Use vaapi to do rgb->yuv420p color conversion

Enable window capture for amd/intel.
Properly check if h264/hevc is supported on amd/intel before using
codec.

3 files changed, 185 insertions, 392 deletions

diff --git a/src/capture/nvfbc.c b/src/capture/nvfbc.c
index d538b5b..a7bb16a 100644
--- a/src/capture/nvfbc.c
+++ b/src/capture/nvfbc.c
@@ -177,8 +177,8 @@ static bool ffmpeg_create_cuda_contexts(gsr_capture_nvfbc *cap_nvfbc, AVCodecCon
         return false;
     }
 
-    video_codec_context->hw_device_ctx = device_ctx;
-    video_codec_context->hw_frames_ctx = frame_context;
+    video_codec_context->hw_device_ctx = av_buffer_ref(device_ctx);
+    video_codec_context->hw_frames_ctx = av_buffer_ref(frame_context);
     return true;
 }
 
@@ -350,9 +350,9 @@ static int gsr_capture_nvfbc_start(gsr_capture *cap, AVCodecContext *video_codec
 
     if(video_codec_context->hw_device_ctx)
         av_buffer_unref(&video_codec_context->hw_device_ctx);
-    // Not needed because the above call to unref device ctx also frees this?
-    //if(video_codec_context->hw_frames_ctx)
-    //    av_buffer_unref(&video_codec_context->hw_frames_ctx);
+    if(video_codec_context->hw_frames_ctx)
+        av_buffer_unref(&video_codec_context->hw_frames_ctx);
+
     gsr_cuda_unload(&cap_nvfbc->cuda);
     return -1;
 }
@@ -438,9 +438,8 @@ static void gsr_capture_nvfbc_destroy(gsr_capture *cap, AVCodecContext *video_co
     gsr_capture_nvfbc_destroy_session(cap);
     if(video_codec_context->hw_device_ctx)
         av_buffer_unref(&video_codec_context->hw_device_ctx);
-    // Not needed because the above call to unref device ctx also frees this?
-    //if(video_codec_context->hw_frames_ctx)
-    //    av_buffer_unref(&video_codec_context->hw_frames_ctx);
+    if(video_codec_context->hw_frames_ctx)
+        av_buffer_unref(&video_codec_context->hw_frames_ctx);
     if(cap_nvfbc) {
         gsr_cuda_unload(&cap_nvfbc->cuda);
         dlclose(cap_nvfbc->library);
diff --git a/src/capture/xcomposite_cuda.c b/src/capture/xcomposite_cuda.c
index a812c12..1ea9923 100644
--- a/src/capture/xcomposite_cuda.c
+++ b/src/capture/xcomposite_cuda.c
@@ -131,8 +131,8 @@ static bool cuda_create_codec_context(gsr_capture_xcomposite_cuda *cap_xcomp, AV
         return false;
     }
 
-    video_codec_context->hw_device_ctx = device_ctx;
-    video_codec_context->hw_frames_ctx = frame_context;
+    video_codec_context->hw_device_ctx = av_buffer_ref(device_ctx);
+    video_codec_context->hw_frames_ctx = av_buffer_ref(frame_context);
     return true;
 }
 
@@ -252,9 +252,8 @@ static void gsr_capture_xcomposite_cuda_stop(gsr_capture *cap, AVCodecContext *v
 
     if(video_codec_context->hw_device_ctx)
         av_buffer_unref(&video_codec_context->hw_device_ctx);
-    // Not needed because the above call to unref device ctx also frees this?
-    //if(video_codec_context->hw_frames_ctx)
-    //    av_buffer_unref(&video_codec_context->hw_frames_ctx);
+    if(video_codec_context->hw_frames_ctx)
+        av_buffer_unref(&video_codec_context->hw_frames_ctx);
 
     if(cap_xcomp->cuda.cu_ctx) {
         CUcontext old_ctx;
diff --git a/src/capture/xcomposite_drm.c b/src/capture/xcomposite_drm.c
index fbd96fc..47cbfcc 100644
--- a/src/capture/xcomposite_drm.c
+++ b/src/capture/xcomposite_drm.c
@@ -1,15 +1,16 @@
 #include "../../include/capture/xcomposite_drm.h"
 #include "../../include/egl.h"
-#include "../../include/vaapi.h"
 #include "../../include/window_texture.h"
 #include "../../include/time.h"
 #include <stdlib.h>
 #include <stdio.h>
 #include <X11/Xlib.h>
+#include <X11/extensions/Xcomposite.h>
 #include <libavutil/hwcontext.h>
 #include <libavutil/hwcontext_vaapi.h>
 #include <libavutil/frame.h>
 #include <libavcodec/avcodec.h>
+//#include <drm_fourcc.h>
 #include <assert.h>
 /* TODO: Proper error checks and cleanups */
 
@@ -27,18 +28,20 @@ typedef struct {
     WindowTexture window_texture;
 
     gsr_egl egl;
-    gsr_vaapi vaapi;
 
