Mpi redistribute
Clean up the redistribute code to avoid excessive memory use and work around an issue in Intel MPI that stops packets larger than 4Gb from being used over Mellanox infiniband (actually that is dapl).
Memory use is reduced by separating the particle types into different sends (so we don't have the overhead of a new copy of parts, xparts, gparts and sparts all at the same time).
Merge request reports
Activity
mentioned in issue #275 (closed)
Added 31 commits:
-
f5707f01...0c4ceb01 - 29 commits from branch
master
- 5efce9c7 - Merge remote-tracking branch 'origin/master' into mpi-redistribute
- c6d9bf45 - Remove debugging code
-
f5707f01...0c4ceb01 - 29 commits from branch
mentioned in issue #98 (closed)
Added 1 commit:
- 0d152f1a - Improve code reuse...
Added 61 commits:
-
0d152f1a...2983cf56 - 60 commits from branch
master
- 3aebc02f - Merge branch 'master' into mpi-redistribute
-
0d152f1a...2983cf56 - 60 commits from branch
Added 1 commit:
- 35d29389 - Don't free g_dest too early, also needed in debug checks
Added 3 commits:
-
35d29389...4b2ed413 - 2 commits from branch
master
- 6db27e70 - Merge remote-tracking branch 'origin/master' into mpi-redistribute
-
35d29389...4b2ed413 - 2 commits from branch
@nnrw56 would you like to have a look before the merge ?
mentioned in commit 6af9170c
mentioned in issue #283 (closed)
Please register or sign in to reply