MantisBT - ParaView
View Issue Details
IDProjectCategoryView StatusDate SubmittedLast Update
0012682ParaView(No Category)public2011-10-26 12:292012-02-08 17:21
ReporterDavid Partyka 
Assigned ToKen Moreland 
PriorityhighSeverityminorReproducibilityhave not tried
StatusclosedResolutionfixed 
PlatformOSOS Version
Product Version3.10.1 
Target VersionFixed in Version3.14 
ProjectSandia
Topic Name
Typeincorrect functionality
Summary0012682: IceT Fails to compile with MPI 1
Description
[ 44%] Building C object Utilities/IceT/src/communication/CMakeFiles/IceTMPI.dir/mpi.c.o
cd /nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2-build/Utilities/IceT/src/communication && /usr/local/bin/mpicc -DIceTMPI_EXPORTS -DVTK_PYTHON_BUILD -I/usr/global/tools/Kitware/Paraview/3.11/chaos_4_x86_64_ib/include -I/usr/gapps/visit/mili/1.10.0/linux-x86_64_gcc-4.1/include -I/usr/gapps/visit/hdf5/1.8.4/linux-x86_64_gcc-4.1/include -g -fPIC -I/nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2-build -I/nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2-build/VTK/Utilities -I/usr/local/include/python2.6 -I/usr/local/tools/mvapich-gnu/include -I/nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2/Utilities/IceT/src/include -I/nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2-build/Utilities/IceT/src/include -ansi -Wall -Wno-long-long -Wcast-align -Wextra -Wformat-security -Wshadow -Wunused -Wreturn-type -Wpointer-arith -Wdeclaration-after-statement -o CMakeFiles/IceTMPI.dir/mpi.c.o -c /nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2/Utilities/IceT/src/communication/mpi.c
/nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2/Utilities/IceT/src/communication/mpi.c: In function 'Gather':
/nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2/Utilities/IceT/src/communication/mpi.c:327: error: 'MPI_IN_PLACE' undeclared (first use in this function)
/nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2/Utilities/IceT/src/communication/mpi.c:327: error: (Each undeclared identifier is reported only once
/nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2/Utilities/IceT/src/communication/mpi.c:327: error: for each function it appears in.)
/nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2/Utilities/IceT/src/communication/mpi.c: In function 'Gatherv':
/nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2/Utilities/IceT/src/communication/mpi.c:348: error: 'MPI_IN_PLACE' undeclared (first use in this function)
/nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2/Utilities/IceT/src/communication/mpi.c: In function 'Allgather':
/nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2/Utilities/IceT/src/communication/mpi.c:366: error: 'MPI_IN_PLACE' undeclared (first use in this function)
make[2]: *** [Utilities/IceT/src/communication/CMakeFiles/IceTMPI.dir/mpi.c.o] Error 1
make[2]: Leaving directory `/nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2-build'
make[1]: *** [Utilities/IceT/src/communication/CMakeFiles/IceTMPI.dir/all] Error 2
make[1]: Leaving directory `/nfs/tmp2/rcook/ParaView/3.12/ParaView-3.12.0-RC2-build'
make: *** [all] Error 2
Steps To Reproduce
Additional Information
TagsNo tags attached.
Relationships
Attached Filespatch 0001-Work-around-when-MPI_IN_PLACE-not-available.patch (4,843) 2011-10-31 14:22
https://www.vtk.org/Bug/file/9054/0001-Work-around-when-MPI_IN_PLACE-not-available.patch
Issue History
Date ModifiedUsernameFieldChange
2011-10-26 12:29David PartykaNew Issue
2011-10-26 12:29David PartykaAssigned To => Ken Moreland
2011-10-26 12:32David PartykaNote Added: 0027611
2011-10-26 15:31Ken MorelandNote Added: 0027613
2011-10-27 16:24David PartykaNote Added: 0027616
2011-10-31 14:18Ken MorelandStatusbacklog => todo
2011-10-31 14:18Ken MorelandStatustodo => active development
2011-10-31 14:22Ken MorelandFile Added: 0001-Work-around-when-MPI_IN_PLACE-not-available.patch
2011-10-31 14:30Ken MorelandNote Added: 0027628
2011-11-05 05:49Takuya OSHIMANote Added: 0027661
2011-11-06 17:10Ken MorelandNote Added: 0027662
2011-11-06 23:08Takuya OSHIMANote Added: 0027664
2011-12-05 13:04Ken MorelandNote Added: 0027771
2011-12-05 13:04Ken MorelandStatusactive development => gatekeeper review
2011-12-05 13:04Ken MorelandFixed in Version => 3.12
2011-12-05 13:04Ken MorelandResolutionopen => fixed
2011-12-09 14:20Utkarsh AyachitStatusgatekeeper review => customer review
2011-12-09 14:20Utkarsh AyachitNote Added: 0027783
2011-12-09 14:21Utkarsh AyachitFixed in Version3.12 => git-master
2012-01-24 19:56Alan ScottNote Added: 0027972
2012-01-24 19:56Alan ScottStatuscustomer review => closed
2012-02-08 17:21Utkarsh AyachitFixed in Versiongit-master => 3.14

Notes
(0027611)
David Partyka   
2011-10-26 12:32   
See this this started by Rich Cook for more details.

http://markmail.org/thread/rggavx4lf4zdf3pt [^]
(0027613)
Ken Moreland   
2011-10-26 15:31   
The issue is that MPI_IN_PLACE is not supported before MPI 2. There needs to be a check for MPI 1 and use a different method in that case.
(0027616)
David Partyka   
2011-10-27 16:24   
Hey Ken, I setup a machine with mpich1 and can recreate the build failures, feel free to send patches to test my way.
(0027628)
Ken Moreland   
2011-10-31 14:30   
I think I have the problem fixed. I would appreciate it if someone (i.e. Dave P.) would verify that it compiles correctly with MPI 1 before I merge with next and move this bug to gatekeeper review.

I've pushed the necessary ParaView changes to the staging area under branch 12682-icet-fails-mpi-1 (SHA 5b6e3fcec86b5b22d4409d111b6609a1f6124e9e). Don't forget to also update the submodules.

Alternately, you could just apply the patch I've attached to this bug (to be applied to the Utilities/IceT subdirectory).
(0027661)
Takuya OSHIMA   
2011-11-05 05:49   
I tested the attached patch with 3.12.0-RC3 on a supercluster that only has MPI-1.2 support and confirmed that IceT compiled correctly. Many thanks for the fix!
(0027662)
Ken Moreland   
2011-11-06 17:10   
Is it possible to confirm that the server also runs with remote rendering turned on? We know the change compiles, but we are having trouble confirming that it runs correctly.
(0027664)
Takuya OSHIMA   
2011-11-06 23:08   
I needed another fix (0012711) in order to get everything compiled since our cluster is headless and thus needed OSMesa under user directory. With the fix applied, the server runs fine with remote rendering turned on (I see subsampled images during interaction).
(0027771)
Ken Moreland   
2011-12-05 13:04   
I believe this patch to be checked in and I am not aware of any further issues.
(0027783)
Utkarsh Ayachit   
2011-12-09 14:20   
merged in master (wherever applicable)
(0027972)
Alan Scott   
2012-01-24 19:56   
Assuming fixed.