Rename rocHPL patch file

projects/rocHPL/rochpl-bw1000-16r.patch → projects/rocHPL/rochpl-scripts-2xbw1000.patch

 diff --git a/scripts/mpirun_rochpl.in b/scripts/mpirun_rochpl.in
-index 155f502..3984e62 100755
+index 155f502..94f7b0d 100755
 --- a/scripts/mpirun_rochpl.in
 +++ b/scripts/mpirun_rochpl.in
 @@ -46,7 +46,7 @@ supported_distro( )
@@ -25,59 +25,65 @@ index 155f502..3984e62 100755
  devices=
  
  # #################################################
-@@ -117,12 +124,13 @@ mpi_args=
- #Check if using OpenMPI
- if [[ $(${mpi_bin} --version | grep "open-mpi") ]]; then
-   mpi_args+=" --map-by node --rank-by slot --bind-to none "
-+  #mpi_args+=" --map-by numa:PE=16 --bind-to core --report-bindings "
- 
-   #Check if this is OpenMPI+UCX
-   ompi_info=$(dirname ${mpi_bin})/ompi_info
-   if [[ $(${ompi_info} | grep "MCA pml: ucx") ]]; then
-     # ucx-specific args
+@@ -111,21 +118,6 @@ num_cpu_cores=$(lscpu | grep "Core(s)" | awk '{print $4}')
+ num_cpu_sockets=$(lscpu | grep Socket | awk '{print $2}')
+ total_cpu_cores=$(($num_cpu_cores*$num_cpu_sockets))
+ 
+-#Default MPI options
+-mpi_args=
+-
+-#Check if using OpenMPI
+-if [[ $(${mpi_bin} --version | grep "open-mpi") ]]; then
+-  mpi_args+=" --map-by node --rank-by slot --bind-to none "
+-
+-  #Check if this is OpenMPI+UCX
+-  ompi_info=$(dirname ${mpi_bin})/ompi_info
+-  if [[ $(${ompi_info} | grep "MCA pml: ucx") ]]; then
+-    # ucx-specific args
 -    mpi_args="--mca pml ucx --mca btl ^vader,tcp,openib,uct ${mpi_args}"
-+    mpi_args="--mca pml ucx ${mpi_args}"
-   fi
- fi
- 
-@@ -153,7 +161,7 @@ while true; do
+-  fi
+-fi
+-
+ # #################################################
+ # Parameter parsing
+ # #################################################
+@@ -153,7 +145,7 @@ while true; do
          exit 0
          ;;
      --version)
 -        ${mpi_bin} -np 1 ${mpi_args} ${rochpl_runscript} --version
-+        ${mpi_bin} --allow-run-as-root -np 1 ${mpi_args} ${rochpl_runscript} --version
++        ${mpi_bin} --allow-run-as-root -np 1 ${rochpl_runscript} --version
          exit 0
          ;;
      -P)
-@@ -218,5 +226,25 @@ if [ ! -z "${devices}" ]; then
+@@ -218,5 +210,24 @@ if [ ! -z "${devices}" ]; then
    rochpl_args+=" --devices=${devices}"
  fi
  
 +echo "Copying files..."
-+scp -P 3333 $0 node02:/workspace/build/
-+scp -P 3333 ./build/run_rochpl node02:/workspace/build/
-+scp -P 3333 ./${filename} node02:/workspace/
++rsync -az -e 'ssh -p 3333' build tpl ${filename} node02:/workspace/
 +
  #run
 -${mpi_bin} -np ${np} ${mpi_args} ${rochpl_runscript} ${rochpl_args}
 +${mpi_bin} --allow-run-as-root \
 +  --prefix ${ompi_prefix} \
-+  ${mpi_args} \
++  --map-by ppr:8:node --bind-to none \
++  --mca pml ucx \
 +  --mca btl ^openib \
 +  --mca btl_tcp_if_include p14p2 \
 +  --mca plm_rsh_args "-p 3333" \
 +  --mca coll_hcoll_enable 0 \
-+  -x UCX_TLS=self,sm,rocm_ipc,rocm_copy,rc_mlx5,ud_mlx5 \
++  -x UCX_TLS=self,sm,rocm,rc \
++  -x UCX_RNDV_SCHEME=put_zcopy \
++  -x UCX_RNDV_FRAG_MEM_TYPE=rocm \
 +  -x UCX_MEMTYPE_CACHE=n \
-+  -x PATH -x LD_LIBRARY_PATH -x OPAL_PREFIX \
-+  -x UCX_NET_DEVICES=mlx5_1:1,mlx5_2:1,mlx5_3:1,mlx5_4:1,mlx5_7:1,mlx5_8:1,mlx5_9:1,mlx5_10:1 \
 +  -x HSA_FORCE_FINE_GRAIN_PCIE=1 \
-+  -x UCX_WARN_UNUSED_ENV_VARS=n \
++  -x PATH -x LD_LIBRARY_PATH -x OPAL_PREFIX \
 +  -np 16 \
 +  -H node01:8,node02:8 \
 +  ${rochpl_runscript} ${rochpl_args}
 diff --git a/scripts/run_rochpl.in b/scripts/run_rochpl.in
-index 1522e5d..81fed8a 100755
+index 1522e5d..68c1958 100755
 --- a/scripts/run_rochpl.in
 +++ b/scripts/run_rochpl.in
 @@ -46,7 +46,7 @@ supported_distro( )
@@ -109,7 +115,7 @@ index 1522e5d..81fed8a 100755
  
  #count the cpus per core
  threads_per_core=$(echo "${cpulist}" | grep -c ".*	0	.*")
-@@ -361,9 +363,24 @@ export OMP_NUM_THREADS=${omp_num_threads}
+@@ -361,9 +363,23 @@ export OMP_NUM_THREADS=${omp_num_threads}
  export OMP_PLACES=${omp_places}
  export OMP_PROC_BIND=true
  
@@ -131,11 +137,10 @@ index 1522e5d..81fed8a 100755
  if [[ $globalRank -lt $size ]]; then
 -  echo "Node Binding: Process $rank [(p,q)=($myp,$myq)] GPU: $mygpu, CPU Cores: $omp_num_threads - $places"
 +  echo "Node Binding: Process $rank [(p,q)=($myp,$myq)] GPU: $mygpu, NUMA: $mynuma, IB: $myib, CPU Cores: $omp_num_threads - $places"
-+#  echo "Node Binding: Process $rank [(p,q)=($myp,$myq)] GPU: $mygpu, NUMA: $mynuma"
  fi
  
  rochpl_args="-P ${P} -Q ${Q} -p ${p} -q ${q} -f ${frac} -it ${it}"
-@@ -374,4 +391,5 @@ else
+@@ -374,4 +390,5 @@ else
  fi
  
  #run