Native Fortran coding options for GPU options for high level GPU coding

Hi all,

Im hoping for some options before i go charging along with what I think is best.

I want to port the compute-intensive parts of a legacy FORTRAN code to the GPU (its a partial differentiation based modelling package)

I have some experience with the Portland’s CUDAF / accelerator which I found to be very good in that it was high level/directive based which made life easy for me - particularly important as I know fortran but am a beginner with CUDA etc. The code is too big to translate to other languages and I would like to avoid mixed language compiling if possible.
I saw the stiky on GPU.NET also…

The important things im looking for are:

  1. I want to use Fortran, I can not learn another language at this time
  2. Im lazy and would ideally like to use directives (like OpenMP style parallelization)
  3. Im cheap, I will buy the Portland compiler if this forum advises it as i found it - and forum support - to be very good but its way beyond my budget so am hoping for an alternative.