Pinned Memory Allocations


I am using Accelerator directives on Fortran. When the “mirror” clause is used does it means that the host allocations would be page-locked?

Thank you,

Hi Sayan,

No. Since physical memory is finite, we can’t pin allocatable memory by default.

If you want, you can mix CUDA Fortran with your PGI Accelerator Model, and use the “pinned” attribute on the host allocatable array. You would then need to also manually create a device array as well as manage the data movement.

  • Mat