sum.simd.hpp
12.4 KB
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
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html
#include "precomp.hpp"
#include "stat.hpp"
namespace cv {
CV_CPU_OPTIMIZATION_NAMESPACE_BEGIN
SumFunc getSumFunc(int depth);
#ifndef CV_CPU_OPTIMIZATION_DECLARATIONS_ONLY
template <typename T, typename ST>
struct Sum_SIMD
{
int operator () (const T *, const uchar *, ST *, int, int) const
{
return 0;
}
};
#if CV_SIMD
template <>
struct Sum_SIMD<uchar, int>
{
int operator () (const uchar * src0, const uchar * mask, int * dst, int len, int cn) const
{
if (mask || (cn != 1 && cn != 2 && cn != 4))
return 0;
len *= cn;
int x = 0;
v_uint32 v_sum = vx_setzero_u32();
int len0 = len & -v_uint8::nlanes;
while (x < len0)
{
const int len_tmp = min(x + 256*v_uint16::nlanes, len0);
v_uint16 v_sum16 = vx_setzero_u16();
for (; x < len_tmp; x += v_uint8::nlanes)
{
v_uint16 v_src0, v_src1;
v_expand(vx_load(src0 + x), v_src0, v_src1);
v_sum16 += v_src0 + v_src1;
}
v_uint32 v_half0, v_half1;
v_expand(v_sum16, v_half0, v_half1);
v_sum += v_half0 + v_half1;
}
if (x <= len - v_uint16::nlanes)
{
v_uint32 v_half0, v_half1;
v_expand(vx_load_expand(src0 + x), v_half0, v_half1);
v_sum += v_half0 + v_half1;
x += v_uint16::nlanes;
}
if (x <= len - v_uint32::nlanes)
{
v_sum += vx_load_expand_q(src0 + x);
x += v_uint32::nlanes;
}
if (cn == 1)
*dst += v_reduce_sum(v_sum);
else
{
uint32_t CV_DECL_ALIGNED(CV_SIMD_WIDTH) ar[v_uint32::nlanes];
v_store_aligned(ar, v_sum);
for (int i = 0; i < v_uint32::nlanes; ++i)
dst[i % cn] += ar[i];
}
v_cleanup();
return x / cn;
}
};
template <>
struct Sum_SIMD<schar, int>
{
int operator () (const schar * src0, const uchar * mask, int * dst, int len, int cn) const
{
if (mask || (cn != 1 && cn != 2 && cn != 4))
return 0;
len *= cn;
int x = 0;
v_int32 v_sum = vx_setzero_s32();
int len0 = len & -v_int8::nlanes;
while (x < len0)
{
const int len_tmp = min(x + 256*v_int16::nlanes, len0);
v_int16 v_sum16 = vx_setzero_s16();
for (; x < len_tmp; x += v_int8::nlanes)
{
v_int16 v_src0, v_src1;
v_expand(vx_load(src0 + x), v_src0, v_src1);
v_sum16 += v_src0 + v_src1;
}
v_int32 v_half0, v_half1;
v_expand(v_sum16, v_half0, v_half1);
v_sum += v_half0 + v_half1;
}
if (x <= len - v_int16::nlanes)
{
v_int32 v_half0, v_half1;
v_expand(vx_load_expand(src0 + x), v_half0, v_half1);
v_sum += v_half0 + v_half1;
x += v_int16::nlanes;
}
if (x <= len - v_int32::nlanes)
{
v_sum += vx_load_expand_q(src0 + x);
x += v_int32::nlanes;
}
if (cn == 1)
*dst += v_reduce_sum(v_sum);
else
{
int32_t CV_DECL_ALIGNED(CV_SIMD_WIDTH) ar[v_int32::nlanes];
v_store_aligned(ar, v_sum);
for (int i = 0; i < v_int32::nlanes; ++i)
dst[i % cn] += ar[i];
}
v_cleanup();
return x / cn;
}
};
template <>
struct Sum_SIMD<ushort, int>
{
int operator () (const ushort * src0, const uchar * mask, int * dst, int len, int cn) const
{
if (mask || (cn != 1 && cn != 2 && cn != 4))
return 0;
len *= cn;
int x = 0;
v_uint32 v_sum = vx_setzero_u32();
for (; x <= len - v_uint16::nlanes; x += v_uint16::nlanes)
{
v_uint32 v_src0, v_src1;
v_expand(vx_load(src0 + x), v_src0, v_src1);
v_sum += v_src0 + v_src1;
