Feel free to edit this repository with questions and answers. I encourage the use of markdown.
Resources for CyEnce/SLURM
Q. I don’t know if we can set up a page of “frequently asked question” for GAMESS. I have posted a bunch of questions on Slack. Though all questions have been answered with very nice instructions, it can be difficult to trace back these questions and instructions after a while.
A. This repository.
Q. I think scripts in GAMESS (comp, compall, lked, config, rungms and gms) are very powerful by covering a variety number of machines, communications, compilers; but it’s difficult to navigate and modify these scripts. I don’t know if we can make templates with minimum information to compile, link and run jobs using these scripts.
A. The config script exists to remove the need to modify comp, compall, and lked - which makes use of the install.info file that is generated from the config process. You should not need to make any modifications to comp, compall, and lked unless you are adding a new target, compiler, math library, or communication library.
An alternative to rungms is the rungms.interactive script that can be found in $GMS_DIR/machines/xeon-phi/. Copy it to your $GMS_DIR folder. The script is initially setup for mpi but if you need sockets or ga then you will need to change:
set TARGET=mpi
within the rungms.interactive script.
An alternative to config makes use of Python3 and Jinja2 templating. Have a look at the following branch. Execute:
./bin/create-install-info.py --help
to see usage information. The script will generate an install.info, compile actvte.x, and produce a rungms script based off install.info.
The gms script is specialized for a given cluster and queuing system. There is no simple template for this file.
Q. Which fortran extension should be used? How does the fortran extension affect OpenMP?
A. A good example: https://software.intel.com/en-us/node/677900
Q. How do I resolve the following linking error for GAMESS?
mkl_semaphore.c:(.text+0x123): undefined reference to `dlopen'
mkl_semaphore.c:(.text+0x152): undefined reference to `dlsym'
...
mkl_memory.c:(.text+0x5b9): undefined reference to `dlopen'
mkl_memory.c:(.text+0x5da): undefined reference to `dlsym'
...
load_library.c:(.text+0x1cd): undefined reference to `dlopen'
load_library.c:(.text+0x1ef): undefined reference to `dlvsym'
...
more undefined references to `dlvsym' follow
A. Hardlinking appears to be the issue. You can switch to soft-linking the Intel math library in GAMESS by adding -so to the end of the setenv GMS_MKL_VERNO line in install.info.
For example if the line reads:
setenv GMS_MKL_VERNO 12
You would change it to:
setenv GMS_MKL_VERNO 12-so
And re-attempt the linking again with either lked or make.
Q. OpenMP sentinels?
A. http://man7.org/linux/man-pages/man1/gfortran.1.html
-fopenacc
Enable the OpenACC extensions. This includes OpenACC "!$acc"
directives in free form and "c$acc", *$acc and "!$acc" directives
in fixed form, "!$" conditional compilation sentinels in free
form and "c$", "*$" and "!$" sentinels in fixed form, and when
linking arranges for the OpenACC runtime library to be linked in.
Note that this is an experimental feature, incomplete, and
subject to change in future versions of GCC. See
<https://gcc.gnu.org/wiki/OpenACC > for more information.
-fopenmp
Enable the OpenMP extensions. This includes OpenMP "!$omp"
directives in free form and "c$omp", *$omp and "!$omp" directives
in fixed form, "!$" conditional compilation sentinels in free
form and "c$", "*$" and "!$" sentinels in fixed form, and when
linking arranges for the OpenMP runtime library to be linked in.
The option -fopenmp implies -frecursive.
Q. Why am I getting inaccurate results for LIBCCHEM runs using LIBINT?
A. The issue is tied to the renormalization.
- Comment out
//renorm()on line 129 inlibint/include/libtin2/shell.h - Inside the
libcchemfolder run:make && make install
- Inside the
$GMS_DIRfolder run:make
- Try the libcchem run again.
