[MPICH2-dev] mpich-test/pt2pt/allpair2 test case issue.

Sapronov, Sergey sergey.sapronov at intel.com
Mon Mar 27 06:22:18 CST 2006


Dear Sirs,

 

allpair2.f test case from mpich-test/pt2pt test suite fails sometimes
with the following output:

 

< Fatal error (code 0x4009430f) in MPI_Test():
< MPI_Test(154): MPI_Test(request=0x60000fffffff7684,
flag=0x60000fffffff7620, status=0x60000fffffff76b0) failed
< MPIDI_CH3U_Handle_recv_pkt(245): Ready send from source 0 and with
tag 3456 had no matching receive
< ABORT - process 1

 

The failure happens in 'Persistent ready sends' section (start line #460
in allpair2.f source).

I guess the test case handles persistent ready send incorrectly.

 

According to standard "A send that uses the ready communication mode may
be started only if the matching receive is already posted. Otherwise,
the operation is erroneous and its outcome is undefined."

 

In our case, process 0 starts ready send operation (line #484 : call
MPI_Startall(2, requests, ierr))

and process 1 starts corresponding receive operation (line #501: call
MPI_Start(requests(2), ierr))

without any synchronizing actions. Thus, the test case does not
guarantee that receive is posted before ready send start.

 

I believe allpair2.f test case should be fixed similar to 'Persistent
ready sends' section fixing in analogous allpair.f test case.

What is your opinion?

 

Perhaps, the issue has been already fixed. And we can only take new
version of the test suite?

Currently we use allpair2.f test case updated 2 Jul. 2004.

 

Thank you.

Best Regards.

 

Sergey Sapronov

Senior Software Engineer,

 

Intel SSG/DPD/Parallel Solution Lab,

Russia, Sarov.

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.mcs.anl.gov/mailman/private/mpich2-dev/attachments/20060327/ab451672/attachment.htm>


More information about the mpich2-dev mailing list