[FFmpeg-devel] [PATCH 2/2] avcodec/x86/mpegenc: support transpose permuation type

James Darnley jdarnley at obe.tv
Fri Jun 16 20:32:18 EEST 2017


On 2017-06-16 17:48, Michael Niedermayer wrote:
> On Fri, Jun 16, 2017 at 03:53:28PM +0200, James Darnley wrote:
>> ---
>>  libavcodec/x86/mpegvideoenc_template.c | 47 +++++++++++++++++++++++++++++++++-
>>  1 file changed, 46 insertions(+), 1 deletion(-)
> 
> Are these 2 with the other patchset intended to give exactly the same
> output ?
> 
> for example: (with the 2 patches and the previously problematic one)
> ./ffmpeg -i ~/tickets/5311/TOON.AVI -flags +bitexact file.avi
> ./ffmpeg-try -i ~/tickets/5311/TOON.AVI -flags +bitexact file-new.avi
> 
> -rw-r----- 1 michael michael 9124822 Jun 16 17:45 file.avi
> -rw-r----- 1 michael michael 9124874 Jun 16 17:46 file-new.avi

Thank you.  The intention was that all 8 patches should give identical
output to either the C or the MMX or both.  Ronald says he has modeled
his "hacks" based on the C because it is much clearer than the MMX.

Barring any remaining bugs in my/our new code we would expect it to be
identical to the C.  With this sample I do infarct get identical
results.  For you I think these should give identical results
> ./ffmpeg-try -i ~/tickets/5311/TOON.AVI -flags +bitexact -idct simple file.avi
> ./ffmpeg-try -i ~/tickets/5311/TOON.AVI -flags +bitexact -idct simpleauto file.avi

The first should use the C and the second should use the new SSE2 (or
AVX depending on your CPU).

This could be an issue with the MMX function raising its head.  MMX was
the old default choice.  Ronald and I are investigating.



More information about the ffmpeg-devel mailing list