forked from scipy/scipy
-
Notifications
You must be signed in to change notification settings - Fork 0
/
correlate_nd.c.src
327 lines (273 loc) · 9.38 KB
/
correlate_nd.c.src
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
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
/*
* vim:syntax=c
* vim:sw=4
*/
#include <Python.h>
#define PY_ARRAY_UNIQUE_SYMBOL _scipy_signal_ARRAY_API
#define NO_IMPORT_ARRAY
#include "numpy/ndarrayobject.h"
#include "sigtools.h"
enum {
CORR_MODE_VALID=0,
CORR_MODE_SAME,
CORR_MODE_FULL
};
static int _correlate_nd_imp(PyArrayIterObject* x, PyArrayIterObject *y,
PyArrayIterObject *z, int typenum, int mode);
PyObject *
scipy_signal_sigtools_correlateND(PyObject *NPY_UNUSED(dummy), PyObject *args)
{
PyObject *x, *y, *out;
PyArrayObject *ax, *ay, *aout;
PyArrayIterObject *itx, *ity, *itz;
int mode, typenum, st;
if (!PyArg_ParseTuple(args, "OOOi", &x, &y, &out, &mode)) {
return NULL;
}
typenum = PyArray_ObjectType(x, 0);
typenum = PyArray_ObjectType(y, typenum);
typenum = PyArray_ObjectType(out, typenum);
ax = (PyArrayObject *)PyArray_FromObject(x, typenum, 0, 0);
if (ax == NULL) {
return NULL;
}
ay = (PyArrayObject *)PyArray_FromObject(y, typenum, 0, 0);
if (ay == NULL) {
goto clean_ax;
}
aout = (PyArrayObject *)PyArray_FromObject(out, typenum, 0, 0);
if (aout == NULL) {
goto clean_ay;
}
if (PyArray_NDIM(ax) != PyArray_NDIM(ay)) {
PyErr_SetString(PyExc_ValueError,
"Arrays must have the same number of dimensions.");
goto clean_aout;
}
if (PyArray_NDIM(ax) == 0) {
PyErr_SetString(PyExc_ValueError, "Cannot convolve zero-dimensional arrays.");
goto clean_aout;
}
itx = (PyArrayIterObject*)PyArray_IterNew((PyObject*)ax);
if (itx == NULL) {
goto clean_aout;
}
ity = (PyArrayIterObject*)PyArray_IterNew((PyObject*)ay);
if (ity == NULL) {
goto clean_itx;
}
itz = (PyArrayIterObject*)PyArray_IterNew((PyObject*)aout);
if (itz == NULL) {
goto clean_ity;
}
st = _correlate_nd_imp(itx, ity, itz, typenum, mode);
if (st) {
goto clean_itz;
}
Py_DECREF(itz);
Py_DECREF(ity);
Py_DECREF(itx);
Py_DECREF(ax);
Py_DECREF(ay);
return PyArray_Return(aout);
clean_itz:
Py_DECREF(itz);
clean_ity:
Py_DECREF(ity);
clean_itx:
Py_DECREF(itx);
clean_aout:
Py_DECREF(aout);
clean_ay:
Py_DECREF(ay);
clean_ax:
Py_DECREF(ax);
return NULL;
}
/*
* Implementation of the type-specific correlation 'kernels'
*/
/**begin repeat
* #fsuf = ubyte, byte, ushort, short, uint, int, ulong,
* long, ulonglong, longlong, float, double, longdouble#
* #type = npy_ubyte, npy_byte, npy_ushort, short, npy_uint, int, npy_ulong,
* long, npy_ulonglong, npy_longlong, float, double, npy_longdouble#
*/
static int _imp_correlate_nd_@fsuf@(PyArrayNeighborhoodIterObject *curx,
PyArrayNeighborhoodIterObject *curneighx, PyArrayIterObject *ity,
PyArrayIterObject *itz)
{
npy_intp i, j;
@type@ acc;
for(i = 0; i < curx->size; ++i) {
acc = 0;
PyArrayNeighborhoodIter_Reset(curneighx);
for(j = 0; j < curneighx->size; ++j) {
acc += *((@type@*)(curneighx->dataptr)) * *((@type@*)(ity->dataptr));
PyArrayNeighborhoodIter_Next(curneighx);
PyArray_ITER_NEXT(ity);
}
PyArrayNeighborhoodIter_Next(curx);
*((@type@*)(itz->dataptr)) = acc;
PyArray_ITER_NEXT(itz);
PyArray_ITER_RESET(ity);
}
return 0;
}
/**end repeat**/
/**begin repeat
* #fsuf = float, double, longdouble#
* #type = float, double, npy_longdouble#
*/
static int _imp_correlate_nd_c@fsuf@(PyArrayNeighborhoodIterObject *curx,
PyArrayNeighborhoodIterObject *curneighx, PyArrayIterObject *ity,
PyArrayIterObject *itz)
{
npy_intp i, j;
@type@ racc, iacc;
@type@ *ptr1, *ptr2;
for(i = 0; i < curx->size; ++i) {
racc = 0;
iacc = 0;
PyArrayNeighborhoodIter_Reset(curneighx);
for(j = 0; j < curneighx->size; ++j) {
ptr1 = ((@type@*)(curneighx->dataptr));
ptr2 = ((@type@*)(ity->dataptr));
racc += ptr1[0] * ptr2[0] + ptr1[1] * ptr2[1];
iacc += ptr1[1] * ptr2[0] - ptr1[0] * ptr2[1];
PyArrayNeighborhoodIter_Next(curneighx);
PyArray_ITER_NEXT(ity);
}
PyArrayNeighborhoodIter_Next(curx);
((@type@*)(itz->dataptr))[0] = racc;
((@type@*)(itz->dataptr))[1] = iacc;
PyArray_ITER_NEXT(itz);
PyArray_ITER_RESET(ity);
}
return 0;
}
/**end repeat**/
static int _imp_correlate_nd_object(PyArrayNeighborhoodIterObject *curx,
PyArrayNeighborhoodIterObject *curneighx, PyArrayIterObject *ity,
PyArrayIterObject *itz)
{
npy_intp i, j;
PyObject *tmp, *tmp2;
char *zero;
PyArray_CopySwapFunc *copyswap = PyArray_DESCR(curx->ao)->f->copyswap;
zero = PyArray_Zero(curx->ao);
for(i = 0; i < curx->size; ++i) {
PyArrayNeighborhoodIter_Reset(curneighx);
copyswap(itz->dataptr, zero, 0, NULL);
for(j = 0; j < curneighx->size; ++j) {
/*
* compute tmp2 = acc + x * y. Not all objects supporting the
* number protocol support inplace operations, so we do it the most
* straightforward way.
