Open ancahamuraru opened 9 years ago
Thanks for filing the issue. I think this is important as in my experience PCI-E transfers are much slower with non page locked memory.
Does OpenCL support page locking memory after allocating it using regular malloc? CUDA does support it and we have been considering removing the complication around allocator function pointers.
Update ocl_pmalloc function to allocate page locked memory.
See also the notes here: https://github.com/StreamComputing/gromacs/wiki/A2.5-log