-    unsigned int target_textures[2];
-
-    unsigned int framebuffer_y;
-    unsigned int framebuffer_uv;
-    unsigned int vao;
+    int fourcc;
+    int num_planes;
+    uint64_t modifiers;
+    int dmabuf_fd;
+    int32_t pitch;
+    int32_t offset;
 
-    unsigned int shader_y;
-    unsigned int shader_uv;
+    unsigned int target_textures[2];
 
     VADisplay va_dpy;
+    VAConfigID config_id;
+    VAContextID context_id;
+    VASurfaceID input_surface;
 } gsr_capture_xcomposite_drm;
 
 static int max_int(int a, int b) {
@@ -47,7 +50,7 @@ static int max_int(int a, int b) {
 
 static bool drm_create_codec_context(gsr_capture_xcomposite_drm *cap_xcomp, AVCodecContext *video_codec_context) {
     AVBufferRef *device_ctx;
-    if(av_hwdevice_ctx_create(&device_ctx, AV_HWDEVICE_TYPE_VAAPI, "/dev/dri/card0", NULL, 0) < 0) {
+    if(av_hwdevice_ctx_create(&device_ctx, AV_HWDEVICE_TYPE_VAAPI, "/dev/dri/renderD128", NULL, 0) < 0) {
         fprintf(stderr, "Error: Failed to create hardware device context\n");
         return false;
     }
@@ -63,7 +66,7 @@ static bool drm_create_codec_context(gsr_capture_xcomposite_drm *cap_xcomp, AVCo
         (AVHWFramesContext *)frame_context->data;
     hw_frame_context->width = video_codec_context->width;
     hw_frame_context->height = video_codec_context->height;
-    hw_frame_context->sw_format = AV_PIX_FMT_NV12;
+    hw_frame_context->sw_format = AV_PIX_FMT_NV12;//AV_PIX_FMT_0RGB32;//AV_PIX_FMT_YUV420P;//AV_PIX_FMT_0RGB32;//AV_PIX_FMT_NV12;
     hw_frame_context->format = video_codec_context->pix_fmt;
     hw_frame_context->device_ref = device_ctx;
     hw_frame_context->device_ctx = (AVHWDeviceContext*)device_ctx->data;
@@ -74,227 +77,20 @@ static bool drm_create_codec_context(gsr_capture_xcomposite_drm *cap_xcomp, AVCo
     cap_xcomp->va_dpy = vactx->display;
 
     if (av_hwframe_ctx_init(frame_context) < 0) {
-        fprintf(stderr, "Error: Failed to initialize hardware frame context (note: ffmpeg version needs to be > 4.0)\n");
+        fprintf(stderr, "Error: Failed to initialize hardware frame context "
+                        "(note: ffmpeg version needs to be > 4.0)\n");
         av_buffer_unref(&device_ctx);
         //av_buffer_unref(&frame_context);
         return false;
     }
 
-    video_codec_context->hw_device_ctx = device_ctx; // TODO: av_buffer_ref? and in more places
-    video_codec_context->hw_frames_ctx = frame_context;
+    video_codec_context->hw_device_ctx = av_buffer_ref(device_ctx);
+    video_codec_context->hw_frames_ctx = av_buffer_ref(frame_context);
     return true;
 }
 
