Page 1 of 1

VASP.6.4.2 installation in GPU

Posted: Thu Apr 11, 2024 5:46 pm
by subrata_kundu1
Hi - I am facing issues installing VASP.6.4.2 in GPU. Following are my loaded modules -
/usr/local/apps/cuda/hpc_sdk-12.2/modulefiles/nvhpc/23.9
gcc/10.2.0
PrgEnv-intel/2022.1.0

I am using makefile.include.nvhpc_acc and changed the mpif90 to mpiifort

FC = mpiifort -gpu=cc60,cc70,cc80,cuda12.0
FCL = mpiifort -gpu=cc60,cc70,cc80,cuda12.0 -c++libs


BUT seeing following error :

ifort: command line warning #10006: ignoring unknown option '-gpu=cc60,cc70,cc80,cuda12.0'

Warning: the -fast option forces static linkage method for the Intel(R) MPI Library.
ifort: command line warning #10006: ignoring unknown option '-gpu=cc60,cc70,cc80,cuda12.0'
ifort: command line warning #10006: ignoring unknown option '-Mfree'
ifort: command line warning #10006: ignoring unknown option '-Mbackslash'
ifort: command line warning #10006: ignoring unknown option '-Mlarge_arrays'
ifort: command line warning #10159: invalid argument for option '-tp'

Warning: the -fast option forces static linkage method for the Intel(R) MPI Library.
ifort: command line warning #10006: ignoring unknown option '-gpu=cc60,cc70,cc80,cuda12.0'
ifort: command line warning #10006: ignoring unknown option '-Mfree'
ifort: command line warning #10006: ignoring unknown option '-Mbackslash'
ifort: command line warning #10006: ignoring unknown option '-Mlarge_arrays'
ifort: command line warning #10159: invalid argument for option '-tp'
nccl2for.F(34): error #7002: Error in opening the compiled module file. Check INCLUDE paths. [CUDAFOR]
use cudaFor
----^
How to resolve that issue?

Re: VASP.6.4.2 installation in GPU

Posted: Thu Apr 11, 2024 10:34 pm
by subrata_kundu1
I am seeing following error now - my nvfortran and mpifort is working

But the compilation exit out after 25-30 minutes with this line
NVFORTRAN-S-0926-Syntax error in #if (main.F: 969)
NVFORTRAN-S-0918-Missing #endif (main.F: 6151)

Re: VASP.6.4.2 installation in GPU

Posted: Fri Apr 12, 2024 8:54 am
by michael_wolloch
Dear subrate_kundu,

For compilation with openACC support for NVIDIA GPUs, you need to use the NVIDIA Fortran compiler (nvfortran), not Intel's Fortran compiler (ifort or ifx)

Can you change back to "mpif90" in your makefile.include, and make sure that the right compiler is used when calling mpif90? Do this by executing:

Code: Select all

mpif90 --version
You should see something like:

Code: Select all

bash-4.4$ mpif90 --version

nvfortran 22.9-0 64-bit target on x86-64 Linux -tp skylake-avx512 
NVIDIA Compilers and Tools
Copyright (c) 2022, NVIDIA CORPORATION & AFFILIATES.  All rights reserved.
Please consult our wiki article on the openACC port and the links therein for more information.

If you already changed your compiler back to mpif90 (I am not quite sure after your second post), please post your toolchain and makefile.include so I can help you further.

All the best, Michael