1
0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-19 08:17:05 +01:00

Almost working on Aurora

This commit is contained in:
2024-09-23 09:43:50 +00:00
parent e637fbacae
commit 02c8178f16
7 changed files with 118 additions and 98 deletions

View File

@ -15,13 +15,13 @@
# 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_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=10485760
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
@ -30,20 +30,22 @@ 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=1
export GRID_CHECKSUM_SEND_BUF=1
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
#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
cp $PBS_NODEFILE nodefile
DIR=reproBigJob.$PBS_JOBID
mkdir -p $DIR
@ -51,10 +53,19 @@ cd $DIR
cp $PBS_NODEFILE nodefile
BINARY=../Test_dwf_mixedcg_prec
echo > pingjob <<EOF
while read node ;
do
echo ssh $node killall -s USR1 -- ../Test_dwf_mixedcg_prec
done < nodefile
EOF
CMD="mpiexec -np 384 -ppn 12 -envall --hostfile nodefile \
../gpu_tile_compact.sh \
../Test_dwf_mixedcg_prec --mpi 4.4.4.6 --grid 128.128.128.96 \
--shm-mpi 1 --comms-overlap --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 6000 --debug-stdout --log Message --debug-signals"
$BINARY --mpi 4.4.4.6 --grid 128.128.128.96 \
--shm-mpi 1 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 6000 --debug-stdout --log Message --debug-signals"
echo $CMD > command-line
env > environment