Q. I get the following error: semget errno=ENOSPC -- check system limit for sysv semaphores.
A. Increase the semaphore limits for the max number of arrays.
You can check the current limit by issuing:
ipcs -l
You can change the limits (as sudo or root by doing) replacing:
MAX_SEMAPHORE_PER_ARRAY_VALUE
MAX_SEMAPHORE_SYSTEM_WIDE
MAX_OPS_PER_SEMO_CALL
MAX_NUMBER_OF_ARRAYS
below with the numeric values you desire.
printf 'MAX_SEMAPHORE_PER_ARRAY_VALUE\tMAX_SEMAPHORE_SYSTEM_WIDE\tMAX_OPS_PER_SEMO_CALL\tMAX_NUMBER_OF_ARRAYS' >/proc/sys/kernel/sem
In this case increasing the value for MAX_NUMBER_OF_ARRAYS.
OR
sudo sysctl -w kernel.sem="496 32767 256 512"
which sets:
------ Messages Limits --------
max queues system wide = 32768
max size of message (bytes) = 8192
default max size of queue (bytes) = 16384
------ Shared Memory Limits --------
max number of segments = 4096
max seg size (kbytes) = 18014398509465599
max total shared memory (kbytes) = 18014398442373116
min seg size (bytes) = 1
------ Semaphore Limits --------
max number of arrays = 512
max semaphores per array = 496
max semaphores system wide = 32767
max ops per semop call = 256
semaphore max value = 32767
Q. None of the GAMESS scripts work on my cluster.
A. Several possible solutions:
- Check that
cshortcshis installed on your system.- Often times the C shell is not installed by default.
- Try changing the first line of each script from
#!/bin/cshto#!/bin/tcshcshmay not exist on your system buttcshmay be available.
- Try adding a
-fto the first line of each script:#!/bin/csh -for#!/bin/tcsh -f- It is possible that your local or system-wide .cshrc file is causing problems. The
-fwill tell shell to not load or startup and resource files.
- It is possible that your local or system-wide .cshrc file is causing problems. The
Q. How do I run the validation *.json files?
A. Resources for this are in: https://github.com/gms-bbg/test-validation
Q. I am running into a compilation problem of boost while building libcchem
A. This could be solved by going to /libcchem/boost and commenting out lines 27 and 28 in the Makefile
Q. GAMESS builds fine, but when I try to run an input file that I copied from a different computer to a Linux cluster, I get very strange errors in the ouput like
*** ERROR, ILLEGAL POINT GROUP=C1
CHOSEN.
A. This could be due to copying the file from a Windows machine to a Linux machine. The line breaks are different in Windows and Linux, so the input file might not be read in by GAMESS properly. Invoking the program dos2unix file.inp can often fix the issue.
Q. How can I reduce file I/O in GAMESS runs?
A. There are various options, depending on the type of calculation you're running and what output you need. All of these are documented in INPUT.DOC, but a brief overview is below.
- set NPRINT=-5 in $CONTRL
- set MEM10=7000000 in $SYSTEM (to get a better estimate of what value to use for MEM10, do a run, and
look at the bottom of the file for "In-memory F10 used 125262 words out of 7000000 allocated."
This puts .F10 file in memory, so it isn't written to disk.)
- set MODIO=31 in $SYSTEM
- set DIRSCF=.t. in $SCF (This avoids writing integrals to disk and opening .F08 file.
- set NPUNCH=0 in $SCF
- set NPUN=-2 in $STATPT
Q. I am trying to run RI-MP2 in GAMESS (for example this uracil input) with the OpenMP/MPI build enabled (by changing GMS_OPENMP to true in install.info and Makefile). I get a segfault or stack error or other strange memory error.
A.
This could happen for a variety of reasons, but one thing to try is increasing the OpenMP thread stack size. This can be done by setting the environment variable OMP_STACKSIZE to a large value in your runscript or environment. For example, in bash: export OMP_STACKSIZE=2000M. KMP_STACKSIZE can also be set if the Intel OpenMP runtime is used. GOMP_STACKSIZE is similar environment variable.
