[FFmpeg-devel] [PATCH] x86/vc1dsp: Port vc1_*_hor_16b_shift2 to NASM format

[Timothy Gu]

On Sun, Feb 14, 2016 at 01:21:57PM +0100, Christophe Gisquet wrote:
> Hi,
> 
> 2016-02-14 6:49 GMT+01:00 Timothy Gu <timothygu99 at gmail.com>:
> >  %if HAVE_MMX_INLINE
> 
> Isn't that macro meant for C code (and in config.asm without much of a
> purpose)?

Yes, but this code isn't used unless inline asm is enabled so I don't want to
fill the binary up with bloat.

> >  ; Compute the rounder 32-r or 8-r and unpacks it to m7
> >  %macro LOAD_ROUNDER_MMX 1 ; round
> > -    movd      m7, %1
> > +    movh      m7, %1
> 
> Same here: until there's a SSE2 version for some of those functions,
> which will require specific code (like a shufps), that's unrelated.

Okay, removed.

> 
> > +cglobal vc1_%2_hor_16b_shift2, 4, 5, 0, dst, stride, src, rnd, h
> 
> Do you intend to port more?

Yes, when I get the time.

> Because in that case, having a SSE2
> version of this one would be possible.
> And the wider version would be straightforward in SSE2.
> And that would allow you to put at last a (c) in this code you've
> spent so much time with.

I do intend on writing a SSE2 version but I'd rather port them all first.

> 
> Otherwise looks OK, but I haven't tested.

FATE passes, so pushed.

Timothy