My Community

I'm experienceing wierd problem when vasp.6.2.0 vasp_std running with nvidia hpc-sdk compilers (hpc-sdk 21.3 version).

The cpu usage blows up for each mpi processes like the picture below.
stdout also shows these messages I have not seen in gpu compiled version, nor vasp5.



To run, I exported two LD_LIBRARY_PATHs ;
/opt/nvidia/hpc_sdk/Linux_x86_64/21.3/comm_libs/openmpi/openmpi-3.1.5/lib
/opt/nvidia/hpc_sdk/Linux_x86_64/21.3/compilers/extras/qd/lib

Is this normal?
I post on this board since I think it might be compiler related?
The job does run until the end.
Any advice would be appreciated.

Thank you in advance.
------------------------
makefile.include
------------------------
# Precompiler options
CPP_OPTIONS= -DHOST=\"LinuxPGI\" \
-DMPI -DMPI_BLOCK=8000 -DMPI_INPLACE -Duse_collective \
-DscaLAPACK \
-DCACHE_SIZE=4000 \
-Davoidalloc \
-Dvasp6 \
-Duse_bse_te \
-Dtbdyn \
-Dqd_emulate \
-Dfock_dblbuf

nvpath = /opt/nvidia/hpc_sdk/Linux_x86_64/21.3/compilers/bin

CPP = $(nvpath)/nvfortran -Mpreprocess -Mfree -Mextend -E $(CPP_OPTIONS) $*$(FUFFIX) > $*$(SUFFIX)

FC = mpif90
FCL = mpif90 -c++libs

FREE = -Mfree

FFLAGS = -Mbackslash -Mlarge_arrays

OFLAG = -fast

DEBUG = -Mfree -O0 -traceback

# Specify your NV HPC-SDK installation, try to set NVROOT automatically
NVROOT =$(shell which nvfortran | awk -F /compilers/bin/nvfortran '{ print $$1 }')
# ...or set NVROOT manually
NVHPC ?= /opt/nvidia/hpc_sdk
NVVERSION = 21.3
NVROOT = $(NVHPC)/Linux_x86_64/$(NVVERSION)

# Use NV HPC-SDK provided BLAS and LAPACK libraries
BLAS = -lblas
LAPACK = -llapack

BLACS =
SCALAPACK = -Mscalapack

LLIBS = $(SCALAPACK) $(LAPACK) $(BLAS)

# Software emulation of quadruple precsion
QD ?= $(NVROOT)/compilers/extras/qd
LLIBS += -L$(QD)/lib -lqdmod -lqd
INCS += -I$(QD)/include/qd

# Use the FFTs from fftw
FFTW ?= /xtmp/khs/fftw3/fftw-3.3.8
LLIBS += -L$(FFTW)/lib -lfftw3
INCS += -I$(FFTW)/include

OBJECTS = fftmpiw.o fftmpi_map.o fftw3d.o fft3dlib.o

# Redefine the standard list of O1 and O2 objects
SOURCE_O1 := pade_fit.o
SOURCE_O2 := pead.o

# For what used to be vasp.5.lib
CPP_LIB = $(CPP)
FC_LIB = nvfortran
CC_LIB = nvc
CFLAGS_LIB = -O
FFLAGS_LIB = -O1 -Mfixed
FREE_LIB = $(FREE)

OBJECTS_LIB= linpack_double.o getshmem.o

# For the parser library
CXX_PARS = nvc++ --no_warnings

# Normally no need to change this
SRCDIR = ../../src
BINDIR = ../../bin

#================================================
# GPU Stuff

CPP_GPU = -DCUDA_GPU -DRPROMU_CPROJ_OVERLAP -DCUFFT_MIN=28 -UscaLAPACK -Ufock_dblbuf # -DUSE_PINNED_MEMORY

OBJECTS_GPU= fftmpiw.o fftmpi_map.o fft3dlib.o fftw3d_gpu.o fftmpiw_gpu.o

CC = nvc
CXX = nvc++
CFLAGS = -fPIC -DADD_ -mp -cuda -DMAGMA_WITH_MKL -DMAGMA_SETAFFINITY -DGPUSHMEM=300 -DHAVE_CUBLAS

## Use a custom CUDA installation: minimal requirement is CUDA >= 10.X. For "sm_80" you need CUDA >= 11.X.
#CUDA_ROOT ?= /usr/local/cuda
#NVCC := $(CUDA_ROOT)/bin/nvcc
#CUDA_LIB := -L$(CUDA_ROOT)/lib64 -lnvToolsExt -lcudart -lcuda -lcufft -lcublas

# Or use the CUDA installation from the NV HPC-SDK
#CUDA_ROOT ?= $(NVROOT)/cuda/11.0
NVCC := nvcc
CUDA_LIB = -cudalib=cublas,cufft -lnvToolsExt

GENCODE_ARCH := -gencode=arch=compute_60,code=\"sm_60,compute_60\" \
-gencode=arch=compute_70,code=\"sm_70,compute_70\" \
-gencode=arch=compute_80,code=\"sm_80,compute_80\"

MPI_INC = $(NVROOT)/comm_libs/mpi/include

Hi!

This is certainly not normal but at this point I can only speculate what the reason could be... what kind of machine are you using? Is this an 8 core machine? Am I assuming correctly that you started this run with 8 MPI tasks? Although you are obviously compiling without OpenMP or OpenACC support it could be that some library called from VASP is creating multiple threads. Can you try to run again but before set the number of threads manually to 1, e.g. by The stdout output is particularly weird, can you comment if you get this only on the screen or also in some files?

Hi.

Thanks for the advice, the solution (setting OMP_NUM_THREADS=1) works! The load no longer explodes.

But the stdout remains as above, outputting strange lines.. Though I cut off the calculation within few electronic steps, I have attached the file, in case it may be of use.

My machine is running Ubuntu 18, and indeed I ran with "mpirun -np 8" on AMD threadripper 2990wx(Total 32 cores).