My Community

We compiled the GPU version of VASP 6.2.0, and if we run it with 1 MPI task and 1 GPU, all tests are passes. However, if we run it with more than 1 (We tried with 2 MPI tasks (2 GPU) and 4 MPI tasks (4 GPU)) then 16 tests jobs fail with segmentation fault (no core dumped), the rest passes (at least no wrong results). We are not really sure how to proceed.


We uploaded the failed test jobs here: https://file.io/CzdnfRbSpowr (without WAVECAR, etc, it said file type not supported here), and the output of the whole test package (run.out). Any suggestions?


Below you can find the textual details:
Hardware:
2x Intel(R) Xeon(R) Gold 6242 CPU @ 2.80GHz
12x 32Gb DDR4 2933 MT/s DIMM
4x NVIDIA Corporation GV100GL [Tesla V100 SXM2 32GB] (rev a1)

Toolchains (All components compiled with NVHPC):
NVHPC 20.9
CUDA 11.0.2
OpenMPI 4.0.3 (NVHPC bundled OpenMPI is not an option, due to the incompatible UCX)
QD 2.3.17
Intel MKL 2020.1.217

NVHPC 21.2
CUDA 11.2.1
OpenMPI 4.0.5 (NVHPC bundled OpenMPI is not an option, due to the incompatible UCX)
QD 2.3.17
Intel MKL 2020.4.304

makefile:
# 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 \
-D_OPENACC \
-DUSENCCL -DUSENCCLP2P

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

FC = mpif90 -acc -gpu=cc70,cuda11.0
FCL = mpif90 -acc -gpu=cc70,cuda11.0 -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 = 20.9
#NVROOT = $(NVHPC)/Linux_x86_64/$(NVVERSION)

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

BLACS =
SCALAPACK = -Mscalapack

CUDA = -cudalib=cublas,cusolver,cufft,nccl -cuda

#LLIBS = $(SCALAPACK) $(LAPACK) $(BLAS) $(CUDA)
LLIBS = -L${MKLROOT}/lib/intel64 -lmkl_scalapack_lp64 -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lmkl_blacs_openmpi_lp64 -lpthread -lm -ldl $(CUDA)

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

# Use the FFTs from fftw
#FFTW ?= /opt/gnu/fftw-3.3.6-pl2-GNU-5.4.0
#LLIBS += -L$(FFTW)/lib -lfftw3
#INCS += -I$(FFTW)/include
INCS += -I$(MKLROOT)/include/fftw

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

Hello!

Could you please reupload the test results, the files in the link seem to be deleted, thank you! Did you run the "fast" or the full testsuite?

One potential source of the problem could be the OpenMPI build: Since you explicitly mentioned that you could not use the OpenMPI delivered with NVHPC.. did you compile a cuda-aware OpenMPI?

Hey Andreas,

You can find the files here:
https://we.tl/t-m3F6sx87t0

Thanks,

Michael