portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-09 13:40:46 +01:00
Code Releases Activity

Device mem

Browse Source
This commit is contained in:
Peter Boyle 2022-11-08 13:22:57 -08:00
parent 1873101362
commit 5c75aa5008
1 changed files with 9 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
systems/PVC/benchmarks/run-2tile-mpi.sh
View File

@ -8,7 +8,6 @@ source /nfs/site/home/paboylex/ATS/GridNew/Grid/systems/PVC-nightly/setup.sh
export NT=16 export NT=16
# export IGC_EnableLSCFenceUGMBeforeEOT=0 # export IGC_EnableLSCFenceUGMBeforeEOT=0
# export SYCL_PROGRAM_COMPILE_OPTIONS="-ze-opt-large-register-file=False" # export SYCL_PROGRAM_COMPILE_OPTIONS="-ze-opt-large-register-file=False"
#export IGC_ShaderDumpEnable=1 #export IGC_ShaderDumpEnable=1
@ -20,14 +19,16 @@ export SYCL_DEVICE_FILTER=gpu,level_zero
export I_MPI_OFFLOAD_CELL=tile export I_MPI_OFFLOAD_CELL=tile
export EnableImplicitScaling=0 export EnableImplicitScaling=0
export EnableWalkerPartition=0 export EnableWalkerPartition=0
#export SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=1 export SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=1
export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
#export SYCL_PI_LEVEL_ZERO_USE_COPY_ENGINE=0 export SYCL_PI_LEVEL_ZERO_USE_COPY_ENGINE=0
mpiexec -launcher ssh -n 2 -host localhost ./wrap4gpu.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.2 --grid 32.32.32.64 --accelerator-threads $NT --shm-mpi 1 > dw.2tile.1x2.log for i in 0
mpiexec -launcher ssh -n 2 -host localhost ./wrap4gpu.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --shm-mpi 1 > dw.2tile.2x1.log do
mpiexec -launcher ssh -n 2 -host localhost ./wrap4gpu.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.2 --grid 32.32.32.64 --accelerator-threads $NT --shm-mpi 1 --device-mem 32768
mpiexec -launcher ssh -n 2 -host localhost ./wrap4gpu.sh ./Benchmark_halo --mpi 1.1.1.2 --grid 32.32.32.64 --accelerator-threads $NT --shm-mpi 1 > halo.2tile.1x2.log mpiexec -launcher ssh -n 2 -host localhost ./wrap4gpu.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --shm-mpi 1 --device-mem 32768
mpiexec -launcher ssh -n 2 -host localhost ./wrap4gpu.sh ./Benchmark_halo --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --shm-mpi 1 > halo.2tile.2x1.log done
#mpiexec -launcher ssh -n 2 -host localhost ./wrap4gpu.sh ./Benchmark_halo --mpi 1.1.1.2 --grid 32.32.32.64 --accelerator-threads $NT --shm-mpi 1 > halo.2tile.1x2.log
#mpiexec -launcher ssh -n 2 -host localhost ./wrap4gpu.sh ./Benchmark_halo --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --shm-mpi 1 > halo.2tile.2x1.log