August 2011 - Mpich2-core - mailman3.cels.anl.gov

Nemesis tcp netmod hang on Windows
by Jayesh Krishna 31 Aug '11

31 Aug '11

I was performing sanity checks on mpich2 1.4.1 x64 Windows build and found that Nemesis tcp netmod hangs in MPI_Finalize() (Sounds like a bug that a user reported a week or so back. He was suggested to see if sock works.) . Using Nemesis shared memory OR sock channel works. So my guess is that the problem existed from the previous version onwards. Looking into this problem. -Jayesh

4 4

nightly tests
by Rajeev Thakur 25 Aug '11

25 Aug '11

I have switched the nightly tests back to the trunk. Rajeev

1 1

1.4.1
by Rajeev Thakur 24 Aug '11

24 Aug '11

Should we release 1.4.1 today? The tests seem to be running. Rajeev

3 2

Re: [mpich2-core] Skip PBS launcher in 1.4.1
by Gropp, William D 17 Aug '11

17 Aug '11

It also adds it to the MPH module - which may be a binary object. We've been very conservative about this in the past. Bill Connected by DROID on Verizon Wireless -----Original message----- From: Anthony Chan <chan(a)mcs.anl.gov> To: "mpich2-core(a)mcs.anl.gov" <mpich2-core(a)mcs.anl.gov> Sent: Mon, Aug 15, 2011 18:21:46 GMT+00:00 Subject: Re: [mpich2-core] Skip PBS launcher in 1.4.1 buildiface in the patch added MPI_INTEGER_KIND (which is defined in 2.2) as requested by this ticket: http://trac.mcs.anl.gov/projects/mpich2/ticket/1515 So this is a bugfix. Given this patch only added MPI_INTEGER_KIND to mpif.h, so old MPI code should still work. I think it shouldn't affect ABI compatibility. A.Chan ----- Original Message ----- > The release creation script flagged this commit as possibly changing > ABI: > https://trac.mcs.anl.gov/projects/mpich2/changeset/8811/mpich2/trunk/src/bi… > > It's Fortran voodoo. Does some one know if it has any impact on ABI? > > -- Pavan > > On 08/15/2011 12:51 PM, Rajeev Thakur wrote: > > I think so. > > > > Rajeev > > > > > > On Aug 15, 2011, at 12:44 PM, Pavan Balaji wrote: > > > >> > >> Should we push out 1.4.1rc1 today? > >> > >> -- Pavan > >> > >> On 08/14/2011 10:47 PM, Rajeev Thakur wrote: > >>>> Rajeev: you'll need to restart the nightly tests once I revert > >>>> this. > >>> > >>> Done. > >>> > >>> On Aug 14, 2011, at 10:18 PM, Pavan Balaji wrote: > >>> > >>>> Hi Anthony, > >>>> > >>>> I was testing the PBS code and some corner cases don't seem to > >>>> have been handled (e.g., when the user passes the -ppn option to > >>>> Hydra). I tried to look into it, but there seem to be some subtle > >>>> bugs in there. > >>>> > >>>> Since 1.4.1 is only a bug-fix release, the PBS changes seem too > >>>> risky to include at this point. I'm going to revert r8905 from > >>>> the 1.4.x branch, so the PBS launcher won't be enabled. I'll > >>>> leave it alone in the trunk so you can look into it. > >>>> > >>>> Rajeev: you'll need to restart the nightly tests once I revert > >>>> this. > >>>> > >>>> -- Pavan > >>>> > >>>> -- > >>>> Pavan Balaji > >>>> http://www.mcs.anl.gov/~balaji > >>> > >> > >> -- > >> Pavan Balaji > >> http://www.mcs.anl.gov/~balaji > > > > -- > Pavan Balaji > http://www.mcs.anl.gov/~balaji

5 7

Skip PBS launcher in 1.4.1
by Pavan Balaji 15 Aug '11

15 Aug '11

Hi Anthony, I was testing the PBS code and some corner cases don't seem to have been handled (e.g., when the user passes the -ppn option to Hydra). I tried to look into it, but there seem to be some subtle bugs in there. Since 1.4.1 is only a bug-fix release, the PBS changes seem too risky to include at this point. I'm going to revert r8905 from the 1.4.x branch, so the PBS launcher won't be enabled. I'll leave it alone in the trunk so you can look into it. Rajeev: you'll need to restart the nightly tests once I revert this. -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji

4 10