diff --git a/systems/PVC/benchmarks/run-2tile-mpi.sh b/systems/PVC/benchmarks/run-2tile-mpi.sh index 34deac2c..cefab776 100755 --- a/systems/PVC/benchmarks/run-2tile-mpi.sh +++ b/systems/PVC/benchmarks/run-2tile-mpi.sh @@ -8,7 +8,6 @@ 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 @@ -20,14 +19,16 @@ 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_DEVICE_SCOPE_EVENTS=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 -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 - -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 +for i in 0 +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_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --shm-mpi 1 --device-mem 32768 +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