-
Notifications
You must be signed in to change notification settings - Fork 0
/
benchmark.py
199 lines (161 loc) · 6.85 KB
/
benchmark.py
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
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
import itertools
import math
import os
import sys
import copy
import eigen
# This file generates multiple transpositions for benchmarking reasons.
# The generated transpositions differ in: Size, dimensionality and order.
# The permutations are choosen such that no index can be fused.
if len(sys.argv) < 3:
print "usage: <number of threads> <thread affinity> <compiler>"
print ""
print "thread affinity:"
print " The thread affinity needs to specified for the given system because it can effect the performance severely."
print " Thes specified value respectively sets the 'KMP_AFFINITY' or 'GOMP_CPU_AFFINITY' environment variable for Intel's ICPC or g++."
print " Examples:"
print " ICPC: 'compact,1'"
print " g++: '0,1,2,3'"
print ""
print "compiler: this value can be either 'g++' or 'icpc'"
print ""
print "Example: 'python benchmark.py 24 compact,1 icpc'"
print "Example: 'python benchmark.py 2 0,2 g++'"
exit(0)
_EigenRoot = "/home/ps072922/projects/eigen-3.3.3/"
_affinity = sys.argv[2]
_compiler = sys.argv[3]
#######################
# Default settings
######################
_floatType = "float"
_beta = 1.0
_numThreads = int(sys.argv[1])
_minBlock = 16
_sizeMB = 200 # in MB
_leadingDimMultiple = 16 #the stride-1 index of both 'A' and 'B' have to be a multiple of this value (useful if a special alignment is required)
if _floatType == "float":
_floatTypeSize = 4.
elif _floatType == "double":
_floatTypeSize = 8
elif _floatType == "complex":
_floatTypeSize = 8
elif _floatType == "doubleComplex":
_floatTypeSize = 16
###################
# DO NOT MODIFY:
###################
_permutations = [
[[1,0]],
[[0,2,1],[1,0,2],[2,1,0]],
[[0,3,2,1],[2,1,3,0],[2,0,3,1],[1,0,3,2],[3,2,1,0]],
[[0,4,2,1,3],[3,2,1,4,0],[2,0,4,1,3],[1,3,0,4,2],[4,3,2,1,0]],
[[0,3,2,5,4,1],[3,2,0,5,1,4],[2,0,4,1,5,3],[3,2,5,1,0,4],[5,4,3,2,1,0]]
]
_genString = "ttc --beta=%f --maxImplementations=500 --numThreads=%d --compiler=%s --architecture=avx --affinity=%s"%(_beta,_numThreads,_compiler,_affinity)
def output(size, perm, fileHandle, fileHandleEigen, counter):
sizeStr = ""
for s in size:
sizeStr += str(s)+","
sizeStr = sizeStr[0:-1] #delete last ','
permStr = ""
for s in perm:
permStr += str(s)+","
permStr = permStr[0:-1] #delete last ','
fileHandle.write(_genString +" --size="+sizeStr + " --perm="+permStr+"\n")
print permStr, "&", sizeStr
filename = "eigen%d.cpp"%counter
fileHandle = open(filename,"w")
code = eigen.genEigen(size, perm, _floatType, _floatTypeSize, _numThreads)
fileHandle.write(code)
fileHandle.close()
fileHandleEigen.write("icpc -O3 -I%s -std=c++14 -qopenmp -xHost %s\n"%(_EigenRoot,filename)) #O0 is used to avoid that the compiler removes trashCache()
fileHandleEigen.write("KMP_AFFINITY=compact,1 OMP_NUM_THREADS=%d numactl --interleave=all ./a.out >>eigen.dat\n"%_numThreads)
counter += 1
fileHandle = open("benchmark.sh","w")
fileHandleEigen = open("benchmarkEigen.sh","w")
fileHandleEigen.write("rm -f eigen.dat\n")
counter = 0
for dim in range(2,7):
numElements = _sizeMB/_floatTypeSize * 2.**20 #size in elements
#determine the value for which base**dim yields roughly an array of size 'sizeMB'
base = int(math.pow(numElements, 1./dim))
for perm in _permutations[dim-2]:
#make the first dimension a multiple of the minBlockSize
size0 = (int(base + _minBlock - 1) / _minBlock) * _minBlock
#the first 0-dim as well as the perm[0]-dimension will have size0 elements each
numElementsTmp = numElements/(size0*size0)
if(dim > 2):
baseTmp = int(math.pow(numElementsTmp, 1./(dim-2)))
else:
baseTmp = 1
size = [ baseTmp for i in range(dim)]
if( perm[0] == 0):
size[1] = size0
size[perm[1]] = size0
else:
