[mpich2-dev] MPI_Aint testcase development

Michael Blocksome blocksom at us.ibm.com
Mon Mar 10 15:15:01 CDT 2008


I'd like to start a discussion about MPI_Aint testcase development. 

IBM is running the following test suites as each new internal development 
builds become available:
1. Intelanl2006
2. SKAMPI5
3. POKMPISYS
4. POKMPIFVT

What is missing from this list are *targeted* tests for the MPI_Aint 
problem. The only known test (to my knowledge) that hits the problem is 
the b_eff_io program .. yet even this test was not written specifically to 
the MPI_Aint problem.

Digging through my inbox, I found this from a summary email sent by Bryan 
Hartlen (IBM) on 2/19:

--------------------------------------------------------
ROMIO issue: 

The Plan
1) Both IBM and ANL agree that the best way to proceed is to use what was 
originally called option B (recasting)
2) Both IBM and ANL agree that IBM will assume the lead for implementing 
the change (making the change(s) as required, testing and releasing)
3) Both IBM and ANL agree that IBM's effort above would benefit from any 
test case development that ANL would be willing to provide. 
4) Both IBM and ANL agree that IBM's scheduled release for this work in 
the V1R2 release cycle (May 16) is acceptable. 
5) Both IBM and ANL would like DOE's agreement that this plan above is 
sufficient to remove the ROMIO issue as a gate to starting 500T acceptance
--------------------------------------------------------

In regards to item #3, is ANL able to write testcase code for this 
problem, given your knowledge and understanding of the code paths 
involved?  Specifically, what code paths should we consider for test case 
development?

Let's discuss on this mailing list the targeted tests that need to be 
developed in order to fully test the solution to the MPI_Aint problem.


Thanks!

Michael Blocksome
Blue Gene Messaging Team Lead
Advanced Systems SW Development
blocksom at us.ibm.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.mcs.anl.gov/mailman/private/mpich2-dev/attachments/20080310/e6f51a9d/attachment.htm>


More information about the mpich2-dev mailing list