[visit-developers] [visit-commits] Updates to 1.12RC (r716[5-9]) and trunk(r7171) -- deadlock fix

Cyrus D. Harrison cyrush at llnl.gov
Mon May 4 11:23:58 EDT 2009


So you are opting out of an all reduce on processors without data - I 
would think this would hang b/c not all processors are participating?
-Cyrus

tom fogal wrote:
> A while ago a user here reported they were hitting a deadlock using the
> subset plot.  Turns out they were hitting a bug that only showed up at
> with a very particular setup:
>
>   * They were running in parallel
>   * A plot which needed ghost data was executing
>   * They had an AMR database
>   * There were more processors than domains
>
> The fix turned out to be incredibly small and simple: we no longer try
> to calculate domain nesting, or exchange ghost data, in the case that
> there are no domains.
>
> I also ran into a compile error, a fix for a warning which was
> obviously a typo, and a runtime error I'd already fixed on the trunk.
> The compile error is a bit sketchy, but I sent a private mail out to
> those more familiar with the code so we can figure out a good solution
> this week.
>
> -tom
>
> trunk:
>
>  1 files changed, 12 insertions(+), 8 deletions(-)
> commit 0c6a03142b8fbbb5479df88c4195b422db65c7b0
> Author: fogal1 <fogal1 at 18c085ea-50e0-402c-830e-de6fd14e8384>
> Date:   Mon May 4 03:21:08 2009 +0000
>
>     Merge from 1.12RC.
>     
>     Fixes:
>       * a compile error in the VLI database,
>       * a minor bug in printing filenames along an error path,
>       * a parallel deadlock in special circumstances.
>     
>     
>     git-svn-id: svn+ssh://svn.nersc.gov/svn/visit/trunk@7171 18c085ea-50e0-402c-830e-de6fd14e8384
>
>  src/avt/Database/Database/avtGenericDatabase.C     |    9 +++-
>  .../Pipeline/Pipeline/avtNamedSelectionManager.C   |   20 ++++++----
>  src/databases/VLI/avtVLIFileFormat.C               |    3 -
>  3 files changed, 18 insertions(+), 14 deletions(-)
>
> RC:
>
> commit 9a22e69eb9f077d3242974b845780cb453bc46bb
> Author: fogal1 <fogal1 at 18c085ea-50e0-402c-830e-de6fd14e8384>
> Date:   Mon May 4 03:16:27 2009 +0000
>
>     Don't re-use the iterator after an erase.
>     
>     This adds a new bool to detail whether or not we found the
>     timestate.  We can't use the `i' iterator to compare to anything
>     in `sil' here, because after the erase() it will be invalidated.
>     
>     (Manual backport from trunk.)
>     
>     git-svn-id: svn+ssh://svn.nersc.gov/svn/visit/branches/1.12RC@7169 18c085ea-50e0-402c-830e-de6fd14e8384
>
>  src/avt/Database/Database/avtDatabase.C |    7 ++++++-
>  1 files changed, 6 insertions(+), 1 deletions(-)
>
> commit 1e31e731377927f95f32f56cea1af9f465107904
> Author: fogal1 <fogal1 at 18c085ea-50e0-402c-830e-de6fd14e8384>
> Date:   Mon May 4 03:16:25 2009 +0000
>
>     Don't re-use the iterator after the find.
>     
>     This adds a new bool to detail whether or not we found the
>     timestate.  We can't use the `i' iterator to compare to anything
>     in metadata here, because after the erase() it will be
>     invalidated.
>     
>     (Manual backport from trunk.)
>     
>     git-svn-id: svn+ssh://svn.nersc.gov/svn/visit/branches/1.12RC@7168 18c085ea-50e0-402c-830e-de6fd14e8384
>
>  src/avt/Database/Database/avtDatabase.C |    9 +++++++--
>  1 files changed, 7 insertions(+), 2 deletions(-)
>
> commit 5d500f235393a0f78123dfb49a364272b9e557ba
> Author: fogal1 <fogal1 at 18c085ea-50e0-402c-830e-de6fd14e8384>
> Date:   Mon May 4 03:16:24 2009 +0000
>
>     Don't do any ghost calculations when there are no domains.
>     
>     This fixes a deadlock which occurs in some extreme circumstances.
>     
>     git-svn-id: svn+ssh://svn.nersc.gov/svn/visit/branches/1.12RC@7167 18c085ea-50e0-402c-830e-de6fd14e8384
>
>  src/avt/Database/Database/avtGenericDatabase.C |    9 ++++++---
>  1 files changed, 6 insertions(+), 3 deletions(-)
>
> commit 89405174cc0a2080b1d42ce2d51c9366f2e6dabe
> Author: fogal1 <fogal1 at 18c085ea-50e0-402c-830e-de6fd14e8384>
> Date:   Mon May 4 03:16:22 2009 +0000
>
>     Don't undef SEEK_* macros.
>     
>     OpenMPI is doing sketchy things with them, and #undef'ing them
>     here is a bit early; breaks compilation with this MPI
>     implementation.
>     
>     git-svn-id: svn+ssh://svn.nersc.gov/svn/visit/branches/1.12RC@7166 18c085ea-50e0-402c-830e-de6fd14e8384
>
>  src/databases/VLI/avtVLIFileFormat.C |    3 ---
>  1 files changed, 0 insertions(+), 3 deletions(-)
>
> commit 34d52ee5ae9ba7fcd64ff67160f6d901f5c29e26
> Author: fogal1 <fogal1 at 18c085ea-50e0-402c-830e-de6fd14e8384>
> Date:   Mon May 4 03:16:20 2009 +0000
>
>     Fix string formatting when an exception occurs.
>     
>     There was a mismatch between formatter format type.  Looked to be
>     a typo in the format argument: we were printing the length of the
>     filename instead of the filename.
>     
>     This uses a stringstream to make sure we don't mess up the types.
>     
>     git-svn-id: svn+ssh://svn.nersc.gov/svn/visit/branches/1.12RC@7165 18c085ea-50e0-402c-830e-de6fd14e8384
>
>  .../Pipeline/Pipeline/avtNamedSelectionManager.C   |   20 ++++++----
>
>   



More information about the visit-developers mailing list