-
Notifications
You must be signed in to change notification settings - Fork 1
/
CITATION.cff
111 lines (110 loc) · 4.05 KB
/
CITATION.cff
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
102
103
104
105
106
107
108
109
110
111
cff-version: 1.2.0
title: Efficient Algorithm Design of Optimizing SpMV on GPU
message: "If you use this software, please cite it as below."
authors:
- family-names: Chu
given-names: Genshen
orcid: 'https://orcid.org/0000-0003-0374-1894'
- family-names: He
given-names: Yuanjie
orcid: 'https://orcid.org/0009-0003-7115-6846'
- family-names: Dong
given-names: Lingyu
orcid: 'https://orcid.org/0000-0003-0919-553X'
- family-names: Ding
given-names: Zhezhao
orcid: 'https://orcid.org/0000-0003-3437-8151'
- family-names: Chen
given-names: Dandan
orcid: 'https://orcid.org/0000-0002-9847-5092'
- family-names: Bai
given-names: He
orcid: 'https://orcid.org/0000-0001-5418-0375'
- family-names: Wang
given-names: Xuesong
orcid: 'https://orcid.org/0009-0000-2811-557X'
- family-names: Hu
given-names: Changjun
orcid: 'https://orcid.org/0000-0003-3857-7262'
identifiers:
- type: doi
value: 10.1145/3588195.3593002
repository-code: 'https://github.com/hpcde/spmv-acc'
abstract: >-
Sparse matrix-vector multiplication (SpMV) is a
fundamental build- ing block for various numerical
computing applications. However, most existing GPU-SpMV
approaches may suffer from either long preprocessing
overhead, load imbalance, format conversion, bad memory
access patterns. In this paper, we proposed two new SpMV
algorithms: flat and line-enhance, as well as their
implementations, for GPU systems to overcome the above
shortcomings. Our algorithms work directly on the CSR
sparse matrix format. To achieve high performance: 1) for
load balance, the flat algorithm uses non- zero splitting
and line-enhance uses a mix of row and non-zero splitting;
2) memory access patterns are designed for both algorithms
for data loading, storing and reduction steps; and 3) an
adaptive approach is proposed to select appropriate
algorithm and parameters based on matrix characteristics.
We evaluate our methods using the SuiteSparse Matrix
Collec- tion on AMD and NVIDIA GPU platforms. Average
performance improvements of 424%, 741%, 49%, 46%, 72% are
achieved when comparing our adaptive approach with
CSR-Vector, CSR-Adaptive, HOLA, cuSparse and merge-based
SpMV, respectively. In bandwidth tests, our approach can
also achieve a high memory bandwidth, which is very close
to the peak memory bandwidth.
keywords:
- SpMV
- GPU
- linear algebra
- sparse matrix
- CSR
license: Apache-2.0
version: 0.6.0
doi: 10.1145/3588195.3593002
date-released: 2022-04-18
url: "https://github.com/hpcde/spmv-acc"
preferred-citation:
type: conference-paper
authors:
- family-names: Chu
given-names: Genshen
orcid: 'https://orcid.org/0000-0003-0374-1894'
- family-names: He
given-names: Yuanjie
orcid: 'https://orcid.org/0009-0003-7115-6846'
- family-names: Dong
given-names: Lingyu
orcid: 'https://orcid.org/0000-0003-0919-553X'
- family-names: Ding
given-names: Zhezhao
orcid: 'https://orcid.org/0000-0003-3437-8151'
- family-names: Chen
given-names: Dandan
orcid: 'https://orcid.org/0000-0002-9847-5092'
- family-names: Bai
given-names: He
orcid: 'https://orcid.org/0000-0001-5418-0375'
- family-names: Wang
given-names: Xuesong
orcid: 'https://orcid.org/0009-0000-2811-557X'
- family-names: Hu
given-names: Changjun
orcid: 'https://orcid.org/0000-0003-3857-7262'
doi: 10.1145/3588195.3593002
title: Efficient Algorithm Design of Optimizing SpMV on GPU
isbn: 979-8-4007-0155-9/23/06
url: http://doi.org/10.1145/3588195.3593002
language: en
urldate: 2023-6-20
booktitle: Proceedings of the 32nd International Symposium on High-Performance Parallel and Distributed Computing (HPDC '23), June 16--23, 2023, Orlando, FL, USA
publisher: ACM Press
# author: Chu, Genshen and He, Yuanjie and Dong, Lingyu and Ding, Zhezhao and Chen, Dandan and Bai, He and Wang, Xuesong and Hu, Changjun
numpages: 14
series: HPDC '23
address: Orlando, Florida
location: Orlando, FL, USA
pages: 1--14
year: 2023