Hi, On Wed, Feb 15, 2012 at 1:54 PM, Martin Storsjö <[email protected]> wrote: > On Wed, 15 Feb 2012, Martin Storsjö wrote: > >> On Wed, 15 Feb 2012, Kostya Shishkov wrote: >> >>> Module: libav >>> Branch: master >>> Commit: 1a265f6187e9036b649fc08f8293b5e1fcd8dfbe >>> >>> Author: Kostya Shishkov <[email protected]> >>> Committer: Kostya Shishkov <[email protected]> >>> Date: Thu Feb 2 20:54:53 2012 +0100 >>> >>> prores encoder >>> >>> --- >>> >>> diff --git a/libavcodec/proresdsp.h b/libavcodec/proresdsp.h >>> index 8b864fa..f657825 100644 >>> --- a/libavcodec/proresdsp.h >>> +++ b/libavcodec/proresdsp.h >>> @@ -30,7 +30,10 @@ >>> typedef struct { >>> int idct_permutation_type; >>> uint8_t idct_permutation[64]; >>> + int dct_permutation_type; >>> + uint8_t dct_permutation[64]; >>> void (* idct_put) (uint16_t *out, int linesize, DCTELEM *block, const >>> int16_t *qmat); >>> + void (* fdct) (const uint16_t *src, int linesize, DCTELEM *block); >>> } ProresDSPContext; >> >> >> This seems to break fate-prores-422 and fate-prores-422_hq on my laptop >> (OS X 64 bit with gcc 4.2.1). But if built without optimizations, it works. >> The fate configuration that is similar to this seems to work though. >> >> If I comment out the dct_permutation array, it works. Does this indicate >> it might be some overread somewhere? Also, if I disable using >> ff_prores_idct_put_10_sse2, it also works. > > > I'm not able to reproduce this any longer. Either something in my build was > broken, or it's a heisenbug.
Sounds to me like a missing dependency may have caused your x86 dsp function init file to not b rebuilt, causing it to assign wrong functions to wrong addresses and random crap happening. May imply a build system bug or so (missing dependency somewhere). Ronald _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
