi) {
+    int init = x11_init(std || vdpau, std || vaapi);
+    #ifdef HAVE_VAAPI
+    if (init == INIT_VAAPI)
+      return X11_VAAPI;
+    #endif
     #ifdef HAVE_VDPAU
-    if (strcmp(name, "x11") != 0 && x11 == 0)
+    if (init == INIT_VDPAU)
       return X11_VDPAU;
     #endif
     return X11;
@@ -182,6 +189,8 @@ char* platform_name(enum platform system) {
     return "AMLogic VPU";
   case X11:
     return "X Window System (software decoding)";
+  case X11_VAAPI:
+    return "X Window System (VAAPI)";
   case X11_VDPAU:
     return "X Window System (VDPAU)";
   case SDL:

src/platform.h

@@ -26,7 +26,7 @@
 
 #define IS_EMBEDDED(SYSTEM) SYSTEM != SDL
 
-enum platform { NONE, SDL, X11, X11_VDPAU, PI, IMX, AML, FAKE };
+enum platform { NONE, SDL, X11, X11_VDPAU, X11_VAAPI, PI, IMX, AML, FAKE };
 
 enum platform platform_check(char*);
 PDECODER_RENDERER_CALLBACKS platform_get_video(enum platform system);

src/video/ffmpeg.c

@@ -22,6 +22,9 @@
 #ifdef HAVE_VDPAU
 #include "ffmpeg_vdpau.h"
 #endif
+#ifdef HAVE_VAAPI
+#include "ffmpeg_vaapi.h"
+#endif
 
 #include <Limelight.h>
 
@@ -53,8 +56,7 @@ int ffmpeg_init(int videoFormat, int width, int height, int perf_lvl, int buffer
 
   av_init_packet(&pkt);
 
-  #ifdef HAVE_VDPAU
-  if (perf_lvl & HARDWARE_ACCELERATION) {
+  if (perf_lvl & VDPAU_ACCELERATION) {
     switch (videoFormat) {
       case VIDEO_FORMAT_H264:
         decoder = avcodec_find_decoder_by_name("h264_vdpau");
@@ -64,13 +66,9 @@ int ffmpeg_init(int videoFormat, int width, int height, int perf_lvl, int buffer
         break;
     }
 
-    if (decoder != NULL)
     ffmpeg_decoder = VDPAU;
-  }
-  #endif
-
-  if (decoder == NULL) {
-    ffmpeg_decoder = SOFTWARE;
+  } else {
+    ffmpeg_decoder = perf_lvl & VAAPI_ACCELERATION ? VAAPI : SOFTWARE;
     switch (videoFormat) {
       case VIDEO_FORMAT_H264:
         decoder = avcodec_find_decoder_by_name("h264");
@@ -79,11 +77,12 @@ int ffmpeg_init(int videoFormat, int width, int height, int perf_lvl, int buffer
         decoder = avcodec_find_decoder_by_name("hevc");
         break;
     }
+  }
+
   if (decoder == NULL) {
     printf("Couldn't find decoder\n");
     return -1;
   }
-  }
 
   decoder_ctx = avcodec_alloc_context3(decoder);
   if (decoder_ctx == NULL) {
@@ -131,6 +130,11 @@ int ffmpeg_init(int videoFormat, int width, int height, int perf_lvl, int buffer
     }
   }
 
+  #ifdef HAVE_VAAPI
+  if (ffmpeg_decoder == VAAPI)
+    vaapi_init(decoder_ctx);
+  #endif
+
   #ifdef HAVE_VDPAU
   if (ffmpeg_decoder == VDPAU)
     vdpau_init(decoder_ctx, width, height);

src/video/ffmpeg.h

@@ -34,9 +34,10 @@
 // Uses a faster bilinear filtering with lower image quality
 #define FAST_BILINEAR_FILTERING 0x20
 // Uses hardware acceleration
-#define HARDWARE_ACCELERATION 0x40
+#define VDPAU_ACCELERATION 0x40
+#define VAAPI_ACCELERATION 0x80
 
-enum decoders {SOFTWARE, VDPAU};
+enum decoders {SOFTWARE, VDPAU, VAAPI};
 extern enum decoders ffmpeg_decoder;
 
 int ffmpeg_init(int videoFormat, int width, int height, int perf_lvl, int buffer_count, int thread_count);

