I have a code which compiles and runs fine using the Cray compiler. I’m trying to get the same application to build using the PGI complier.
When I compile I have added the following parameters: -ta=nvidia -acc -Minfo=accel
Everything appears to compile fine and messages showing parallelization seem correct. However, when I run I get the failure: ‘cuMemHostRegister returned error 999: Unknown’
I have tried building against both cuda 5.0 and 5.5 and both have the same issue. I am using pgi’s compiler version 13.7
Any idea what might be going wrong?