[FFmpeg-devel] [RFC] avx2 and scalarproduct

Ronald S. Bultje rsbultje at gmail.com
Sun Oct 11 12:50:30 CEST 2015


Hi Kieran,

On Sun, Oct 11, 2015 at 5:02 AM, Kieran Kunhya <kierank at obe.tv> wrote:

> On 9 Oct 2015 10:13 pm, "Paul B Mahol" <onemda at gmail.com> wrote:
> >
> > diff --git a/libavcodec/x86/audiodsp.asm b/libavcodec/x86/audiodsp.asm
> > index 3ffb27f..246e945 100644
> > --- a/libavcodec/x86/audiodsp.asm
> > +++ b/libavcodec/x86/audiodsp.asm
> > @@ -41,7 +41,14 @@ cglobal scalarproduct_int16, 3,3,3, v1, v2, order
> >      add     orderq, mmsize*2
> >      jl .loop
> >      HADDD   m2, m0
> > +%if cpuflag(avx2)
> > +    movd   eax, xm2
> > +%if mmsize > 16
> > +    xor   eax, eax
>
> huh?


This was some test code discussed on IRC, Paul seems to have an issue where
mmsize=16 after INIT_YMM avx2, or something like that.

Ronald


More information about the ffmpeg-devel mailing list