src/video/ffmpeg_vaapi.c

@@ -0,0 +1,75 @@
+/*
+ * This file is part of Moonlight Embedded.
+ *
+ * Copyright (C) 2017 Iwan Timmer
+ *
+ * Moonlight is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * Moonlight is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with Moonlight; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <va/va.h>
+#include <va/va_x11.h>
+#include <libavcodec/avcodec.h>
+#include <libavutil/hwcontext.h>
+#include <libavutil/hwcontext_vaapi.h>
+#include <X11/Xlib.h>
+
+#define MAX_SURFACES 16
+
+static AVBufferRef* device_ref;
+
+static enum AVPixelFormat va_get_format(AVCodecContext* context, const enum AVPixelFormat* pixel_format) {
+  AVBufferRef* hw_ctx = av_hwframe_ctx_alloc(device_ref);
+  if (hw_ctx == NULL) {
+    fprintf(stderr, "Failed to initialize Vaapi buffer\n");
+    return AV_PIX_FMT_NONE;
+  }
+
+  AVHWFramesContext* fr_ctx = (AVHWFramesContext*) hw_ctx->data;
+  fr_ctx->format = AV_PIX_FMT_VAAPI;
+  fr_ctx->sw_format = AV_PIX_FMT_NV12;
+  fr_ctx->width = context->coded_width;
+  fr_ctx->height = context->coded_height;
+  fr_ctx->initial_pool_size = MAX_SURFACES + 1;
+
+  if (av_hwframe_ctx_init(hw_ctx) < 0) {
+    fprintf(stderr, "Failed to initialize VAAPI frame context");
+    return AV_PIX_FMT_NONE;
+  }
+  
+  context->pix_fmt = AV_PIX_FMT_VAAPI;
+  context->hw_device_ctx = device_ref;
+  context->hw_frames_ctx = hw_ctx;
+  context->slice_flags = SLICE_FLAG_CODED_ORDER | SLICE_FLAG_ALLOW_FIELD;
+  return AV_PIX_FMT_VAAPI;
+}
+
+static int va_get_buffer(AVCodecContext* context, AVFrame* frame, int flags) {
+  return av_hwframe_get_buffer(context->hw_frames_ctx, frame, 0);
+}
+
+int vaapi_init_lib() {
+  return av_hwdevice_ctx_create(&device_ref, AV_HWDEVICE_TYPE_VAAPI, ":0", NULL, 0);
+}
+
+int vaapi_init(AVCodecContext* decoder_ctx) {
+  decoder_ctx->get_format = va_get_format;
+  decoder_ctx->get_buffer2 = va_get_buffer;
+}
+
+void vaapi_queue(AVFrame* dec_frame, Window win, int width, int height) {
+  VASurfaceID surface = (VASurfaceID)(uintptr_t)dec_frame->data[3];
+  AVHWDeviceContext* device = (AVHWDeviceContext*) device_ref->data;
+  AVVAAPIDeviceContext *va_ctx = device->hwctx;
+  vaPutSurface(va_ctx->display, surface, win, 0, 0, dec_frame->width, dec_frame->height, 0, 0, width, height, NULL, 0, 0);
+}

src/video/ffmpeg_vaapi.h

@@ -0,0 +1,24 @@
+/*
+ * This file is part of Moonlight Embedded.
+ *
+ * Copyright (C) 2017 Iwan Timmer
+ *
+ * Moonlight is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * Moonlight is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with Moonlight; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <va/va.h>
+
+int vaapi_init_lib();
+int vaapi_init(AVCodecContext* decoder_ctx);
+void vaapi_queue(AVFrame* dec_frame, Window win, int width, int height);

src/video/video.h

@@ -22,11 +22,19 @@
 #include <stdbool.h>
 
 #define DISPLAY_FULLSCREEN 1
-#define ENABLE_HARDWARE_ACCELERATION 2
+#define ENABLE_HARDWARE_ACCELERATION_1 2
+#define ENABLE_HARDWARE_ACCELERATION_2 4
+
+#define INIT_EGL 1
+#define INIT_VDPAU 2
+#define INIT_VAAPI 3
 
 #ifdef HAVE_X11
-int x11_init(bool vdpau);
+int x11_init(bool vdpau, bool vaapi);
 extern DECODER_RENDERER_CALLBACKS decoder_callbacks_x11;
+#ifdef HAVE_VAAPI
+extern DECODER_RENDERER_CALLBACKS decoder_callbacks_x11_vaapi;
+#endif
 #ifdef HAVE_VDPAU
 extern DECODER_RENDERER_CALLBACKS decoder_callbacks_x11_vdpau;
 #endif

src/video/x11.c

@@ -23,6 +23,9 @@
 #ifdef HAVE_VDPAU
 #include "ffmpeg_vdpau.h"
 #endif
+#ifdef HAVE_VAAPI
+#include "ffmpeg_vaapi.h"
+#endif
 
 #include "../input/x11.h"
 #include "../loop.h"
@@ -36,13 +39,19 @@
 #include <poll.h>
 
 #define DECODER_BUFFER_SIZE 92*1024
+#define X11_VDPAU_ACCELERATION ENABLE_HARDWARE_ACCELERATION_1
+#define X11_VAAPI_ACCELERATION ENABLE_HARDWARE_ACCELERATION_2
 
 static char* ffmpeg_buffer = NULL;
 
 static Display *display = NULL;
+static Window window;
 
 static int pipefd[2];
 
+static int display_width;
+static int display_height;
+
 static int frame_handle(int pipefd) {
   AVFrame* frame = NULL;
   while (read(pipefd, &frame, sizeof(void*)) > 0);
@@ -52,18 +61,23 @@ static int frame_handle(int pipefd) {
   return LOOP_OK;
 }
 
