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
112
113
114
115
116
117
118
119
120
121
122
123
124
|
#!/usr/bin/env python
##############################################################################
#
# Copyright (c) 2003-2016 by The University of Queensland
# http://www.uq.edu.au
#
# Primary Business: Queensland, Australia
# Licensed under the Apache License, version 2.0
# http://www.apache.org/licenses/LICENSE-2.0
#
# Development until 2012 by Earth Systems Science Computational Center (ESSCC)
# Development 2012-2013 by School of Earth Sciences
# Development from 2014 by Centre for Geoscience Computing (GeoComp)
#
##############################################################################
from __future__ import print_function, division
import os,csv
device_id = '0' # index of the device to use
binary_filename = '../spmv' # command used to run the tests
output_file = 'benchmark_output.log' # file where results are stored
# The unstructured matrices are available online:
# http://www.nvidia.com/content/NV_Research/matrices.zip
mats = []
unstructured_path = '~/scratch/Matrices/williams/mm/'
unstructured_mats = [('Dense','dense2.mtx'),
('Protein','pdb1HYS.mtx'),
('FEM/Spheres','consph.mtx'),
('FEM/Cantilever','cant.mtx'),
('Wind Tunnel','pwtk.mtx'),
('FEM/Harbor','rma10.mtx'),
('QCD','qcd5_4.mtx'),
('FEM/Ship','shipsec1.mtx'),
('Economics','mac_econ_fwd500.mtx'),
('Epidemiology','mc2depi.mtx'),
('FEM/Accelerator','cop20k_A.mtx'),
('Circuit','scircuit.mtx'),
('Webbase','webbase-1M.mtx'),
('LP','rail4284.mtx') ]
unstructured_mats = [ mat + (unstructured_path,) for mat in unstructured_mats]
structured_path = '~/scratch/Matrices/stencil/'
structured_mats = [('Laplacian_3pt_stencil', '3pt_1000000.mtx'),
('Laplacian_5pt_stencil', '5pt_1000x1000.mtx'),
('Laplacian_7pt_stencil', '7pt_100x100x100.mtx'),
('Laplacian_9pt_stencil', '9pt_1000x1000.mtx'),
('Laplacian_27pt_stencil', '27pt_100x100x100.mtx')]
structured_mats = [ mat + (structured_path,) for mat in structured_mats]
# assemble suite of matrices
trials = unstructured_mats + structured_mats
def run_tests(value_type):
# remove previous result (if present)
open(output_file,'w').close()
# run benchmark for each file
for matrix,filename,path in trials:
matrix_filename = path + filename
# setup the command to execute
cmd = binary_filename
cmd += ' ' + matrix_filename # e.g. pwtk.mtx
cmd += ' --device=' + device_id # e.g. 0 or 1
cmd += ' --value_type=' + value_type # e.g. float or double
# execute the benchmark on this file
os.system(cmd)
# process output_file
matrices = {}
results = {}
kernels = set()
#
fid = open(output_file)
for line in fid.readlines():
tokens = dict( [tuple(part.split('=')) for part in line.split()] )
if 'file' in tokens:
file = os.path.split(tokens['file'])[1]
matrices[file] = tokens
results[file] = {}
else:
kernel = tokens['kernel']
results[file][kernel] = tokens
kernels.add(tokens['kernel'])
## put CPU results before GPU results
#kernels = ['csr_serial'] + sorted(kernels - set(['csr_serial']))
kernels = sorted(kernels)
# write out CSV formatted results
def write_csv(field):
fid = open('bench_' + value_type + '_' + field + '.csv','w')
writer = csv.writer(fid)
writer.writerow(['matrix','file','rows','cols','nonzeros'] + kernels)
for (matrix,file,path) in trials:
line = [matrix, file, matrices[file]['rows'], matrices[file]['cols'], matrices[file]['nonzeros']]
matrix_results = results[file]
for kernel in kernels:
if kernel in matrix_results:
line.append( matrix_results[kernel][field] )
else:
line.append(' ')
writer.writerow( line )
fid.close()
write_csv('gflops') #GFLOP/s
write_csv('gbytes') #GBytes/s
run_tests('float')
run_tests('double')
|
