[FFmpeg-devel] Patch: Inline asm fixes for Intel compiler on Windows
protogonoi at gmail.com
Mon Apr 7 08:29:32 CEST 2014
> this patch breaks fate-vp6a-skip_alpha, fate-vp6f and fate-vp6a
> possibly also others, didnt test with -k
Interesting, I always run a local FATE test before submitting anything and
it worked fine on mine so that would suggest it is compiler dependent.
I think all the in-out parameters must be +& instead of just +.
> Note that this was incorrect even before, but it probably
> shows up only now since now it is trivial for the compiler
> to figure out that c->code_word and code_word are the same thing,
> and without early clobber it can place them onto the same register.
> This of course invalidates all previous benchmarks...
Thanks, I was scratching my head trying to work out what was going on (as
stated previously fate worked for me). I was trying to keep the patch to a
minimum change wise so it never occurred to me to check that the
constraints were correct even in the current git master.
I have updated the constraints and, fate still tests as correct using my
compilers. I reran the benchmarks and got a slight change in performance
but its still 4.6% faster.
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 1434 bytes
Desc: not available
More information about the ffmpeg-devel