Linking to Thrust (cuda) compiled code.

[dayalsoap]

Hi All. Basically, all I want to do is figure out how to link this thrust code (compiled with nvcc) from a code that uses an MPI c++ compiler. Here is the code:

thrust/histogram.cu at master * thrust/thrust * GitHub

Basically, I need to be able to invoke dense_histogram and sparse_histogram.

This histogram.cu file has to be compiled with the nvcc compiler (nvcc histogram.cu -c)

This will produce histogram.o

Then, I'd like to have another file, compiled with an MPI compiler be able to call those functions.

I'm at a loss on what I need to move into a header file, how to construct the header file, etc.

Does anyone have any advice? Thanks

[MutantJohn]

`nvcc` uses `gcc` under the hood for the linking stage.

I think all you have to do is use `nvcc` for the actual CUDA compilations to generate the object file and then you can link it normally like you would with gcc.

Edit: Nvidia also has a CUDA developers forum where you can ask this question as well. They're far more knowledgeable than I am in this instance.