bf-blender | linux standard & pentium4 builds (060617)

User-contributed CVS development builds. Please test and give feedback!

Moderators: jesterKing, stiv

Post Reply
ebrain
Posts: 0
Joined: Mon Sep 01, 2003 2:48 pm
Location: Nantes [FRANCE]
Contact:

bf-blender | linux standard & pentium4 builds (060617)

Post by ebrain »

Hi dear linux users :)

Here is the builds repository : http://blenderclan.tuxfamily.org/perso/builds/
(compiled on gcc, ubuntu dapper drake with i686 kernel, 2GHz Pentium4 & 1Go RAM)

There are two versions : one totally standard, one hard-optimized for pentium4 (in my tests, the optimized one renders 2.5x faster than the other one).

Don't forget you may not use the optimized build if you have not a pentium4 cpu.
Thank's for feedback ;).
ebrain
www.blenderclan.org [French Blender Community]

vizzy
Posts: 0
Joined: Tue Feb 28, 2006 12:36 am

options?

Post by vizzy »

hi,

unfortunately i got python 2.3, did you build with scons? make built actually builts a more static binary and is python 2.3/2.4 independend :)

can you post the optimization options maybe? (for scons)

i tried optimizing at blender, but got bad results from time to time, depends on the floating simulation blender uses i heard. so i was told to not optimize to get expected results. I really like to test your options :)

(maybe, contact me at #vizzy , i got a testsuite from zr once to test different builts (compare output / rendertime etc. with a set of demo-blendfiles))

thx in advance,
yours vizzy

ebrain
Posts: 0
Joined: Mon Sep 01, 2003 2:48 pm
Location: Nantes [FRANCE]
Contact:

Post by ebrain »

Here is the flags section :

Code: Select all

CCFLAGS = ['-pipe','-fPIC','-funsigned-char','-fno-strict-aliasing']

CPPFLAGS = ['-DXP_UNIX']
CXXFLAGS = ['-pipe','-fPIC','-funsigned-char','-fno-strict-aliasing']

### ajout flags d'optim ##
CCFLAGS.extend(   [ '-march=pentium4', '-O3', '-ffast-math', '-fomit-frame-pointer', '-mmmx', '-mfpmath=sse,387' ] )
CXXFLAGS.extend( [ '-march=pentium4', '-O3', '-ffast-math','-fomit-frame-pointer', '-mmmx', '-mfpmath=sse,387' ] )
##########################

REL_CFLAGS = ['-O3']
REL_CCFLAGS = ['-O3']
You're right it's build with scons. I'll try with 'make' as you said, to have a less dependant build :).
On my computer I have no crash and no computing error (I've compared ouput pics with and without optim, and they were exactly identical).

Bye bye and thanks for your advice :D

PS: for speed improvement, it seems that my test scene was more optim compliant than another... I've tried some others test files, and the optim one was 1.5x faster :!:
ebrain
www.blenderclan.org [French Blender Community]

vizzy
Posts: 0
Joined: Tue Feb 28, 2006 12:36 am

Post by vizzy »

ebrain wrote:Here is the flags section :

### ajout flags d'optim ##
CCFLAGS.extend( [ '-march=pentium4', '-O3', '-ffast-math', '-fomit-frame-pointer', '-mmmx', '-mfpmath=sse,387' ] )
CXXFLAGS.extend( [ '-march=pentium4', '-O3', '-ffast-math','-fomit-frame-pointer', '-mmmx', '-mfpmath=sse,387' ] )
##########################
hi,
many thanks!
i can agrree now :) with your options blender renders at least 1.1 to 1.5x faster.
i also added -msse3 , but cant say if its needed or not ...

I also can see no more relevant differences between 242rc1 and optimized cvs, but the refract monkey always looks a bit different. non optimized version seems to have nicer shading.
anyway, its faster :)

yours vizzy / verblendet

ps: for a make built i would sugest to disable gamengine, for me it breaks compiling with make and gamengine, but when disable the 'nearly realtime 3d simulation interactive console' (ge) its fine :) theres a trick to compile ge into blender when building with scons first, i cant explain.

Post Reply