-#define GL_COMPILE_STATUS                 0x8B81
-#define GL_INFO_LOG_LENGTH                0x8B84
-
-unsigned int esLoadShader ( gsr_capture_xcomposite_drm *cap_xcomp, unsigned int type, const char *shaderSrc ) {
-   unsigned int shader;
-   int compiled;
-   
-   // Create the shader object
-   shader = cap_xcomp->egl.glCreateShader ( type );
-
-   if ( shader == 0 )
-   	return 0;
-
-   // Load the shader source
-   cap_xcomp->egl.glShaderSource ( shader, 1, &shaderSrc, NULL );
-   
-   // Compile the shader
-   cap_xcomp->egl.glCompileShader ( shader );
-
-   // Check the compile status
-   cap_xcomp->egl.glGetShaderiv ( shader, GL_COMPILE_STATUS, &compiled );
-
-   if ( !compiled ) 
-   {
-      int infoLen = 0;
-
-      cap_xcomp->egl.glGetShaderiv ( shader, GL_INFO_LOG_LENGTH, &infoLen );
-      
-      if ( infoLen > 1 )
-      {
-         char* infoLog = malloc (sizeof(char) * infoLen );
-
-         cap_xcomp->egl.glGetShaderInfoLog ( shader, infoLen, NULL, infoLog );
-         fprintf (stderr, "Error compiling shader:\n%s\n", infoLog );            
-         
-         free ( infoLog );
-      }
-
-      cap_xcomp->egl.glDeleteShader ( shader );
-      return 0;
-   }
-
-   return shader;
-
-}
-
-#define GL_FRAGMENT_SHADER                0x8B30
-#define GL_VERTEX_SHADER                  0x8B31
-#define GL_COMPILE_STATUS                 0x8B81
-#define GL_LINK_STATUS                    0x8B82
-
-
-//
-///
-/// \brief Load a vertex and fragment shader, create a program object, link program.
-//         Errors output to log.
-/// \param vertShaderSrc Vertex shader source code
-/// \param fragShaderSrc Fragment shader source code
-/// \return A new program object linked with the vertex/fragment shader pair, 0 on failure
-//
-unsigned int esLoadProgram ( gsr_capture_xcomposite_drm *cap_xcomp, const char *vertShaderSrc, const char *fragShaderSrc )
-{
-   unsigned int vertexShader;
-   unsigned int fragmentShader;
-   unsigned int programObject;
-   int linked;
-
-   // Load the vertex/fragment shaders
-   vertexShader = esLoadShader ( cap_xcomp, GL_VERTEX_SHADER, vertShaderSrc );
-   if ( vertexShader == 0 )
-      return 0;
-
-   fragmentShader = esLoadShader ( cap_xcomp, GL_FRAGMENT_SHADER, fragShaderSrc );
-   if ( fragmentShader == 0 )
-   {
-      cap_xcomp->egl.glDeleteShader( vertexShader );
-      return 0;
-   }
-
-   // Create the program object
-   programObject = cap_xcomp->egl.glCreateProgram ( );
-   
-   if ( programObject == 0 )
-      return 0;
-
-   cap_xcomp->egl.glAttachShader ( programObject, vertexShader );
-   cap_xcomp->egl.glAttachShader ( programObject, fragmentShader );
-
-   // Link the program
-   cap_xcomp->egl.glLinkProgram ( programObject );
-
-   // Check the link status
-   cap_xcomp->egl.glGetProgramiv ( programObject, GL_LINK_STATUS, &linked );
-
-   if ( !linked ) 
-   {
-      int infoLen = 0;
-
-      cap_xcomp->egl.glGetProgramiv ( programObject, GL_INFO_LOG_LENGTH, &infoLen );
-      
-      if ( infoLen > 1 )
-      {
-         char* infoLog = malloc (sizeof(char) * infoLen );
-
-         cap_xcomp->egl.glGetProgramInfoLog ( programObject, infoLen, NULL, infoLog );
-         fprintf (stderr, "Error linking program:\n%s\n", infoLog );            
-         
-         free ( infoLog );
-      }
-
-      cap_xcomp->egl.glDeleteProgram ( programObject );
-      return 0;
-   }
-
-   // Free up no longer needed shader resources
-   cap_xcomp->egl.glDeleteShader ( vertexShader );
-   cap_xcomp->egl.glDeleteShader ( fragmentShader );
-
-   return programObject;
-}
-
-#define RGB_TO_YUV "const mat4 RGBtoYUV = mat4(0.257,  0.439, -0.148, 0.0,\n" \
-                   "                           0.504, -0.368, -0.291, 0.0,\n" \
-                   "                           0.098, -0.071,  0.439, 0.0,\n" \
-                   "                           0.0625, 0.500,  0.500, 1.0);"
-
-static unsigned int LoadShadersY(gsr_capture_xcomposite_drm *cap_xcomp) {
-	char vShaderStr[] =
-        "#version 300 es                                   \n"
-        "in vec2 pos;                                      \n"
-        "in vec2 texcoords;                                \n"
-        "out vec2 texcoords_out;                           \n"
-		"void main()                                       \n"
-		"{                                                 \n"
-        "   texcoords_out = texcoords;                     \n"
-		"   gl_Position = vec4(pos.x, pos.y, 0.0, 1.0);    \n"
-		"}                                                 \n";
-
-    char fShaderStr[] =
-        "#version 300 es                                                                 \n"
-		"precision mediump float;                                                        \n"
-        "in vec2 texcoords_out;                                                          \n"
-        "uniform sampler2D tex1;                                                         \n"
-        "out vec4 FragColor;                                                             \n"
-        RGB_TO_YUV
-		"void main()                                                                     \n"
-		"{                                                                               \n"
-        "   FragColor.x = (RGBtoYUV * vec4(texture(tex1, texcoords_out).rgb, 1.0)).x;    \n"
-		"}                                                                               \n";
-
-    unsigned int shader_program = esLoadProgram(cap_xcomp, vShaderStr, fShaderStr);
-	if (shader_program == 0) {
-        fprintf(stderr, "failed to create shader!\n");
-        return 0;
-    }
-
-    cap_xcomp->egl.glBindAttribLocation(shader_program, 0, "pos");
-    cap_xcomp->egl.glBindAttribLocation(shader_program, 1, "texcoords");
-	return shader_program;
-}
-
-static unsigned int LoadShadersUV(gsr_capture_xcomposite_drm *cap_xcomp) {
-	char vShaderStr[] =
-        "#version 300 es                                 \n"
-        "in vec2 pos;                                    \n"
-        "in vec2 texcoords;                              \n"
-        "out vec2 texcoords_out;                         \n"
-		"void main()                                     \n"
-		"{                                               \n"
-        "  texcoords_out = texcoords;                    \n"
-		"  gl_Position = vec4(pos.x, pos.y, 0.0, 1.0);   \n"
-		"}                                               \n";
-
-    char fShaderStr[] =
-        "#version 300 es                                                                       \n"
-		"precision mediump float;                                                              \n"
-        "in vec2 texcoords_out;                                                                \n"
-        "uniform sampler2D tex1;                                                               \n"
-        "out vec4 FragColor;                                                                   \n"
-        RGB_TO_YUV
-		"void main()                                                                           \n"
-		"{                                                                                     \n"
-        "   FragColor.xy = (RGBtoYUV * vec4(texture(tex1, texcoords_out*2.0).rgb, 1.0)).zy;    \n"
-		"}                                                                                     \n";
-
-    unsigned int shader_program = esLoadProgram(cap_xcomp, vShaderStr, fShaderStr);
-	if (shader_program == 0) {
-        fprintf(stderr, "failed to create shader!\n");
-        return 0;
-    }
-
-    cap_xcomp->egl.glBindAttribLocation(shader_program, 0, "pos");
-    cap_xcomp->egl.glBindAttribLocation(shader_program, 1, "texcoords");
-	return shader_program;
-}
-
-#define GL_FLOAT				0x1406
-#define GL_FALSE				0
-#define GL_TRUE					1
-#define GL_TRIANGLES				0x0004
 #define DRM_FORMAT_MOD_INVALID 72057594037927935
 
