[FFmpeg-devel] [PATCH 2/2] x86/dsputilenc: implement SSE2 versions of pix_{sum16, norm1}

[Christophe Gisquet]

2014-05-27 18:34 GMT+02:00 James Almer <jamrial at gmail.com>:
> Ah, I did not intend to remove the "and eax, 0xffff" at all. Good thing you
> noticed, thanks.

It's not like removing it would change much anything, but on the hand,
capping the output at 65535 (255*256=65280 so why does it matter?) is
weird.

> Actually, they do use mmx. And good think you  mention this because now i see
> they have been using HADD* macros in mmx functions for a while now.
> Guess nobody tried x264 encoding with pre-SSE2 CPUs to experience a crash.

I think they prefer not to clutter their version with things they
don't care. But you can try and submit them and see whether it is the
case.