-
Notifications
You must be signed in to change notification settings - Fork 13
/
Copy pathArrayArithmetic.h
515 lines (473 loc) · 16.1 KB
/
ArrayArithmetic.h
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
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
/*
* Copyright 2010,2011,2012 Reality Jockey, Ltd.
* http://rjdj.me/
*
* This file is part of ZenGarden.
*
* ZenGarden is free software: you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* ZenGarden is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public License
* along with ZenGarden. If not, see <http://www.gnu.org/licenses/>.
*
*/
#ifndef _ARRAY_ARITHMETIC_H_
#define _ARRAY_ARITHMETIC_H_
#if __APPLE__
// The Accelerate framework is a library of tuned vector operations
#include <Accelerate/Accelerate.h>
#endif
#if __SSE__
#include <xmmintrin.h>
#elif __ARM_NEON__
// __ARM_NEON__ is defined by the compiler if the arguments "-mfloat-abi=softfp -mfpu=neon" are passed.
#include <arm_neon.h>
#endif
/**
* This class offers static inline functions for computing basic arithmetic with float arrays.
* It offers a central place for optimised implementations of common compute-intensive operations.
* In all SSE cases, input vectors can be (16-byte) unaligned, but output vectors must be aligned.
*/
class ArrayArithmetic {
public:
static inline void add(float *input0, float *input1, float *output, int startIndex, int endIndex) {
#if __APPLE__
vDSP_vadd(input0+startIndex, 1, input1+startIndex, 1, output+startIndex, 1, endIndex-startIndex);
#elif __SSE__
input0 += startIndex;
input1 += startIndex;
output += startIndex;
int n = endIndex - startIndex;
// align buffer to 16-byte boundary
switch (startIndex & 0x3) {
case 0: default: break;
case 1: *output++ = *input0++ + *input1++; --n;
case 2: *output++ = *input0++ + *input1++; --n;
case 3: *output++ = *input0++ + *input1++; --n;
}
int n4 = n & 0xFFFFFFFC;
while (n4) {
_mm_store_ps(output, _mm_add_ps(_mm_load_ps(input0), _mm_load_ps(input1)));
n4 -= 4; input0 += 4; input1 += 4; output += 4;
}
switch (n & 0x3) {
case 3: *output++ = *input0++ + *input1++;
case 2: *output++ = *input0++ + *input1++;
case 1: *output++ = *input0++ + *input1++;
case 0: default: break;
}
#elif __ARM_NEON__
input0 += startIndex;
input1 += startIndex;
output += startIndex;
int n = endIndex - startIndex;
int n4 = n & 0xFFFFFFFC;
float32x4_t inVec0, inVec1, res;
while (n4) {
inVec0 = vld1q_f32((const float32_t *) input0);
inVec1 = vld1q_f32((const float32_t *) input1);
res = vaddq_f32(inVec0, inVec1);
vst1q_f32((float32_t *) output, res);
n4 -= 4;
input0 += 4;
input1 += 4;
output += 4;
}
switch (n & 0x3) {
case 3: *output++ = *input0++ + *input1++;
case 2: *output++ = *input0++ + *input1++;
case 1: *output++ = *input0++ + *input1++;
default: break;
}
#else
for (int i = startIndex; i < endIndex; i++) {
output[i] = input0[i] + input1[i];
}
#endif
}
static inline void add(float *input, float constant, float *output, int startIndex, int endIndex) {
#if __APPLE__
vDSP_vsadd(input+startIndex, 1, &constant, output+startIndex, 1, endIndex-startIndex);
#elif __SSE__
input += startIndex;
output += startIndex;
int n = endIndex - startIndex;
// align buffer to 16-byte boundary
switch (startIndex & 0x3) {
case 0: default: break;
case 1: *output++ += constant; --n;
