Thinking about renderants and bsgp..

See renderants.org
webpage has bsgp 2.0.2 says fixes large code generation and obj issues
removes info in 2.0.1 release notes saying multigpu support, and renderants code, etc..
Also renderants 2.0.1 has renderants code..
In 2.0.2 removed renderants so you have to demand in renderants.org but has scenes of the paper but doesn't work.. in exe I have..
I have to send bugs..

Also I want to congratulate you and your students for doing BSGP seems a very good tool for easing CUDA programming..
now I have a bunch of questions being so interested in the work..
I have been lucky enough to find BSGP 2.0 released on Microsoft Research which includes Renderants source code..
I have reading the manual and seeing the release notes, I have some questions:

Says it has "Debug support (uses bsgpdbg)" it's related to the GPU debugging paper (which I have read) i.e. GPU debugging of kernels?.. etc..
in the paper it's named CUDAdb it's the same?.. also I don't seem to find how to use it.. i.e in folder bin I have the editor and compiler (ctc)..
if not present are they going to release CUDAdb?

Also says Multi-GPU support (-DmultiGPU) and I think this has to be supported on source (Renderants only?) i.e. it's not a compiler feature right?..

I also find BSGP good to use but I see some limitations..

1. It compiles directly to PTX,.. it isn't an option for "translating" to CUDA kernel files and source translating to C source..
If I say that is becase it would allow using BSGP in Windows x64 ,Linux and Mac where I have CUDA stack..
I also related see in Topcoder CUDA contest that second winner said it programmed in BSGP and translated to CUDA.. is one of BSGP workers and had such a translator?..
if not it's really easy to do by hand.. for example translate Renderants to CUDA?..
reading BSGP paper I remember to get the idea that the motivation was taking care of global synchronization points and automatically creating multiple kernels for achieving it with maximal data reuse between them..
2. Also now with OpenCL coming I see BSGP could have cross vendor supporting OpenCL insted of CUDA, is BSGP upgrading to that? since concepts are aplicable to other APIs and if not it's their interest it's really possible to get access to the code for trying to add that as a backend?..i.e are they going to post BSGP on google code for example?.. at least the core parts..

In case someone cares would be intereseting to add to a GPU bench if something does a scientific GPGPU benchmark (mainly OpenCL) for testing GPUs implementing key building blocks as:

*LU
*FFTs
*Matmul
*Sparse math
*Sorting
*Peak flops

etc..
mainly leveraging research codes published in the web... and translating to OpenCL if some wants.. for example Renderants work..
I have successfully compiled the RenderAnts binary but I'm restricted on Win 32 and a benchmark would wanted to make Win,Lin and Mac compatible..
Also having BSGP CUDA translated code would allow trying to port to OpenCL more or less..
another possibilty is using as a benchmark a raytracer in BSGP but it has the same OS limitations..

assuming I use the binary I got then I have tried to get for example Big Bunny open source Blender stuff but it don't work with Renderants with current Blender and Mosaic versions..
so it's would be really useful for someone of bsgp of sending key steps in obtaining the Elephant RIB scenes you use the paper or pointing me to some way to obtain it..

Finally if they don't have really plans on whether if release BSGP as open source or close source would be any problem with using code in it if I don't want to relase code using it?