GPU-implementation

[17barbieri]

Hi,
I am a PhD student working on tomosynthesis and I would like to know if you have yet implemented the GPU compatibility for the toolbox. If not, do you still plan to do it in the foreseeable future ?
Otherwise, have you tried other GPU-based libraries ?
Thank you a lot,

[rodrigovimieiro]

Hello,

The toolbox already works with GPU. There are instructions how to compile the codes on the readme file.

Let me know if you have any problems.

[17barbieri]

Hi, thank you very much for your swift response!
My bad, everything is complied correctly, I just had to switch from the default "projectionDDb" to "projectionDDb_cuda" and so on.
I'm getting a lot of warnings/errors though. They look like this:
"cudaMemset Padding Volume"
"cudaMemset Padding Projections"
"cudaMemset Zeros"
and
"cudaMemcpy Final".
Should I worry about those ?
Actually, all the results of "projectionDDb_cuda" are 0.0 everywhere...

[rodrigovimieiro]

The CUDA codes were not properly compiled.

It might be related to item 5 on readme:

If you have problems with arch=sm_XX, modify it in the setup.py code according to your NVIDIA-GPU architecture. This link has some references.

[17barbieri]

Thanks a lot, works great now !