case 2: *output++ += constant; --n;
case 3: *output++ += constant; --n;
}
int n4 = n & 0xFFFFFFFC;
const __m128 constVec = _mm_set1_ps(constant);
while (n4) {
_mm_store_ps(output, _mm_add_ps(_mm_load_ps(input), constVec));
n4 -= 4; input += 4; output += 4;
}
switch (n & 0x3) {
case 3: *output++ += constant;
case 2: *output++ += constant;
case 1: *output++ += constant;
case 0: default: break;
}
#elif __ARM_NEON__
input += startIndex;
output += startIndex;
int n = endIndex - startIndex;
int n4 = n & 0xFFFFFFFC;
float32x4_t inVec, res;
while (n4) {
inVec = vld1q_f32((const float32_t *) input);
res = vaddq_f32(inVec, constant);
vst1q_f32((float32_t *) output, res);
n4 -= 4;
input += 4;
output += 4;
}
switch (n & 0x3) {
case 3: *output++ += constant;
case 2: *output++ += constant;
case 1: *output++ += constant;
default: break;
}
#else
for (int i = startIndex; i < endIndex; i++) {
output[i] = input[i] + constant;
}
#endif
}
// output = input0 - input1
static inline void subtract(float *input0, float *input1, float *output, int startIndex, int endIndex) {
#if __APPLE__
vDSP_vsub(input1+startIndex, 1, input0+startIndex, 1, output+startIndex, 1, endIndex-startIndex);
#elif __SSE__
input0 += startIndex;
input1 += startIndex;
output += startIndex;
int n = endIndex - startIndex;
switch (startIndex & 0x3) {
case 0: default: break;
case 1: *output++ = *input0++ - *input1++; --n;
case 2: *output++ = *input0++ - *input1++; --n;
case 3: *output++ = *input0++ - *input1++; --n;
}
int n4 = n & 0xFFFFFFFC;
while (n4) {
_mm_store_ps(output, _mm_sub_ps(_mm_load_ps(input0), _mm_load_ps(input1)));
n4 -= 4; input0 += 4; input1 += 4; output += 4;
}
switch (n & 0x3) {
case 3: *output++ = *input0++ - *input1++;
case 2: *output++ = *input0++ - *input1++;
case 1: *output++ = *input0++ - *input1++;
case 0: default: break;
}
#elif __ARM_NEON__
input0 += startIndex;
input1 += startIndex;
output += startIndex;
int n = endIndex - startIndex;
int n4 = n & 0xFFFFFFFC;
float32x4_t inVec0, inVec1, res;
while (n4) {
inVec0 = vld1q_f32((const float32_t *) input0);
inVec1 = vld1q_f32((const float32_t *) input1);
res = vsubq_f32(inVec0, inVec1);
vst1q_f32((float32_t *) output, res);
n4 -= 4;
input0 += 4;
input1 += 4;
output += 4;
}
switch (n & 0x3) {
case 3: *output++ = *input0++ - *input1++;
case 2: *output++ = *input0++ - *input1++;
case 1: *output++ = *input0++ - *input1++;
default: break;
}
#else
for (int i = startIndex; i < endIndex; i++) {
output[i] = input0[i] - input1[i];
}
#endif
}
static inline void subtract(float *input, float constant, float *output, int startIndex, int endIndex) {
#if __APPLE__
float negation = -1.0f * constant;
vDSP_vsadd(input+startIndex, 1, &negation, output+startIndex, 1, endIndex-startIndex);
#elif __SSE__
input += startIndex;
output += startIndex;
int n = endIndex - startIndex;
switch (startIndex & 0x3) {
case 0: default: break;
case 1: *output++ -= constant; --n;
case 2: *output++ -= constant; --n;
case 3: *output++ -= constant; --n;
}
int n4 = n & 0xFFFFFFFC;
const __m128 constVec = _mm_set1_ps(constant);
while (n4) {
_mm_store_ps(output, _mm_sub_ps(_mm_load_ps(input), constVec));
n4 -= 4; input += 4; output += 4;
}
switch (n & 0x3) {
case 3: *output++ -= constant;
case 2: *output++ -= constant;
case 1: *output++ -= constant;
case 0: default: break;
