summaryrefslogtreecommitdiff
path: root/BS/dimes-hetsim-hbm.sh
diff options
context:
space:
mode:
authorBirte Kristina Friesel <birte.friesel@uos.de>2024-07-04 15:46:18 +0200
committerBirte Kristina Friesel <birte.friesel@uos.de>2024-07-04 15:46:18 +0200
commit207a5fcf7f2d482f117a3d4a54484daf1fae92e2 (patch)
treea2c90238faefecc605cc17f431e1ee20860d3001 /BS/dimes-hetsim-hbm.sh
parent308d09c666faa489398236f140fdfd7db24ed7d4 (diff)
BS: Add separate files for HBM and NMC benchmarks
Diffstat (limited to 'BS/dimes-hetsim-hbm.sh')
-rwxr-xr-xBS/dimes-hetsim-hbm.sh65
1 files changed, 65 insertions, 0 deletions
diff --git a/BS/dimes-hetsim-hbm.sh b/BS/dimes-hetsim-hbm.sh
new file mode 100755
index 0000000..b043e13
--- /dev/null
+++ b/BS/dimes-hetsim-hbm.sh
@@ -0,0 +1,65 @@
+#!/bin/bash
+
+cd baselines/cpu
+make -B NUMA=1
+
+mkdir -p log/$(hostname)
+fn=log/$(hostname)/$(date +%Y%m%d)
+
+# upstream DPU version uses 2048576 * uint64 ≈ 16 MiB (DPU max: 64 MiB)
+# upstream DPU version uses 2 queries
+# * uint64 == 16 MiB -- fits into cache
+input_size_dpu=$(perl -E 'say 2 ** 21')
+# * uint64 == 8 MiB
+num_queries_dpu=1048576
+
+# * uint64 == 4 GiB
+input_size_hbm=$(perl -E 'say 2 ** 29')
+# * uint64 == 128 MiB
+num_queries_hbm=16777216
+
+run_benchmark() {
+ local "$@"
+ OMP_NUM_THREADS=${nr_threads} ./bs_omp ${input_size} ${num_queries} $ram $cpu
+ return $?
+}
+
+export -f run_benchmark
+
+(
+
+parallel -j1 --eta --joblog ${fn}.1.joblog --resume --header : \
+ run_benchmark i={i} nr_threads={nr_threads} ram={ram} cpu={cpu} \
+ input_size=${input_size_dpu} num_queries=${num_queries_dpu} \
+ ::: i $(seq 1 20) \
+ ::: nr_threads 1 2 4 8 12 16 \
+ ::: cpu $(seq 0 7) \
+ ::: ram $(seq 0 15)
+
+parallel -j1 --eta --joblog ${fn}.2.joblog --resume --header : \
+ run_benchmark i={i} nr_threads={nr_threads} ram={ram} cpu={cpu} \
+ input_size=${input_size_hbm} num_queries=${num_queries_hbm} \
+ ::: i $(seq 1 20) \
+ ::: nr_threads 1 2 4 8 12 16 \
+ ::: cpu $(seq 0 7) \
+ ::: ram $(seq 0 15)
+
+parallel -j1 --eta --joblog ${fn}.3.joblog --resume --header : \
+ run_benchmark i={i} nr_threads={nr_threads} ram={ram} cpu={cpu} \
+ input_size=${input_size_dpu} num_queries=${num_queries_dpu} \
+ ::: i $(seq 1 20) \
+ ::: nr_threads 32 48 64 96 128 \
+ ::: cpu -1 \
+ ::: ram $(seq 0 15)
+
+parallel -j1 --eta --joblog ${fn}.4.joblog --resume --header : \
+ run_benchmark i={i} nr_threads={nr_threads} ram={ram} cpu={cpu} \
+ input_size=${input_size_hbm} num_queries=${num_queries_hbm} \
+ ::: i $(seq 1 20) \
+ ::: nr_threads 32 48 64 96 128 \
+ ::: cpu -1 \
+ ::: ram $(seq 0 15)
+
+) > ${fn}.txt
+
+xz -f -v -9 -M 800M ${fn}.txt