[Ipopt] Ipopt release 3.11.8
Tony Kelman
kelman at berkeley.edu
Fri Apr 11 04:25:38 EDT 2014
HAVE_MPI_INITIALIZED should get checked during Ipopt’s configure (check config.log in the Ipopt subfolder), then the value gets populated into Ipopt/src/Common/config.h. The MPI_Init might also get skipped if MUMPS_MPI_H is defined somewhere?
From: Dominique Orban
Sent: Friday, April 11, 2014 12:55 AM
To: Stefan Vigerske
Cc: ipopt mailing list
Subject: Re: [Ipopt] Ipopt release 3.11.8
Thanks Stefan. I tried upgrading the Homebrew formula to 3.11.8 but I'm still getting the test failures due to MPI_Init():
---
./run_unitTests
Running unitTests...
Testing AMPL Solver Executable...
---- 8< ---- Start of test program output ---- 8< ----
==> Warning: Treating 1 binary and 1 integer variables as continous.
*** The MPI_Comm_rank() function was called before MPI_INIT was invoked.
*** This is disallowed by the MPI standard.
*** Your MPI job will now abort.
[Aragorn:91158] Local abort before MPI_INIT completed successfully; not able to aggregate error messages, and not able to guarantee that all other processes were killed!
---
I looked through the updated source code but it's not clear to me who's supposed to set HAVE_MPI_INITIALIZED?
Thanks
Dominique
On Tue, Apr 8, 2014 at 3:30 AM, Stefan Vigerske <stefan at math.hu-berlin.de> wrote:
Hi,
I just created a new Ipopt maintenance release.
This release includes an important fix for a bug that has been introduced (by me) with Ipopt 3.11.0 and could be responsible for difficult to debug convergence problems. If you were using Ipopt 3.11, it is suggested to update.
Here is a list of changes (r1234 stands for a related svn changeset, #123 for a ticket number in the trac bugtracking system):
- fixed a bug, introduced with Ipopt 3.11.0, where the tag in the
Ipopt's caching mechanism was not unique over time, which lead to
failures of Ipopt that were difficult to debug or recognize
(e.g., Ipopt may have stopped with an restoration failure for
instances that solved fine with Ipopt 3.10) [r2472, r2473]
I'm very thankful to Gabriel Hackebeil and Kurt Majewski for their
debugging effort on this issue.
- building Mumps with pthreads is now disabled by default [#229]
- fixed setting of LD on Windows (now set to link only iff using
MS/Intel compilers) [#230]
- fixed download link for Gnumex [r2471]
- for some messages about too-few-degrees-of-freedom and restoration
failure, the message level changed from error to strong-warning
[r2460, r2469]
- revised calls to MPI_Init and MPI_Finalize in Mumps interface [r2467]
(MPI_Init is now called only if function MPI_Initialized is available
and MPI has not been initialized already; MPI_Finalize is only called
if Ipopt also called MPI_Init; ...)
The source tarballs should be available tomorrow.
Enjoy,
Stefan
_______________________________________________
Ipopt mailing list
Ipopt at list.coin-or.org
http://list.coin-or.org/mailman/listinfo/ipopt
_______________________________________________
Ipopt mailing list
Ipopt at list.coin-or.org
http://list.coin-or.org/mailman/listinfo/ipopt
--------------------------------------------------------------------------------
_______________________________________________
Ipopt mailing list
Ipopt at list.coin-or.org
http://list.coin-or.org/mailman/listinfo/ipopt
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20140411/094dd615/attachment.html>
More information about the Ipopt
mailing list