[FFmpeg-devel] [PATCH] Altivec vector_fmul_scalar
Tue Jan 6 15:49:00 CET 2009
On Tuesday 06 January 2009, David Conrad wrote:
> ~7-9% faster vorbis, aac, and ac3.
> I have no clue why it's not bitexact to the C version; I tried not
> using the add of madd, and even enabling denormal handling to match
> the C version. The differences are only a very occasional +/- 1 however.
Sorry for barging it, but can you also try to benchmark an Altivec variant of
the following patch (or even benchmark it as-is without Altivec)?
It should be much more useful for PPC platform than for the others because of
1. Vorbis uses "nonnative" bitstream byte order on PPC and UPDATE_CACHE
operations are supposed to be more expensive there. As the patch optimizes
bitstream reading to vastly reduce the number of such operations, benefit on
PPC should be better.
2. As PPC uses intrinsics anyway, it avoids the controversy of mixing C and
inline assembly :)
I'm not expecting replies posted in *this* thread and hijacking it was not my
PS. This is also some kind of ping, asking for review/comments of that old
patch, because it seems to be forgotten now. The performance improvement for
vorbis is too good to be ignored.
More information about the ffmpeg-devel