portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-08-17 19:51:54 +01:00
Code Releases Activity

Assertion updates to macros (mostly) with backtrace.

Browse Source 
WIlson flow to include options for DBW2, Iwasaki, Symanzik.
View logging for data assurance
This commit is contained in:
Peter Boyle
2025-08-07 15:48:38 +00:00
parent 41f344bbd3
commit 9e6a4a4737
306 changed files with 2013 additions and 1928 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
systems/Aurora/config-command
View File

@@ -2,7 +2,7 @@ export MPFR=`spack find --paths mpfr | grep ^mpfr | awk '{print $2}' `
export GMP=`spack find --paths gmp | grep ^gmp | awk '{print $2}' `
export CLIME=`spack find --paths c-lime | grep ^c-lime | awk '{print $2}' `
export UNWIND=`spack find --paths libunwind | grep ^libunwind | awk '{print $2}' `
export HDF5=/opt/cray/pe/hdf5/1.12.2.3/gnu/9.1
../../configure \
--enable-simd=GPU \
--enable-gen-simd-width=64 \
@@ -16,6 +16,7 @@ export UNWIND=`spack find --paths libunwind | grep ^libunwind | awk '{print $2
--enable-accelerator-aware-mpi=no \
--enable-unified=no \
--with-lime=$CLIME \
--with-hdf5=${HDF5} \
--with-gmp=$GMP \
--with-mpfr=$MPFR \
--with-unwind=$UNWIND \

38
systems/Aurora/tests/reproBigJob.pbs
View File

@@ -1,30 +1,21 @@
#!/bin/bash
#PBS -l select=32
#PBS -q EarlyAppAccess
#PBS -q prod
#PBS -A LatticeQCD_aesp_CNDA
#PBS -l walltime=02:00:00
#PBS -N reproBigJob
#PBS -k doe
#PBS -l filesystems=flare
#PBS -l filesystems=home
#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=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
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"
@@ -33,17 +24,12 @@ 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
source ../source.sh
cp $PBS_NODEFILE nodefile
DIR=reproBigJob.$PBS_JOBID
@@ -58,17 +44,17 @@ BINARY=../Test_dwf_mixedcg_prec
echo > pingjob <<EOF
while read node ;
do
echo ssh $node killall -s USR1 -- ../Test_dwf_mixedcg_prec
echo ssh $node killall -HUP Test_dwf_mixedcg_prec
done < nodefile
EOF
CMD="mpiexec -np 384 -ppn 12 -envall --hostfile nodefile \
../gpu_tile_compact.sh \
$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"
../gpu_tile.sh \
$BINARY --mpi 4.4.4.6 --grid 64.64.64.96 \
--shm-mpi 0 --comms-overlap --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 6000 --log Message "
echo $CMD > command-line
env > environment
$CMD
grep Oops Grid.stderr.* > failures.$PBS_JOBID
rm core.*
grep Oops */Grid.stderr.* > failures.$PBS_JOBID