决赛现场脚本

docs/contest_runners.md

@@ -1,30 +1,40 @@
 # TN
 ```bash
-# qibotn目录下
-I_MPI_FABRICS=shm:ofi \
-I_MPI_OFI_PROVIDER=tcp \
-FI_PROVIDER=tcp \
+# search + contract，Open MPI 多节点：每节点 2 rank，每 rank 绑定 1 个 NUMA。
+# MPI_HOSTS 里每个节点写 :2，MPI_RANKS = 节点数 * 2。
+# 每个 rank 使用 MPI_PE 个 core；这台 2-NUMA AMD 节点用 MPI_PE=128。
+
+NQUBITS=40 \
+TN_DEBUG_TRIALS=1 \
+SCHEDULER_HOST=10.20.1.100 \
+DASK_ADDRESS=tcp://10.20.1.100:8786 \
+WORKER_HOSTS="10.20.1.100 10.20.1.101 10.20.1.102 10.20.1.103" \
 CASE=main1 \
 OBSERVABLES=long_z_string \
-NQUBITS=34 \
-NLAYERS=20 \
-TORCH_THREADS=48 \
-SEARCH_REPEATS=2048 \
-SEARCH_TIME=300 \
-SCHEDULER_HOST=10.20.1.103 \
-WORKER_HOSTS="10.20.1.103 10.20.6.101" \
-DASK_ADDRESS="tcp://10.20.1.103:8786" \
-NWORKERS=84 \
-NTHREADS=1 \
-MPIEXEC_FULL="mpirun -np 4 -hostfile /home/yx/qibotn/hostfile -perhost 2" \
+TORCH_THREADS=80 \
+MPI_PE=80 \
+MPI_MAP_BY=ppr:1:numa:PE=80 \
+MPI_BIND_TO=core \
+OMP_NUM_THREADS=80 \
+MKL_NUM_THREADS=80 \
+BLIS_NUM_THREADS=80 \
+MPI_HOSTS="node-0:2,node-1:2,node-2:2,node-3:2" \
+MPI_RANKS=8 \
+NWORKERS=96 \
+TN_TARGET_SIZE=17179869184 \
 tools/run_tn_dask_mpi_all.sh
 
 # 单独缩并contract计算
 
-I_MPI_FABRICS=shm:ofi \
-I_MPI_OFI_PROVIDER=tcp \
-FI_PROVIDER=tcp \
-mpirun -np 4 -hostfile /home/yx/qibotn/hostfile -perhost 2 \
+mpirun --map-by ppr:1:numa:PE=80 --bind-to core --report-bindings \
+  -x LD_PRELOAD=/home/aocc/aocl/5.2.0/aocc/lib_LP64/libblis-mt.so.5 \
+  -x BLIS_NUM_THREADS=80 \
+  -x OMP_NUM_THREADS=80 \
+  -x MKL_NUM_THREADS=80 \
+  -x OMP_PROC_BIND=close \
+  -x OMP_PLACES=cores \
+  -np 8 \
+  -host node-0:2,node-1:2,node-2:2,node-3:2 \
   .venv/bin/python -u tools/tn_contest_runner.py contract \
   --mpi \
   --case main1 \
@@ -32,22 +42,47 @@ mpirun -np 4 -hostfile /home/yx/qibotn/hostfile -perhost 2 \
   --nlayers 20 \
   --observables long_z_string \
   --tree-dir trees/contest_tn \
-  --torch-threads 48 \
+  --torch-threads 80 \
   --dtype complex64
 ```
 
