portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-09-20 01:05:38 +01:00
Code Releases Activity

Deprecate teh PVC directory and make a PVC-OEM generic PVC target with

Browse Source 
no queueing system dependency -- just interactive scripts
This commit is contained in:
Peter Boyle 2023-10-03 17:04:20 +00:00
parent 7786ea9921
commit 6d0c2de399
10 changed files with 0 additions and 138 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

0
systems/OEM/README → systems/PVC-OEM/README
View File

0
systems/OEM/benchmarks/bench.sh → systems/PVC-OEM/benchmarks/bench.sh
View File

0
systems/OEM/benchmarks/select_gpu.sh → systems/PVC-OEM/benchmarks/select_gpu.sh
View File

0
systems/OEM/config-command → systems/PVC-OEM/config-command
View File

0
systems/OEM/setup.sh → systems/PVC-OEM/setup.sh
View File

62
systems/PVC/benchmarks/run-1tile.sh
View File

@ -1,62 +0,0 @@
#!/bin/sh
##SBATCH -p PVC-SPR-QZEH
##SBATCH -p PVC-ICX-QZNW
#SBATCH -p QZ1J-ICX-PVC
##SBATCH -p QZ1J-SPR-PVC-2C
#source /nfs/site/home/paboylex/ATS/GridNew/Grid/systems/PVC-nightly/setup.sh
export NT=8
export I_MPI_OFFLOAD=1
export I_MPI_OFFLOAD_TOPOLIB=level_zero
export I_MPI_OFFLOAD_DOMAIN_SIZE=-1
# export IGC_EnableLSCFenceUGMBeforeEOT=0
# export SYCL_PROGRAM_COMPILE_OPTIONS="-ze-opt-large-register-file=False"
export SYCL_DEVICE_FILTER=gpu,level_zero
#export IGC_ShaderDumpEnable=1
#export IGC_DumpToCurrentDir=1
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 --device-mem 32768
export ZE_AFFINITY_MASK=0
export I_MPI_OFFLOAD_CELL=device
export EnableImplicitScaling=1
export EnableWalkerPartition=1
#mpiexec -launcher ssh -n 2 -host localhost vtune -collect gpu-hotspots -knob gpu-sampling-interval=1 -data-limit=0 -r ./vtune_run4 -- ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-overlap --shm-mpi 1
#mpiexec -launcher ssh -n 1 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-overlap --shm-mpi 1
#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
#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-overlap --shm-mpi 1
#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
#mpirun -np 2 ./wrap.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.2 --grid 16.32.32.64 --accelerator-threads $NT --comms-sequential --shm-mpi 0
#mpirun -np 2 ./wrap.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.2 --grid 32.32.32.64 --accelerator-threads $NT --comms-sequential --shm-mpi 1

33
systems/PVC/benchmarks/run-2tile-mpi.sh
View File

@ -1,33 +0,0 @@
#!/bin/bash
##SBATCH -p PVC-SPR-QZEH
##SBATCH -p PVC-ICX-QZNW
#SBATCH -p QZ1J-ICX-PVC
#source /nfs/site/home/paboylex/ATS/GridNew/Grid/systems/PVC-nightly/setup.sh
export NT=16
# export IGC_EnableLSCFenceUGMBeforeEOT=0
# export SYCL_PROGRAM_COMPILE_OPTIONS="-ze-opt-large-register-file=False"
#export IGC_ShaderDumpEnable=1
#export IGC_DumpToCurrentDir=1
export I_MPI_OFFLOAD=1
export I_MPI_OFFLOAD_TOPOLIB=level_zero
export I_MPI_OFFLOAD_DOMAIN_SIZE=-1
export SYCL_DEVICE_FILTER=gpu,level_zero
export I_MPI_OFFLOAD_CELL=tile
export EnableImplicitScaling=0
export EnableWalkerPartition=0
#export SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=1
#export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
export SYCL_PI_LEVEL_ZERO_USE_COPY_ENGINE=0
for i in 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
do
mpiexec -launcher ssh -n 2 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.2 --grid 32.32.32.64 --accelerator-threads $NT --shm-mpi 0 --device-mem 32768 > 1.1.1.2.log$i
mpiexec -launcher ssh -n 2 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --shm-mpi 0 --device-mem 32768 > 2.1.1.1.log$i
done
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

9
systems/PVC/benchmarks/wrap.sh
View File

@ -1,9 +0,0 @@
#!/bin/sh
export ZE_AFFINITY_MASK=0.$MPI_LOCALRANKID
echo Ranke $MPI_LOCALRANKID ZE_AFFINITY_MASK is $ZE_AFFINITY_MASK
$@

16
systems/PVC/config-command
View File

@ -1,16 +0,0 @@
INSTALL=/nfs/site/home/paboylx/prereqs/
../../configure \
--enable-simd=GPU \
--enable-gen-simd-width=64 \
--enable-comms=mpi-auto \
--disable-accelerator-cshift \
--disable-gparity \
--disable-fermion-reps \
--enable-shm=nvlink \
--enable-accelerator=sycl \
--enable-unified=no \
MPICXX=mpicxx \
CXX=dpcpp \
LDFLAGS="-fsycl-device-code-split=per_kernel -fsycl-device-lib=all -lze_loader -L$INSTALL/lib" \
CXXFLAGS="-fsycl-unnamed-lambda -fsycl -no-fma -I$INSTALL/include -Wno-tautological-compare"

18
systems/PVC/setup.sh
View File

@ -1,18 +0,0 @@
export https_proxy=http://proxy-chain.intel.com:911
#export LD_LIBRARY_PATH=/nfs/site/home/azusayax/install/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=$HOME/prereqs/lib/:$LD_LIBRARY_PATH
module load intel-release
module load intel-comp-rt/embargo-ci-neo
#source /opt/intel/oneapi/PVC_setup.sh
#source /opt/intel/oneapi/ATS_setup.sh
#module load intel-nightly/20230331
#module load intel-comp-rt/ci-neo-master/026093
#module load intel/mpich
module load intel/mpich/pvc45.3
export PATH=~/ATS/pti-gpu/tools/onetrace/:$PATH
#clsh embargo-ci-neo-022845
#source /opt/intel/vtune_amplifier/amplxe-vars.sh