Dear all,
on my machine, sander from AmberTools20 run much slower when compiled using the cmake system versus the old configure script. In fact, on the same input, the speed is literally halved. I have tested
this fairly extensively, and I can exclude any effect due to choice of BLA vendor as well as system vs. bundled libraries (including all sort of combinations). At this stage, I really can't figure out
the origin of the problem. However, by comparing the output of ldd on the sander executables obtained with the two method, I notice that the cmake one links some libraries that are not found in the
"traditionally compiled" one. Specifically, these are
libfftw3
libemil
libsff_fortran
libnetcdff
libnetcdf
libarpack
liblapack
libblas
and their MPI versions, wherever appropriate. Am I right in assuming these are compiled in the traditional executable statically? If so, could this be the origin of the problem? If not, does anyone
have any idea of what causes it?
By the way, why doesn't libatlas get linked even though I explicitly set -DBLA_VENDOR=ATLAS?
Cheers,
Charo
--
Dr. Charo I. del Genio
Senior Lecturer in Statistical Physics
Applied Mathematics Research Centre (AMRC)
Design Hub
Coventry University Technology Park
Coventry CV1 5FB
UK
https://charodelgenio.weebly.com
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed May 06 2020 - 05:30:02 PDT