回答編集履歴
1
edit
test
CHANGED
|
@@ -27,3 +27,163 @@
|
|
|
27
27
|
|
|
28
28
|
|
|
29
29
|
「直したソースコード」が誤っていますが、実行すべきコードを実行できていますか？
|
|
30
|
+
|
|
31
|
+
|
|
32
|
+
|
|
33
|
+
---
|
|
34
|
+
|
|
35
|
+
|
|
36
|
+
|
|
37
|
+
私の手元では以下のコードのuseCPUを変えると正常にスケールします。
|
|
38
|
+
|
|
39
|
+
|
|
40
|
+
|
|
41
|
+
```python
|
|
42
|
+
|
|
43
|
+
import concurrent.futures
|
|
44
|
+
|
|
45
|
+
import numpy as np
|
|
46
|
+
|
|
47
|
+
import time
|
|
48
|
+
|
|
49
|
+
|
|
50
|
+
|
|
51
|
+
img = np.ones(shape=(500,5000,3))
|
|
52
|
+
|
|
53
|
+
|
|
54
|
+
|
|
55
|
+
def main():
|
|
56
|
+
|
|
57
|
+
useCPU = 4
|
|
58
|
+
|
|
59
|
+
step = int(len(img) / useCPU )
|
|
60
|
+
|
|
61
|
+
mulchProcess(useCPU=useCPU, step=step)
|
|
62
|
+
|
|
63
|
+
|
|
64
|
+
|
|
65
|
+
def changeToGray(number: int, length: int):
|
|
66
|
+
|
|
67
|
+
s = time.time()
|
|
68
|
+
|
|
69
|
+
endPioint = number + length if number + length < len(img) else len(img) - 1
|
|
70
|
+
|
|
71
|
+
part_height = img[number:endPioint-1]
|
|
72
|
+
|
|
73
|
+
for width in part_height:
|
|
74
|
+
|
|
75
|
+
for pixel in width:
|
|
76
|
+
|
|
77
|
+
gray = int(pixel[0]*0.3) + int(pixel[1]*0.59) + int(pixel[2]*0.11)
|
|
78
|
+
|
|
79
|
+
pixel[0] = gray
|
|
80
|
+
|
|
81
|
+
pixel[1] = gray
|
|
82
|
+
|
|
83
|
+
pixel[2] = gray
|
|
84
|
+
|
|
85
|
+
t = time.time() - s
|
|
86
|
+
|
|
87
|
+
return number, part_height, t
|
|
88
|
+
|
|
89
|
+
|
|
90
|
+
|
|
91
|
+
def mulchProcess(useCPU: int, step: int):
|
|
92
|
+
|
|
93
|
+
index_list = [i for i in range(0, len(img), step)]
|
|
94
|
+
|
|
95
|
+
with concurrent.futures.ProcessPoolExecutor(max_workers=useCPU) as executer:
|
|
96
|
+
|
|
97
|
+
fs = [executer.submit(changeToGray, i, step) for i in index_list]
|
|
98
|
+
|
|
99
|
+
for future in concurrent.futures.as_completed(fs):
|
|
100
|
+
|
|
101
|
+
line_number = future.result()[0]
|
|
102
|
+
|
|
103
|
+
part_height = future.result()[1]
|
|
104
|
+
|
|
105
|
+
t = future.result()[2]
|
|
106
|
+
|
|
107
|
+
print(t)
|
|
108
|
+
|
|
109
|
+
for i, height in zip(range(line_number, line_number+len(part_height)), part_height):
|
|
110
|
+
|
|
111
|
+
img[i] = height
|
|
112
|
+
|
|
113
|
+
|
|
114
|
+
|
|
115
|
+
s = time.time()
|
|
116
|
+
|
|
117
|
+
main()
|
|
118
|
+
|
|
119
|
+
print(time.time()-s)
|
|
120
|
+
|
|
121
|
+
```
|
|
122
|
+
|
|
123
|
+
|
|
124
|
+
|
|
125
|
+
時間は以下のようになります。
|
|
126
|
+
|
|
127
|
+
手元のPCは4coreなので、8個使おうとしても一度に4つずつしか実行できません。
|
|
128
|
+
|
|
129
|
+
それらがキューに乗せられて順番に実行されるわけではなく、スイッチしながら実行されるので、プロセスあたり2倍の時間がかかり、結局全体の時間は同じになります。
|
|
130
|
+
|
|
131
|
+
むしろスイッチした分だけオーバーヘッドがあって全体の実行時間が長くなっています。
|
|
132
|
+
|
|
133
|
+
```
|
|
134
|
+
|
|
135
|
+
useCPU = 1
|
|
136
|
+
|
|
137
|
+
3.6020491123199463
|
|
138
|
+
|
|
139
|
+
total 4.26886773109436
|
|
140
|
+
|
|
141
|
+
|
|
142
|
+
|
|
143
|
+
useCPU = 2
|
|
144
|
+
|
|
145
|
+
1.7662358283996582
|
|
146
|
+
|
|
147
|
+
1.7804999351501465
|
|
148
|
+
|
|
149
|
+
total 2.166551113128662
|
|
150
|
+
|
|
151
|
+
|
|
152
|
+
|
|
153
|
+
useCPU = 4
|
|
154
|
+
|
|
155
|
+
0.9876649379730225
|
|
156
|
+
|
|
157
|
+
0.9972729682922363
|
|
158
|
+
|
|
159
|
+
1.0097663402557373
|
|
160
|
+
|
|
161
|
+
1.0141608715057373
|
|
162
|
+
|
|
163
|
+
total 1.2486088275909424
|
|
164
|
+
|
|
165
|
+
|
|
166
|
+
|
|
167
|
+
useCPU = 8
|
|
168
|
+
|
|
169
|
+
0.9724259376525879
|
|
170
|
+
|
|
171
|
+
0.9717621803283691
|
|
172
|
+
|
|
173
|
+
0.973222017288208
|
|
174
|
+
|
|
175
|
+
0.97617506980896
|
|
176
|
+
|
|
177
|
+
0.9814469814300537
|
|
178
|
+
|
|
179
|
+
0.9806389808654785
|
|
180
|
+
|
|
181
|
+
0.9841761589050293
|
|
182
|
+
|
|
183
|
+
0.9821760654449463
|
|
184
|
+
|
|
185
|
+
0.015022039413452148
|
|
186
|
+
|
|
187
|
+
total 1.2429370880126953
|
|
188
|
+
|
|
189
|
+
```
|