diff options
| author | Birte Kristina Friesel <birte.friesel@uos.de> | 2024-07-22 16:20:00 +0200 |
|---|---|---|
| committer | Birte Kristina Friesel <birte.friesel@uos.de> | 2024-07-22 16:20:00 +0200 |
| commit | 32f8ed0eabb3e21d8d5315b11c3c8e7e06e668af (patch) | |
| tree | 4ba212e6eb19011021e919199aeae69b4c3fae0c /VA/dimes-hetsim-nmc.sh | |
| parent | 27765c0a341ccdb8772481645f5e9b4fcf9c6abf (diff) | |
VA baseline: configurable memcpy NUMA binding
Diffstat (limited to 'VA/dimes-hetsim-nmc.sh')
| -rwxr-xr-x | VA/dimes-hetsim-nmc.sh | 24 |
1 files changed, 18 insertions, 6 deletions
diff --git a/VA/dimes-hetsim-nmc.sh b/VA/dimes-hetsim-nmc.sh index cabae57..a176027 100755 --- a/VA/dimes-hetsim-nmc.sh +++ b/VA/dimes-hetsim-nmc.sh @@ -40,20 +40,20 @@ make -B NUMA=1 ( -echo "CPU single-node operation (1/3)" >&2 +echo "CPU single-node operation (1/4)" >&2 parallel -j1 --eta --joblog ${fn}.1.joblog --resume --header : \ - ./va -i {input_size} -a {ram_in} -b {ram_out} -c {cpu} -t {nr_threads} -w 0 -e 40 \ + ./va -i {input_size} -a {ram_in} -b {ram_out} -c {cpu} -t {nr_threads} -w 0 -e 20 \ ::: ram_in 0 1 \ ::: ram_out 0 1 \ ::: cpu 0 1 \ ::: nr_threads 1 2 4 8 12 16 \ ::: input_size 167772160 -echo "CPU multi-node operation (2/3)" >&2 +echo "CPU multi-node operation (2/4)" >&2 parallel -j1 --eta --joblog ${fn}.2.joblog --resume --header : \ - ./va -i {input_size} -a {ram_in} -b {ram_out} -c {cpu} -t {nr_threads} -w 0 -e 40 \ + ./va -i {input_size} -a {ram_in} -b {ram_out} -c {cpu} -t {nr_threads} -w 0 -e 20 \ ::: ram_in 0 1 \ ::: ram_out 0 1 \ ::: cpu -1 \ @@ -66,11 +66,23 @@ make -B NUMA=1 NUMA_MEMCPY=1 ( -echo "CPU single-node operation with setup cost, cpu/out on same node (3/3)" >&2 +echo "CPU single-node operation with setup cost, memcpy node == input node, cpu node == output node (3/4)" >&2 parallel -j1 --eta --joblog ${fn}.3.joblog --resume --header : \ - ./va -i {input_size} -a {ram_in} -b {ram_out} -c {cpu} -C {cpu} -t {nr_threads} -w 0 -e 40 \ + ./va -i {input_size} -a {ram_in} -b {ram_out} -c {cpu} -C {cpu} -M {cpu_memcpy} -t {nr_threads} -w 0 -e 20 \ + ::: ram_in 0 1 \ + :::+ cpu_memcpy 0 1 \ + ::: cpu 0 1 \ + :::+ ram_out 0 1 \ + ::: nr_threads 1 2 4 8 12 16 \ + ::: input_size 167772160 + +echo "CPU single-node operation with setup cost, memcpy node == 0, cpu node == output node (4/4)" >&2 + +parallel -j1 --eta --joblog ${fn}.4.joblog --resume --header : \ + ./va -i {input_size} -a {ram_in} -b {ram_out} -c {cpu} -C {cpu} -M {cpu_memcpy} -t {nr_threads} -w 0 -e 20 \ ::: ram_in 0 1 \ + ::: cpu_memcpy 0 \ ::: cpu 0 1 \ :::+ ram_out 0 1 \ ::: nr_threads 1 2 4 8 12 16 \ |