llIn the code I found the following functions:
xchange_deri.c (BGL OK, non BGL OK)
xchange_field.c (non-block OK, shmem NO , naive-Sendrecv NO)
xchange_gauge.c (non-block OK, naive-Sendrec NO)
xchange_halffield.c (shmem NO, persistent NO, non-block OK, 32-shmem NO, 32-non-block NO)
xchange_c.c (non-block OK, naive-Sendrec NO)
xchange_2fields.c (non-block OK)
Where are the communications in the GPU part?!?
personally I have only used:
_deri non BGL
even if not used much, I would keep the naive because it is good to have a simple version.
Please add your own