aboutsummaryrefslogtreecommitdiff
path: root/README
diff options
context:
space:
mode:
authorknarf <knarf@b61c5cb5-eaca-4651-9a7a-d64986f99364>2009-09-18 19:55:30 +0000
committerknarf <knarf@b61c5cb5-eaca-4651-9a7a-d64986f99364>2009-09-18 19:55:30 +0000
commita6345d6e112cf83d6c1a60f9049524b30aa1a0ae (patch)
tree041f814e54726450956e7740dd2628988477d6ad /README
parent268936b8f93bec3a18681d2650cac01d4395dd76 (diff)
bugfix for the "automatic" topologyy generator:
It used to fall back to the old method, when some of the spcifications for nprocs have already been set, because it should not overwrite something e.g. specified by parameters. If this is not the case, nprocs is usually initialized with 0es. This was good. PUGH also tries to be smart and for very small extends in one direction (smaller than 2*ghostsize) it presets nprocs in that direction with 1. This is also good. Now, PUGH presets this before the "automatic" topology generator kicks in, which means this sees the "1" and falls back to the old method. This old method however has problems with exactly this type of grid and gets it wrong (overwriting that "1"). This patch now lets the automatic method only fall back to the old method if it arrives at a different answer than preset values (if set). (This fallback is neccessary because it cannot handle all cases which the old method does.) This fixes the problem because the automatic method also arrives at the preset value of "1" (and can set the other dimentions which have not been preset). git-svn-id: http://svn.cactuscode.org/arrangements/CactusPUGH/PUGH/trunk@497 b61c5cb5-eaca-4651-9a7a-d64986f99364
Diffstat (limited to 'README')
0 files changed, 0 insertions, 0 deletions