Two small bug fixes in the openmp part (release 2019-06-30 R1)

[ivr900]

Two small bug fixes are reported here in part of the OpenMP GAMESS build 2019-06-30 R1. Relevant to all compiler environments.

1. Small fix in file source/ompgrd2.src to remove overlap in the parameter definition caused by introduction 'efp_module.src'. Line 1058 must be changed as - mxatm, mxgrid, mxfrg, mxdppt + mxatm, mxgrid

2. Small fix in Makefile to make the base modules created before the OpenMP modules section, where efm_module.mod (one of the base modules) it is already required. Line 59 must be changed as - MODULES := $(shell sed -n '/openmp-start-section/,/openmp-end-section/p' $(COMPALL) | grep ".\/comp" | awk '{printf "%s ",$$2}') $(MODULES) + MODULES := $(MODULES) $(shell sed -n '/openmp-start-section/,/openmp-end-section/p' $(COMPALL) | grep ".\/comp" | awk '{printf "%s ",$$2}')

[saromleang]

Thank you for bringing this to our attention.

We discovered the same issue in 1. after making the public release. The solution we took is the following around line 1062: - USE comm_FRGINF + USE comm_FRGINF, ONLY: nfrg

[ivr900]

Hi Sarom,

also I noticed that the version printout in gamess.src has not been updated. This potentially may confuse users.

Best regards, Ivan

[saromleang]

Thanks for catching that!
The version printout should be updated now.

[saromleang]

Issue has been resolved with the latest patched released of GAMESS.

We did not make any changes with the build system.

Please use:

make ddi
make modules
make -j $(nproc)

[ivr900]

Sarom,

the second bug (see the top post) has not been fixed in the latest patch 1 of 2019-06-30 Release. Compiling of ompgrd2.src must be after compiling of efpmodule.src, as the former uses the module comm_FRGINF from the latter. If using the make procedure, it can be done as I suggested in the top post.

Best regards, Ivan

[saromleang]

Can you detail how you are compiling GAMESS.

We do not have any problems using the approach that we prescribed for compiling GAMESS.

make ddi
make modules
make -j $(nproc)

[ivr900]

Sarom, I follow the same procedure

config
make ddi

is fine. But 'make modules' stucks, if unmodifed 'Makefile.in' is used, giving:

gfortran -c -fdefault-integer-8 -g -O0 -fbacktrace -fopenmp -march=armv8-a -ffree-line-length-none -w -fno-aggressive-loop-optimizations ompgrd2.F90
ompgrd2.F90:1062.10:

      USE comm_FRGINF, ONLY: nfrg
          1
Fatal Error: Can't open module file 'comm_frginf.mod' for reading at (1): No such file or directory

The module comm_FRGINF is located in source code file 'efpmodule.src', which as diagnostic above shows, must be compiled before 'ompgrd2.src'.

Ivan

[ivr900]

This is relevant to the case when we specify

setenv GMS_OPENMP          true

in 'install.info'. For the alternative choice it is irrelevant, as the OMP source codes are not compiled.

[saromleang]

I must have overlooked the fact that you were building threaded GAMESS. We will have a look into this. Thanks.

[colleeneb]

It looks like this was fixed in the "GAMESS patched release 2019 R1 Patch 1" commit to master. ./comp efpmodule wad added after line 304 in compall, which ensures that it's compiled before ompgrd2.

[ivr900]

It looks like this was fixed in the "GAMESS patched release 2019 R1 Patch 1" commit to master. ./comp efpmodule wad added after line 304 in compall, which ensures that it's compiled before ompgrd2.

Sorry, overlooked this line. :( Still, I see more logic to compile the optional OMP source codes after the general GAMESS source code modules, so not to have the line ./comp efpmodule repeated twice.

[colleeneb]

You're right, it's good to avoid unnecessary repetition, and anything that simplifies the build is good. Will have to think about it. It looks like prec, mx_limits, efpmodule, and mod_nosp_basis are all listed twice.