On 05/27/2011 10:34 AM, Ronald S. Bultje wrote:

> Hi,
> 
> On Fri, May 27, 2011 at 10:33 AM, Ronald S. Bultje <[email protected]> wrote:
>> On Sun, May 15, 2011 at 2:40 PM, Justin Ruggles
>> <[email protected]> wrote:
>> [..]
>>> +    movzx    tmp2d, word [mant_cntq+4]
>>> +    imul     tmp2d, 43691
>>> +    shr      tmp2d, 17
>>
>> Can you use pmulhw here? I'm thinking of loading 4x2 words at once in
>> a SSE2 register, pmuluw by a register holding 8x word 43691, then
>> right-shift by 1.
> 
> Can't type today, I mean pmulhuw.


Hmmm... I didn't consider that. The low 16 bits are shifted out anyway.
I think I can rewrite the whole thing now to do something similar to the
sse4.1 version, but I don't think it will need to use sse4.1 since it
can operate on words rather than dwords.

Thanks,
Justin
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to