| commit | 1e84d284ef773006e24748ad1ba096a32653335f | |
| author | Mark Abraham <mark.j.abraham@gmail.com> | |
| Fri, 3 Nov 2017 02:20:21 +0000 (3 03:20 +0100) | ||
| committer | Mark Abraham <mark.j.abraham@gmail.com> | |
| Fri, 24 Nov 2017 15:53:52 +0000 (24 16:53 +0100) | ||
| tree | 5a950779d1f163c42745890b990591b037c506be | treesnapshot (tar.gz zip) |
| parent | a0142c2c456a3cc9d52e7b0bf5f79a52df81f1f1 | commitdiff |
Activate PME on GPUs
Adds support for a run with a single rank, running NB and PME on the
same GPU.
Fixed bug in thread-MPI gather and gatherv functions. Real MPI does
not require that the send buffer is non-NULL if the send count is
zero, and the thread-MPI docs have not documented that this is an
intended extra requirement. This would be a problem only if e.g.
memcpy(dest,src,0) itself dereferenced src.
TODO user docs
Change-Id: Iace9e720c0958eaacaaa81307f1fe7324fa4c9ac
Adds support for a run with a single rank, running NB and PME on the
same GPU.
Fixed bug in thread-MPI gather and gatherv functions. Real MPI does
not require that the send buffer is non-NULL if the send count is
zero, and the thread-MPI docs have not documented that this is an
intended extra requirement. This would be a problem only if e.g.
memcpy(dest,src,0) itself dereferenced src.
TODO user docs
Change-Id: Iace9e720c0958eaacaaa81307f1fe7324fa4c9ac