[x264-devel] Windows x64 support

[BugMaster]

On Sat, 17 Jan 2009 15:44:10 +0000 (UTC), Loren Merritt wrote:
> On Sat, 17 Jan 2009, BugMaster wrote:
>> On Fri, 16 Jan 2009 20:53:03 +0000 (UTC), Loren Merritt wrote:
>>
>>> I think x264_checkasm_call could be simplified if you keep the random
>>> constants in static const memory rather than in %defines.
>>
>> Did you mean something like this (couldn't test it now but it compiles
>> at least):
>> http://stashbox.org/363808/x264_win64_support.02mod.r1080.diff

> That's part of it. The rest is:

> ALIGN 16
> n4:  dq 0xa77809bf11b239d1
> n5:  dq 0x2ba9bf3d2f05b389
> x6:  dq 0x1a1b2550a612b48c, 0x79445c159ce79064
> ...

>      pxor   xmm0, xmm0
> %assign i 6
> %rep 16-6
>      pxor   xmm %+ i, [x %+ i GLOBAL]
>      por    xmm0, xmm %+ i
>      %assign i i+1
> %endrep
>      packsswb xmm0, xmm0
>      movq   r5, xmm0
>      or     r4, r5


> --Loren Merritt
> _______________________________________________
> x264-devel mailing list
> x264-devel at videolan.org
> http://mailman.videolan.org/listinfo/x264-devel

Thanks, I am not very good experienced with SSE instructions (so
"packsswb" hint was very useful).

Here is new patch (could really test it only on Monday):
http://stashbox.org/364601/x264_win64_support.02mod2.r1081.diff

Loren, can you tell me for what "xor  eax, eax" is used before "call
printf" in 32 bit version of x264_checkasm_call. May be I can remove "xor  eax, eax" from
both 32 bit and 64 bit versions?