Hi,

as you can see, our fate is yellow since I committed H264/10bit. While
debugging, I found that h264pred appears to have some aliasing
problems, in the form of:
(pixel4=uint64_t)

((pixel4 *) dst)[x] = ((pixel4 *) src)[x];

... and this is partially responsible for the fate failures. Basically
what happens is that very ocasionally, some pixels are simply not
stored in memory. Unfortunately I didn't get to test this patch
because Kostylev's machine died while I was debugging the issue. Seems
to me that replacing each occurrence of the above with something like
AV_WN4PA(dst, AV_RN4PA(src)); should fix it. This patch doesn't break
10bit H264 on my system. It may fix fate on the currently failing
machines, not sure...

Ronald

Attachment: alias.patch
Description: Binary data

_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to