-#define EGL_TRUE                          1
-#define EGL_IMAGE_PRESERVED_KHR           0x30D2
-#define EGL_NATIVE_PIXMAP_KHR             0x30B0
-
-static uint32_t fourcc(uint32_t a, uint32_t b, uint32_t c, uint32_t d) {
-    return (d << 24) | (c << 16) | (b << 8) | a;
-}
-
 static int gsr_capture_xcomposite_drm_start(gsr_capture *cap, AVCodecContext *video_codec_context) {
     gsr_capture_xcomposite_drm *cap_xcomp = cap->priv;
 
@@ -317,15 +113,51 @@ static int gsr_capture_xcomposite_drm_start(gsr_capture *cap, AVCodecContext *vi
         return -1;
     }
 
-    if(!gsr_vaapi_load(&cap_xcomp->vaapi)) {
-        fprintf(stderr, "gsr error: gsr_capture_xcomposite_drm_start: failed to load vaapi\n");
+    if(!cap_xcomp->egl.eglExportDMABUFImageQueryMESA) {
+        fprintf(stderr, "gsr error: gsr_capture_xcomposite_drm_start: could not find eglExportDMABUFImageQueryMESA\n");
+        gsr_egl_unload(&cap_xcomp->egl);
+        return -1;
+    }
+
+    if(!cap_xcomp->egl.eglExportDMABUFImageMESA) {
+        fprintf(stderr, "gsr error: gsr_capture_xcomposite_drm_start: could not find eglExportDMABUFImageMESA\n");
         gsr_egl_unload(&cap_xcomp->egl);
         return -1;
     }
 
     /* Disable vsync */
     cap_xcomp->egl.eglSwapInterval(cap_xcomp->egl.egl_display, 0);
+#if 0
+    // TODO: Fallback to composite window
+    if(window_texture_init(&cap_xcomp->window_texture, cap_xcomp->dpy, cap_xcomp->params.window, &cap_xcomp->gl) != 0) {
+        fprintf(stderr, "gsr error: gsr_capture_xcomposite_start: failed get window texture for window %ld\n", cap_xcomp->params.window);
+        gsr_egl_unload(&cap_xcomp->egl);
+        return -1;
+    }
+
+    cap_xcomp->egl.glBindTexture(GL_TEXTURE_2D, window_texture_get_opengl_texture_id(&cap_xcomp->window_texture));
+    cap_xcomp->texture_size.x = 0;
+    cap_xcomp->texture_size.y = 0;
+    cap_xcomp->egl.glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_WIDTH, &cap_xcomp->texture_size.x);
+    cap_xcomp->egl.glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_HEIGHT, &cap_xcomp->texture_size.y);
+    cap_xcomp->egl.glBindTexture(GL_TEXTURE_2D, 0);
+
+    cap_xcomp->texture_size.x = max_int(2, cap_xcomp->texture_size.x & ~1);
+    cap_xcomp->texture_size.y = max_int(2, cap_xcomp->texture_size.y & ~1);
+
+    cap_xcomp->target_texture_id = gl_create_texture(cap_xcomp, cap_xcomp->texture_size.x, cap_xcomp->texture_size.y);
+    if(cap_xcomp->target_texture_id == 0) {
+        fprintf(stderr, "gsr error: gsr_capture_xcomposite_start: failed to create opengl texture\n");
+        gsr_capture_xcomposite_stop(cap, video_codec_context);
+        return -1;
+    }
 
