blob: 14ed35c37accf52d9db83d6613b429bbe589f60d (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
|
#!/bin/bash
mkdir -p log/$(hostname) baselines/cpu/log/$(hostname)
fn=log/$(hostname)/$(date +%Y%m%d)
# Upstream DPU version uses ts_size = 33554432 elements and query_length = 256 with int32_t data
# Upstream CPU version uses inputs/randomlist33M.txt with 33618177 elements and query_length = 256 with double
# This benchmark uses int32 and 2²⁵ == 33554432 elements for both.
run_benchmark_nmc() {
local "$@"
sudo limit_ranks_to_numa_node ${numa_rank}
if make -B NR_DPUS=${nr_dpus} NR_TASKLETS=${nr_tasklets} BL=8 WITH_ALLOC_OVERHEAD=1 WITH_LOAD_OVERHEAD=1 WITH_FREE_OVERHEAD=1; then
bin/ts_host -w 0 -e 100 -n ${input_size}
fi
return $?
}
export -f run_benchmark_nmc
run_benchmark() {
local "$@"
OMP_NUM_THREADS=${nr_threads} ./streamp_openmp inputs/randomlistDPU.txt 256 ${ram} ${cpu}
}
export -f run_benchmark
(
echo "NMC single-node operation (1/2)" >&2
parallel -j1 --eta --joblog ${fn}.1.joblog --header : \
run_benchmark_nmc nr_dpus={nr_dpus} nr_tasklets=8 input_size={input_size} numa_rank={numa_rank} \
::: numa_rank 0 1 \
::: nr_dpus 64 128 256 512 768 1024 \
::: input_size 33554432
echo "NMC multi-node operation (2/2)" >&2
parallel -j1 --eta --joblog ${fn}.2.joblog --header : \
run_benchmark_nmc nr_dpus={nr_dpus} nr_tasklets=8 input_size={input_size} numa_rank={numa_rank} \
::: numa_rank any \
::: nr_dpus 1536 2048 2304 \
::: input_size 33554432
) > ${fn}.txt
xz -f -v -9 -M 800M ${fn}.txt
cd baselines/cpu
make -B NUMA=1 DTYPE=int32_t
make inputs/randomlistDPU.txt
(
echo "CPU single-node operation (1/2)" >&2
parallel -j1 --eta --joblog ${fn}.1.joblog --header : \
run_benchmark nr_threads={nr_threads} ram={ram} cpu={cpu} \
::: i $(seq 0 20) \
::: ram 0 1 \
::: cpu 0 1 \
::: nr_threads 1 2 4 8 12 16
echo "CPU multi-node operation (1/2)" >&2
parallel -j1 --eta --joblog ${fn}.2.joblog --header : \
run_benchmark nr_threads={nr_threads} ram={ram} cpu={cpu} \
::: i $(seq 0 20) \
::: ram 0 1 \
::: cpu -1 \
::: nr_threads 24 32
) > ${fn}.txt
xz -f -v -9 -M 800M ${fn}.txt
|