*/
tmp = PyNumber_Multiply(*((PyObject**)curneighx->dataptr),
*((PyObject**)ity->dataptr));
tmp2 = PyNumber_Add(*((PyObject**)itz->dataptr), tmp);
Py_DECREF(tmp);
/* Update current output item (acc) */
Py_DECREF(*((PyObject**)itz->dataptr));
*((PyObject**)itz->dataptr) = tmp2;
PyArrayNeighborhoodIter_Next(curneighx);
PyArray_ITER_NEXT(ity);
}
PyArrayNeighborhoodIter_Next(curx);
PyArray_ITER_NEXT(itz);
PyArray_ITER_RESET(ity);
}
PyDataMem_FREE(zero);
return 0;
}
static int _correlate_nd_imp(PyArrayIterObject* itx, PyArrayIterObject *ity,
PyArrayIterObject *itz, int typenum, int mode)
{
PyArrayNeighborhoodIterObject *curneighx, *curx;
npy_intp i, nz, nx;
npy_intp bounds[NPY_MAXDIMS*2];
/* Compute boundaries for the neighborhood iterator curx: curx is used to
* traverse x directly, such as each point of the output is the
* innerproduct of y with the neighborhood around curx */
switch(mode) {
case CORR_MODE_VALID:
/* Only walk through the input points such as the corresponding
* output will not depend on 0 padding */
for(i = 0; i < PyArray_NDIM(itx->ao); ++i) {
bounds[2*i] = PyArray_DIMS(ity->ao)[i] - 1;
bounds[2*i+1] = PyArray_DIMS(itx->ao)[i] - 1;
}
break;
case CORR_MODE_SAME:
/* Only walk through the input such as the output will be centered
relatively to the output as computed in the full mode */
for(i = 0; i < PyArray_NDIM(itx->ao); ++i) {
nz = PyArray_DIMS(itx->ao)[i];
/* Recover 'original' nx, before it was zero-padded */
nx = nz - PyArray_DIMS(ity->ao)[i] + 1;
if ((nz - nx) % 2 == 0) {
bounds[2*i] = (nz - nx) / 2;
} else {
bounds[2*i] = (nz - nx - 1) / 2;
}
bounds[2*i+1] = bounds[2*i] + nx - 1;
}
break;
case CORR_MODE_FULL:
for(i = 0; i < PyArray_NDIM(itx->ao); ++i) {
bounds[2*i] = 0;
bounds[2*i+1] = PyArray_DIMS(itx->ao)[i] - 1;
}
break;
default:
PyErr_BadInternalCall();
return -1;
}
curx = (PyArrayNeighborhoodIterObject*)PyArray_NeighborhoodIterNew(itx,
bounds, NPY_NEIGHBORHOOD_ITER_ZERO_PADDING, NULL);
if (curx == NULL) {
PyErr_SetString(PyExc_SystemError, "Could not create curx ?");
return -1;
}
/* Compute boundaries for the neighborhood iterator: the neighborhood for x
should have the same dimensions as y */
for(i = 0; i < PyArray_NDIM(ity->ao); ++i) {
bounds[2*i] = -PyArray_DIMS(ity->ao)[i] + 1;
bounds[2*i+1] = 0;
}
curneighx = (PyArrayNeighborhoodIterObject*)PyArray_NeighborhoodIterNew(
(PyArrayIterObject*)curx, bounds, NPY_NEIGHBORHOOD_ITER_ZERO_PADDING, NULL);
if (curneighx == NULL) {
goto clean_curx;
}
switch(typenum) {
/**begin repeat
* #TYPE = UBYTE, BYTE, USHORT, SHORT, UINT, INT, ULONG, LONG, ULONGLONG,
* LONGLONG, FLOAT, DOUBLE, LONGDOUBLE, CFLOAT, CDOUBLE, CLONGDOUBLE#
* #type = ubyte, byte, ushort, short, uint, int, ulong, long, ulonglong,
* longlong, float, double, longdouble, cfloat, cdouble, clongdouble#
*/
case NPY_@TYPE@:
_imp_correlate_nd_@type@(curx, curneighx, ity, itz);
break;
/**end repeat**/
/* The object array case does not worth being optimized, since most of
the cost is numerical operations, not iterators moving in this case ? */
case NPY_OBJECT:
_imp_correlate_nd_object(curx, curneighx, ity, itz);
break;
default:
PyErr_SetString(PyExc_ValueError, "Unsupported type");
goto clean_curneighx;
}
Py_DECREF((PyArrayIterObject*)curx);
Py_DECREF((PyArrayIterObject*)curneighx);
return 0;
clean_curneighx:
Py_DECREF((PyArrayIterObject*)curneighx);
clean_curx:
Py_DECREF((PyArrayIterObject*)curx);
return -1;
}