+    video_codec_context->width = cap_xcomp->texture_size.x;
+    video_codec_context->height = cap_xcomp->texture_size.y;
+
+    cap_xcomp->window_resize_timer = clock_get_monotonic_seconds();
+    return 0;
+#else
     // TODO: Fallback to composite window
     if(window_texture_init(&cap_xcomp->window_texture, cap_xcomp->dpy, cap_xcomp->params.window, &cap_xcomp->egl) != 0) {
         fprintf(stderr, "gsr error: gsr_capture_xcomposite_drm_start: failed get window texture for window %ld\n", cap_xcomp->params.window);
@@ -346,13 +178,47 @@ static int gsr_capture_xcomposite_drm_start(gsr_capture *cap, AVCodecContext *vi
     video_codec_context->width = cap_xcomp->texture_size.x;
     video_codec_context->height = cap_xcomp->texture_size.y;
 
+    {
+        const intptr_t pixmap_attrs[] = {
+            EGL_IMAGE_PRESERVED_KHR, EGL_TRUE,
+            EGL_NONE,
+        };
+
+        EGLImage img = cap_xcomp->egl.eglCreateImage(cap_xcomp->egl.egl_display, cap_xcomp->egl.egl_context, EGL_GL_TEXTURE_2D, (EGLClientBuffer)(uint64_t)window_texture_get_opengl_texture_id(&cap_xcomp->window_texture), pixmap_attrs);
+        if(!img) {
+            fprintf(stderr, "eglCreateImage failed\n");
+            return -1;
+        }
+
+        if(!cap_xcomp->egl.eglExportDMABUFImageQueryMESA(cap_xcomp->egl.egl_display, img, &cap_xcomp->fourcc, &cap_xcomp->num_planes, &cap_xcomp->modifiers)) {
+            fprintf(stderr, "eglExportDMABUFImageQueryMESA failed\n"); 
+            return -1;
+        }
+
+        if(cap_xcomp->num_planes != 1) {
+            // TODO: FAIL!
+            fprintf(stderr, "Blablalba\n");
+            return -1;
+        }
+
+        if(!cap_xcomp->egl.eglExportDMABUFImageMESA(cap_xcomp->egl.egl_display, img, &cap_xcomp->dmabuf_fd, &cap_xcomp->pitch, &cap_xcomp->offset)) {
+            fprintf(stderr, "eglExportDMABUFImageMESA failed\n");
+            return -1;
+        }
+
+        fprintf(stderr, "texture: %u, dmabuf: %d, pitch: %d, offset: %d\n", window_texture_get_opengl_texture_id(&cap_xcomp->window_texture), cap_xcomp->dmabuf_fd, cap_xcomp->pitch, cap_xcomp->offset);
+        fprintf(stderr, "fourcc: %d, num planes: %d, modifiers: %zu\n", cap_xcomp->fourcc, cap_xcomp->num_planes, cap_xcomp->modifiers);
+    }
+
     if(!drm_create_codec_context(cap_xcomp, video_codec_context)) {
         fprintf(stderr, "failed to create hw codec context\n");
         gsr_egl_unload(&cap_xcomp->egl);
         return -1;
     }
 
+    //fprintf(stderr, "sneed: %u\n", cap_xcomp->FramebufferName);
     return 0;
+#endif
 }
 
 static void gsr_capture_xcomposite_drm_tick(gsr_capture *cap, AVCodecContext *video_codec_context, AVFrame **frame) {
@@ -372,11 +238,7 @@ static void gsr_capture_xcomposite_drm_tick(gsr_capture *cap, AVCodecContext *vi
         (*frame)->format = video_codec_context->pix_fmt;
         (*frame)->width = video_codec_context->width;
         (*frame)->height = video_codec_context->height;
-        (*frame)->color_range = video_codec_context->color_range;
-        (*frame)->color_primaries = video_codec_context->color_primaries;
-        (*frame)->color_trc = video_codec_context->color_trc;
-        (*frame)->colorspace = video_codec_context->colorspace;
-        (*frame)->chroma_location = video_codec_context->chroma_sample_location;
+        (*frame)->color_range = AVCOL_RANGE_JPEG;
 
         int res = av_hwframe_get_buffer(video_codec_context->hw_frames_ctx, *frame, 0);
         if(res < 0) {
@@ -384,143 +246,63 @@ static void gsr_capture_xcomposite_drm_tick(gsr_capture *cap, AVCodecContext *vi
             return;
         }
 
