[FFmpeg-devel] [PATCH] Faster SSE FFT/MDCT

[Michael Niedermayer]

Hi

On Fri, May 11, 2007 at 04:51:21PM +0800, Zuxy Meng wrote:
> Hi,
> 
> 2007/5/11, Zuxy Meng <zuxy.meng at gmail.com>:
> >Hi,
> >
> >The patch unrolls some loops, utilizing all 8 xmm registers. fft-test
> >shows ~10% speed up in (I)FFT and ~5% speed up in (I)MDCT on my
> >Dothan. Of course with x86-64 we can unroll one more time but I don't
> >have a test bench....
> >
> >Full test passed on x86, and a test on x86-64 would be prudent: I used
> >xmm8 to save a memory access.
> 
> I just unrolled another loop in imdct. Now IMDCT is about ~8% faster
> compared to SVN head. Please ignore the last patch and try this one
> instead:-)

looks ok if it passes regression tests 

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

There will always be a question for which you do not know the correct awnser.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20070512/49e40ff0/attachment.pgp>