portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-13 20:57:06 +01:00
Code Releases Activity

Merge branch 'feature/dirichlet' of https://github.com/paboyle/Grid into feature/dirichlet

Browse Source
This commit is contained in:
Peter Boyle
2022-05-25 12:38:10 -04:00
parent 136d843ce7 18028f4309
commit 60f4cb0ffd
63 changed files with 2295 additions and 1120 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
systems/Perlmutter/config-command
View File

@ -1,9 +1,14 @@
DIR=`pwd`
PREFIX=$DIR/../Prequisites/install/
../../configure \
--enable-comms=mpi \
--enable-simd=GPU \
--enable-shm=nvlink \
--enable-gen-simd-width=64 \
--enable-accelerator=cuda \
--enable-setdevice \
--disable-accelerator-cshift \
--with-gmp=$PREFIX \
--disable-fermion-reps \
--disable-unified \
--disable-gparity \

31
systems/Perlmutter/dwf4.slurm
View File

@ -1,24 +1,27 @@
#!/bin/bash
#SBATCH -A mp13
#SBATCH -A m3886_g
#SBATCH -C gpu
#SBATCH -q regular
#SBATCH -q debug
#SBATCH -t 0:20:00
#SBATCH -n 16
#SBATCH --ntasks-per-node=4
#SBATCH -c 32
#SBATCH --exclusive
#SBATCH -N 1
#SBATCH -n 4
#SBATCH --ntasks-per-node=4
#SBATCH --gpus-per-task=1
#SBATCH --gpu-bind=map_gpu:0,1,2,3
#SBATCH --exclusive
#SBATCH --gpu-bind=none
export SLURM_CPU_BIND="cores"
export MPICH_RDMA_ENABLED_CUDA=1
export MPICH_GPU_SUPPORT_ENABLED=1
srun ./benchmarks/Benchmark_comms_host_device --mpi 2.2.2.2 --accelerator-threads 8 > comms.4node
export MPICH_RDMA_ENABLED_CUDA=1
export MPICH_GPU_IPC_ENABLED=1
export MPICH_GPU_EAGER_REGISTER_HOST_MEM=0
export MPICH_GPU_NO_ASYNC_MEMCPY=0
#export MPICH_SMP_SINGLE_COPY_MODE=CMA
OPT="--comms-overlap --comms-concurrent --shm-mpi 0"
srun ./benchmarks/Benchmark_dwf_fp32 --mpi 2.2.2.2 --grid 64.64.64.64 --accelerator-threads 8 --shm 2048 $OPT > dwf.64.64.64.64.4node.opt0
srun ./benchmarks/Benchmark_dwf_fp32 --mpi 2.2.2.2 --grid 48.48.48.48 --accelerator-threads 8 --shm 2048 $OPT > dwf.48.48.48.48.4node.opt0
OPT="--comms-overlap --shm-mpi 1"
VOL=64.64.32.32
srun ./benchmarks/Benchmark_dwf_fp32 --mpi 2.2.1.1 --grid $VOL --accelerator-threads 8 --shm 2048 $OPT
#srun ./benchmarks/Benchmark_dwf_fp32 --mpi 2.1.1.4 --grid $VOL --accelerator-threads 8 --shm 2048 $OPT
#srun ./benchmarks/Benchmark_dwf_fp32 --mpi 1.1.1.8 --grid $VOL --accelerator-threads 8 --shm 2048 $OPT
OPT="--comms-overlap --comms-concurrent --shm-mpi 1"
srun ./benchmarks/Benchmark_dwf_fp32 --mpi 2.2.2.2 --grid 64.64.64.64 --accelerator-threads 8 --shm 2048 $OPT > dwf.64.64.64.64.4node.opt1
srun ./benchmarks/Benchmark_dwf_fp32 --mpi 2.2.2.2 --grid 48.48.48.48 --accelerator-threads 8 --shm 2048 $OPT > dwf.48.48.48.48.4node.opt1

2
systems/Perlmutter/sourceme.sh
View File

@ -1,4 +1,4 @@
export CRAY_ACCEL_TARGET=nvidia80
module load PrgEnv-gnu cpe-cuda cuda
module load PrgEnv-gnu cpe-cuda cudatoolkit/11.4