new CVS code since 0.0.8

[Lynx]

I decided to make a new thread to encourage testing of new CVS code, even though the number of "backported" improvements/features will probably be quite limited.

I'd like people to give feedback on compile issues, bugs and performance regressions.
For informations on CVS access go to CVS repository page

April 30th 2006:
New kd-tree code with SAH construction algorithm.
Some may already have played with the code/testbuilds i posted earlier, but i fixed a couple of bugs since then and also tweaked it a bit. But it's still no O(n*logn) compiler, that means it can be quite slow on large meshes.

But the old one was quite expensive in all aspects too. The new one should reduce memory usage on high-poly scenes quite noticeably (~200-300MB for one million triangles), and improve renderspeed on many scenes.

May 18th 2006:
Improved the image loading code so it doesn't crash that easily on files with unknown file extension. Yafray tried to identify the file format by probing all known ones, but didn't check carefully enough before attempting to load HDR and TGA files.
Also became a bit more verbose on loading failures/wrong file extensions.

June 4th 2006:
Added -p option for Linux to specify an alternative plugin path when rendering XML (yafray executable).
Also added code to print used plugin path and number of plugins loaded to console

Forgot Mac OS specific code for aligned allocations...though it actually ignores alignment, but so far nothing depends on it (i hope...)

June 6th 2006:
Thanks to eeshlo who added a new render parameter "threads" to plugin export it is finally possible to change the number of threads (still called "Processors" in interface and "CPUs" internally...) anytime the user wants to, not only before the first rendering of a blender session.

Also finally officially killed a bug causing wrong colour output with gcc 3.4 and later when Clamp RGB is disabled and saving to LDR image formats. Colours brighter than 1.0 were set to 127 instead of 255 because the whole expression was incorrectly casted to char instead of unsigned char and the compiler noticed the constant 255 is beyond the
range of char and smacked it to 127 upon compilation.

[Claws]

I noticed the commit today when I pulled the lates CVS, No compilations errors for me on std. slackware 10.2. Thats all I can say because I don't know how to test it.

Great to see you adding stuff!

[afecelis]

I'm getting:
"cvs checkout: dying gasps from cvs.blender.org unexpected"
error-warning message

[Lynx]

Hm Claws, i don't know what compiler you use, but gcc 4.0.3 refused the code because a stupid mistake that MSVC ignores slipped in again...now for me kd-tree code builds without warnings even...
How to test it? Just use YafRay and compare rendertimes and memory usage
You may also launch Blender from console to see a bit more info, though i reduced output from kd-trees unless you set a #define to put out more details.

@afecelis: not my fault then, i don't touch the cvs server setup...using cvs-version from stone-age maybe?

-edit-
btw. the "arch=" argument is back on linux, e.g. "scons arch=athlon-xp" works again...am also thinking about adding a -mtune=i686 when no arch is given, so it goes a bit faster on newer PCs without breaking compatibility for you poor 386...like...you can render anything with a whooping 8 or 16MB RAM....

[Jedrzej_s]

Nice Lynx !!! We still waiting for new YafRay .

[kam]

Very good news!

If anyone compile one for windows, I will definate test it and start to support in with my YA(G)Max!

[lguillaume]

Very good news!

If anyone compile one for windows, I will definate test it and start to support in with my YA(G)Max!

Hello, I make a yafray compilation for windows, you can download from :
- 7zip (471KB)
- zip (897KB)

Normally I have put all DLL

[VividHazE]

Cool does this mean YafRay development has started again? Or is this a limited project to improve it, short-term?

[Hassan]

Yes, we are waiting your new Yafray version excitemently....

Regards.

Hasan

[afecelis]

using cvs-version from stone-age maybe?

On the contrary; I was using CVS' latest version (1.11.21) and got the gasp errors with it. I used an older version (1.11.14) and everything downloaded ok.
In case anyone's interested, here's my build for windows:
http://afecelis.gdlib.net/Blender/YafRay20060505.zip

[Jedrzej_s]

Lynx <- This news in CVS do it mean that YafRay development really still alive ??? Will You still work on YafRay and on future will release new version ??? This would be awesome... .

[Claws]

Now that I've finished my project I can do some small benchmarks for you Lynx
Heres a small benchmark:

50mb XML file - tell me if you need more info. - full GI (low qual.)

2300mhz Barton - 1 gig ram.

Both 0.0.8 and CVS compiled with "scons arch=athlon-xp"

Yafray CVS

real 10m11.208s
user 9m17.083s
sys 0m4.164s

Yafray 0.0.8

real 14m21.082s
user 13m23.366s
sys 0m4.792s

640x512 render

...and I think I was using 0.0.8 during my project due to a faulty install of CVS 'doh'

edit: so that means your stuff is right on - great work!

[Lynx]

Thx...
though i'm somewhat out of ideas right now...working a bit on my new design, but not very extensively either.
I just fixed a few bugs with image loading, it just crashed too easily with unknown files formats.

[Claws]

Well post if theres anything we non coders can help you out with

[Lynx]

Just fixed and added a few more details.
Testing desired, especially with current bf-blender code, since eeshlo did some nice work bringing back blender integration after orange merge!