X264/AVC SSE3 Rev.221

[cakuhnen]

Dear Friends

I have make this only for test if yours have a Pentium 3.0 or above
you can try this build, but sorry only VFW not CLI, i'm working in CLI
to add more fast compression assembler CPU optimizations but i have big problems, so if yours expect the CLI release wait, i need to fix a bug that Intel Compiler don't compile, unresolved external symbol.

To install, first uninstall the current yours have installed then
extract the contents of rar arquive to a directory and right click in the file x264vfw.inf and select install, this will install the codec in your windows system folder, if XP ask you for not certification (WHQL) simple ignore this and select Yes.
Buid from last revision of X264, test the codec and report in this forum the current speed...
You need any Zip program like Winzip and the 7Zip to extract the arquive.
If you have any site to me to add my buids, thank's, i add P2,P3,P4,P4 SSE3 HT, P4 Extreme Edition buids.

Thank's

cakuhnen

[Sirber]

Is it compiled with SSE3 flag or SSE3 asm? If it's just compiler flags, don't bother.

[JoeBG]

I´m hoping for the CLI version :)

[Sirber]

Originally posted by JoeBG
I´m hoping for the CLI version :)

If you have any site to me to add my buids, thank's, i add P2,P3,P4,P4 SSE3 HT, P4 Extreme Edition buids.

Get one in the sticky. I wouldn't think he made that much optimized assembly code.

[bond]

cakuhnen, i hope we will see the source for your optimisations too soon! remember the gpl needs you to release the source too!

[bill_baroud]

if he added some code though :rolleyes:

P2,P3,P4,P4 SSE3 HT, P4 Extreme Edition

would mean :
P2 > added MMX (already here ?)
P3 > added iSSE (already here ?)
P4 > added SSE2 (already here ?)
P4 SSE3 HT > wow he multithreaded the codec and added SSE3 ?
P4 EE > ??? what's the difference with the latter proc, just more cache perhaps some EM64T ?


allow me to be very septic about it. But it would be a great news if he could confirm that he added some code.

Is there a daily snapshot of the sources somewhere ? i don't have a svn client but i've the intel compiler installed, i could try to build the codec with it and compare the speed.

[Sharktooth]

Originally posted by bill_baroud
Is there a daily snapshot of the sources somewhere ? i don't have a svn client but i've the intel compiler installed, i could try to build the codec with it and compare the speed.
No, but i can post the source in the usual thread whenever i make a build.

EDIT: http://www.webalice.it/f.corriga/x264/X264_rev225_src.rar

[bill_baroud]

thanks ... x264 library compiled (with /03 /QxP), some loop was vectorized.
Now, i'm searching what i need to have stdint.h or _strcasecmp...the intel linker doesn't like .a file ;)


[edit]
here you go :
http://moodub.free.fr/x264/x264_rev225_O3_QxP.zip
http://moodub.free.fr/x264/x264vfw_rev225__03_QxP.zip

absolutly no warranty and untested (no SSE3 heh ;))... your computer can explode while using this, etc, etc ...

[Sharktooth]

so? anyone tested those builds?

[Sirber]

I have no SSE3 :o

[BBugsBunny]

OK I've tested the builds of bill_baroud (rev 225) and cakuhnen (rev 221) from this tread using the same configuration & test video as in this tread:
http://forum.doom9.org/showthread.php?threadid=94065

Results:
1. pass (fast first pass disabled):
cakuhnen: 10.8 fps
bill_baroud: 15 fps

2. pass:
cakuhnen: 18.4 fps
bill_baroud: 16.9 fps

@bill_baroud: you got an error in the file names:
the wvf file is the cli build and the cli is the wvf - the names are swaped.

[Sirber]

So, his build is only compiled for, no ASM. :angry:

[bill_baroud]

@BBugsBunny : huh, yeah, my bad. Anyway, the rev233b (same place) are correctly named ;) and i played with other options... i wonder if it did something good.
(oups, i compiled it for any processor with SSE2, so it won't be the same as this pseudo-sse3 build).
Anybody to test this latest build with sse2 ? i'm curious as i've only AMD at home and a crappy celeron at work.

@Sirber : well his build looks different... the first pass is slower but the 2nd pass is faster.
And even with hand made optimization, you wouldn't get a blazing speed increase.