Branch: refs/heads/master
Home: https://github.com/STEllAR-GROUP/hpx
Commit: a6dccfa1acd02cf3462276a8a67b344af72cfc71
https://github.com/STEllAR-GROUP/hpx/commit/a6dccfa1acd02cf3462276a8a67b344af72cfc71
Author: Mikael Simberg <[email protected]>
Date: 2021-04-06 (Tue, 06 Apr 2021)
Changed paths:
M libs/core/thread_pools/include/hpx/thread_pools/scheduling_loop.hpp
M libs/core/threading_base/include/hpx/threading_base/scheduler_base.hpp
M libs/full/async_cuda/src/cuda_future.cpp
M libs/full/async_mpi/include/hpx/async_mpi/mpi_future.hpp
M libs/full/async_mpi/src/mpi_future.cpp
Log Message:
-----------
Return status enum from scheduler polling functions
Polling functions can signal if they still have work to
do (polling_status::busy) or if they are idle (polling_status::idle).
When polling_status::busy is returned the scheduling loop does not
decrement the idle loop counter. This allows running a thread pool which
only does polling without adjusting its idling behaviour (prior to this
change a pool that only did polling would sleep by default even it needs
to poll since it doesn't have any regular HPX tasks to execute).
Commit: f08b511b5907468d1571e9c6b8749e05602c1c4c
https://github.com/STEllAR-GROUP/hpx/commit/f08b511b5907468d1571e9c6b8749e05602c1c4c
Author: Mikael Simberg <[email protected]>
Date: 2021-04-06 (Tue, 06 Apr 2021)
Changed paths:
M libs/full/async_cuda/src/cuda_future.cpp
Log Message:
-----------
Don't poll forever for CUDA events that are not ready
Commit: dfecbe97844f1f3ba6c7323fe02da9e494c8b6cc
https://github.com/STEllAR-GROUP/hpx/commit/dfecbe97844f1f3ba6c7323fe02da9e494c8b6cc
Author: Mikael Simberg <[email protected]>
Date: 2021-04-06 (Tue, 06 Apr 2021)
Changed paths:
M libs/full/async_cuda/src/cuda_future.cpp
Log Message:
-----------
Avoid potentially quadratic behaviour in CUDA future polling
Commit: 198d8d453c537e969362c37621cdb7988846afe2
https://github.com/STEllAR-GROUP/hpx/commit/198d8d453c537e969362c37621cdb7988846afe2
Author: Mikael Simberg <[email protected]>
Date: 2021-04-07 (Wed, 07 Apr 2021)
Changed paths:
M libs/full/async_cuda/include/hpx/async_cuda/custom_gpu_api.hpp
M libs/full/async_cuda/tests/unit/cuda_future.cpp
Log Message:
-----------
Use CUDA host pinned memory in cuda_future test
Pinned memory is required for asynchronous memory transfers.
Commit: 886e4e3a12aef3bffe31f954e68b006095f8a8c1
https://github.com/STEllAR-GROUP/hpx/commit/886e4e3a12aef3bffe31f954e68b006095f8a8c1
Author: Mikael Simberg <[email protected]>
Date: 2021-04-08 (Thu, 08 Apr 2021)
Changed paths:
M libs/core/thread_pools/include/hpx/thread_pools/scheduling_loop.hpp
M libs/core/threading_base/include/hpx/threading_base/scheduler_base.hpp
M libs/full/async_cuda/include/hpx/async_cuda/custom_gpu_api.hpp
M libs/full/async_cuda/src/cuda_future.cpp
M libs/full/async_cuda/tests/unit/cuda_future.cpp
M libs/full/async_mpi/include/hpx/async_mpi/mpi_future.hpp
M libs/full/async_mpi/src/mpi_future.cpp
Log Message:
-----------
Merge pull request #5277 from msimberg/cuda-polling-fixes
Small fixes and improvements to CUDA/MPI polling
Compare:
https://github.com/STEllAR-GROUP/hpx/compare/7aaba63ac92e...886e4e3a12ae
_______________________________________________
hpx-commits mailing list
[email protected]
https://mail.cct.lsu.edu/mailman/listinfo/hpx-commits
