I tested this 46060 windows 64bit build against the same in Linux and:
Windows: 6min 18sec
Linux: 6min 07sec
So Linux yet ahead, but much less.
@psy-fi:
i have changed it, but now node_edit.c sends to the same ED_image_draw_info a 0 int instead if info->profile!
one question though, why it compiles ok in msvcsp1 but not in mingw, and i tested the functionality and it works 100%?
Psy-fi 46060 (Mingw-64 With BVH Optomizations)
7.49 Seconds!
Psy-fi 45947 (Without BVH build optimizations)
32seconds build bvh
Tomato 45946
9.9Seconds build bvh
Official Blender 2.63
49 Seconds Build BVH
the above is a kitchen scene i made, i will do the regular benchmark and report in a little bit.
The “MikePan car Blender benchmark” time really isnt effected
by this BVH time the BVH already build fast on that scene…
I was getting
Psy-fi (min64 without BHV optimizations)
3.6GHZ
3min 29sec
Now its actually a little slower(its probably the same)
3min 31sec(2 sec difference)
But on a large scene the BVH does build faster by alot.
my personal benchmark a kitchen scene.
Kitchen scene 50 samples
psi-fi MinGW-64 no BHV fix(45947)
rendertime
3min 39sec
Kitchen scene 50 samples
psi-fi 46060 (Mingw-64 With BVH Optomizations)
rendertime
3min 6sec
Official Blender 2.63 From blender.org 50 samples
rendertime
7min 16sec
Hi every one,
Here comes the average “wannabe” user that wants to build his own Blender without qualification. Hehe…
I have been able to do that in the past though, during the 2.4x period. I used scons/mingw.
When i stumbled upon that thread i thought it could be time for me to try again with the new 2.6x series on my 64 bit system (previously win 32).
In my case, I wanted to try with 2.62.
But what version of Python do i have to have installed ? 2.7.3 or 3.2.3 ?
At the moment I have both (under C:\Python32 and C:\ Python27 and in the path variable)
I have installed mingw64 as well and used the user-config.py that is mentionned in the wiki.
Alas, I get that line everytime:
E:\Bureau\Source blender-2.62\blender>python scons/scons.py BF_TOOLSET=mingw
scons: Reading SConscript files …
Command-line arguments
BF_TOOLSET = mingw
Command-line targets
No targets given, using default
Using mingw
Using config file: build_files\scons\config\win32-mingw-config.py
Using user-config file: user-config.py
Build with parallel jobs: 4
Build with debug symbols: False
Missing: “#…/lib/windows/python/include/python3.2\Python.h”,
Set ‘BF_PYTHON_INC’ to point to a valid python include path.
Containing Python.h for python version “3.2”
E:\Bureau\Source blender-2.62\blender>
I’ve got that “lib” folder alongside the “blender” one. It contains the mingw64 folder.
In this folder i could find the Python.h file incriminated.
Someone knows what i am doing wrong ?
it would be cool if someone would make a video tutorial, covering all the steps,
including the update process, then the steps needed to compile.
actually it’s simple, all you have to do is reading!
http://wiki.blender.org/index.php/Dev:2.5/Doc/Building_Blender/Windows/MinGW/Scons
@ bat3a: yes, that page is very nice and clean.
The steps i used are very similar to what we could find on mpan3’s blog a while ago.
Apparently scons uses Python 2.7 doesn’t it ?
I am a bit puzzled as it points to some python 3.2 thing. So I have both installed (both 64 bits) and no success.
I have tested some mingw64 builds and, yes they are soooo much faster than official builds, I didn’t know that was so spectacular !
Apparently scons uses Python 2.7 doesn’t it ?
it should work but i didn’t test it, i used python 2.6. no need for 3.2
mingw is very strict with the gcc version, choose the version EXACTLY as it is specified in the wiki.
For those who aren’t aware yet, I’ve added some time ago a buildslave to do mingw64 builds: see http://builder.blender.org/builders/mingw_win64_scons So if you’re not into setting up your own environment, but want to have regular new builds, you can get them from the builder server. Nightly builds in the very least. Enjoy! /Nathan
Also, for builders: I have found a method to build CUDA kernels with MinGW-w64, be sure to check the wiki.
Are the MinGW builds as stable and full-featured as the official ones?
The buildbot has done that all the time (w. the MinGW builds) or am I missing something?
In my experience, absolutely. 
I’m using to mingw builds - one to model with the other to render whilst still modelling.
Thanks, Farmfield.
As soon as the MinGW build for 2.63a is available, I’m on it.
Noticed that these builds have stopped popping up on graphicall. Is there any reason why you all stopped building with it?
Also note that the Mingw64 builds don’t have COLLADA support yet - not sure what actually is still missing compared to official releases.
Anyway, building OpenCOLLADA with mingw64 is WIP, and I haven’t succeeded in compiling it fully yet. Enough errors that needs patching 
/Nathan
Hi,
When I try to build without CUDA it works fine but I get errors with.
This is my build log (look at #279) : http://www.pasteall.org/31778
Thank you in advance.