On 04/22/2011 08:10 PM, Loren Merritt wrote:

> On Fri, 22 Apr 2011, Justin Ruggles wrote:
> 
>> Could someone test the patch below with a modern Intel CPU other than Atom?
>> I'm getting slower results for the SSE version than the MMX version on
>> Athlon64, but SSE is faster on Atom.  I'm guessing it's another Athlon issue,
>> but it could be something else...
> 
> penryn:
> 3656 +/- 3  6ch c
> 1475 +/- 3  6ch mmx
> 1516 +/- 1  6ch sse
> 1091 +/- 5  6ch sse patched
>   908 +/- 1  2ch c
>   468 +/- 2  2ch mmx
>   314 +/- 1  2ch sse
> 
> conroe:
> 3221 +/- 6  6ch c
> 1295 +/- 5  6ch mmx
> 1672 +/- 3  6ch sse
> 1010 +/- 5  6ch sse patched
>   818 +/- 1  2ch c
>   403 +/- 2  2ch mmx
>   296 +/- 3  2ch sse
> 
> I don't know why conroe got lower totals than penryn, when every single 
> instruction is equal or slower in isolation.


awesome. thanks for the patch too!
here are my athlon64 and atom benchmarks w/ your patch included

athlon64:
5391  6ch c
2095  6ch mmx
2106  6ch sse
1067  2ch c
 643  2ch mmx
 636  2ch sse

atom:
7288  6ch c
4568  6ch mmx
2796  6ch sse
2884  2ch c
1413  2ch mmx
 715  2ch sse

The athlon64 speeds for mmx vs sse versions are close enough now that we
don't really need a special case, and sse is significantly faster on intel.

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

Reply via email to