}
if (x <= len - v_uint32::nlanes)
{
v_sum += vx_load_expand(src0 + x);
x += v_uint32::nlanes;
}
if (cn == 1)
*dst += v_reduce_sum(v_sum);
else
{
uint32_t CV_DECL_ALIGNED(CV_SIMD_WIDTH) ar[v_uint32::nlanes];
v_store_aligned(ar, v_sum);
for (int i = 0; i < v_uint32::nlanes; ++i)
dst[i % cn] += ar[i];
}
v_cleanup();
return x / cn;
}
};
template <>
struct Sum_SIMD<short, int>
{
int operator () (const short * src0, const uchar * mask, int * dst, int len, int cn) const
{
if (mask || (cn != 1 && cn != 2 && cn != 4))
return 0;
len *= cn;
int x = 0;
v_int32 v_sum = vx_setzero_s32();
for (; x <= len - v_int16::nlanes; x += v_int16::nlanes)
{
v_int32 v_src0, v_src1;
v_expand(vx_load(src0 + x), v_src0, v_src1);
v_sum += v_src0 + v_src1;
}
if (x <= len - v_int32::nlanes)
{
v_sum += vx_load_expand(src0 + x);
x += v_int32::nlanes;
}
if (cn == 1)
*dst += v_reduce_sum(v_sum);
else
{
int32_t CV_DECL_ALIGNED(CV_SIMD_WIDTH) ar[v_int32::nlanes];
v_store_aligned(ar, v_sum);
for (int i = 0; i < v_int32::nlanes; ++i)
dst[i % cn] += ar[i];
}
v_cleanup();
return x / cn;
}
};
#if CV_SIMD_64F
template <>
struct Sum_SIMD<int, double>
{
int operator () (const int * src0, const uchar * mask, double * dst, int len, int cn) const
{
if (mask || (cn != 1 && cn != 2 && cn != 4))
return 0;
len *= cn;
int x = 0;
v_float64 v_sum0 = vx_setzero_f64();
v_float64 v_sum1 = vx_setzero_f64();
for (; x <= len - 2 * v_int32::nlanes; x += 2 * v_int32::nlanes)
{
v_int32 v_src0 = vx_load(src0 + x);
v_int32 v_src1 = vx_load(src0 + x + v_int32::nlanes);
v_sum0 += v_cvt_f64(v_src0) + v_cvt_f64(v_src1);
v_sum1 += v_cvt_f64_high(v_src0) + v_cvt_f64_high(v_src1);
}
#if CV_SIMD256 || CV_SIMD512
double CV_DECL_ALIGNED(CV_SIMD_WIDTH) ar[v_float64::nlanes];
v_store_aligned(ar, v_sum0 + v_sum1);
for (int i = 0; i < v_float64::nlanes; ++i)
dst[i % cn] += ar[i];
#else
double CV_DECL_ALIGNED(CV_SIMD_WIDTH) ar[2 * v_float64::nlanes];
v_store_aligned(ar, v_sum0);
v_store_aligned(ar + v_float64::nlanes, v_sum1);
for (int i = 0; i < 2 * v_float64::nlanes; ++i)
dst[i % cn] += ar[i];
#endif
v_cleanup();
return x / cn;
}
};
template <>
struct Sum_SIMD<float, double>
{
int operator () (const float * src0, const uchar * mask, double * dst, int len, int cn) const
{
if (mask || (cn != 1 && cn != 2 && cn != 4))
return 0;
len *= cn;
int x = 0;
v_float64 v_sum0 = vx_setzero_f64();
v_float64 v_sum1 = vx_setzero_f64();
for (; x <= len - 2 * v_float32::nlanes; x += 2 * v_float32::nlanes)
{
v_float32 v_src0 = vx_load(src0 + x);
v_float32 v_src1 = vx_load(src0 + x + v_float32::nlanes);
v_sum0 += v_cvt_f64(v_src0) + v_cvt_f64(v_src1);
v_sum1 += v_cvt_f64_high(v_src0) + v_cvt_f64_high(v_src1);
}
#if CV_SIMD256 || CV_SIMD512
double CV_DECL_ALIGNED(CV_SIMD_WIDTH) ar[v_float64::nlanes];
v_store_aligned(ar, v_sum0 + v_sum1);
for (int i = 0; i < v_float64::nlanes; ++i)
dst[i % cn] += ar[i];
#else
double CV_DECL_ALIGNED(CV_SIMD_WIDTH) ar[2 * v_float64::nlanes];
v_store_aligned(ar, v_sum0);
v_store_aligned(ar + v_float64::nlanes, v_sum1);
for (int i = 0; i < 2 * v_float64::nlanes; ++i)
dst[i % cn] += ar[i];
#endif
v_cleanup();
return x / cn;
}
