mirror of
https://github.com/paboyle/Grid.git
synced 2024-11-09 23:45:36 +00:00
Files for SYCL
This commit is contained in:
parent
bb0a0da47a
commit
17d7177105
62
systems/PVC/benchmarks/run-1tile.sh
Normal file
62
systems/PVC/benchmarks/run-1tile.sh
Normal file
@ -0,0 +1,62 @@
|
||||
#!/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=16
|
||||
|
||||
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 --cacheblocking 8.8.8.8
|
||||
|
||||
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
|
||||
|
26
systems/PVC/benchmarks/run-2tile-mpi.sh
Executable file
26
systems/PVC/benchmarks/run-2tile-mpi.sh
Executable file
@ -0,0 +1,26 @@
|
||||
#!/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
|
||||
|
||||
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 2 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-sequential --shm-mpi 1 > 2tile.log
|
||||
|
14
systems/PVC/benchmarks/wrap.sh
Executable file
14
systems/PVC/benchmarks/wrap.sh
Executable file
@ -0,0 +1,14 @@
|
||||
#!/bin/sh
|
||||
|
||||
export ZE_AFFINITY_MASK=0.$MPI_LOCALRANKID
|
||||
|
||||
echo Ranke $MPI_LOCALRANKID ZE_AFFINITY_MASK is $ZE_AFFINITY_MASK
|
||||
|
||||
|
||||
if [ $MPI_LOCALRANKID = "0" ]
|
||||
then
|
||||
# ~psteinbr/build_pti/ze_tracer -c $@
|
||||
onetrace --chrome-kernel-timeline $@
|
||||
else
|
||||
$@
|
||||
fi
|
15
systems/PVC/config-command
Normal file
15
systems/PVC/config-command
Normal file
@ -0,0 +1,15 @@
|
||||
INSTALL=/nfs/site/home/azusayax/install
|
||||
../../configure \
|
||||
--enable-simd=GPU \
|
||||
--enable-gen-simd-width=64 \
|
||||
--enable-comms=mpi \
|
||||
--disable-accelerator-cshift \
|
||||
--disable-gparity \
|
||||
--disable-fermion-reps \
|
||||
--enable-shm=nvlink \
|
||||
--enable-accelerator=sycl \
|
||||
--enable-unified=yes \
|
||||
CXX=mpicxx \
|
||||
LDFLAGS="-fsycl-device-code-split=per_kernel -fsycl-device-lib=all -lze_loader -L$INSTALL/lib" \
|
||||
CXXFLAGS="-cxx=dpcpp -fsycl-unnamed-lambda -fsycl -no-fma -I$INSTALL/include -Wtautological-constant-compare"
|
||||
|
11
systems/PVC/setup.sh
Normal file
11
systems/PVC/setup.sh
Normal file
@ -0,0 +1,11 @@
|
||||
export https_proxy=http://proxy-chain.intel.com:911
|
||||
export LD_LIBRARY_PATH=/nfs/site/home/azusayax/install/lib:$LD_LIBRARY_PATH
|
||||
|
||||
module load intel-release
|
||||
source /opt/intel/oneapi/PVC_setup.sh
|
||||
#source /opt/intel/oneapi/ATS_setup.sh
|
||||
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
|
Loading…
Reference in New Issue
Block a user