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

Aurora testing

This commit is contained in:
Peter Boyle 2024-09-10 18:11:52 +00:00
parent 160969a758
commit 11be10d2c0
4 changed files with 78 additions and 14 deletions

View File

@ -2,7 +2,8 @@
## qsub -q EarlyAppAccess -A Aurora_Deployment -I -l select=1 -l walltime=60:00
#PBS -l select=16:system=sunspot,place=scatter
#PBS -l select=16
#PBS -q EarlyAppAccess
#PBS -A LatticeQCD_aesp_CNDA
#PBS -l walltime=01:00:00
#PBS -N dwf
@ -13,19 +14,14 @@
cd $PBS_O_WORKDIR
#source ../sourceme.sh
source ../sourceme.sh
cat $PBS_NODEFILE
#export MPICH_COLL_SYNC=1
#export MPICH_ENV_DISPLAY=1
export MPICH_
export OMP_NUM_THREADS=3
export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1
module load oneapi/eng-compiler/2023.05.15.003
module load mpich/51.2/icc-all-deterministic-pmix-gpu
#export LD_LIBRARY_PATH=/soft/restricted/CNDA/updates/2023.05.15.001/oneapi/compiler/eng-20230512/compiler/linux/lib/:$LD_LIBRARY_PATH
#module load mpich/51.2/icc-all-deterministic-pmix-gpu
#unset MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE
#unset MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE
#unset MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST

View File

@ -1,6 +1,7 @@
#!/bin/bash
#PBS -l select=16:system=sunspot,place=scatter
#PBS -l select=16
#PBS -q EarlyAppAccess
#PBS -A LatticeQCD_aesp_CNDA
#PBS -l walltime=02:00:00
#PBS -N repro1gpu
@ -9,8 +10,9 @@
#export OMP_PROC_BIND=spread
#unset OMP_PLACES
module load oneapi/eng-compiler/2023.05.15.003
module load mpich/51.2/icc-all-deterministic-pmix-gpu
#module load oneapi/eng-compiler/2023.05.15.003
#module load mpich/51.2/icc-all-deterministic-pmix-gpu
# 56 cores / 6 threads ~9
export OMP_NUM_THREADS=6
@ -34,6 +36,8 @@ export SYCL_PROGRAM_COMPILE_OPTIONS="-ze-opt-large-register-file"
cd $PBS_O_WORKDIR
source ../sourceme.sh
NN=`cat $PBS_NODEFILE | wc -l`
echo $PBS_NODEFILE
cat $PBS_NODEFILE

View File

@ -0,0 +1,63 @@
#!/bin/bash
#PBS -l select=16
#PBS -q EarlyAppAccess
#PBS -A LatticeQCD_aesp_CNDA
#PBS -l walltime=02:00:00
#PBS -N reproBigJob
#PBS -k doe
#export OMP_PROC_BIND=spread
#unset OMP_PLACES
#module load oneapi/eng-compiler/2023.05.15.003
#module load mpich/51.2/icc-all-deterministic-pmix-gpu
# 56 cores / 6 threads ~9
export OMP_NUM_THREADS=6
export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1
#export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE=0
#export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE=0
#export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_BUFFER_SZ=1048576
#export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_THRESHOLD=131072
#export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_NUM_BUFFERS_PER_CHUNK=16
#export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_MAX_NUM_BUFFERS=16
#export MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST=1
export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
export SYCL_PI_LEVEL_ZERO_USE_COPY_ENGINE=1
export SYCL_PI_LEVEL_ZERO_USE_COPY_ENGINE_FOR_D2D_COPY=1
export SYCL_PROGRAM_COMPILE_OPTIONS="-ze-opt-large-register-file"
export GRID_PRINT_ENTIRE_LOG=0
export GRID_CHECKSUM_RECV_BUF=0
export GRID_CHECKSUM_SEND_BUF=0
export MPICH_OFI_NIC_POLICY=GPU
export MPIR_CVAR_ALLREDUCE_DEVICE_COLLECTIVE=0
export MPIR_CVAR_REDUCE_DEVICE_COLLECTIVE=0
export MPIR_CVAR_ALLREDUCE_INTRA_ALGORITHM=recursive_doubling
unset MPIR_CVAR_CH4_COLL_SELECTION_TUNING_JSON_FILE
unset MPIR_CVAR_COLL_SELECTION_TUNING_JSON_FILE
unset MPIR_CVAR_CH4_POSIX_COLL_SELECTION_TUNING_JSON_FILE
cd $PBS_O_WORKDIR
DIR=reproBigJob.$PBS_JOBID
mkdir -p $DIR
cd $DIR
cp $PBS_NODEFILE nodefile
CMD="mpiexec -np 192 -ppn 12 -envall --hostfile nodefile \
../gpu_tile_compact.sh \
../Test_dwf_mixedcg_prec --mpi 4.4.4.3 --grid 128.128.128.96 \
--shm-mpi 0 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 6000 --debug-stdout --log Message --comms-overlap"
echo $CMD > command-line
env > environment
$CMD
grep Oops Grid.stderr.* > failures.$PBS_JOBID
rm core.*

View File

@ -1,6 +1,7 @@
#!/bin/bash
#PBS -l select=32:system=sunspot,place=scatter
#PBS -l select=16
#PBS -q EarlyAppAccess
#PBS -A LatticeQCD_aesp_CNDA
#PBS -l walltime=02:00:00
#PBS -N reproN
@ -9,8 +10,8 @@
#export OMP_PROC_BIND=spread
#unset OMP_PLACES
module load oneapi/eng-compiler/2023.05.15.003
module load mpich/51.2/icc-all-deterministic-pmix-gpu
#module load oneapi/eng-compiler/2023.05.15.003
#module load mpich/51.2/icc-all-deterministic-pmix-gpu
# 56 cores / 6 threads ~9
export OMP_NUM_THREADS=6