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

Updated scripts for crusher

This commit is contained in:
Peter Boyle 2022-02-14 17:55:16 -05:00
parent a3b022d469
commit f49d5c2d22
5 changed files with 24 additions and 19 deletions

View File

@ -8,5 +8,5 @@
--disable-gparity \
CXX=hipcc MPICXX=mpicxx \
CXXFLAGS="-fPIC -I/opt/rocm-4.5.0/include/ -std=c++14 -I${MPICH_DIR}/include " \
--prefix=/ccs/home/chulwoo/Grid \
LDFLAGS=" -L${MPICH_DIR}/lib -lmpi -L${CRAY_MPICH_ROOTDIR}/gtl/lib -lmpi_gtl_hsa "
HIPFLAGS = --amdgpu-target=gfx90a

View File

@ -3,24 +3,28 @@
#SBATCH -A LGT104
#SBATCH -t 01:00:00
##SBATCH -U openmpThu
#SBATCH -p ecp
##SBATCH -p ecp
#SBATCH -J DWF
#SBATCH -o DWF.%J
#SBATCH -e DWF.%J
#SBATCH -N 1
#SBATCH -n 1
#SBATCH --exclusive
DIR=.
module list
export MPIR_CVAR_GPU_EAGER_DEVICE_MEM=0
#export MPIR_CVAR_GPU_EAGER_DEVICE_MEM=0
export MPICH_GPU_SUPPORT_ENABLED=1
#export MPICH_SMP_SINGLE_COPY_MODE=XPMEM
export MPICH_SMP_SINGLE_COPY_MODE=XPMEM
#export MPICH_SMP_SINGLE_COPY_MODE=NONE
export MPICH_SMP_SINGLE_COPY_MODE=CMA
export OMP_NUM_THREADS=8
#export MPICH_SMP_SINGLE_COPY_MODE=CMA
export OMP_NUM_THREADS=1
AT=8
echo MPICH_SMP_SINGLE_COPY_MODE $MPICH_SMP_SINGLE_COPY_MODE
PARAMS=" --accelerator-threads ${AT} --grid 32.32.32.32 --mpi 1.1.1.1 --comms-overlap"
srun -n1 --label -c$OMP_NUM_THREADS --gpus-per-task=1 ./mpiwrapper.sh ./benchmarks/Benchmark_dwf_fp32 $PARAMS
PARAMS=" --accelerator-threads ${AT} --grid 24.24.24.24 --shm-mpi 0 --mpi 1.1.1.1"
srun --gpus-per-task 1 -n1 ./benchmarks/Benchmark_dwf_fp32 $PARAMS

View File

@ -3,12 +3,12 @@
#SBATCH -A LGT104
#SBATCH -t 01:00:00
##SBATCH -U openmpThu
#SBATCH -p ecp
#SBATCH -J DWF
#SBATCH -o DWF.%J
#SBATCH -e DWF.%J
#SBATCH -N 1
#SBATCH -n 4
#SBATCH --exclusive
DIR=.
module list
@ -19,8 +19,9 @@ export MPICH_SMP_SINGLE_COPY_MODE=NONE
#export MPICH_SMP_SINGLE_COPY_MODE=CMA
export OMP_NUM_THREADS=4
AT=8
echo MPICH_SMP_SINGLE_COPY_MODE $MPICH_SMP_SINGLE_COPY_MODE
PARAMS=" --accelerator-threads ${AT} --grid 32.32.64.64 --mpi 1.1.2.2 --comms-overlap --shm 2048 --shm-mpi 0"
srun -n4 --label -c$OMP_NUM_THREADS --gpus-per-task=1 ./mpiwrapper.sh ./benchmarks/Benchmark_dwf_fp32 $PARAMS
PARAMS=" --accelerator-threads 8 --grid 32.32.64.64 --mpi 1.1.2.2 --comms-overlap --shm 2048 --shm-mpi 0"
srun --gpus-per-task 1 -n4 ./mpiwrapper.sh ./benchmarks/Benchmark_dwf_fp32 $PARAMS

View File

@ -1,14 +1,14 @@
#!/bin/bash
# Begin LSF Directives
#SBATCH -A LGT104
#SBATCH -t 00:20:00
#SBATCH -p ecp
#SBATCH -t 01:00:00
##SBATCH -U openmpThu
#SBATCH -J DWF
#SBATCH -o DWF.%J
#SBATCH -e DWF.%J
#SBATCH -N 1
#SBATCH -n 8
#SBATCH --exclusive
DIR=.
module list
@ -19,9 +19,9 @@ export MPICH_SMP_SINGLE_COPY_MODE=NONE
#export MPICH_SMP_SINGLE_COPY_MODE=CMA
export OMP_NUM_THREADS=1
AT=8
echo MPICH_SMP_SINGLE_COPY_MODE $MPICH_SMP_SINGLE_COPY_MODE
PARAMS=" --accelerator-threads ${AT} --grid 32.64.64.64 --mpi 1.2.2.2 --comms-overlap --shm 2048 --shm-mpi 0"
PARAMS=" --accelerator-threads 8 --grid 32.64.64.64 --mpi 1.2.2.2 --comms-overlap --shm 2048 --shm-mpi 0"
srun --gpus-per-task 1 -n8 ./mpiwrapper.sh ./benchmarks/Benchmark_dwf_fp32 $PARAMS
srun -n8 -c 1 --gpus-per-task=1 --gpu-bind=closest --label -c$OMP_NUM_THREADS --gpus-per-task=1 ./mpiwrapper.sh ./benchmarks/Benchmark_dwf_fp32 $PARAMS

View File

@ -2,4 +2,4 @@ module load PrgEnv-gnu
module load rocm/4.5.0
module load gmp
module load cray-fftw
module load craype-accel-amd-gfx908
module load craype-accel-amd-gfx90a