[XviD-devel] [PATCH] Faster direct/interpolate bvop block interpolation

[Edouard Gomez]

Hey hey,

as promised yesterday, here is a preleminary patch for faster
direct/interpolation bvop block interpolation.

I striped down the file referenced in:
http://edu.bnhof.de/pipermail/xvid-devel/2002-August/000363.html

To keep only 8x8 add versions. This is just to validate the
idea and see how much i could expect from this trick.

The current patch:
http://ed.gomez.free.fr/vrac/xvid-patches/faster-bvop-interpolation.diff

implements a C backend and a mmx-ext backend (usage of pavgb).
So everyone using a >=pII or >K6-3dnow2 CPUs is invited to
test.

First results were good, but i expected much more from this :-(
BENCHMARKs: VC: 148,335s <-- new code
BENCHMARKs: VC: 168,374s <-- old code
(12% improvement)
(BENCHMARKs: VC:  89,375s <-- ffmpeg)

As i'm using a ck kernel, these numbers are maybe not exact,
so if another dev could benchmark the patch... it'll be great.

My next task is to make sure the C backend is not buggy, i
haven't tested it yet.

-- 
Edouard Gomez