Seemingly renderD128 works where card0 does not.

[nageru] / h264encode.cpp

diff --git a/h264encode.cpp b/h264encode.cpp
index 96e0165746ef71928ca1ca3b72dfd3bccf877fb4..a478c2a1a2459a2fa24d3ed06c34a598ea1bf8d1 100644 (file)
--- a/h264encode.cpp
+++ b/h264encode.cpp
@@ -205,7 +205,6 @@ private:
                EGLImage y_egl_image, cbcr_egl_image;
 
                // Only if use_zerocopy == false.
-               RefCountedGLsync readback_done_fence;
                GLuint pbo;
                uint8_t *y_ptr, *cbcr_ptr;
                size_t y_offset, cbcr_offset;
@@ -925,7 +924,7 @@ int H264EncoderImpl::init_va(const string &va_display)
     
     if (support_encode == 0) {
         printf("Can't find VAEntrypointEncSlice for H264 profiles. If you are using a non-Intel GPU\n");
-        printf("but have one in your system, try launching Nageru with --va-display /dev/dri/card0\n");
+        printf("but have one in your system, try launching Nageru with --va-display /dev/dri/renderD128\n");
         printf("to use VA-API against DRM instead of X11.\n");
         exit(1);
     } else {
@@ -1112,10 +1111,10 @@ int H264EncoderImpl::setup_encode()
             glBindBuffer(GL_PIXEL_PACK_BUFFER, gl_surfaces[i].pbo);
             glBufferStorage(GL_PIXEL_PACK_BUFFER, frame_width * frame_height * 2, nullptr, GL_MAP_READ_BIT | GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT);
             uint8_t *ptr = (uint8_t *)glMapBufferRange(GL_PIXEL_PACK_BUFFER, 0, frame_width * frame_height * 2, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
-            gl_surfaces[i].y_ptr = ptr;
-            gl_surfaces[i].cbcr_ptr = ptr + frame_width * frame_height;
             gl_surfaces[i].y_offset = 0;
             gl_surfaces[i].cbcr_offset = frame_width * frame_height;
+            gl_surfaces[i].y_ptr = ptr + gl_surfaces[i].y_offset;
+            gl_surfaces[i].cbcr_ptr = ptr + gl_surfaces[i].cbcr_offset;
             glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
         }
     }
@@ -1950,7 +1949,12 @@ void H264EncoderImpl::encode_frame(H264EncoderImpl::PendingFrame frame, int enco
                                    int frame_type, int64_t pts, int64_t dts)
 {
        // Wait for the GPU to be done with the frame.
-       glClientWaitSync(frame.fence.get(), 0, 0);
+       GLenum sync_status;
+       do {
+               sync_status = glClientWaitSync(frame.fence.get(), 0, 1000000000);
+               check_error();
+       } while (sync_status == GL_TIMEOUT_EXPIRED);
+       assert(sync_status != GL_WAIT_FAILED);
 
        // Release back any input frames we needed to render this frame.
        frame.input_frames.clear();
@@ -1967,11 +1971,11 @@ void H264EncoderImpl::encode_frame(H264EncoderImpl::PendingFrame frame, int enco
                unsigned char *surface_p = nullptr;
                vaMapBuffer(va_dpy, surf->surface_image.buf, (void **)&surface_p);
 
-               unsigned char *y_ptr = (unsigned char *)surface_p + surf->surface_image.offsets[0];
-               memcpy_with_pitch(y_ptr, surf->y_ptr, frame_width, surf->surface_image.pitches[0], frame_height);
+               unsigned char *va_y_ptr = (unsigned char *)surface_p + surf->surface_image.offsets[0];
+               memcpy_with_pitch(va_y_ptr, surf->y_ptr, frame_width, surf->surface_image.pitches[0], frame_height);
 
-               unsigned char *cbcr_ptr = (unsigned char *)surface_p + surf->surface_image.offsets[1];
-               memcpy_with_pitch(cbcr_ptr, surf->cbcr_ptr, (frame_width / 2) * sizeof(uint16_t), surf->surface_image.pitches[1], frame_height / 2);
+               unsigned char *va_cbcr_ptr = (unsigned char *)surface_p + surf->surface_image.offsets[1];
+               memcpy_with_pitch(va_cbcr_ptr, surf->cbcr_ptr, (frame_width / 2) * sizeof(uint16_t), surf->surface_image.pitches[1], frame_height / 2);
 
                va_status = vaUnmapBuffer(va_dpy, surf->surface_image.buf);
                CHECK_VASTATUS(va_status, "vaUnmapBuffer");