回答編集履歴
3
bsdfanさんのロジックを取り入れて修正。私の環境ではnumba版と比べてcython版は数割速くなるようです。
test
CHANGED
|
@@ -46,7 +46,19 @@
|
|
|
46
46
|
|
|
47
47
|
|
|
48
48
|
|
|
49
|
+
start = t.time()
|
|
50
|
+
|
|
51
|
+
result3 = cy.test_20190524_changed2(a, b, c, x, y)
|
|
52
|
+
|
|
53
|
+
print('time : ' + str(round((t.time() - start),5)) + ' [sec]')
|
|
54
|
+
|
|
55
|
+
|
|
56
|
+
|
|
57
|
+
|
|
58
|
+
|
|
49
|
-
print((result1 == result2).all()
|
|
59
|
+
print((result1 == result2).all() and
|
|
60
|
+
|
|
61
|
+
(result2 == result3).all())
|
|
50
62
|
|
|
51
63
|
|
|
52
64
|
|
|
@@ -56,9 +68,11 @@
|
|
|
56
68
|
|
|
57
69
|
""" =>
|
|
58
70
|
|
|
59
|
-
time : 0.097
|
|
71
|
+
time : 0.09572 [sec]
|
|
72
|
+
|
|
60
|
-
|
|
73
|
+
time : 0.01765 [sec]
|
|
74
|
+
|
|
61
|
-
time : 0.01
|
|
75
|
+
time : 0.01645 [sec]
|
|
62
76
|
|
|
63
77
|
True
|
|
64
78
|
|
|
@@ -66,6 +80,8 @@
|
|
|
66
80
|
|
|
67
81
|
|
|
68
82
|
|
|
83
|
+
|
|
84
|
+
|
|
69
85
|
```
|
|
70
86
|
|
|
71
87
|
|
|
@@ -134,6 +150,56 @@
|
|
|
134
150
|
|
|
135
151
|
@cython.wraparound(False)
|
|
136
152
|
|
|
153
|
+
cpdef np.ndarray[np_float_t, ndim=2] test_20190524_changed2(
|
|
154
|
+
|
|
155
|
+
np.ndarray[np_float_t, ndim=3] a,
|
|
156
|
+
|
|
157
|
+
np.ndarray[np_float_t, ndim=2] b,
|
|
158
|
+
|
|
159
|
+
np.ndarray[np_int_t, ndim=1] c,
|
|
160
|
+
|
|
161
|
+
np.ndarray[np_int_t, ndim=1] x,
|
|
162
|
+
|
|
163
|
+
np.ndarray[np_int_t, ndim=1] y):
|
|
164
|
+
|
|
165
|
+
|
|
166
|
+
|
|
167
|
+
cdef np.ndarray[np_float_t, ndim=3] d
|
|
168
|
+
|
|
169
|
+
cdef np.ndarray[np_float_t, ndim=2] img
|
|
170
|
+
|
|
171
|
+
cdef int i, j, k, num
|
|
172
|
+
|
|
173
|
+
|
|
174
|
+
|
|
175
|
+
num = 10000
|
|
176
|
+
|
|
177
|
+
|
|
178
|
+
|
|
179
|
+
d = a * b[c, 0]
|
|
180
|
+
|
|
181
|
+
|
|
182
|
+
|
|
183
|
+
img = np.zeros((10000, 10000))
|
|
184
|
+
|
|
185
|
+
|
|
186
|
+
|
|
187
|
+
for i in range(num):
|
|
188
|
+
|
|
189
|
+
for j in range(2):
|
|
190
|
+
|
|
191
|
+
for k in range(2):
|
|
192
|
+
|
|
193
|
+
img[x[i] + j , y[i] + k] += d[j, k, i]
|
|
194
|
+
|
|
195
|
+
return img
|
|
196
|
+
|
|
197
|
+
|
|
198
|
+
|
|
199
|
+
@cython.boundscheck(False)
|
|
200
|
+
|
|
201
|
+
@cython.wraparound(False)
|
|
202
|
+
|
|
137
203
|
cpdef np.ndarray[np_int_t, ndim=1] test_20190524_origin(
|
|
138
204
|
|
|
139
205
|
np.ndarray[np_float_t, ndim=3] a,
|
2
改行の編集
test
CHANGED
|
@@ -128,6 +128,8 @@
|
|
|
128
128
|
|
|
129
129
|
return img
|
|
130
130
|
|
|
131
|
+
|
|
132
|
+
|
|
131
133
|
@cython.boundscheck(False)
|
|
132
134
|
|
|
133
135
|
@cython.wraparound(False)
|
1
デコレータの修正。速度はほとんど変化なし
test
CHANGED
|
@@ -92,8 +92,6 @@
|
|
|
92
92
|
|
|
93
93
|
@cython.wraparound(False)
|
|
94
94
|
|
|
95
|
-
|
|
96
|
-
|
|
97
95
|
cpdef np.ndarray[np_float_t, ndim=2] test_20190524_changed(
|
|
98
96
|
|
|
99
97
|
np.ndarray[np_float_t, ndim=3] a,
|
|
@@ -130,7 +128,9 @@
|
|
|
130
128
|
|
|
131
129
|
return img
|
|
132
130
|
|
|
131
|
+
@cython.boundscheck(False)
|
|
133
132
|
|
|
133
|
+
@cython.wraparound(False)
|
|
134
134
|
|
|
135
135
|
cpdef np.ndarray[np_int_t, ndim=1] test_20190524_origin(
|
|
136
136
|
|