[FFmpeg-devel] [patch][OpenHEVC]added ASM functions for epel + qpel

Christophe Gisquet christophe.gisquet at gmail.com
Mon Mar 3 20:45:35 CET 2014


Hi,

2014-03-03 20:28 GMT+01:00 Pierre Edouard Lepere
<Pierre-Edouard.Lepere at insa-rennes.fr>:
> for the SBUTTERFLY, it's to have coeff1 and coeff 2 in the same register. the explanation of Ronald is more or less spot on.
> this allows the use of maddubs and add instructions instead of mul and hadd which are more costly IIRC.

It makes sense. It's more about me not noticing it was ssse3 only and
these butterflies were fancy names for mostly unpacks.

> I'll try to find a good coeff to use the pmulhrw instruction instead of the current shifts.

I'm a bit lost here, because I was expecting some kind of rounding,
iirc in the last stage when going from internal 16bits to output
8bits. pmulhrw usually helps doing something like (a+OFFSET)>>s but
there seems to be no OFFSET here, so maybe I was mistaken.

Thanks,
-- 
Christophe


More information about the ffmpeg-devel mailing list