Dear Mailing list members
I'm trying to compile Amber16 with AmberTools19 using intel compilers
2017.4 and CUDA 9.0 (I also tried with GNU 5.0). At the moment It seems
like I am able to compile Serial, MPI and CUDA Serial but CUDA MPI fails
with error (a similar error appears when compiling with GNU):
*Error Message:*
mpif90 -DMPI -DMKL -DBINTRAJ -DEMIL -DPUBFFT -ip -O3 -no-prec-div -xHost
-DCUDA -DGTI -DMPI -DMPICH_IGNORE_CXX_SEEK
-I/apps/chemistry/amber/16/ambertools19/el7/AVX512/intel-2017/intel-2017/amber16/include
-c binrestart.F90
binrestart.F90(102): error #6631: A non-optional actual argument must be
present when invoking a procedure with an explicit interface. [ISMREMD]
if (NC_defineRemdIndices(ncid, remd_dimension, remd_indices_var_id,&
--------^
binrestart.F90(102): error #6631: A non-optional actual argument must be
present when invoking a procedure with an explicit interface.
[REMD_VALUESVID]
if (NC_defineRemdIndices(ncid, remd_dimension, remd_indices_var_id,&
--------^
binrestart.F90(102): error #6631: A non-optional actual argument must be
present when invoking a procedure with an explicit interface.
[REMD_TYPESVID]
if (NC_defineRemdIndices(ncid, remd_dimension, remd_indices_var_id,&
--------^
compilation aborted for binrestart.F90 (code 1)
make[4]: *** [binrestart.o] Error 1
make[4]: Leaving directory
`/apps/chemistry/amber/16/ambertools19/el7/AVX512/intel-2017/intel-2017/amber16/src/pmemd/src'
make[3]: *** [cuda_parallel_SPFP] Error 2
make[3]: Leaving directory
`/apps/chemistry/amber/16/ambertools19/el7/AVX512/intel-2017/intel-2017/amber16/src/pmemd/src'
make[2]: *** [cuda_parallel] Error 2
make[2]: Leaving directory
`/apps/chemistry/amber/16/ambertools19/el7/AVX512/intel-2017/intel-2017/amber16/src/pmemd'
make[1]: *** [cuda_parallel] Error 2
make[1]: Leaving directory
`/apps/chemistry/amber/16/ambertools19/el7/AVX512/intel-2017/intel-2017/amber16/src'
make: *** [install] Error 2
*Possible reason:*
Looking at the source code I saw:
$AMBERHOME/src/pmemd/src/binrestart.F90
100 # ifdef MPI
101 ! REMD indices
102 if (NC_defineRemdIndices(ncid, remd_dimension, remd_indices_var_id,&
103 repidx_var_id, crdidx_var_id,&
104 remd_types, .true., (remd_method .eq.
-1),&
105 remd_groupsVID=remd_groups_var_id ))&
106 call mexit(mdout,1)
And in $AMBERHOME/AmberTools/src/lib/AmberNetcdf.F90
783 !> MODULE AMBERNETCDF FUNCTION NC_DEFINEREMDINDICES
784 !> .brief Create M-REMD index info in traj/restart file.
785 logical function NC_defineRemdIndices(ncid, remd_dimension,
indicesVID, &
786 repidxVID, crdidxVID, &
787 remd_types, isRestart, isREMD, &
788 isMREMD, frameDID,
remd_groupsVID, &
789 remd_valuesVID, remd_typesVID)
790 use netcdf
And notice that there is a difference in the number of arguments in both
functions. I wonder if I'm missing some patch, or if AmberTools19 is
incompatible with Amber16?
Any help is highly appreciated.
*Installation steps:*
./configure intel
#(say yes when prompted to install minicuda)
make -j install
./configure -mpi intel
make -j install
./configure -cuda intel
make -j install
./configure -mpi -cuda intel
make -j install
*More info:*
I patched $AMBERHOME/AmberTools/src/configure_python as advised here:
http://archive.ambermd.org/201907/0060.html
Output of: ./update_amber --version
Version is reported as <version>.<patches applied>
AmberTools version 16.24
Amber version 16.15
Not sure how Amber is figuring out the patch number for AmberTools19 but
the version number I know is hardcoded $AMBERHOME/updateutils/main.py, so
that's why the 16 in that output.
Regards
Jose
--
José Javier Muñoz Criollo
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Aug 14 2019 - 09:00:03 PDT