};
#endif
#endif
template<typename T, typename ST>
static int sum_(const T* src0, const uchar* mask, ST* dst, int len, int cn )
{
const T* src = src0;
if( !mask )
{
Sum_SIMD<T, ST> vop;
int i = vop(src0, mask, dst, len, cn), k = cn % 4;
src += i * cn;
if( k == 1 )
{
ST s0 = dst[0];
#if CV_ENABLE_UNROLLED
for(; i <= len - 4; i += 4, src += cn*4 )
s0 += src[0] + src[cn] + src[cn*2] + src[cn*3];
#endif
for( ; i < len; i++, src += cn )
s0 += src[0];
dst[0] = s0;
}
else if( k == 2 )
{
ST s0 = dst[0], s1 = dst[1];
for( ; i < len; i++, src += cn )
{
s0 += src[0];
s1 += src[1];
}
dst[0] = s0;
dst[1] = s1;
}
else if( k == 3 )
{
ST s0 = dst[0], s1 = dst[1], s2 = dst[2];
for( ; i < len; i++, src += cn )
{
s0 += src[0];
s1 += src[1];
s2 += src[2];
}
dst[0] = s0;
dst[1] = s1;
dst[2] = s2;
}
for( ; k < cn; k += 4 )
{
src = src0 + i*cn + k;
ST s0 = dst[k], s1 = dst[k+1], s2 = dst[k+2], s3 = dst[k+3];
for( ; i < len; i++, src += cn )
{
s0 += src[0]; s1 += src[1];
s2 += src[2]; s3 += src[3];
}
dst[k] = s0;
dst[k+1] = s1;
dst[k+2] = s2;
dst[k+3] = s3;
}
return len;
}
int i, nzm = 0;
if( cn == 1 )
{
ST s = dst[0];
for( i = 0; i < len; i++ )
if( mask[i] )
{
s += src[i];
nzm++;
}
dst[0] = s;
}
else if( cn == 3 )
{
ST s0 = dst[0], s1 = dst[1], s2 = dst[2];
for( i = 0; i < len; i++, src += 3 )
if( mask[i] )
{
s0 += src[0];
s1 += src[1];
s2 += src[2];
nzm++;
}
dst[0] = s0;
dst[1] = s1;
dst[2] = s2;
}
else
{
for( i = 0; i < len; i++, src += cn )
if( mask[i] )
{
int k = 0;
#if CV_ENABLE_UNROLLED
for( ; k <= cn - 4; k += 4 )
{
ST s0, s1;
s0 = dst[k] + src[k];
s1 = dst[k+1] + src[k+1];
dst[k] = s0; dst[k+1] = s1;
s0 = dst[k+2] + src[k+2];
s1 = dst[k+3] + src[k+3];
dst[k+2] = s0; dst[k+3] = s1;
}
#endif
for( ; k < cn; k++ )
dst[k] += src[k];
nzm++;
}
}
return nzm;
}
static int sum8u( const uchar* src, const uchar* mask, int* dst, int len, int cn )
{ CV_INSTRUMENT_REGION(); return sum_(src, mask, dst, len, cn); }
static int sum8s( const schar* src, const uchar* mask, int* dst, int len, int cn )
{ CV_INSTRUMENT_REGION(); return sum_(src, mask, dst, len, cn); }
static int sum16u( const ushort* src, const uchar* mask, int* dst, int len, int cn )
{ CV_INSTRUMENT_REGION(); return sum_(src, mask, dst, len, cn); }
static int sum16s( const short* src, const uchar* mask, int* dst, int len, int cn )
{ CV_INSTRUMENT_REGION(); return sum_(src, mask, dst, len, cn); }
static int sum32s( const int* src, const uchar* mask, double* dst, int len, int cn )
{ CV_INSTRUMENT_REGION(); return sum_(src, mask, dst, len, cn); }
static int sum32f( const float* src, const uchar* mask, double* dst, int len, int cn )
{ CV_INSTRUMENT_REGION(); return sum_(src, mask, dst, len, cn); }
static int sum64f( const double* src, const uchar* mask, double* dst, int len, int cn )
{ CV_INSTRUMENT_REGION(); return sum_(src, mask, dst, len, cn); }
SumFunc getSumFunc(int depth)
{
static SumFunc sumTab[] =
{
(SumFunc)GET_OPTIMIZED(sum8u), (SumFunc)sum8s,
(SumFunc)sum16u, (SumFunc)sum16s,
(SumFunc)sum32s,
(SumFunc)GET_OPTIMIZED(sum32f), (SumFunc)sum64f,
0
};
return sumTab[depth];
}
#endif
CV_CPU_OPTIMIZATION_NAMESPACE_END
} // namespace