[Bf-committers] Raytracer Optimization

[Yves Poissant]

Hello André,

I am quite excited to see you take on the raytracer optimization project. 
>From your bio, you seem to be the right person to tackle this.

The code I left in the patch tracker was not designed to be implemented in 
Blender as-is. I was experimenting with different structures during spare 
time and the code was rather to just let me experiment in a way that I felt 
comfortable with. But then I got a job offer programming a render engine and 
I feared that I would not be allowed to continue working on that code from 
my employer. I did not want this work to just disapear so I took the 
decision to put the code in the tracker in the state it was just before I 
signed the NDA with my employer. That is what you get in the tracker.

I'm glad to see that your goal is to go further than just optimizing the 
acceleration structure. In my experience, concentrating the efforts only 
onto the acceleration structure did not gain as much performance as I 
expected. I tabulated the render times for some 35 scenes. Some scenes had 
dramatic improvements while other did not improve at all. Overall, the BVH 
performance was not slower than the octree and sometimes much faster but the 
results were still quite disapointing. As for the BIH, it was also globally 
faster than the octree but for some scene, it was significantly slower than 
the octree. In light of this, I think that your approach of analyzing the 
scene and selecting the most appropriate acceleration structure may be 
interesting. My best improvement stat was 720% (7.2 times faster) using the 
BIH and my worst slowdown was -96% using a non SAH-BVH. I would have liked 
to invest more time into it and discover what was causing that but now you 
know the rest of the story.

I wish you good luck in your BSoC. If you want to discuss acceleration 
structure with me, then just leave me an email and I will discuss. Of course 
there might be areas where I will not be able to venture due to my NDA but I 
can discuss any public knowledge so fell free.

Yves

----- Original Message ----- 
From: "André Pinto" <andresusanopinto at gmail.com>
To: "bf-blender developers" <bf-committers at blender.org>
Sent: Monday, April 20, 2009 3:51 PM
Subject: Re: [Bf-committers] Raytracer Optimization


> Hello Timothy,
>
> GSoC results are out, and Raytrace Optimization did got accepted ;)
> I am the student of that proposal.. and for the following months I will be
> working on raytrace code.
>
> You can check my full proposal here: 
> http://andresp.no-ip.org/gsoc/raytree/
>
> And indeed it goes in some directions you said, namely try different data
> structures, and make it more modular to make it easier to try new stuff.
> Although it does runs away from things like CUDA optimized (lets wait for
> openCL :p?)
>
> Btw testing with real user cases is part of that proposal:
> so it could be useful to share your test.blend (if you would like to do 
> so)
>
> And of course any feedback and help/hints for SoC will be welcome ;)
>
>
> Best regards,
> André
>
>
> 2009/4/20 GSR <gsr.b3d at infernal-iceberg.com>
>
>> Hi,
>> tbaldridge at gmail.com (2009-04-20 at 1034.51 -0500):
>> > octree algorithm. So, I'm interested in performing some tests of
>> > optimization of this code by either implementing a new sorting
>> > algorithm such as a BIH or kd-tree.
>>
>> There have been some patches that could be a guide or just need clean
>> ups. BIH and BVH at least, look in tracker and list.
>>
>> GSR
>>
>> _______________________________________________
>> Bf-committers mailing list
>> Bf-committers at blender.org
>> http://lists.blender.org/mailman/listinfo/bf-committers
>>
> _______________________________________________
> Bf-committers mailing list
> Bf-committers at blender.org
> http://lists.blender.org/mailman/listinfo/bf-committers
>