From now, when working with Fortran and CUDA, i am used to do some kind of C interface to mix them together. Due to differences between C and Fortran, these interfaces may need to do some extra work on tables. One can also change Fortran tables earlier in the code in prevision of their utilisations on a CUDA kernel, but then their are not efficient anymore on the Fortran code on CPU, which is, I believe, not the case in C since the order of loops and of the tables don’t change.
Well, my point is that i like to have some feedback from users who had been using or testing the PGI Fortran Compiler. How does it works ? What is your impression on it ?
And talking money, if you bought it, does it worth the investment ?
Hi, the PGI Fortran compiler works fine. They also have a product - Accelerator - that can really help the development process. I am working through it now. If you are a Fortran person, then I recommend it as the way to go.
Well its not free (PGI) … and for students thats a bad thing… :( … I wish nvidia could have added some basic support of FORTRAN “like in a fnvcc compiler… ? anybody”… oh well…
I am not especially a Fortran person, but I need to work on old scientific code so…
I took a look at the accelerator, and I am not particularly interested in. It looks like HMPP except it only produces code for CUDA whereas HMPP can output CUDA, SSE, C, …