[FFmpeg-devel] [PATCH] NEON add_signed_pixels_clamped
Sat Apr 4 15:27:19 CEST 2009
David Conrad <lessen42 at gmail.com> writes:
> 3% overall wmv3 decoding speedup.
> Also, is it possible to have something like
> .macro reg=2
> evaluate to d4? Or any other ideas to put the repeated sections in a
> macro that isn't ugly?
I don't think there is a simple way, but read on.
The attached version is 1 cycle faster on Cortex-A8, 5 cycles faster
on A9. Unfortunately it's somewhat more difficult to read than the
original. That's the price you pay for speed.
mans at mansr.com
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 3313 bytes
Desc: not available
More information about the ffmpeg-devel