Hi,
I recommend trying new compilers.
Generally, recent GNU compilers produce executables that
perform as good as Intel compilers. Of course, details
can matter, so if you have benchmarks for Amber18 using your system
that compare compiler performance, we'd be interested in the results.
I made a brief attempt to build sander with intel/15.0.3 and mkl;
there was a compiler error because intel 15 mkl is so old it does not
have fftw3.f03
scott
ps snippets below show the failure and compare mkl 15 vs 16 :
===
...
[SANDER] FC rmsgrd.F90
[SANDER] FC ../lib/AmberNetcdf.F90
cd ../rism && \
if [ -n "-DRISMSANDER" ]; then \
make amber_rism_interface.SANDER.o; \
else \
test -e amber_rism_interface.SANDER.o || touch amber_rism_interface.SANDER.o; \
fi
make[3]: Entering directory `/tmp/amber/AmberTools/src/rism'
[RISM] FC rism_report_c.F90
[RISM] FC fftw3.F90
fftw3.F90(36): error #5102: Cannot open include file 'fftw3.f03'
include 'fftw3.f03'
----------^
compilation aborted for fftw3.F90 (code 1)
make[3]: *** [fftw3.o] Error 1
make[3]: Leaving directory `/tmp/amber/AmberTools/src/rism'
make[2]: *** [../rism/amber_rism_interface.SANDER.o] Error 2
make[2]: Leaving directory `/tmp/amber/AmberTools/src/sander'
===
===
ll /usr/local/intel/composer_xe_2015.3.187/mkl/include/fftw
total 124
4 -rw-r--r-- 1 root root 2788 Apr 10 2015 fftw3.f
28 -rw-r--r-- 1 root root 22978 Apr 10 2015 fftw3.h
4 -rw-r--r-- 1 root root 1731 Apr 10 2015 fftw3_mkl.f
28 -rw-r--r-- 1 root root 22263 Apr 10 2015 fftw3_mkl_f77.h
4 -rw-r--r-- 1 root root 2623 Apr 10 2015 fftw3_mkl.h
12 -rw-r--r-- 1 root root 9346 Apr 10 2015 fftw3-mpi.h
4 -rw-r--r-- 1 root root 2835 Apr 10 2015 fftw3-mpi_mkl.h
4 -rw-r--r-- 1 root root 1436 Apr 10 2015 fftw_f77.i
12 -rw-r--r-- 1 root root 11897 Apr 10 2015 fftw.h
8 -rw-r--r-- 1 root root 6467 Apr 10 2015 fftw_mpi.h
4 -rw-r--r-- 1 root root 1201 Apr 10 2015 fftw_threads.h
4 -rw-r--r-- 1 root root 3285 Apr 10 2015 rfftw.h
4 -rw-r--r-- 1 root root 2966 Apr 10 2015 rfftw_mpi.h
4 -rw-r--r-- 1 root root 1585 Apr 10 2015 rfftw_threads.h
===
===
lll /usr/local/intel/compilers_and_libraries_2016/linux/mkl/include/fftw/
total 220
4 -rw-r--r-- 1 root root 3833 Jan 18 2016 fftw3.f
60 -rw-r--r-- 1 root root 55115 Mar 17 2016 fftw3.f03
20 -rw-r--r-- 1 root root 17896 Mar 17 2016 fftw3.h
4 -rw-r--r-- 1 root root 1661 Jan 18 2016 fftw3_mkl.f
28 -rw-r--r-- 1 root root 22172 Jan 18 2016 fftw3_mkl_f77.h
4 -rw-r--r-- 1 root root 2589 Jan 18 2016 fftw3_mkl.h
44 -rw-r--r-- 1 root root 38059 Mar 17 2016 fftw3-mpi.f03
12 -rw-r--r-- 1 root root 9621 Mar 17 2016 fftw3-mpi.h
4 -rw-r--r-- 1 root root 2468 Jan 18 2016 fftw3-mpi_mkl.h
4 -rw-r--r-- 1 root root 1436 Oct 22 2014 fftw_f77.i
12 -rw-r--r-- 1 root root 11897 Oct 22 2014 fftw.h
8 -rw-r--r-- 1 root root 6462 Mar 17 2016 fftw_mpi.h
4 -rw-r--r-- 1 root root 1201 Oct 22 2014 fftw_threads.h
4 -rw-r--r-- 1 root root 3280 Mar 17 2016 rfftw.h
4 -rw-r--r-- 1 root root 2966 Jan 22 2016 rfftw_mpi.h
4 -rw-r--r-- 1 root root 1585 Oct 22 2014 rfftw_threads.h
===
On Mon, Sep 16, 2019 at 03:32:25PM +0300, viktor drobot wrote:
> https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fftp.desolve.ru%2Fftp%2Fviktor%2Ftest_case.tar.gz&data=02%7C01%7Csbrozell%40iqb.rutgers.edu%7Cfa7911f483654b16639d08d73aa1f7ac%7Cb92d2b234d35447093ff69aca6632ffe%7C1%7C1%7C637042339640121956&sdata=yDfFVx%2F25DaQKki6IVVeKE0lElpKQ30P3S6PLQv6Chk%3D&reserved=0
>
> Here are my test case with 4 variants for dynamics - plain dynamics,
> metadynamics, qm/mm dynamics and qm/mm metadynamics. Running on our
> Lomonosov-2 cluster with GCC-4.8.5/CUDA-8.0 works just fine for all
> modules (sander and sander.MPI [all types of dynamics]; pmemd,
> pmemd.MPI, pmemd.cuda, pmemd.cuda.MPI [dynamics without QM/MM]).
> However, using Intel-15.0.3 compilers with MKL 2015.3.187-lp64 leads to
> fails with sander.MPI module (silent segfault) and nice working with
> other modules.
>
> 16.09.2019 15:21, David A Case ??????????:
> > On Sun, Sep 15, 2019, viktor drobot wrote:
> >
> >> It's very pity because I can't use QM/MM with high-performance Intel
> >> compilers and MKL library which gives some boost. What I can do to find
> >> out the cause of this bug? Thank you!
> > Can you create a small test case, and post the files involved? We would
> > need to be able to reproduce the problem before trying to debug it.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Sep 18 2019 - 13:30:02 PDT