-int x11_init(bool vdpau) {
+int x11_init(bool vdpau, bool vaapi) {
   XInitThreads();
   display = XOpenDisplay(NULL);
   if (!display)
-    return -1;
+    return 0;
 
-  #ifdef HAVE_VDPAU
-  if (vdpau && vdpau_init_lib(display) != 0)
-    return -2;
+  #ifdef HAVE_VAAPI
+  if (vaapi && vaapi_init_lib(display) == 0)
+    return INIT_VAAPI;
   #endif
 
-  return 0;
+  #ifdef HAVE_VDPAU
+  if (vdpau && vdpau_init_lib(display) == 0)
+    return INIT_VDPAU;
+  #endif
+
+  return INIT_EGL;
 }
 
 int x11_setup(int videoFormat, int width, int height, int redrawRate, void* context, int drFlags) {
@@ -78,8 +92,6 @@ int x11_setup(int videoFormat, int width, int height, int redrawRate, void* cont
     return -1;
   }
 
-  int display_width;
-  int display_height;
   if (drFlags & DISPLAY_FULLSCREEN) {
     Screen* screen = DefaultScreenOfDisplay(display);
     display_width = WidthOfScreen(screen);
@@ -91,7 +103,7 @@ int x11_setup(int videoFormat, int width, int height, int redrawRate, void* cont
 
   Window root = DefaultRootWindow(display);
   XSetWindowAttributes winattr = { .event_mask = PointerMotionMask | ButtonPressMask | ButtonReleaseMask | KeyPressMask | KeyReleaseMask };
-  Window window = XCreateWindow(display, root, 0, 0, display_width, display_height, 0, CopyFromParent, InputOutput, CopyFromParent, CWEventMask, &winattr);
+  window = XCreateWindow(display, root, 0, 0, display_width, display_height, 0, CopyFromParent, InputOutput, CopyFromParent, CWEventMask, &winattr);
   XMapWindow(display, window);
   XStoreName(display, window, "Moonlight");
 
@@ -113,10 +125,10 @@ int x11_setup(int videoFormat, int width, int height, int redrawRate, void* cont
   XFlush(display);
 
   int avc_flags = SLICE_THREADING;
-  #ifdef HAVE_VDPAU
-  if (drFlags & ENABLE_HARDWARE_ACCELERATION)
-    avc_flags |= HARDWARE_ACCELERATION;
-  #endif
+  if (drFlags & X11_VDPAU_ACCELERATION)
+    avc_flags |= VDPAU_ACCELERATION;
+  else if (drFlags & X11_VAAPI_ACCELERATION)
+    avc_flags |= VAAPI_ACCELERATION;
 
   if (ffmpeg_init(videoFormat, width, height, avc_flags, 2, 2) < 0) {
     fprintf(stderr, "Couldn't initialize video decoding\n");
@@ -143,7 +155,11 @@ int x11_setup(int videoFormat, int width, int height, int redrawRate, void* cont
 }
 
 int x11_setup_vdpau(int videoFormat, int width, int height, int redrawRate, void* context, int drFlags) {
-  return x11_setup(videoFormat, width, height, redrawRate, context, drFlags | ENABLE_HARDWARE_ACCELERATION);
+  return x11_setup(videoFormat, width, height, redrawRate, context, drFlags | X11_VDPAU_ACCELERATION);
+}
+
+int x11_setup_vaapi(int videoFormat, int width, int height, int redrawRate, void* context, int drFlags) {
+  return x11_setup(videoFormat, width, height, redrawRate, context, drFlags | X11_VAAPI_ACCELERATION);
 }
 
 void x11_cleanup() {
@@ -167,6 +183,8 @@ int x11_submit_decode_unit(PDECODE_UNIT decodeUnit) {
         write(pipefd[1], &frame, sizeof(void*));
       else if (ffmpeg_decoder == VDPAU)
         vdpau_queue(frame);
+      else if (ffmpeg_decoder == VAAPI)
+        vaapi_queue(frame, window, display_width, display_height);
     }
   }
 
@@ -186,3 +204,10 @@ DECODER_RENDERER_CALLBACKS decoder_callbacks_x11_vdpau = {
   .submitDecodeUnit = x11_submit_decode_unit,
   .capabilities = CAPABILITY_DIRECT_SUBMIT,
 };
+
+DECODER_RENDERER_CALLBACKS decoder_callbacks_x11_vaapi = {
+  .setup = x11_setup_vaapi,
+  .cleanup = x11_cleanup,
+  .submitDecodeUnit = x11_submit_decode_unit,
+  .capabilities = CAPABILITY_SLICES_PER_FRAME(4) | CAPABILITY_REFERENCE_FRAME_INVALIDATION_AVC | CAPABILITY_REFERENCE_FRAME_INVALIDATION_HEVC | CAPABILITY_DIRECT_SUBMIT,
+};