On 4/5/2010, at 5:45 p.m., Ben Roberts wrote:
> Unfortunately, I've just switched back to using 11.1.084, and I still get the same complaints. I don't know how I could have overlooked them earlier.
>
> A bigger problem I ran into with the 11.1.088 was that MPI executables failed rather comprehensively (in the sense that nothing at all worked - everything segfaulted). So I'm seeing if downgrading fixed that problem too. If not, I'll at least see if I can find out what's going on, though I suspect nothing good or easy to track down.
Nope, downgrading didn't fix the problem with the parallel programs, to wit, the following:
./Run.sff: Program error
make[2]: *** [sff_test] Error 1
Running test to do simple minimization with shake
(this tests the molecular mechanics interface)
[qtpclnt070:22754] *** Process received signal ***
[qtpclnt070:22754] Signal: Segmentation fault (11)
[qtpclnt070:22754] Signal code: Address not mapped (1)
[qtpclnt070:22754] Failing at address: 0x362e6cb68
[qtpclnt070:22754] *** End of error message ***
[qtpclnt070:22755] *** Process received signal ***
[qtpclnt070:22755] Signal: Segmentation fault (11)
[qtpclnt070:22755] Signal code: Address not mapped (1)
[qtpclnt070:22755] Failing at address: 0x362e6cb68
[qtpclnt070:22755] *** End of error message ***
and analogous messages for other parts of AmberTools parallel and Amber parallel.
Nor did compiling OpenMPI with --with-memory-manager=none, per a suggestion I found online.
I'll look further into this. If the problem seems to be Amber code-related, I'll move this part of the discussion to amber-developers. If on my system, I plan on doing a follow-up post here for discovery by other inquiring users, along with a post at the Amber on Mac blog detailing what needs to be done.
Ben
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue May 04 2010 - 17:00:04 PDT