-        VADRMPRIMESurfaceDescriptor prime;
-
-        VASurfaceID surface_id = (uintptr_t)(*frame)->data[3];
-        VAStatus va_status = cap_xcomp->vaapi.vaExportSurfaceHandle(cap_xcomp->va_dpy, surface_id, VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME_2, VA_EXPORT_SURFACE_READ_WRITE | VA_EXPORT_SURFACE_SEPARATE_LAYERS, &prime); // TODO: Composed layers
+        fprintf(stderr, "fourcc: %u\n", cap_xcomp->fourcc);
+        fprintf(stderr, "va surface id: %u\n", (VASurfaceID)(uintptr_t)(*frame)->data[3]);
+
+        int xx = 0, yy = 0;
+        cap_xcomp->egl.glBindTexture(GL_TEXTURE_2D, window_texture_get_opengl_texture_id(&cap_xcomp->window_texture));
+        cap_xcomp->egl.glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_WIDTH, &xx);
+        cap_xcomp->egl.glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_HEIGHT, &yy);
+        cap_xcomp->egl.glBindTexture(GL_TEXTURE_2D, 0);
+
+        uintptr_t dmabuf = cap_xcomp->dmabuf_fd;
+
+        VASurfaceAttribExternalBuffers buf = {0};
+        buf.pixel_format = VA_FOURCC_BGRX; // TODO: VA_FOURCC_XRGB?
+        buf.width = xx;
+        buf.height = yy;
+        buf.data_size = yy * cap_xcomp->pitch;
+        buf.num_planes = 1;
+        buf.pitches[0] = cap_xcomp->pitch;
+        buf.offsets[0] = cap_xcomp->offset;
+        buf.buffers = &dmabuf;
+        buf.num_buffers = 1;
+        buf.flags = 0;
+        buf.private_data = 0;
+
+        #define VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME        0x20000000
+
+        VASurfaceAttrib attribs[2] = {0};
+        attribs[0].type = VASurfaceAttribMemoryType;
+        attribs[0].flags = VA_SURFACE_ATTRIB_SETTABLE;
+        attribs[0].value.type = VAGenericValueTypeInteger;
+        attribs[0].value.value.i = VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME; // TODO: prime1 instead?
+        attribs[1].type = VASurfaceAttribExternalBufferDescriptor;
+        attribs[1].flags = VA_SURFACE_ATTRIB_SETTABLE;
+        attribs[1].value.type = VAGenericValueTypePointer;
+        attribs[1].value.value.p = &buf;
+        
+        VAStatus va_status = vaCreateSurfaces(cap_xcomp->va_dpy, VA_RT_FORMAT_RGB32, xx, yy, &cap_xcomp->input_surface, 1, attribs, 2);
         if(va_status != VA_STATUS_SUCCESS) {
-            fprintf(stderr, "vaExportSurfaceHandle failed\n");
+            fprintf(stderr, "failed to create surface: %d\n", va_status);
+            abort();
             return;
         }
-        cap_xcomp->vaapi.vaSyncSurface(cap_xcomp->va_dpy, surface_id);
-
-        fprintf(stderr, "fourcc: %u, width: %u, height: %u\n", prime.fourcc, prime.width, prime.height);
-        for(int i = 0; i < prime.num_layers; ++i) {
-            fprintf(stderr, "  drm format: %u, num planes: %u\n", prime.layers[i].drm_format, prime.layers[i].num_planes);
-            for(int j = 0; j < prime.layers[i].num_planes; ++j) {
-                const uint32_t object_index = prime.layers[i].object_index[j];
-                fprintf(stderr, "    object index: %u, offset: %u, pitch: %u, fd: %d, size: %u, drm format mod: %lu\n", object_index, prime.layers[i].offset[j], prime.layers[i].pitch[j], prime.objects[object_index].fd, prime.objects[object_index].size, prime.objects[object_index].drm_format_modifier);
-            }
-        }
-
-        #define EGL_LINUX_DRM_FOURCC_EXT          0x3271
-        #define EGL_WIDTH                         0x3057
-        #define EGL_HEIGHT                        0x3056
-        #define EGL_DMA_BUF_PLANE0_FD_EXT         0x3272
-        #define EGL_DMA_BUF_PLANE0_OFFSET_EXT     0x3273
-        #define EGL_DMA_BUF_PLANE0_PITCH_EXT      0x3274
-        #define EGL_LINUX_DMA_BUF_EXT             0x3270
-
-        #define GL_TEXTURE0				0x84C0
-        #define GL_COLOR_ATTACHMENT1              0x8CE1
-
-        #define FOURCC_NV12 842094158
-
-        if(prime.fourcc == FOURCC_NV12) { // This happens on AMD
-            while(cap_xcomp->egl.glGetError()) {}
-            while(cap_xcomp->egl.eglGetError() != EGL_SUCCESS){}
-
-            EGLImage images[2];
-            cap_xcomp->egl.glGenTextures(2, cap_xcomp->target_textures);
-            assert(cap_xcomp->egl.glGetError() == 0);
-            for(int i = 0; i < 2; ++i) {
-                const uint32_t formats[2] = { fourcc('R', '8', ' ', ' '), fourcc('G', 'R', '8', '8') };
-                const int layer = i;
-                const int plane = 0;
-
-                const intptr_t img_attr[] = {
-                    EGL_LINUX_DRM_FOURCC_EXT,   formats[i],
-                    EGL_WIDTH,                  prime.width / (1 + i), // half size
-                    EGL_HEIGHT,                 prime.height / (1 + i), // for chroma
-                    EGL_DMA_BUF_PLANE0_FD_EXT,  prime.objects[prime.layers[layer].object_index[plane]].fd,
-                    EGL_DMA_BUF_PLANE0_OFFSET_EXT,  prime.layers[layer].offset[plane],
-                    EGL_DMA_BUF_PLANE0_PITCH_EXT,  prime.layers[layer].pitch[plane],
-                    EGL_NONE
-                };
-                images[i] = cap_xcomp->egl.eglCreateImage(cap_xcomp->egl.egl_display, 0, EGL_LINUX_DMA_BUF_EXT, NULL, img_attr); // TODO: Cleanup at the end of this for loop
-                assert(images[i]);
-                assert(cap_xcomp->egl.eglGetError() == EGL_SUCCESS);
-
-                //cap_xcomp->egl.glActiveTexture(GL_TEXTURE0 + i);
-                cap_xcomp->egl.glBindTexture(GL_TEXTURE_2D, cap_xcomp->target_textures[i]);
-                assert(cap_xcomp->egl.glGetError() == 0);
-
-                cap_xcomp->egl.glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-                cap_xcomp->egl.glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
-                cap_xcomp->egl.glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
-                cap_xcomp->egl.glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
-                assert(cap_xcomp->egl.glGetError() == 0);
-
-                cap_xcomp->egl.glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, images[i]);
-                assert(cap_xcomp->egl.glGetError() == 0);
-                assert(cap_xcomp->egl.eglGetError() == EGL_SUCCESS);
-            }
-            //cap_xcomp->egl.glActiveTexture(GL_TEXTURE0);
-            cap_xcomp->egl.glBindTexture(GL_TEXTURE_2D, 0);
 