}
#elif __ARM_NEON__
input += startIndex;
output += startIndex;
int n = endIndex - startIndex;
int n4 = n & 0xFFFFFFFC;
float32x4_t inVec, res;
while (n4) {
inVec = vld1q_f32((const float32_t *) input);
res = vsubq_f32(inVec, constant);
vst1q_f32((float32_t *) output, res);
n4 -= 4;
input += 4;
output += 4;
}
switch (n & 0x3) {
case 3: *output++ -= constant;
case 2: *output++ -= constant;
case 1: *output++ -= constant;
default: break;
}
#else
for (int i = startIndex; i < endIndex; i++) {
output[i] = input[i] - constant;
}
#endif
}
static inline void multiply(float *input0, float *input1, float *output, int startIndex, int endIndex) {
#if __APPLE__
vDSP_vmul(input0+startIndex, 1, input1+startIndex, 1, output+startIndex, 1, endIndex-startIndex);
#elif __SSE__
input0 += startIndex;
input1 += startIndex;
output += startIndex;
int n = endIndex - startIndex;
switch (startIndex & 0x3) {
case 0: default: break;
case 1: *output++ = *input0++ * *input1++; --n;
case 2: *output++ = *input0++ * *input1++; --n;
case 3: *output++ = *input0++ * *input1++; --n;
}
int n4 = n & 0xFFFFFFFC;
while (n4) {
_mm_store_ps(output, _mm_mul_ps(_mm_load_ps(input0), _mm_load_ps(input1)));
n4 -= 4; input0 += 4; input1 += 4; output += 4;
}
switch (n & 0x3) {
case 3: *output++ = *input0++ * *input1++;
case 2: *output++ = *input0++ * *input1++;
case 1: *output++ = *input0++ * *input1++;
case 0: default: break;
}
#elif __ARM_NEON__
input0 += startIndex;
input1 += startIndex;
output += startIndex;
int n = endIndex - startIndex;
int n4 = n & 0xFFFFFFFC;
float32x4_t inVec0, inVec1, res;
while (n4) {
inVec0 = vld1q_f32((const float32_t *) input0); // use VLD1 as data is NOT interleaved
inVec1 = vld1q_f32((const float32_t *) input1); // load
res = vmulq_f32(inVec0, inVec1); // compute
vst1q_f32((float32_t *) output, res); // store
n4 -= 4;
input0 += 4;
input1 += 4;
output += 4;
}
switch (n & 0x3) {
case 3: *output++ = *input0++ * *input1++;
case 2: *output++ = *input0++ * *input1++;
case 1: *output++ = *input0++ * *input1++;
default: break;
}
#else
for (int i = startIndex; i < endIndex; i++) {
output[i] = input0[i] * input1[i];
}
#endif
}
static inline void multiply(float *input, float constant, float *output, int startIndex, int endIndex) {
#if __APPLE__
vDSP_vsmul(input+startIndex, 1, &constant, output+startIndex, 1, endIndex-startIndex);
#elif __SSE__
input += startIndex;
output += startIndex;
int n = endIndex - startIndex;
switch (startIndex & 0x3) {
case 0: default: break;
case 1: *output++ *= constant; --n;
case 2: *output++ *= constant; --n;
case 3: *output++ *= constant; --n;
}
int n4 = n & 0xFFFFFFFC;
const __m128 constVec = _mm_set1_ps(constant);
while (n4) {
_mm_store_ps(output, _mm_mul_ps(_mm_load_ps(input), constVec));
n4 -= 4; input += 4; output += 4;
}
switch (n & 0x3) {
case 3: *output++ *= constant;
case 2: *output++ *= constant;
case 1: *output++ *= constant;
case 0: default: break;
}
#elif __ARM_NEON__
input += startIndex;
output += startIndex;
int n = endIndex - startIndex;
int n4 = n & 0xFFFFFFFC;
float32x4_t inVec, res;
while (n4) {
inVec = vld1q_f32((const float32_t *) input);
res = vmulq_n_f32(inVec, constant);
vst1q_f32((float32_t *) output, res);
n4 -= 4;
input += 4;
output += 4;
}
switch (n & 0x3) {
case 3: *output++ *= constant;
case 2: *output++ *= constant;
