portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-09 23:45:36 +00:00
Code Releases Activity

Faster

Browse Source
This commit is contained in:
Peter Boyle 2022-09-23 16:49:14 -04:00
parent a0d682687e
commit a2cefaa53a
3 changed files with 23 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
systems/Crusher/config-command
View File

@ -6,10 +6,12 @@
--enable-simd=GPU \
--disable-fermion-reps \
--with-gmp=$OLCF_GMP_ROOT \
--with-fftw=$FFTW_DIR/.. \
--with-mpfr=/opt/cray/pe/gcc/mpfr/3.1.4/ \
--enable-gparity \
CXX=hipcc MPICXX=mpicxx \
CXXFLAGS="-fPIC -I{$ROCM_PATH}/include/ -std=c++14 -I${MPICH_DIR}/include " \
LDFLAGS=" -L${MPICH_DIR}/lib -lmpi -L${CRAY_MPICH_ROOTDIR}/gtl/lib -lmpi_gtl_hsa -lamdhip64 "
CXXFLAGS="-fPIC -I{$ROCM_PATH}/include/ -std=c++14 -I${MPICH_DIR}/include -L/lib64 " \
LDFLAGS="-L/lib64 -L/opt/rocm-5.2.0/lib/ -L${MPICH_DIR}/lib -lmpi -L${CRAY_MPICH_ROOTDIR}/gtl/lib -lmpi_gtl_hsa -lamdhip64 "

23
systems/Crusher/dwf.slurm
View File

@ -10,23 +10,26 @@
#SBATCH -n 8
#SBATCH --exclusive
#SBATCH --gpu-bind=map_gpu:0,1,2,3,7,6,5,4
#export BIND="--cpu-bind=verbose,map_ldom:3,3,1,1,2,2,0,0"
DIR=.
source sourceme.sh
export MPIR_CVAR_GPU_EAGER_DEVICE_MEM=0
export MPIR_CVAR_GPU_EAGER_DEVICE_MEM=16384
export MPICH_GPU_SUPPORT_ENABLED=1
export MPICH_SMP_SINGLE_COPY_MODE=XPMEM
export OMP_NUM_THREADS=1
export MPICH_SMP_SINGLE_COPY_MODE=CMA
export OMP_NUM_THREADS=4
export MPICH_OFI_NIC_POLICY=GPU
echo MPICH_SMP_SINGLE_COPY_MODE $MPICH_SMP_SINGLE_COPY_MODE
#PARAMS=" --accelerator-threads 8 --grid 64.64.32.16 --mpi 2.2.2.1 --comms-sequential --shm 2048 --shm-mpi 1"
#srun --gpus-per-task 1 -n8 $BIND ./wrap.sh ./benchmarks/Benchmark_dwf_fp32 $PARAMS
echo working directory
pwd
PARAMS=" --accelerator-threads 8 --grid 16.16.32.32 --mpi 1.1.1.1 --comms-sequential --shm 2048 --shm-mpi 1"
srun --gpus-per-task 1 -n1 $BIND ./wrap.sh ./benchmarks/Benchmark_dwf_fp32 $PARAMS
PARAMS=" --accelerator-threads 8 --grid 32.32.32.32 --mpi 1.1.1.1 --comms-sequential --shm 2048 --shm-mpi 0"
srun --gpus-per-task 1 -n1 ./benchmarks/Benchmark_dwf_fp32 $PARAMS
PARAMS=" --accelerator-threads 8 --grid 32.16.32.32 --mpi 1.1.1.2 --comms-sequential --shm 2048 --shm-mpi 1"
srun --gpus-per-task 1 -n2 $BIND ./wrap.sh ./benchmarks/Benchmark_dwf_fp32 $PARAMS
PARAMS=" --accelerator-threads 8 --grid 64.64.64.32 --mpi 2.2.2.1 --comms-sequential --shm 2048 --shm-mpi 0"
srun --gpus-per-task 1 -n8 ./benchmarks/Benchmark_dwf_fp32 $PARAMS
PARAMS=" --accelerator-threads 8 --grid 32.32.32.64 --mpi 1.2.2.2 --comms-sequential --shm 2048 --shm-mpi 1"
srun --gpus-per-task 1 -n8 $BIND ./wrap.sh ./benchmarks/Benchmark_dwf_fp32 $PARAMS

8
systems/Crusher/sourceme.sh
View File

@ -1,8 +1,12 @@
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/sw/crusher/spack-envs/base/opt/cray-sles15-zen3/gcc-11.2.0/gperftools-2.9.1-72ubwtuc5wcz2meqltbfdb76epufgzo2/lib
module load emacs
#module load gperftools
module load PrgEnv-gnu
module load rocm/5.1.0
module load rocm/5.2.0
module load cray-mpich/8.1.16
#module load cray-mpich/8.1.17
module load gmp
#module load cray-fftw
module load cray-fftw
module load craype-accel-amd-gfx90a
export LD_LIBRARY_PATH=/opt/gcc/mpfr/3.1.4/lib:$LD_LIBRARY_PATH
#Hack for lib