Hi everybody, I wanted to report an issue I ran into in pmemd. I hesitate
to call this a bug, as I think it has something to do with compiler
versions, etc., but on my machine the igb7 and igb8 tests in pmemd on the
GPU fail. After some investigation this appears to be due to the neckMaxPos
and neckMaxVal arrays not being passed correctly to the GPU code. It seems
that the fortran code is not arranging the numbers in memory in the order
that the c++ code expects, so passing a pointer to the initial element of
the array and using pointer math to load the data fails. It does not appear
to be a column-vs-row arrangement issue, rather the fortran code seems to
arrange the values in some arbitrary order. When I wrote new functions to
pass these arrays element by element the tests succeeded again. I am using
gfortran and g++ versions 9.4.0 and cuda version 11.4. I am not sure what
the best approach to fixing this would be, so I was raising the issue here.
--
*Julie Anne Harris | **Manager of Product Development, Drug Design*
*ComputChem* | 1100 Wicomico St, Baltimore, MD 21230, USA
*e:* julie.harris.computchem.com | *w:* ComputChem.com
*l: *www.linkedin.com/in/julie-harris-2b244b21a
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Mar 22 2022 - 10:30:02 PDT
