dcmplx pgfortran problem - intrinsic funcition and radeons


I can’t include a dcmplx fortran function in a kernel loop using as target: -ta=radeon:hawaii,nollvm using -ta=tesla the compilation goes without errors.

Anyone has a issue like that with intrinsic fortran functions?

best regards

Maicon Faria

Hi Maicon Faria,

While we strive to have feature parity across all target devices, it’s not always possible. I don’t exactly why dcomplex was not available on Radeon, but would suspect a hardware limitation or lack of software support in the AMD back-end OpenCL compiler.