[mpich2-dev] MPE and collcheck

[chan at mcs.anl.gov]

Hi Lisandro,

collchk has not been tested extensively on intercommunicator,
so I'm not surprised to see bugs.  I think same_int() checks to see if
the input communciator is intercomm, if so, merge the intercomm
as intracomm and do check on the merged intracomm.  If you do see bugs,
send the bug report to mpich2-maint at mcs.anl.gov.

Thanks,
A.Chan

----- "Lisandro Dalcin" <dalcinl at gmail.com> wrote:

> After getting errors when running my mpi4py's testsuite, I took a
> look
> at  src/mpe2/src/collchk/src/ ... Perhpas I got confused, but ...
> regarding the implementation of CollChk_same_int(), it seems that
> other calls are wrong for the case of INTERcommunicators:
> 
> 
> 1) CollChk_same_high_low
> 2) CollChk_same_root()
> 
> For (1), the CollChk_same_int() should be done at each side of the
> intercomm.
> 
> For (2), it is even more complicated, as it should make different
> checks at each side, plus a global check for compatibility of the
> root=MPI_ROOT at one side with the root=someint at the other side.
> 
> 
> I could provide patches, though I'm still not sure about the best way
> to implement (2).
> 
> BTW, Why in CollChk_same_int() the 'usecomm' is not
> PMPI_Comm_free()'ed ??
> 
> -- 
> Lisandro Dalcín
> ---------------
> Centro Internacional de Métodos Computacionales en Ingeniería (CIMEC)
> Instituto de Desarrollo Tecnológico para la Industria Química (INTEC)
> Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET)
> PTLC - Güemes 3450, (3000) Santa Fe, Argentina
> Tel/Fax: +54-(0)342-451.1594