portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2026-05-21 17:44:16 +01:00
Code Releases Activity

Perlmutter compile update

Browse Source
This commit is contained in:
Peter Boyle
2026-04-27 13:53:28 -07:00
parent 2a8084d569
commit e4404efe5a
2 changed files with 23 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files
systems/Perlmutter/config-command
+4 -5
View File
@@ -1,17 +1,16 @@
DIR=`pwd` DIR=`pwd`
PREFIX=$DIR/../Prequisites/install/ PREFIX=$HOME/DDHMC/Grid/systems/Prerequisites/install/
../../configure \ ../../configure \
--enable-comms=mpi \ --enable-comms=mpi \
--enable-simd=GPU \ --enable-simd=GPU \
--enable-shm=nvlink \ --enable-shm=nvlink \
--enable-gen-simd-width=64 \ --enable-gen-simd-width=64 \
--enable-accelerator=cuda \
--enable-setdevice \
--disable-accelerator-cshift \
--with-gmp=$PREFIX \ --with-gmp=$PREFIX \
--with-mpfr=$PREFIX \
--enable-accelerator=cuda \
--disable-fermion-reps \ --disable-fermion-reps \
--disable-unified \ --disable-unified \
--disable-gparity \ --disable-gparity \
CXX=nvcc \ CXX=nvcc \
LDFLAGS="-cudart shared " \ LDFLAGS="-cudart shared " \
CXXFLAGS="-ccbin CC -gencode arch=compute_80,code=sm_80 -std=c++14 -cudart shared" CXXFLAGS="-ccbin CC -gencode arch=compute_80,code=sm_80 -std=c++17 -cudart shared"
systems/Perlmutter/dwf4.slurm
+19 -6
View File
@@ -1,5 +1,6 @@
#!/bin/bash #!/bin/bash
#SBATCH -A m3886_g #SBATCH -A m5294_g
#m3886_g
#SBATCH -C gpu #SBATCH -C gpu
#SBATCH -q debug #SBATCH -q debug
#SBATCH -t 0:20:00 #SBATCH -t 0:20:00
@@ -19,9 +20,21 @@ export MPICH_GPU_EAGER_REGISTER_HOST_MEM=0
export MPICH_GPU_NO_ASYNC_MEMCPY=0 export MPICH_GPU_NO_ASYNC_MEMCPY=0
#export MPICH_SMP_SINGLE_COPY_MODE=CMA #export MPICH_SMP_SINGLE_COPY_MODE=CMA
OPT="--comms-sequential --shm-mpi 1" cat << EOF > select_gpu
VOL=64.64.64.64 #!/bin/bash
srun ./benchmarks/Benchmark_dwf_fp32 --mpi 2.2.1.1 --grid $VOL --accelerator-threads 8 --shm 2048 $OPT export GPU_MAP=(0 1 2 3)
#srun ./benchmarks/Benchmark_dwf_fp32 --mpi 2.1.1.4 --grid $VOL --accelerator-threads 8 --shm 2048 $OPT export NUMA_MAP=( 0 1 2 3 )
#srun ./benchmarks/Benchmark_dwf_fp32 --mpi 1.1.1.8 --grid $VOL --accelerator-threads 8 --shm 2048 $OPT export GPU=\$SLURM_LOCALID
export NUMA=\$SLURM_LOCALID
export CUDA_VISIBLE_DEVICES=\$GPU
exec numactl -m \$NUMA -N \$NUMA \$*
EOF
chmod +x ./select_gpu
OPT="--comms-sequential --shm-mpi 0"
VOL=64.64.32.32
srun ./select_gpu ./benchmarks/Benchmark_dwf_fp32 --mpi 2.2.1.1 --grid $VOL --device-mem 16000 --accelerator-threads 8 --shm 2048 $OPT
OPT="--comms-overlap --shm-mpi 0"
srun ./select_gpu ./benchmarks/Benchmark_dwf_fp32 --mpi 2.2.1.1 --grid $VOL --device-mem 16000 --accelerator-threads 8 --shm 2048 $OPT