[Bf-committers] CUDA backend implementation for GSoC?

[Miguel A. Figueroa-Villanueva]

Hello Devs,

First of all let me thank you for a great product!

Second, I would like to make the disclaimer that I'm less than a
newbie regarding the blender code, since I plan to build it soon but I
still haven't even tried it.

Now, I have been thinking about a nice Google Summer of Code project
that one of my students could get involved in. I immediately thought
that it would be a good contribution to the Blender community if
someone took advantage of the GPU on modern cards as well as a great
learning experience for the students. In particular, I have access to
a few NVidia cards, which support CUDA.

My thoughts were to create a layer above the rendering backend, which
can dynamically select whether to use the CPU-based backend (current
stuff) or a different GPU-based backend (to be developed) depending on
whether or not there is a CUDA supported card. This would be done so
that in the future other backends could be implemented to support the
ATI stuff or the OpenCL, etc...

However, when I started looking at the Blender Architecture docs, it
seems this wouldn't be as easy given the many inter dependencies... My
question are:

1. Does any of this make sense?

2. Would there be interest from the community to see something like
this implemented? Or is anyone working on something like this?

3. Which modules would be the most to benefit from CUDA support?

Any other thoughts?

Thanks again for a solid Open Source piece!

--Miguel