GitLab

explicit match on "profile" field.

and codec setup (probe). This fixes a memory leak (avctx, pctx)
on destroy and most interestingly makes it possible to
detect unsupported codecs.

e.g. VC-1 sequence headers for elementary streams.

- Simplify parsing with an AVCodeParserContext.

Improve gst_vaapi_profile_from_caps() for H.264 & caps with "codec-data".

Otherwise, we might still have to seek into the stream.
i.e. keep the data longer.

several times. This fixes decoding of MPEG-2 and H.264
because those created a GstVaapiContext later through
avcodec_decode_video() that was a protected call.

XXX: only VC-1 decoding works at this time because of awful
bugs left in GstVaapiDecoderFfmpeg et al.

There are tons of bugs left:
- Decoder API not nice enough with error conditions
- FFmpeg parser is sometimes broken
- Packets queue can be lost

and that returns the surface to that context on destruction.

to create another texture (and storage) for the TFP, simply a new texture name.

where TFP+FBO are used, thus avoiding the need fully preserve the states
and call into glGet*() functions that need synchronization.