Jun 2011
1 / 5
Jun 2011
Jun 2011
nyiotis
Jun '11

Hey all,
I want to write a plug-in with CUDA 4.0 on linux that should be OS independent. I am not sure as for the best strategy: in particular, I prefer coding using the CUDA Runtime API instead of the CUDA Driver API. Is this an option in Maya? Apart from others, assuming I code with CUDA Driver API, then the similar-in-functionality-to STL Thrust O/O lib, which was integrated in CUDA v4.0, is not an option anymore.     

As for working examples, I 've spotted 22https://github.com/redpawfx/cudaFluid22 (so far) which I haven't compiled (yet) for linux, 
but it inspires me w.r.t the design, although I'd prefer to avoid MEL. Any opinions or experience with GPU-Maya coding are more than welcome. 

Cheers

N 

  • created

    Jun '11

  • last reply

    Jun '11
  • 4

    replies

  • 4.1k

    views

  • 1

    user

  • 2

    links

joojaa
Jun '11
katisss

Well have you actually built a naive application allrewady and actually profiled your code? I mean the cude cose is still going to need to ask the data form within maya. Theres is a a slight chance that the extra effort you put into the cuda code is not worth it, because the actual reading of the data and writing it back might still be the actual bottleneck. So its only if your wrap deformer code is actually very heavy. Your not going to get any stellar performance either way.

nyiotis
Jun '11
katisss

thanks joojaa, even without having a working protoype yet you are probably right:
since within maya's api you have to re-write almost everything, it probably does
not worth the effort. profiling the code might giveme some hints, cheers

N