portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-13 12:47:05 +01:00
Code Releases Activity

More britney logging improvements

Browse Source
This commit is contained in:
Peter Boyle
2024-03-19 14:36:21 +00:00
parent 660eb76d93
commit fab1efb48c
7 changed files with 253 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
systems/Aurora/tests/repro16.pbs
View File

@ -2,26 +2,39 @@
## qsub -q EarlyAppAccess -A Aurora_Deployment -I -l select=1 -l walltime=60:00
#PBS -q EarlyAppAccess
#PBS -l select=16
#PBS -l walltime=02:00:00
#PBS -l select=16:system=sunspot,place=scatter
#PBS -A LatticeQCD_aesp_CNDA
#PBS -l walltime=01:00:00
#PBS -N dwf
#PBS -k doe
#export OMP_PROC_BIND=spread
#unset OMP_PLACES
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
#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
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_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE=0
export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE=0
@ -32,10 +45,17 @@ export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_NUM_BUFFERS_PER_CHUNK=16
export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_MAX_NUM_BUFFERS=16
export MPICH_OFI_NIC_POLICY=GPU
# 12 ppn, 16 nodes, 192 ranks
DIR=repro.$PBS_JOBID
mkdir $DIR
cd $DIR
CMD="mpiexec -np 192 -ppn 12 -envall \
./gpu_tile_compact.sh \
./Test_dwf_mixedcg_prec --mpi 2.4.4.6 --grid 64.128.128.192 \
--shm-mpi 1 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 6000 "
../gpu_tile_compact.sh \
../Test_dwf_mixedcg_prec --mpi 2.4.4.6 --grid 64.128.128.192 \
--shm-mpi 1 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 3000 --debug-stdout --log Message,Iterative"
#--comms-overlap
$CMD
grep Oops Grid.stderr.* > failures.$PBS_JOBID
rm core.*

81
systems/Aurora/tests/repro1gpu.pbs Normal file
View File

@ -0,0 +1,81 @@
#!/bin/bash
#PBS -l select=16:system=sunspot,place=scatter
#PBS -A LatticeQCD_aesp_CNDA
#PBS -l walltime=02:00:00
#PBS -N repro1gpu
#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_GPU_USE_IMMEDIATE_COMMAND_LIST=1
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 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
NN=`cat $PBS_NODEFILE | wc -l`
echo $PBS_NODEFILE
cat $PBS_NODEFILE
echo $NN nodes in node file
for n in `eval echo {1..$NN}`
do
THIS_NODE=`head -n$n $PBS_NODEFILE | tail -n1 `
echo Node $n is $THIS_NODE
for g in {0..11}
do
export NUMA_MAP=(0 0 0 1 1 1 0 0 0 1 1 1 )
export TILE_MAP=(0 0 0 0 0 0 1 1 1 1 1 1 )
export GPU_MAP=(0 1 2 3 4 5 0 1 2 3 4 5 )
export numa=${NUMA_MAP[$g]}
export gpu_id=${GPU_MAP[$g]}
export tile_id=${TILE_MAP[$g]}
export gpu=$gpu_id.$tile_id
cd $PBS_O_WORKDIR
DIR=repro.1gpu.$PBS_JOBID/node-$n-$THIS_NODE-GPU-$gpu
mkdir -p $DIR
cd $DIR
echo $THIS_NODE > nodefile
echo $gpu > gpu
export ZE_AFFINITY_MASK=$gpu
export ONEAPI_DEVICE_FILTER=gpu,level_zero
CMD="mpiexec -np 1 -ppn 1 -envall --hostfile nodefile \
numactl -N $numa -m $numa ../../Test_dwf_mixedcg_prec --mpi 1.1.1.1 --grid 16.16.32.32 \
--shm-mpi 0 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 6000 --debug-stdout --log Message"
echo $CMD
$CMD &
done
done
wait

78
systems/Aurora/tests/reproN.pbs Normal file
View File

@ -0,0 +1,78 @@
#!/bin/bash
#PBS -l select=16:system=sunspot,place=scatter
#PBS -A LatticeQCD_aesp_CNDA
#PBS -l walltime=02:00:00
#PBS -N reproN
#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_GPU_USE_IMMEDIATE_COMMAND_LIST=1
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 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
NN=`cat $PBS_NODEFILE | wc -l`
echo $PBS_NODEFILE
cat $PBS_NODEFILE
echo $NN nodes in node file
for n in `eval echo {1..$NN}`
do
cd $PBS_O_WORKDIR
THIS_NODE=`head -n$n $PBS_NODEFILE | tail -n1 `
echo Node $n is $THIS_NODE
DIR=repro.$PBS_JOBID/node-$n-$THIS_NODE
mkdir -p $DIR
cd $DIR
echo $THIS_NODE > nodefile
CMD="mpiexec -np 12 -ppn 12 -envall --hostfile nodefile \
../../gpu_tile_compact.sh \
../../Test_dwf_mixedcg_prec --mpi 1.2.2.3 --grid 32.64.64.96 \
--shm-mpi 0 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 6000 --debug-stdout --log Message --comms-overlap"
$CMD &
done
wait
for n in ` eval echo {1..$NN} `
do
THIS_NODE=`head -n$n $PBS_NODEFILE | tail -n1 `
DIR=repro.$PBS_JOBID/node-$n-$THIS_NODE
cd $DIR
grep Oops Grid.stderr.* > failures.$PBS_JOBID
rm core.*
done