Asynchronous CUDA transfers are asserted to always use pinned memory
commit82c726dcef63fe0f69c5a58ebda6f04d0be7ff51
authorAleksei Iupinov <a.yupinov@gmail.com>
Tue, 7 Nov 2017 12:40:17 +0000 (7 13:40 +0100)
committerErik Lindahl <erik.lindahl@gmail.com>
Sat, 18 Nov 2017 14:39:21 +0000 (18 15:39 +0100)
tree3928100af7898a91b41748e8c19e164e4d39f7ed
parent385a29debad0d64f360ad35dbba6bbf70dbe6f69
Asynchronous CUDA transfers are asserted to always use pinned memory

A few tests are also added for the memory status checker function.

Change-Id: Ifd6c9b2a0ad8ad8c325dc553534314e81e5d8bd7
src/gromacs/gpu_utils/cudautils.cu
src/gromacs/gpu_utils/tests/CMakeLists.txt
src/gromacs/gpu_utils/tests/pinnedmemorychecker.cpp [new file with mode: 0644]