+        //vaBeginPicture(cap_xcomp->va_dpy, )
 
+        va_status = vaCreateConfig(cap_xcomp->va_dpy, VAProfileNone, VAEntrypointVideoProc, NULL, 0, &cap_xcomp->config_id);
+        if(va_status != VA_STATUS_SUCCESS) {
+            fprintf(stderr, "vaCreateConfig failed: %d\n", va_status);
+            abort();
+            return;
+        }
 
-            cap_xcomp->egl.glGenFramebuffers(1, &cap_xcomp->framebuffer_y);
-            cap_xcomp->egl.glBindFramebuffer(GL_FRAMEBUFFER, cap_xcomp->framebuffer_y);
-
-            cap_xcomp->egl.glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, cap_xcomp->target_textures[0], 0);
-
-            // Set the list of draw buffers.
-            unsigned int DrawBuffers[1] = {GL_COLOR_ATTACHMENT0};
-            cap_xcomp->egl.glDrawBuffers(1, DrawBuffers); // "1" is the size of DrawBuffers
-
-            if(cap_xcomp->egl.glCheckFramebufferStatus(GL_FRAMEBUFFER) != GL_FRAMEBUFFER_COMPLETE) {
-                fprintf(stderr, "Failed to setup framebuffer\n");
-                return;
-            }
-
-            cap_xcomp->egl.glBindFramebuffer(GL_FRAMEBUFFER, 0);
-
-            cap_xcomp->egl.glGenFramebuffers(1, &cap_xcomp->framebuffer_uv);
-            cap_xcomp->egl.glBindFramebuffer(GL_FRAMEBUFFER, cap_xcomp->framebuffer_uv);
-
-            cap_xcomp->egl.glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, cap_xcomp->target_textures[1], 0);
-
-            // Set the list of draw buffers.
-            cap_xcomp->egl.glDrawBuffers(1, DrawBuffers); // "1" is the size of DrawBuffers
-
-            if(cap_xcomp->egl.glCheckFramebufferStatus(GL_FRAMEBUFFER) != GL_FRAMEBUFFER_COMPLETE) {
-                fprintf(stderr, "Failed to setup framebuffer\n");
-                return;
-            }
-
-            cap_xcomp->egl.glBindFramebuffer(GL_FRAMEBUFFER, 0);
-
-
-            cap_xcomp->shader_y = LoadShadersY(cap_xcomp);
-            cap_xcomp->shader_uv = LoadShadersUV(cap_xcomp);
-
-            float vVertices[] = {
-                -1.0f,  1.0f,  0.0f, 1.0f,
-                -1.0f, -1.0f,  0.0f, 0.0f,
-                1.0f, -1.0f,  1.0f, 0.0f,
-
-                -1.0f,  1.0f,  0.0f, 1.0f,
-                1.0f, -1.0f,  1.0f, 0.0f,
-                1.0f,  1.0f,  1.0f, 1.0f
-            };
-
-            unsigned int quadVBO;
-            cap_xcomp->egl.glGenVertexArrays(1, &cap_xcomp->vao);
-            cap_xcomp->egl.glGenBuffers(1, &quadVBO);
-            cap_xcomp->egl.glBindVertexArray(cap_xcomp->vao);
-            cap_xcomp->egl.glBindBuffer(GL_ARRAY_BUFFER, quadVBO);
-            cap_xcomp->egl.glBufferData(GL_ARRAY_BUFFER, sizeof(vVertices), &vVertices, GL_STATIC_DRAW);
-
-            cap_xcomp->egl.glEnableVertexAttribArray(0);
-            cap_xcomp->egl.glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, 4 * sizeof(float), (void*)0);
-
-            cap_xcomp->egl.glEnableVertexAttribArray(1);
-            cap_xcomp->egl.glVertexAttribPointer(1, 2, GL_FLOAT, GL_FALSE, 4 * sizeof(float), (void*)(2 * sizeof(float)));
-
-            cap_xcomp->egl.glBindVertexArray(0);
-        } else {
-            fprintf(stderr, "unexpected fourcc: %u, expected nv12\n", prime.fourcc);
+        VASurfaceID target_surface_id = (uintptr_t)(*frame)->data[3];
+        va_status = vaCreateContext(cap_xcomp->va_dpy, cap_xcomp->config_id, xx, yy, VA_PROGRESSIVE, &target_surface_id, 1, &cap_xcomp->context_id);
+        if(va_status != VA_STATUS_SUCCESS) {
+            fprintf(stderr, "vaCreateContext failed: %d\n", va_status);
+            abort();
             return;
         }
 
