Fastest Avisynth for Athlon XP machine

[ookzDVD]

Dear forum,

There are many AviSynth beta files around here,
but... please someone could pick me the fastest one for
Athlon XP machine. (I think it should be compiled use ICL6)

Thank you.

[Acaila]

ICL = Intel Compiler => Optimized for Intel machines :) (and there's no equivalent for Athlons AFAIK)

I've never noticed any difference in speed between the various Avisynth versions.

[ookzDVD]

@Acaila,

Thank you ;)
yes... ICL6 is optimized for Intel, but since AthlonXP support
MMX and SSE not SSE2 I think ICL6 is the best solution.

[sh0dan]

ICL in general optimizes better than MS VC++, and often a speed increase of 5-10% is the result.
I seem to remember someone trying to compile it with ICL without much success - so there may be some problems.

ICL doesn't generate MMX or SSE - all of that has to be written in assembler.

[Richard Berg]

ICL will generate vectorized code, though it sometimes needs some prodding. Its major restrictions are (1) in order to assemble 3dnow ops you have to put them in a separate file & use MASM (2) price & licensing restrictions in general.

[Didée]

Could it be that the actual 2.02 is much slower than 1.07d ???

Yesterday I ran an avs-script that was originally written for 1.07d through 2.02 ...
There were only some basic filters in it: crop, TomsMoComp, BicubicResize, and DeLogo1.3, requiring YUY2->RGB->YUY2 (Don´t have the exact script handy: I am at work)

Through 1.07d, I got ~7.5 fps

Through 2.02, I got ~3.5 fps

That´s definetly slower, isn´t it?

[Richard Berg]

Can you pinpoint the slowdown to a specific filter? Both versions are OSS so it should be easy to correct any slowdowns.

And yes, I'll be back to hacking on avisynth soon, fear not :)

[Didée]

That should be easy to find out. Alas, I can post the result no earlier than tomorrow afternoon (GMT). At work ATM ...
Could it be that the culprit is using a Virtualdub-filter in general?