portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-13 20:57:06 +01:00
Code Releases Activity

COmms improvements

Browse Source
This commit is contained in:
Peter Boyle
2023-03-21 08:53:56 -07:00
parent a997d24743
commit e1c326558a
9 changed files with 46 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
systems/PVC/benchmarks/run-1tile.sh
View File

@ -21,7 +21,7 @@ export I_MPI_OFFLOAD_CELL=tile
export EnableImplicitScaling=0
export EnableWalkerPartition=0
export ZE_AFFINITY_MASK=0.0
mpiexec -launcher ssh -n 1 -host localhost ./Benchmark_dwf_fp32 --mpi 1.1.1.1 --grid 32.32.32.32 --accelerator-threads $NT --comms-sequential --shm-mpi 1 --cacheblocking 8.8.8.8
mpiexec -launcher ssh -n 1 -host localhost ./Benchmark_dwf_fp32 --mpi 1.1.1.1 --grid 32.32.32.32 --accelerator-threads $NT --comms-sequential --shm-mpi 0
export ZE_AFFINITY_MASK=0
export I_MPI_OFFLOAD_CELL=device

4
systems/PVC/benchmarks/run-2tile-mpi.sh
View File

@ -20,7 +20,7 @@ export I_MPI_OFFLOAD_CELL=tile
export EnableImplicitScaling=0
export EnableWalkerPartition=0
mpiexec -launcher ssh -n 1 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.1 --grid 32.32.32.32 --accelerator-threads $NT --comms-sequential --shm-mpi 1 > 1tile.log
#mpiexec -launcher ssh -n 1 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.1 --grid 32.32.32.32 --accelerator-threads $NT --comms-sequential --shm-mpi 0 > 1tile.log
mpiexec -launcher ssh -n 2 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-sequential --shm-mpi 1 > 2tile.log
mpiexec -launcher ssh -n 2 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-sequential --shm-mpi 0

10
systems/PVC/benchmarks/wrap.sh
View File

@ -5,10 +5,10 @@ export ZE_AFFINITY_MASK=0.$MPI_LOCALRANKID
echo Ranke $MPI_LOCALRANKID ZE_AFFINITY_MASK is $ZE_AFFINITY_MASK
if [ $MPI_LOCALRANKID = "0" ]
then
#if [ $MPI_LOCALRANKID = "0" ]
#then
# ~psteinbr/build_pti/ze_tracer -c $@
onetrace --chrome-kernel-timeline $@
else
# onetrace --chrome-kernel-timeline $@
#else
$@
fi
#fi

6
systems/PVC/config-command
View File

@ -1,4 +1,4 @@
INSTALL=/nfs/site/home/azusayax/install
INSTALL=/nfs/site/home/paboylx/prereqs/
../../configure \
--enable-simd=GPU \
--enable-gen-simd-width=64 \
@ -8,8 +8,8 @@ INSTALL=/nfs/site/home/azusayax/install
--disable-fermion-reps \
--enable-shm=nvlink \
--enable-accelerator=sycl \
--enable-unified=yes \
--enable-unified=no \
CXX=mpicxx \
LDFLAGS="-fsycl-device-code-split=per_kernel -fsycl-device-lib=all -lze_loader -L$INSTALL/lib" \
CXXFLAGS="-cxx=icpx -fsycl-unnamed-lambda -fsycl -I$INSTALL/include -Wno-tautological-constant-compare"
CXXFLAGS="-cxx=icpx -fsycl-unnamed-lambda -fsycl -Wno-tautological-constant-compare -I$INSTALL/include"

3
systems/PVC/setup.sh
View File

@ -1,5 +1,6 @@
export https_proxy=http://proxy-chain.intel.com:911
export LD_LIBRARY_PATH=/nfs/site/home/azusayax/install/lib:$LD_LIBRARY_PATH
#export LD_LIBRARY_PATH=/nfs/site/home/azusayax/install/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=$HOME/prereqs/lib/:$LD_LIBRARY_PATH
module load intel-release
source /opt/intel/oneapi/PVC_setup.sh