Re: [libav-devel] [PATCH] lavr: replace the SSE version of ff_conv_fltp_to_flt_6ch() with AVX

2012-05-08 Thread Justin Ruggles
On 05/08/2012 04:30 PM, Vitor Sessak wrote: > On 05/08/2012 09:59 PM, Justin Ruggles wrote: >> On 05/07/2012 11:40 AM, Vitor Sessak wrote: >>> On 05/07/2012 04:29 PM, Justin Ruggles wrote: On 05/07/2012 08:18 AM, Vitor Sessak wrote: > On 05/03/2012 09:39 PM, Justin Ruggles wrote: >> Th

Re: [libav-devel] [PATCH] lavr: replace the SSE version of ff_conv_fltp_to_flt_6ch() with AVX

2012-05-08 Thread Vitor Sessak
On 05/08/2012 09:59 PM, Justin Ruggles wrote: On 05/07/2012 11:40 AM, Vitor Sessak wrote: On 05/07/2012 04:29 PM, Justin Ruggles wrote: On 05/07/2012 08:18 AM, Vitor Sessak wrote: On 05/03/2012 09:39 PM, Justin Ruggles wrote: The SSE version is slower than the MMX version on Athlon64 and Sand

Re: [libav-devel] [PATCH] lavr: replace the SSE version of ff_conv_fltp_to_flt_6ch() with AVX

2012-05-08 Thread Justin Ruggles
On 05/07/2012 11:40 AM, Vitor Sessak wrote: > On 05/07/2012 04:29 PM, Justin Ruggles wrote: >> On 05/07/2012 08:18 AM, Vitor Sessak wrote: >>> On 05/03/2012 09:39 PM, Justin Ruggles wrote: The SSE version is slower than the MMX version on Athlon64 and Sandy Bridge, but the 3-arg AVX

Re: [libav-devel] [PATCH] lavr: replace the SSE version of ff_conv_fltp_to_flt_6ch() with AVX

2012-05-07 Thread Vitor Sessak
On 05/07/2012 04:29 PM, Justin Ruggles wrote: On 05/07/2012 08:18 AM, Vitor Sessak wrote: On 05/03/2012 09:39 PM, Justin Ruggles wrote: The SSE version is slower than the MMX version on Athlon64 and Sandy Bridge, but the 3-arg AVX version is faster on Sandy Bridge. --- Athlon64 C - 26322 MM

Re: [libav-devel] [PATCH] lavr: replace the SSE version of ff_conv_fltp_to_flt_6ch() with AVX

2012-05-07 Thread Justin Ruggles
On 05/07/2012 08:18 AM, Vitor Sessak wrote: > On 05/03/2012 09:39 PM, Justin Ruggles wrote: >> The SSE version is slower than the MMX version on Athlon64 and Sandy >> Bridge, >> but the 3-arg AVX version is faster on Sandy Bridge. >> --- >> Athlon64 >>C - 26322 >> MMX - 7997 >> SSE - 8662 >>

Re: [libav-devel] [PATCH] lavr: replace the SSE version of ff_conv_fltp_to_flt_6ch() with AVX

2012-05-07 Thread Vitor Sessak
On 05/03/2012 09:39 PM, Justin Ruggles wrote: The SSE version is slower than the MMX version on Athlon64 and Sandy Bridge, but the 3-arg AVX version is faster on Sandy Bridge. --- Athlon64 C - 26322 MMX - 7997 SSE - 8662 Sandy Bridge C - 13011 MMX - 4262 SSE - 4680 AVX - 3960 Note th

Re: [libav-devel] [PATCH] lavr: replace the SSE version of ff_conv_fltp_to_flt_6ch() with AVX

2012-05-06 Thread Justin Ruggles
On 05/03/2012 03:39 PM, Justin Ruggles wrote: > The SSE version is slower than the MMX version on Athlon64 and Sandy Bridge, > but the 3-arg AVX version is faster on Sandy Bridge. > --- > Athlon64 > C - 26322 > MMX - 7997 > SSE - 8662 > > Sandy Bridge > C - 13011 > MMX - 4262 > SSE - 4680

Re: [libav-devel] [PATCH] lavr: replace the SSE version of ff_conv_fltp_to_flt_6ch() with AVX

2012-05-03 Thread Justin Ruggles
On 05/03/2012 03:39 PM, Justin Ruggles wrote: > The SSE version is slower than the MMX version on Athlon64 and Sandy Bridge, > but the 3-arg AVX version is faster on Sandy Bridge. > --- > Athlon64 > C - 26322 > MMX - 7997 > SSE - 8662 > > Sandy Bridge > C - 13011 > MMX - 4262 > SSE - 4680

[libav-devel] [PATCH] lavr: replace the SSE version of ff_conv_fltp_to_flt_6ch() with AVX

2012-05-03 Thread Justin Ruggles
The SSE version is slower than the MMX version on Athlon64 and Sandy Bridge, but the 3-arg AVX version is faster on Sandy Bridge. --- Athlon64 C - 26322 MMX - 7997 SSE - 8662 Sandy Bridge C - 13011 MMX - 4262 SSE - 4680 AVX - 3960 Note that although the new code is simpler, using it as S