blob: 62e4ed41dfe6999b7f676eddac555829e94fbf14 (
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
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
|
#!/bin/bash
mkdir -p log/$(hostname) baselines/cpu/log/$(hostname)
fn=log/$(hostname)/$(date +%Y%m%d)
# Args: -m m -n n -o M_ -p N_
#
# Input: (M_ * m) × (N_ * n) matrix
# Output: (N_* n) × (M_ * m) matrix
# Step 1: transpose (M_ * m) × N_ matrix that consits of tiles of size n
# CPU version: explicit
# DPU version: implicit (appropriate write operations to DPUs)
# Step 2: transpose m × n matrix; this happens N_ * M_ times.
# DPU version: Each tasklet transposes a single m × n matrix / tile.
# (16 × 8 tile takes up 1 KiB WRAM)
# Step 3: Transpose M_ × n matrix that consists of tiles of size m.
#
# Note for DPU version: if M_ > #DPUs, steps 1 through 3 are repeated.
# Number of repetitions == ceil(M_ / #DPUS)
# For Hetsim benchmarks, we set M_ == #DPUs to simplify the task graph (no repetitions that depend on the number of available DPUs).
# Just in case, there is also a configuration with M_ == 2048 independent of #DPUs
#
# input size: uint64(DPU)/double(CPU) * M_ * m * N_ * n
# output size: uint64(DPU)/double(CPU) * M_ * m * N_ * n -- on DPU only; CPU version operates in-place
# Upstream DPU version uses int64_t, -p 2048 -o 12288 -x 1 [implicit -m 16 -n 8]
# Upstream CPU version uses double, -p 2556 -o 4096 -m 16 -n 8 and fails with -o 12288 (allocation error)
#
# -p 2048 -o 2048 -m 16 -n 8 -> matrix size: 4 GiB
# -p [64 .. 2304] -o 2048 -m 16 -n 8 -> matrix size: 128 MiB .. 4.5 GiB
run_benchmark_nmc() {
local "$@"
sudo limit_ranks_to_numa_node ${numa_rank}
if make -B NR_DPUS=${nr_dpus} NR_TASKLETS=${nr_tasklets} WITH_ALLOC_OVERHEAD=1 WITH_LOAD_OVERHEAD=1 WITH_FREE_OVERHEAD=1; then
bin/host_code -w 0 -e 40 -p ${p} -o 2048 -m 16 -n 8 -x 1
fi
return $?
}
export -f run_benchmark_nmc
(
echo "NMC single-node operation (1/4)" >&2
parallel -j1 --eta --joblog ${fn}.1.joblog --header : \
run_benchmark_nmc p={nr_dpus} 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
echo "NMC multi-node operation (2/4)" >&2
parallel -j1 --eta --joblog ${fn}.2.joblog --header : \
run_benchmark_nmc p={nr_dpus} nr_dpus={nr_dpus} nr_tasklets=16 input_size={input_size} numa_rank={numa_rank} \
::: numa_rank any \
::: nr_dpus 1536 2048 2304
echo "NMC single-node operation (3/4)" >&2
parallel -j1 --eta --joblog ${fn}.3.joblog --header : \
run_benchmark_nmc p=2048 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
echo "NMC multi-node operation (4/4)" >&2
parallel -j1 --eta --joblog ${fn}.4.joblog --header : \
run_benchmark_nmc p=2048 nr_dpus={nr_dpus} nr_tasklets=16 input_size={input_size} numa_rank={numa_rank} \
::: numa_rank any \
::: nr_dpus 1536 2048 2304
) > ${fn}.txt
xz -f -v -9 -M 800M ${fn}.txt
cd baselines/cpu
make -B NUMA=1
(
echo "CPU single-node operation (1/2)" >&2
parallel -j1 --eta --joblog ${fn}.1.joblog --header : \
./trns -w 0 -r 40 -p {p} -o 2048 -m 16 -n 8 -t {nr_threads} -a {ram} -b {ram} -c {cpu} \
::: p 64 128 256 512 768 1024 1536 2048 2304 \
::: ram 0 1 \
::: cpu 0 1 \
::: nr_threads 1 2 4 8 12 16
echo "CPU multi-node operation (2/2)" >&2
parallel -j1 --eta --joblog ${fn}.2.joblog --header : \
./trns -w 0 -r 40 -p {p} -o 2048 -m 16 -n 8 -t {nr_threads} -a {ram} -b {ram} -c {cpu} \
::: p 64 128 256 512 768 1024 1536 2048 2304 \
::: ram 0 1 \
::: cpu -1 \
::: nr_threads 24 32
) > ${fn}.txt
xz -f -v -9 -M 800M ${fn}.txt
|
