| commit | f82f9d7b2aa46f48f692512a02e871a58c0eca12 | |
| author | Mark Abraham <mark.j.abraham@gmail.com> | |
| Wed, 25 Mar 2015 13:41:40 +0000 (25 14:41 +0100) | ||
| committer | Gerrit Code Review <gerrit@gerrit.gromacs.org> | |
| Tue, 30 Jun 2015 08:34:52 +0000 (30 10:34 +0200) | ||
| tree | 7077aec1464796cf80773b646dc9df82d7590ea1 | treesnapshot (tar.gz zip) |
| parent | 227d86669e3519edf2952e25f894b07a91f84300 | commitdiff |
Add support for flushing WDDM queue
Relevant only with CUDA on Windows (and profiling?)
On Windows the WDDM driver (default for non-Tesla) can prevent
immediate submission of CUDA tasks to the GPU in an attempt
to try to amortize driver overheads. However, as we need
tasks to start immediately for optimal concurrent execution,
this "feature" will result in large overheads. A well-
documented workaround is implemented by this change.
Change-Id: I69a6bb59dc8cae18fba539de49c977c0ee814d07
Relevant only with CUDA on Windows (and profiling?)
On Windows the WDDM driver (default for non-Tesla) can prevent
immediate submission of CUDA tasks to the GPU in an attempt
to try to amortize driver overheads. However, as we need
tasks to start immediately for optimal concurrent execution,
this "feature" will result in large overheads. A well-
documented workaround is implemented by this change.
Change-Id: I69a6bb59dc8cae18fba539de49c977c0ee814d07
| src/gromacs/mdlib/nbnxn_cuda/nbnxn_cuda.cu | diffblobblamehistory |