On 10/06/16 08:06, Anton Khirnov wrote: > Quoting Mark Thompson (2016-06-10 01:25:58) >> Adds a general mechanism for setting driver quirks, and detects the >> driver name to set them. >> >> When the i965 driver is being used, destroy all parameter buffers >> explicitly after use despite passing them to vaRenderPicture(). >> --- >> An alternative approach to solving the same problem as the previous patch. >> Maybe this is actually better once we start detecting the driver anyway? > > Yes, this looks much better to me.
Hmm, ok. Thinking a bit further, maybe this should be available a bit more generally because <https://git.libav.org/?p=libav.git;a=blob;f=libavcodec/vaapi.c;hb=HEAD#l180> should also use it - maybe a property of the device rather than inside the encoder? A flags field in AVVAAPIDeviceContext would seem to do the right thing, but it is unclear who would initialise it - the user, or inside av_hwdevice_ctx_init(). I should probably acquire a suitable AMD card so that I can actually test this on more than one target. > Btw, do the intel driver devs know about this problem? Yes. I don't know if it's been considered recently, but see <https://lists.freedesktop.org/archives/libva/2012-July/000961.html> for some hand-wringing a while ago. Also compare Intel's libyami behaviour: <https://github.com/01org/libyami/blob/0aef8be5122f7ef917258aafcdb44a22bdb0c1c8/vaapi/vaapipicture.cpp#L65> <https://github.com/01org/libyami/blob/0aef8be5122f7ef917258aafcdb44a22bdb0c1c8/vaapi/VaapiBuffer.cpp#L86> with use in the official encoding example: <https://cgit.freedesktop.org/libva/tree/test/encode/h264encode.c#n1764>. - Mark _______________________________________________ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel