probeskip patch

[puffpio]

Some interesting feedback

using sharktooth's HQ-slowest profile, 512x384 29.97 fps encoding to around 900kbps

on a P4 @ 3ghz w/ HT disabled:
Before probeskip patch: 2.5-2.7 fps 1st pass
After probeskip patch: 2.7-3.0 fps 1st pass

On an Athlon XP 3000+ (2ghz):
Before probeskip patch: 2.6-2.7 fps 1st pass
After probeskip patch: 1.8-2.0 fps 1st pass


so speedup with a p4 but speeddown with an athlon xp? I don't have an athlon 64 to compare against...

[puffpio]

i finished the p4 ones, and the athlon xp before probeskip patch ones

i'm running the athlon xp after probeskip patch encode right now...
it appears to be gaining FPS as it goes along...

[akupenguin]

random numbers.

HQ-Slowest still uses turbo, so probeskip does not affect 1st pass speed or quality at all.

[Sirber]

r386 | pengvado | 2005-12-13 12:24:02 +0100 (Tue, 13 Dec 2005) | 4 lines
Changed paths:
M /trunk/encoder/analyse.c

postpone pskip decision until after p16x16ref0 motion search.
reduces the number of erroneous pskips in low-detail regions.

So it's not "safe" to use it currently (rev 385)?

[Sharktooth]

Why not?
rev 385+probeskip = rev 386...

@pengvado: turbo in MeGUI doesnt actually disable no-fast-pskip.
EDIT: now it does.

[Sirber]

so your rev 385 build does... "postpone pskip decision until after p16x16ref0 motion search"?

[Sharktooth]

that's what does the probeskip patch.

[Sirber]

oh, thanks! :D

Sorry for my noob questions :(

[bond]

my findings on probeskip vs other skips:
http://forum.doom9.org/showthread.php?p=748863#post748863