This week I got the new hardware-accelerated blits for YUV import to GL working again.
The pipeline I have is drm_mmal decoding 360 frames of 1080p Big Buck Bunny trailer using MMAL, importing them to GL as an image_external texture, drawing to the back buffer, and pageflipping.
Switching MMAL from producing linear RGBA buffers to linear NV12 buffers improved FPS by 18.4907% +/- 0.746806% (n=7), and to YV12 by 14.4922% +/- 0.569289%. The improvement is slightly understated, as there’s some fixed overhead of waiting for vcsm to time out to indicate that the stream is complete.
I also polished up Dave’s SAND support patch for KMS, and submitted it. This lets video decode into KMS planes skip a copy of the buffers (I didn’t do performance comparisons of this, though).
Patches are submitted, and the next step will be to implement import of SAND buffers in GL to match the KMS support.