[FFmpeg-devel] [PATCH]: Align branch target for fft_sse.c, fft_3dn.c and fft_3dn2.c

Zuxy Meng zuxy.meng
Sun May 27 08:17:48 CEST 2007


2007/5/27, Loren Merritt <lorenm at u.washington.edu>:
> On Sat, 26 May 2007, Zuxy Meng wrote:
> > This patch aligns the loop entry on 16-byte aligned address, as
> > recommended by Intel's and AMD's manuals. But I don't see any
> > noticeable improvements on my Dothan. Anyone to test it on other CPUs?
> Core2:
> 256-point fft got 11% faster.
> 2048-point fft got 2% slower.
> Total time to decode vorbis is unchanged.

Thanks! Seems that if branch prediction works right, the misaligned
penalty is well hidden in the pipeline, so patch dropped.

Beauty is truth,
While truth is beauty.
PGP KeyID: E8555ED6

More information about the ffmpeg-devel mailing list