[FFmpeg-devel] [PATCH] x86/intmath: allow the source operand for icc/icl ff_ctzll to be a memory location

Henrik Gramner henrik at gramner.com
Sun Oct 25 18:53:42 CET 2015


On Sun, Oct 25, 2015 at 5:39 PM, Matt Oliver <protogonoi at gmail.com> wrote:
> Using tzcnt would be the more interesting option. Changing the asm to tzcnt
> works with ICL. But also the use of the tzcnt intrinsic (_tzcnt_u64) would
> be possible with both intel and msvc. I dont have anything older than
> Haswell or Piledriver to double check but I have seen its use in several
> other projects without issue so in theory it shouldnt be a problem.

IIRC tzcnt (rep bsf) is only slower than plain bsf on really ancient
machines, e.g. 80486 and such, so there's not really any downside of
using it.


More information about the ffmpeg-devel mailing list