Opened 4 years ago
Last modified 3 months ago
#7665 new defect
DJGPP: alignment silently reduced to 16 bytes with questionable reason
Reported by: | michael.kostylev | Owned by: | |
---|---|---|---|
Priority: | important | Component: | avutil |
Version: | git-master | Keywords: | djgpp regression |
Cc: | Blocked By: | ||
Blocking: | Reproduced by developer: | no | |
Analyzed by developer: | no |
Description
Summary of the bug:
Commit 1473afac5d11a0a90810b6cd8107d63640c9fd38 actually broke the
AVX code when the compiler is able to build it. The vast majority of the DJGPP toolchains are unable to maintain even the 16-byte alingment, so the only good workaround for such a toolchain is to use --disable-sse. That in turn could be a a good reason for relaxing alignment requirements.
How to reproduce:
See http://fate.xffm.org/report.cgi?slot=x86_32-freedos-gcc-4.8-avx2&time=20190105220055 The code 255 (SIGSEGV, actually) is caused by reduced alignment, the rest looks as dosemu-related.
Change History (2)
comment:1 by , 4 years ago
Keywords: | djgpp regression added |
---|---|
Priority: | normal → important |
comment:2 by , 3 months ago
Note:
See TracTickets
for help on using tickets.
https://github.com/FFmpeg/FFmpeg/blob/5bad4856035ca5ed571e9d7d9b1d503a5c9ef0a5/libavutil/mem_internal.h#L82
Now.