case 1: *output++ *= constant;
default: break;
}
#else
for (int i = startIndex; i < endIndex; i++) {
output[i] = input[i] * constant;
}
#endif
}
// recipocal: vrecpeq_f32
static inline void divide(float *input0, float *input1, float *output, int startIndex, int endIndex) {
#if __APPLE__
vDSP_vdiv(input1+startIndex, 1, input0+startIndex, 1, output+startIndex, 1, endIndex-startIndex);
#elif __SSE__
input0 += startIndex;
input1 += startIndex;
output += startIndex;
int n = endIndex - startIndex;
switch (startIndex & 0x3) {
case 0: default: break;
case 1: *output++ = *input0++ / *input1++; --n;
case 2: *output++ = *input0++ / *input1++; --n;
case 3: *output++ = *input0++ / *input1++; --n;
}
int n4 = n & 0xFFFFFFFC;
while (n4) {
_mm_store_ps(output, _mm_div_ps(_mm_load_ps(input0), _mm_load_ps(input1)));
n4 -= 4; input0 += 4; input1 += 4; output += 4;
}
switch (n & 0x3) {
case 3: *output++ = *input0++ / *input1++;
case 2: *output++ = *input0++ / *input1++;
case 1: *output++ = *input0++ / *input1++;
case 0: default: break;
}
#else
for (int i = startIndex; i < endIndex; i++) {
output[i] = input0[i] / input1[i];
}
#endif
}
static inline void divide(float *input, float constant, float *output, int startIndex, int endIndex) {
#if __APPLE__
vDSP_vsdiv(input+startIndex, 1, &constant, output+startIndex, 1, endIndex-startIndex);
#elif __SSE__
input += startIndex;
output += startIndex;
int n = endIndex - startIndex;
switch (startIndex & 0x3) {
case 0: default: break;
case 1: *output++ /= constant; --n;
case 2: *output++ /= constant; --n;
case 3: *output++ /= constant; --n;
}
int n4 = n & 0xFFFFFFFC;
const __m128 constVec = _mm_set1_ps(constant);
while (n4) {
_mm_store_ps(output, _mm_div_ps(_mm_load_ps(input), constVec));
n4 -= 4; input += 4; output += 4;
}
switch (n & 0x3) {
case 3: *output++ /= constant;
case 2: *output++ /= constant;
case 1: *output++ /= constant;
case 0: default: break;
}
#else
for (int i = startIndex; i < endIndex; i++) {
output[i] = input[i] / constant;
}
#endif
}
static inline void fill(float *input, float constant, int startIndex, int endIndex) {
#if __APPLE__
vDSP_vfill(&constant, input+startIndex, 1, endIndex-startIndex);
#elif __SSE__
input += startIndex;
int n = endIndex - startIndex;
switch (startIndex & 0x3) {
case 0: default: break;
case 1: *input++ = constant; --n;
case 2: *input++ = constant; --n;
case 3: *input++ = constant; --n;
}
int n4 = n & 0xFFFFFFFC; // force n to be a multiple of 4
const __m128 constVec = _mm_set1_ps(constant);
while (n4) {
_mm_store_ps(input, constVec);
n4 -= 4; input += 4;
}
switch (n & 0x3) {
case 3: *input++ = constant;
case 2: *input++ = constant;
case 1: *input++ = constant;
case 0: default: break;
}
#elif __ARM_NEON__
input += startIndex;
int n = endIndex - startIndex;
int n4 = n & 0xFFFFFFFC; // force n to be a multiple of 4
float32x4_t constVec = vdupq_n_f32(constant);
while (n4) {
vst1q_f32((float32_t *) input, constVec);
n4 -= 4;
input += 4;
}
switch (n & 0x3) {
case 3: *input++ = constant;
case 2: *input++ = constant;
case 1: *input++ = constant;
default: break;
}
#else
for (int i = startIndex; i < endIndex; i++) {
input[i] = constant;
}
#endif
}
private:
ArrayArithmetic(); // no instances of this object are allowed
~ArrayArithmetic();
};
#endif // _ARRAY_ARITHMETIC_H_