Hi Rob,
On 8/6/2014 1:32 PM, Rob Latham wrote:
On 08/06/2014 11:50 AM, Mohamad Chaarawi wrote:
If I use mpich 3.1.2 , I don't see those issues.
What Mohamad forgot to tell you is that he doesn't see those issues
because I patched them on monday.
If I understood right, your patch was to fix the hindexed_block type in
ROMIO.. However the failure in OMPI was in the datatype creation itself,
which doesn't have to do with ROMIO or mpich.
The other issue I reported (with using 0 sized datatypes for some
processes), worked with an unpatched version of mpich 3.1.2 (I think),
since I was using hindexed and hvector (not hindexed_block).
Note that the replicator I have here is slightly different from the one
I sent a few days ago.
Thanks,
Mohamad
ROMIO has some HINDEXED_BLOCK fixes that OMPI should pick up on the
next romio resync. You are on your own with ompio!
==rob