So I looked into this, but I'm not familiar with sander so I don't really know the details of what's going on.
The run_lmod command calls lmodC which later calls a gradient descent that does the sander force call.
In general the bug seems like it stems from an mpi communication issue within the lmodC library ($AMBERHOME/AmberTools/src/sff/lmodC.c) where the single mpi broadcast of the random seed doesn't work. If you comment out this broadcast the code seems to work fine.
My initial thought which I may be wrong is that the lmodC library actual runs serially, as what comes out of it is essentially a set of coordinates to be used in a force calculation, if that is the case then syncing up the random seed probably isn't important as the force calculation does a broadcast of all the forces (since AMBER in general doesn't do domain decomposition and has a replicated data approach to comm).
If someone more experienced in how sander works wants to look into it that'd be great.
-Charlie
On 2/24/19, 5:29 PM, "Istvan Kolossvary" <istvan.kolossvary.hu> wrote:
Hi, I noticed that LMOD tests were disabled in the sander test suite, so
this problem may have been overlooked. I am running tests with both xmin
and lmod in sander using sander.MPI. The xmin minimizer works fine,
however, the lmod minimizer has a problem with the exact same system,
which is very strange since lmod is calling xmin for all force
computations. In $AMBERHOME/AmberTools/src/sander/force.F90 (amber18),
line 258
call mpi_bcast(xx(lcrd), 3*natom, MPI_DOUBLE_PRECISION, 0, &
commsander, ierr)
The call never returns.
! xx: global real array (holding e.g. coordinates at
position
! l15 etc., see locmem.F90)
In this case the xx array holds coords and forces. lcrd=12481,
natom=6240, commsander=0, and ierr=0. As I said the exact same system
works fine with xmin.
The lmod job works fine with serial sander.
Does anyone have any suggestion?
Many thanks,
Istvan
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
This email message is for the sole use of the intended recipient(s) and may contain confidential and privileged information. Any unauthorized review, use, disclosure or distribution is prohibited. If you are not the intended recipient, please contact the sender by reply email and destroy all copies of the original message. If you are the intended recipient, please be advised that the content of this message is subject to access, review and disclosure by the sender's Email System Administrator.
This email message is for the sole use of the intended recipient(s) and may contain confidential and privileged information. Any unauthorized review, use, disclosure or distribution is prohibited. If you are not the intended recipient, please contact the sender by reply email and destroy all copies of the original message. If you are the intended recipient, please be advised that the content of this message is subject to access, review and disclosure by the sender's Email System Administrator.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Mar 08 2019 - 11:30:04 PST
