Branch: refs/heads/cuda-futures
Home: https://github.com/STEllAR-GROUP/hpx
Commit: ab5309d76aab9cc5308fcd0f31a41d28d24081e1
https://github.com/STEllAR-GROUP/hpx/commit/ab5309d76aab9cc5308fcd0f31a41d28d24081e1
Author: John Biddiscombe <[email protected]>
Date: 2020-07-21 (Tue, 21 Jul 2020)
Changed paths:
M libs/async_cuda/src/cuda_future.cpp
M libs/async_mpi/src/mpi_future.cpp
M libs/thread_pools/include/hpx/thread_pools/scheduling_loop.hpp
M libs/threading_base/include/hpx/threading_base/scheduler_base.hpp
M libs/threading_base/src/scheduler_base.cpp
Log Message:
-----------
Allow combined MPI and/or CUDA polling for futures
The polling functions will hardly ever change during normal
operation, so to avoid taking a lock, we use an function pointer
for each of the polling functions for MPI and CUDA separately.
This means they can be set/cleared by any thread and when unchanged
no cache lines are affected and overheads are minimized.
_______________________________________________
hpx-commits mailing list
[email protected]
https://mail.cct.lsu.edu/mailman/listinfo/hpx-commits
