| commit | 3820adc1501c77026250276eb3d770d189d50fde | |
| author | Szilárd Páll <pall.szilard@gmail.com> | |
| Fri, 6 Sep 2019 21:54:52 +0000 (6 23:54 +0200) | ||
| committer | Mark Abraham <mark.j.abraham@gmail.com> | |
| Tue, 10 Sep 2019 11:27:59 +0000 (10 13:27 +0200) | ||
| tree | 67c14d2151b896c42c3589083888ce9fa27bcb78 | treesnapshot (tar.gz zip) |
| parent | 2486a4260e242d444fa71c60b91898751de76fe5 | commitdiff |
Use thread-MPI in the gpucomm matrix
Initial versions of the direct communication do not support MPI and do
not implement graceful fallback to staged communication. Until they do
so we need to avoid library MPI in these matrices.
Change-Id: Ie7ab579f6058831229596fd6ba09a185d3a0685f
Initial versions of the direct communication do not support MPI and do
not implement graceful fallback to staged communication. Until they do
so we need to avoid library MPI in these matrices.
Change-Id: Ie7ab579f6058831229596fd6ba09a185d3a0685f
| admin/builds/gpucomm-matrix.txt | diffblobblamehistory |