[FFmpeg-devel] [PATCH] mmx implementation of vc-1 inverse transformations

Benoit Fouet benoit.fouet
Tue Jul 8 11:24:15 CEST 2008


Michael Niedermayer wrote:
> On Thu, Jul 03, 2008 at 02:51:18PM +0200, Victor Pollex wrote:
>   
>> Michael Niedermayer schrieb:
>>     
>>> On Sat, Jun 28, 2008 at 12:31:41PM +0200, Victor Pollex wrote:
>>>       
> [...]
>   
>>> [...]
>>>
>>>   
>>>       
>>>> +#define LOAD_ADD_CLAMP_STORE_8X1(io,t0,t1,r0,r1)\
>>>> +    "movq      "#io", "#t0"\n\t"\
>>>> +    "movq      "#t0", "#t1"\n\t"\
>>>> +    "punpcklbw %%mm7, "#t0"\n\t"\
>>>> +    "punpckhbw %%mm7, "#t1"\n\t"\
>>>> +    "paddw     "#r0", "#t0"\n\t"\
>>>> +    "paddw     "#r1", "#t1"\n\t"\
>>>> +    "packuswb  "#t1", "#t0"\n\t"\
>>>> +    "movq      "#t0", "#io"\n\t"
>>>>     
>>>>         
>>> some of the movq seem redundant
>>>   
>>>       
>> I'm sorry but I don' see it. Although this is now obsolete, I'd still like 
>> to know which one is redundant and why.
>>     
>
> i thought io was a register, if it is memory then things are different
> sorry
>
>
> [...]
>
> the LOAD4/STORE4 patch is ok
>
> [...]
>
>   

Applied this part

-- 
Benoit Fouet
Purple Labs S.A.
www.purplelabs.com




More information about the ffmpeg-devel mailing list