Just stumbled across this press released. It seems that PGI will be developing the native Fortran equivalent of nvcc for CUDA in partnership with NVIDA.
In a way, this is good news because PGIs Fortran compilers are excellent. In a way it is also bad news, because my experience with PGI has been that their licensing is rather inflexible and expensive compared to the other major compiler vendors. Still, interesting times.
this is bad news for me if this is the only fortran capable cuda compiler nvidia will give :( …
i really like the Intel Fortran compiler. I have yet to find a compiler which optimizes as well as ifort , plus its free fro Linux for personal use.
I am hoping and waiting ( for 1yr now) for NVidia to come out with Fortran version of cuda as they mentioned long time back. I hope am not disappointed…
But anwyas if that doesnt coem and this is the only option then I may stick to CUDA on C only or maybe I will give it a try. Let me know if you try how the compiler compares to latest version of ifort in terms of speed and file size ;)
I recently ‘sat in’ on a PGI webinar. It was awful. Furthermore, it made it sound very difficult, which it shouldn’t have to be. In order to get access to their Fortran/Cuda, one has to buy the compiler as well, and that is expensive. Finally, I got the impression that not all Cuda functionality is supported (yet, they said). No discussion of how to handle complex numbers either. I give this a two thumbsdown.