Re: [PATCH] Convert vector_fmul to yasm and add AVX

On 04/07/2012 07:10 PM, Justin Ruggles wrote:

> On 04/07/2012 07:02 PM, Kieran Kunhya wrote:
> 
>> Yes it was intentional, but if you want to keep it that's of course fine.
> 
> 
> I'll test it on Athlon64 to see if it's faster than SSE.

The speeds are nearly identical.

Best of 5 runs:
SSE:   190 cycles
3DNow: 180 cycles

Mean of the 3 median runs out of 5 runs:
SSE:   212 cycles
3DNow: 216 cycles

I'm ok with not keeping the 3DNow version of this function.

-Justin