PGI CUDA Fortran - Running with Optimizations

I’ve just finished installing the pgfortran CUDA fortran compiler from PGI and I’m trying to compile their example sgemm.cuf code. It compiles and runs beautifully with a standard pgfortran -o SGEMM sgemm.cuf call, but when I try the “with optimizations” option (pgfortran -02 sgemm.cuf) the compilation works fine but at runtime I receive 65536 errors (all handled) and an extremely low (.006ms) processing time, indicating it likely did not run at all.

I’m running a GTS 250, 1.8GHz, 1GB vram on Ubuntu 9.04 64-bit. Any help would be appreciated.

You should probably ask PGI about it. I understand they have support forums for the purpose.