@@ -542,34 +324,47 @@ static bool gsr_capture_xcomposite_drm_should_stop(gsr_capture *cap, bool *err)
 
 static int gsr_capture_xcomposite_drm_capture(gsr_capture *cap, AVFrame *frame) {
     gsr_capture_xcomposite_drm *cap_xcomp = cap->priv;
-    vec2i source_size = cap_xcomp->texture_size;
 
-    cap_xcomp->egl.glBindVertexArray(cap_xcomp->vao);
-    cap_xcomp->egl.glViewport(0, 0, source_size.x, source_size.y);
-    cap_xcomp->egl.glBindTexture(GL_TEXTURE_2D, window_texture_get_opengl_texture_id(&cap_xcomp->window_texture));
+    VASurfaceID target_surface_id = (uintptr_t)frame->data[3];
 
-    {
-        cap_xcomp->egl.glBindFramebuffer(GL_FRAMEBUFFER, cap_xcomp->framebuffer_y);
-        //cap_xcomp->egl.glClear(GL_COLOR_BUFFER_BIT);
+    VAStatus va_status = vaBeginPicture(cap_xcomp->va_dpy, cap_xcomp->context_id, target_surface_id);
+    if(va_status != VA_STATUS_SUCCESS) {
+        fprintf(stderr, "vaBeginPicture failed: %d\n", va_status);
+        abort();
+        return 1;
+    }
 
-        cap_xcomp->egl.glUseProgram(cap_xcomp->shader_y);
-        cap_xcomp->egl.glDrawArrays(GL_TRIANGLES, 0, 6);
+    VAProcPipelineParameterBuffer params = {0};
+    params.surface = cap_xcomp->input_surface;
+    params.surface_region = NULL;
+    params.output_background_color = 0xFF000000;
+    params.filter_flags = VA_FRAME_PICTURE;
+    // TODO: Colors
+
+    VABufferID buffer_id = 0;
+    va_status = vaCreateBuffer(cap_xcomp->va_dpy, cap_xcomp->context_id, VAProcPipelineParameterBufferType, sizeof(params), 1, &params, &buffer_id);
+    if(va_status != VA_STATUS_SUCCESS) {
+        fprintf(stderr, "vaCreateBuffer failed: %d\n", va_status);
+        return 1;
     }
 
-    {
-        cap_xcomp->egl.glBindFramebuffer(GL_FRAMEBUFFER, cap_xcomp->framebuffer_uv);
-        //cap_xcomp->egl.glClear(GL_COLOR_BUFFER_BIT);
+    va_status = vaRenderPicture(cap_xcomp->va_dpy, cap_xcomp->context_id, &buffer_id, 1);
+    if(va_status != VA_STATUS_SUCCESS) {
+        fprintf(stderr, "vaRenderPicture failed: %d\n", va_status);
+        return 1;
+    }
 
-        cap_xcomp->egl.glUseProgram(cap_xcomp->shader_uv);
-        cap_xcomp->egl.glDrawArrays(GL_TRIANGLES, 0, 6);
+    va_status = vaEndPicture(cap_xcomp->va_dpy, cap_xcomp->context_id);
+    if(va_status != VA_STATUS_SUCCESS) {
+        fprintf(stderr, "vaEndPicture failed: %d\n", va_status);
+        return 1;
     }
 
-    cap_xcomp->egl.glBindVertexArray(0);
-    cap_xcomp->egl.glUseProgram(0);
-    cap_xcomp->egl.glBindTexture(GL_TEXTURE_2D, 0);
-    cap_xcomp->egl.glBindFramebuffer(GL_FRAMEBUFFER, 0);
+    // TODO: Needed?
+    //vaSyncSurface(cap_xcomp->va_dpy, target_surface_id);
 
-    cap_xcomp->egl.eglSwapBuffers(cap_xcomp->egl.egl_display, cap_xcomp->egl.egl_surface);
+    // TODO: Remove
+    //cap_xcomp->egl.eglSwapBuffers(cap_xcomp->egl.egl_display, cap_xcomp->egl.egl_surface);
 
     return 0;
 }