 # MPS
 ```
-cd /home/yx/qibotn
+cd /home/qibo/qibotn
 
-I_MPI_FABRICS=shm:ofi \
-I_MPI_OFI_PROVIDER=tcp \
-FI_PROVIDER=tcp \
-MPIEXEC_FULL="mpirun -np 4 -hostfile /home/yx/qibotn/hostfile -perhost 2" \
+MPIEXEC=mpirun \
+MPI_HOSTS="node-2:4,node-3:4" \
+MPI_RANKS=8 \
+MPI_PE=48 \
+MPI_MAP_BY=ppr:2:numa:PE=48 \
+MPI_BIND_TO=core \
+MPI_REPORT_BINDINGS=1 \
 TORCH_THREADS=48 \
+OMP_NUM_THREADS=48 \
+MKL_NUM_THREADS=48 \
+BLIS_NUM_THREADS=48 \
 OBS_FILTER=ring_xz \
 MAIN1_NQ=128 \
 MAIN1_LAYERS=24 \
 MAIN1_BOND=1024 \
 tools/run_vidal_mpi_contest_cases.sh main1
-```
+
+
+
+MPIEXEC=mpirun \
+MPI_HOSTS="node-2:4" \
+MPI_RANKS=4 \
+MPI_PE=48 \
+MPI_MAP_BY=ppr:2:numa:PE=48 \
+MPI_BIND_TO=core \
+MPI_REPORT_BINDINGS=1 \
+TORCH_THREADS=48 \
+OMP_NUM_THREADS=48 \
+MKL_NUM_THREADS=48 \
+BLIS_NUM_THREADS=48 \
+OBS_FILTER=ring_xz \
+MAIN1_NQ=128 \
+MAIN1_LAYERS=24 \
+MAIN1_BOND=1024 \
+tools/run_vidal_mpi_contest_cases.sh main1
+```

docs/xianchang.md

@@ -0,0 +1,42 @@
+mpirun --map-by ppr:1:numa:PE=80 --bind-to core --report-bindings \
+  -x LD_PRELOAD=/home/aocc/aocl/5.2.0/aocc/lib_LP64/libblis-mt.so.5 \
+  -x BLIS_NUM_THREADS=80 \
+  -x OMP_NUM_THREADS=80 \
+  -x MKL_NUM_THREADS=80 \
+  -x OMP_PROC_BIND=close \
+  -x OMP_PLACES=cores \
+  -np 4 \
+  -host node-0:2,node-1:2,node-2:2,node-3:2 \
+  .venv/bin/python -u tools/tn_contest_runner.py contract \
+  --mpi \
+  --case main1 \
+  --nqubits 34 \
+  --nlayers 20 \
+  --observables long_z_string \
+  --tree-dir trees/contest_tn \
+  --torch-threads 80 \
+  --dtype complex64
+
+
+SEARCH_TIME=300  NQUBITS=40 TN_DEBUG_TRIALS=1 SCHEDULER_HOST=10.20.1.102 DASK_ADDRESS=tcp://10.20.1.102:8786 WORKER_HOSTS="10.20.1.102 10.20.1.103" CASE=main1 OBSERVABLES=long_z_string TORCH_THREADS=80 MPI_PE=80 MPI_MAP_BY=ppr:1:numa:PE=80 MPI_BIND_TO=core OMP_NUM_THREADS=80 MKL_NUM_THREADS=80 BLIS_NUM_THREADS=80 MPI_HOSTS="node-2:2,node-3:2" MPI_RANKS=4 NWORKERS=128  TN_TARGET_SIZE=17179869184 tools/run_tn_dask_mpi_all.sh
+
+
+NQUBITS=40 \
+TN_DEBUG_TRIALS=1 \
+SCHEDULER_HOST=10.20.1.102 \
+DASK_ADDRESS=tcp://10.20.1.102:8786 \
+WORKER_HOSTS="10.20.1.102 10.20.1.103" \
+CASE=main1 \
+OBSERVABLES=long_z_string \
+TORCH_THREADS=80 \
+MPI_PE=80 \
+MPI_MAP_BY=ppr:1:numa:PE=80 \
+MPI_BIND_TO=core \
+OMP_NUM_THREADS=80 \
+MKL_NUM_THREADS=80 \
+BLIS_NUM_THREADS=80 \
+MPI_HOSTS="node-2:2,node-3:2" \
+MPI_RANKS=4 \
+NWORKERS=96 \
+TN_TARGET_SIZE=17179869184 \
+tools/run_tn_dask_mpi_all.sh