portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-09 23:45:36 +00:00
Code Releases Activity

More blasted shell variables

Browse Source
This commit is contained in:
Peter Boyle 2024-03-06 00:03:59 +00:00
parent 9b5f741e85
commit 7435315d50
4 changed files with 38 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
systems/Aurora/benchmarks/bench1024.pbs
View File

@ -25,12 +25,16 @@ export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1
export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE=0 export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE=0
export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE=0 export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE=0
export MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST=1 #export MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST=1
export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_BUFFER_SZ=1048576 export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_BUFFER_SZ=1048576
export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_THRESHOLD=131072 export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_THRESHOLD=131072
export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_NUM_BUFFERS_PER_CHUNK=16 export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_NUM_BUFFERS_PER_CHUNK=16
export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_MAX_NUM_BUFFERS=16 export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_MAX_NUM_BUFFERS=16
export MPICH_OFI_NIC_POLICY=GPU export MPICH_OFI_NIC_POLICY=GPU
export FI_CXI_CQ_FILL_PERCENT=10
export FI_CXI_DEFAULT_CQ_SIZE=262144
#export FI_CXI_DEFAULT_CQ_SIZE=131072
#export FI_CXI_CQ_FILL_PERCENT=20
# 12 ppn, 32 nodes, 384 ranks # 12 ppn, 32 nodes, 384 ranks
# #
@ -45,12 +49,12 @@ CMD="mpiexec -np 12288 -ppn 12 -envall \
./gpu_tile_compact.sh \ ./gpu_tile_compact.sh \
./Benchmark_dwf_fp32 --mpi 8.8.8.24 --grid 128.128.128.384 \ ./Benchmark_dwf_fp32 --mpi 8.8.8.24 --grid 128.128.128.384 \
--shm-mpi 1 --shm 2048 --device-mem 32000 --accelerator-threads 32 --comms-overlap" --shm-mpi 1 --shm 2048 --device-mem 32000 --accelerator-threads 32 --comms-overlap"
$CMD | tee 1024node.dwf.small $CMD | tee 1024node.dwf.small.cq
CMD="mpiexec -np 12288 -ppn 12 -envall \ CMD="mpiexec -np 12288 -ppn 12 -envall \
./gpu_tile_compact.sh \ ./gpu_tile_compact.sh \
./Benchmark_dwf_fp32 --mpi 16.8.8.12 --grid 256.256.256.384 \ ./Benchmark_dwf_fp32 --mpi 16.8.8.12 --grid 256.256.256.384 \
--shm-mpi 1 --shm 2048 --device-mem 32000 --accelerator-threads 32 --comms-overlap" --shm-mpi 1 --shm 2048 --device-mem 32000 --accelerator-threads 32 --comms-overlap"
$CMD | tee 1024node.dwf $CMD | tee 1024node.dwf.cq

17
systems/Aurora/benchmarks/bench12.pbs
View File

@ -17,6 +17,7 @@ source ../sourceme.sh
export OMP_NUM_THREADS=3 export OMP_NUM_THREADS=3
export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1 export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1
#unset MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE #unset MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE
#unset MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE #unset MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE
#unset MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST #unset MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST
@ -35,11 +36,25 @@ CMD="mpiexec -np 24 -ppn 12 -envall \
./Benchmark_comms_host_device --mpi 2.3.2.2 --grid 32.24.32.192 \ ./Benchmark_comms_host_device --mpi 2.3.2.2 --grid 32.24.32.192 \
--shm-mpi 1 --shm 2048 --device-mem 32000 --accelerator-threads 32" --shm-mpi 1 --shm 2048 --device-mem 32000 --accelerator-threads 32"
$CMD #$CMD
CMD="mpiexec -np 24 -ppn 12 -envall \ CMD="mpiexec -np 24 -ppn 12 -envall \
./gpu_tile_compact.sh \ ./gpu_tile_compact.sh \
./Benchmark_dwf_fp32 --mpi 2.3.2.2 --grid 64.96.64.64 --comms-overlap \ ./Benchmark_dwf_fp32 --mpi 2.3.2.2 --grid 64.96.64.64 --comms-overlap \
--shm-mpi 1 --shm 2048 --device-mem 32000 --accelerator-threads 32" --shm-mpi 1 --shm 2048 --device-mem 32000 --accelerator-threads 32"
#$CMD
CMD="mpiexec -np 1 -ppn 1 -envall \
./gpu_tile_compact.sh \
./Benchmark_dwf --mpi 1.1.1.1 --grid 16.32.32.32 --comms-sequential \
--shm-mpi 1 --shm 2048 --device-mem 32000 --accelerator-threads 32"
$CMD
CMD="mpiexec -np 1 -ppn 1 -envall \
./gpu_tile_compact.sh \
./Benchmark_dwf_fp32 --mpi 1.1.1.1 --grid 16.32.32.32 --comms-sequential \
--shm-mpi 1 --shm 2048 --device-mem 32000 --accelerator-threads 32"
$CMD $CMD

4
systems/Aurora/config-command
View File

@ -11,6 +11,6 @@ TOOLS=$HOME/tools
--enable-unified=no \ --enable-unified=no \
MPICXX=mpicxx \ MPICXX=mpicxx \
CXX=icpx \ CXX=icpx \
LDFLAGS="-fiopenmp -fsycl -fsycl-device-code-split=per_kernel -fsycl-device-lib=all -lze_loader -L$TOOLS/lib64/" \ LDFLAGS="-fiopenmp -fsycl -fsycl-device-code-split=per_kernel -fsycl-device-lib=all -lze_loader -L$TOOLS/lib64/ -L${MKLROOT}/lib -qmkl=parallel " \
CXXFLAGS="-fiopenmp -fsycl-unnamed-lambda -fsycl -I$INSTALL/include -Wno-tautological-compare -I$HOME/ -I$TOOLS/include" CXXFLAGS="-fiopenmp -fsycl-unnamed-lambda -fsycl -I$INSTALL/include -Wno-tautological-compare -I$HOME/ -I$TOOLS/include -qmkl=parallel"

13
systems/Aurora/sourceme.sh
View File

@ -3,6 +3,19 @@
module use /soft/modulefiles module use /soft/modulefiles
module load intel_compute_runtime/release/agama-devel-682.22 module load intel_compute_runtime/release/agama-devel-682.22
export FI_CXI_DEFAULT_CQ_SIZE=131072
export FI_CXI_CQ_FILL_PERCENT=20
export SYCL_PROGRAM_COMPILE_OPTIONS="-ze-opt-large-register-file"
#export SYCL_PROGRAM_COMPILE_OPTIONS="-ze-intel-enable-auto-large-GRF-mode"
#
# -ftarget-register-alloc-mode=pvc:default
# -ftarget-register-alloc-mode=pvc:small
# -ftarget-register-alloc-mode=pvc:large
# -ftarget-register-alloc-mode=pvc:auto
#
export HTTP_PROXY=http://proxy.alcf.anl.gov:3128 export HTTP_PROXY=http://proxy.alcf.anl.gov:3128
export HTTPS_PROXY=http://proxy.alcf.anl.gov:3128 export HTTPS_PROXY=http://proxy.alcf.anl.gov:3128
export http_proxy=http://proxy.alcf.anl.gov:3128 export http_proxy=http://proxy.alcf.anl.gov:3128