size[0] = size0
size[perm[0]] = size0
################
# at this point we have a size which is fairly similar in each dimension.
#
# Now let's generate three different sizes for each permutation.
################
sizeTmp = copy.deepcopy(size)
if( sizeTmp[0] % _leadingDimMultiple != 0):
sizeTmp[0] += (_leadingDimMultiple - sizeTmp[0]%_leadingDimMultiple)
if( sizeTmp[perm[0]] % _leadingDimMultiple != 0):
sizeTmp[perm[0]] += (_leadingDimMultiple - sizeTmp[perm[0]]%_leadingDimMultiple)
### 1) everything pretty equal
output(sizeTmp, perm, fileHandle, fileHandleEigen, counter)
counter+=1
if dim >= 6:
scewFactor = 3
else:
scewFactor = 6
sizeTmp = copy.deepcopy(size)
### 2) skewed in 0-dim
sizeTmp[0] *= scewFactor
for i in range(1,dim):
if( sizeTmp[i] > scewFactor and (perm[0] != i or dim == 2)):
sizeTmp[i] /= scewFactor
break
# ensure that the sizeTmp is close to the desired value
totalElements = 1
for s in sizeTmp:
totalElements *= s
sizeTmpMax = int(numElements / (totalElements / max(sizeTmp)))
sizeTmp[sizeTmp.index(max(sizeTmp))] = sizeTmpMax
if( sizeTmp[0] % _leadingDimMultiple != 0):
sizeTmp[0] += (_leadingDimMultiple - sizeTmp[0]%_leadingDimMultiple)
if( sizeTmp[perm[0]] % _leadingDimMultiple != 0):
sizeTmp[perm[0]] += (_leadingDimMultiple - sizeTmp[perm[0]]%_leadingDimMultiple)
output(sizeTmp, perm, fileHandle, fileHandleEigen, counter)
counter+=1
### 3) skewed in perm[0]-dim
sizeTmp = copy.deepcopy(size) #restore old size
idx = 0
if( perm[0] == 0 ):
idx = 1
sizeTmp[perm[idx]] *= scewFactor
if( dim == 2 ):
sizeTmp[0] /= scewFactor
else:
for i in range(1,dim):
if( sizeTmp[i] > scewFactor and (perm[idx] != i)):
sizeTmp[i] /= scewFactor
break
# ensure that the sizeTmp is close to the desired value
totalElements = 1
for s in sizeTmp:
totalElements *= s
sizeTmpMax = int(numElements / (totalElements / max(sizeTmp)))
sizeTmp[sizeTmp.index(max(sizeTmp))] = sizeTmpMax
if( sizeTmp[0] % _leadingDimMultiple != 0):
sizeTmp[0] += (_leadingDimMultiple - sizeTmp[0]%_leadingDimMultiple)
if( sizeTmp[perm[0]] % _leadingDimMultiple != 0):
sizeTmp[perm[0]] += (_leadingDimMultiple - sizeTmp[perm[0]]%_leadingDimMultiple)
output(sizeTmp, perm, fileHandle, fileHandleEigen, counter)
counter+=1
fileHandle.close()
fileHandleEigen.close()