summaryrefslogtreecommitdiff
path: root/SEL/dimes-hetsim-nmc.sh
diff options
context:
space:
mode:
authorBirte Kristina Friesel <birte.friesel@uos.de>2024-07-12 08:02:20 +0200
committerBirte Kristina Friesel <birte.friesel@uos.de>2024-07-12 08:02:20 +0200
commit33f08b50d6ecf84ad482b0f6b3883ffc6bebe52b (patch)
tree4523274e9f9fe1ed9e18268f2d52dd013f6d3d67 /SEL/dimes-hetsim-nmc.sh
parentfcd2655ecfb662309bf31d3681d971bca7aeae0e (diff)
SEL: Add DIMES NMC script
Diffstat (limited to 'SEL/dimes-hetsim-nmc.sh')
-rwxr-xr-xSEL/dimes-hetsim-nmc.sh58
1 files changed, 58 insertions, 0 deletions
diff --git a/SEL/dimes-hetsim-nmc.sh b/SEL/dimes-hetsim-nmc.sh
new file mode 100755
index 0000000..e6752a6
--- /dev/null
+++ b/SEL/dimes-hetsim-nmc.sh
@@ -0,0 +1,58 @@
+#!/bin/bash
+
+mkdir -p log/$(hostname) baselines/cpu/log/$(hostname)
+fn=log/$(hostname)/$(date +%Y%m%d)
+
+# 2^28 elem == 2 GiB data (upstream version uses 1.875 GiB)
+
+run_benchmark_nmc() {
+ local "$@"
+ sudo limit_ranks_to_numa_node ${numa_rank}
+ if make -B NR_DPUS=${nr_dpus} NR_TASKLETS=${nr_tasklets} BL=10 WITH_ALLOC_OVERHEAD=1 WITH_LOAD_OVERHEAD=1 WITH_FREE_OVERHEAD=1 PARALLEL_READ=1; then
+ bin/host_code -w 0 -e 100 -i ${input_size} -x 1
+ fi
+ return $?
+}
+
+export -f run_benchmark_nmc
+
+(
+
+parallel -j1 --eta --joblog ${fn}.1.joblog --header : \
+ run_benchmark_nmc nr_dpus={nr_dpus} nr_tasklets=16 input_size={input_size} numa_rank={numa_rank} \
+ ::: numa_rank 0 1 \
+ ::: nr_dpus 64 128 256 512 768 1024 \
+ ::: input_size $(( 2 ** 28 ))
+
+parallel -j1 --eta --joblog ${fn}.2.joblog --header : \
+ run_benchmark_nmc nr_dpus={nr_dpus} nr_tasklets=16 input_size={input_size} numa_rank={numa_rank} \
+ ::: numa_rank any \
+ ::: nr_dpus 1536 2048 2304 \
+ ::: input_size $(( 2 ** 28 ))
+
+) > ${fn}.txt
+
+xz -f -v -9 -M 800M ${fn}.txt
+
+cd baselines/cpu
+make -B NUMA=1
+
+(
+
+parallel -j1 --eta --joblog ${fn}.1.joblog --header : \
+ ./sel -a {ram} -b {ram} -c {cpu} -i {input_size} -t {nr_threads} -w 0 -e 40 \
+ ::: ram 0 1 \
+ ::: cpu 0 1 \
+ ::: nr_threads 1 2 4 8 12 16 \
+ ::: input_size $(( 2 ** 28 ))
+
+parallel -j1 --eta --joblog ${fn}.2.joblog --header : \
+ ./sel -a {ram} -b {ram} -c {cpu} -i {input_size} -t {nr_threads} -w 0 -e 40 \
+ ::: ram any \
+ ::: cpu -1 \
+ ::: nr_threads 24 32 \
+ ::: input_size $(( 2 ** 28 ))
+
+) > ${fn}.txt
+
+xz -f -v -9 -M 800M ${fn}.txt