[MPlayer-dev-eng] AMD64 assembler support

[Aurelien Jacobs]

On Sun, 31 Jul 2005 19:15:47 +0200
Reimar Döffinger <Reimar.Doeffinger at stud.uni-karlsruhe.de> wrote:

> Hi,
> On Sun, Jul 31, 2005 at 06:11:55PM +0200, Aurelien Jacobs wrote:
> > I've reworked your patch (cleaned, fixed...).
> > It seems pretty good now, so I will probably commit.
> [...]
> > +# define REG_si "rsi"
> > +# define REG_di "rdi"
> 
> This is still not consistent, it's REG_S and REG_D everywhere else.

Right.

> > +# define REG_bp "rbp"
> 
> That one's REG_BP everywhere else (well, cpudetect.h actually).

Right.

> > -#ifdef ARCH_X86
> > +#if defined ARCH_X86 || defined ARCH_X86_64
> 
> hmm.. I think in other places we use brackets:
> #if defined(ARCH_X86) || defined(ARCH_X86_64)

Right.

> > -static uint64_t attribute_used __attribute__((aligned(8)))
> > magicF2W= 0x43c0000043c00000LL; -static uint64_t attribute_used
> > __attribute__((aligned(8))) wm1010= 0xFFFF0000FFFF0000LL; -static
> > uint64_t attribute_used __attribute__((aligned(8))) wm0101=
> > 0x0000FFFF0000FFFFLL; -static uint64_t attribute_used
> > __attribute__((aligned(8))) wm1100= 0xFFFFFFFF00000000LL; +#include
> > "a52_internal.h" +
> > +
> > +static uint64_t attribute_used __attribute__((aligned(8)))
> > magicF2W= 0x43c0000043c00000ULL; +static uint64_t attribute_used
> > __attribute__((aligned(8))) wm1010= 0xFFFF0000FFFF0000ULL; +static
> > uint64_t attribute_used __attribute__((aligned(8))) wm0101=
> > 0x0000FFFF0000FFFFULL; +static uint64_t attribute_used
> > __attribute__((aligned(8))) wm1100= 0xFFFFFFFF00000000ULL;
> 
> Why this LL -> ULL change?

I don't know. That was in the original patch. But indeed that's not
needed. So it's removed from my new patch.

> > Index: liba52/downmix.c
> > ===================================================================
> > RCS file: /cvsroot/mplayer/main/liba52/downmix.c,v
> > retrieving revision 1.17
> > diff -u -1 -r1.17 downmix.c
> > --- liba52/downmix.c	22 Mar 2005 23:27:18 -0000	1.17
> > +++ liba52/downmix.c	31 Jul 2005 15:57:18 -0000
> > @@ -60,4 +60,4 @@
> >      if(mm_accel & MM_ACCEL_X86_MMX) upmix= upmix_MMX;
> > -    if(mm_accel & MM_ACCEL_X86_SSE) downmix= downmix_SSE;
> >      if(mm_accel & MM_ACCEL_X86_3DNOW) downmix= downmix_3dnow;
> > +    if(mm_accel & MM_ACCEL_X86_SSE) downmix= downmix_SSE;
> 
> That's unrelated isn't it? Also it _might_ be intended...

Yes, it is. That's why it's in a separate patch !
That's only for benchmarking 3dnow vs SSE on CPU which support both,
so that we can decide which one to prefer.

Aurel
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: liba52_amd64.diff
URL: <http://lists.mplayerhq.hu/pipermail/mplayer-dev-eng/attachments/20050731/2330e8f6/attachment.txt>