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
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
|
/*
* Copyright (c) 2015 The WebM project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <arm_neon.h>
#include <assert.h>
#include "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"
#include "vpx/vpx_integer.h"
#include "vpx_dsp/arm/mem_neon.h"
#include "vpx_dsp/arm/sum_neon.h"
static INLINE uint8x8_t load_unaligned_2_buffers(const void *const buf0,
const void *const buf1) {
uint32_t a;
uint32x2_t aa;
memcpy(&a, buf0, 4);
aa = vdup_n_u32(a);
memcpy(&a, buf1, 4);
aa = vset_lane_u32(a, aa, 1);
return vreinterpret_u8_u32(aa);
}
static INLINE void sad4x_4d(const uint8_t *const src_ptr, const int src_stride,
const uint8_t *const ref_array[4],
const int ref_stride, const int height,
uint32_t sad_array[4]) {
int i;
uint16x8_t abs[2] = { vdupq_n_u16(0), vdupq_n_u16(0) };
#if !defined(__aarch64__)
uint16x4_t a[2];
#endif
uint32x4_t r;
assert(!((intptr_t)src_ptr % sizeof(uint32_t)));
assert(!(src_stride % sizeof(uint32_t)));
for (i = 0; i < height; ++i) {
const uint8x8_t s = vreinterpret_u8_u32(
vld1_dup_u32((const uint32_t *)(src_ptr + i * src_stride)));
const uint8x8_t ref01 = load_unaligned_2_buffers(
ref_array[0] + i * ref_stride, ref_array[1] + i * ref_stride);
const uint8x8_t ref23 = load_unaligned_2_buffers(
ref_array[2] + i * ref_stride, ref_array[3] + i * ref_stride);
abs[0] = vabal_u8(abs[0], s, ref01);
abs[1] = vabal_u8(abs[1], s, ref23);
}
#if defined(__aarch64__)
abs[0] = vpaddq_u16(abs[0], abs[1]);
r = vpaddlq_u16(abs[0]);
#else
a[0] = vpadd_u16(vget_low_u16(abs[0]), vget_high_u16(abs[0]));
a[1] = vpadd_u16(vget_low_u16(abs[1]), vget_high_u16(abs[1]));
r = vpaddlq_u16(vcombine_u16(a[0], a[1]));
#endif
vst1q_u32(sad_array, r);
}
void vpx_sad4x4x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
sad4x_4d(src_ptr, src_stride, ref_array, ref_stride, 4, sad_array);
}
void vpx_sad4x8x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
sad4x_4d(src_ptr, src_stride, ref_array, ref_stride, 8, sad_array);
}
////////////////////////////////////////////////////////////////////////////////
// Can handle 512 pixels' sad sum (such as 16x32 or 32x16)
static INLINE void sad_512_pel_final_neon(const uint16x8_t sum[4],
uint32_t sad_array[4]) {
#if defined(__aarch64__)
const uint16x8_t a0 = vpaddq_u16(sum[0], sum[1]);
const uint16x8_t a1 = vpaddq_u16(sum[2], sum[3]);
const uint16x8_t b0 = vpaddq_u16(a0, a1);
const uint32x4_t r = vpaddlq_u16(b0);
#else
const uint16x4_t a0 = vadd_u16(vget_low_u16(sum[0]), vget_high_u16(sum[0]));
const uint16x4_t a1 = vadd_u16(vget_low_u16(sum[1]), vget_high_u16(sum[1]));
const uint16x4_t a2 = vadd_u16(vget_low_u16(sum[2]), vget_high_u16(sum[2]));
const uint16x4_t a3 = vadd_u16(vget_low_u16(sum[3]), vget_high_u16(sum[3]));
const uint16x4_t b0 = vpadd_u16(a0, a1);
const uint16x4_t b1 = vpadd_u16(a2, a3);
const uint32x4_t r = vpaddlq_u16(vcombine_u16(b0, b1));
#endif
vst1q_u32(sad_array, r);
}
#if defined(__arm__) || !defined(__ARM_FEATURE_DOTPROD)
// Can handle 1024 pixels' sad sum (such as 32x32)
static INLINE void sad_1024_pel_final_neon(const uint16x8_t sum[4],
uint32_t sad_array[4]) {
#if defined(__aarch64__)
const uint16x8_t a0 = vpaddq_u16(sum[0], sum[1]);
const uint16x8_t a1 = vpaddq_u16(sum[2], sum[3]);
const uint32x4_t b0 = vpaddlq_u16(a0);
const uint32x4_t b1 = vpaddlq_u16(a1);
const uint32x4_t r = vpaddq_u32(b0, b1);
vst1q_u32(sad_array, r);
#else
const uint16x4_t a0 = vpadd_u16(vget_low_u16(sum[0]), vget_high_u16(sum[0]));
const uint16x4_t a1 = vpadd_u16(vget_low_u16(sum[1]), vget_high_u16(sum[1]));
const uint16x4_t a2 = vpadd_u16(vget_low_u16(sum[2]), vget_high_u16(sum[2]));
const uint16x4_t a3 = vpadd_u16(vget_low_u16(sum[3]), vget_high_u16(sum[3]));
const uint32x4_t b0 = vpaddlq_u16(vcombine_u16(a0, a1));
const uint32x4_t b1 = vpaddlq_u16(vcombine_u16(a2, a3));
const uint32x2_t c0 = vpadd_u32(vget_low_u32(b0), vget_high_u32(b0));
const uint32x2_t c1 = vpadd_u32(vget_low_u32(b1), vget_high_u32(b1));
vst1q_u32(sad_array, vcombine_u32(c0, c1));
#endif
}
// Can handle 2048 pixels' sad sum (such as 32x64 or 64x32)
static INLINE void sad_2048_pel_final_neon(const uint16x8_t sum[4],
uint32_t sad_array[4]) {
#if defined(__aarch64__)
const uint32x4_t a0 = vpaddlq_u16(sum[0]);
const uint32x4_t a1 = vpaddlq_u16(sum[1]);
const uint32x4_t a2 = vpaddlq_u16(sum[2]);
const uint32x4_t a3 = vpaddlq_u16(sum[3]);
const uint32x4_t b0 = vpaddq_u32(a0, a1);
const uint32x4_t b1 = vpaddq_u32(a2, a3);
const uint32x4_t r = vpaddq_u32(b0, b1);
vst1q_u32(sad_array, r);
#else
const uint32x4_t a0 = vpaddlq_u16(sum[0]);
const uint32x4_t a1 = vpaddlq_u16(sum[1]);
const uint32x4_t a2 = vpaddlq_u16(sum[2]);
const uint32x4_t a3 = vpaddlq_u16(sum[3]);
const uint32x2_t b0 = vadd_u32(vget_low_u32(a0), vget_high_u32(a0));
const uint32x2_t b1 = vadd_u32(vget_low_u32(a1), vget_high_u32(a1));
const uint32x2_t b2 = vadd_u32(vget_low_u32(a2), vget_high_u32(a2));
const uint32x2_t b3 = vadd_u32(vget_low_u32(a3), vget_high_u32(a3));
const uint32x2_t c0 = vpadd_u32(b0, b1);
const uint32x2_t c1 = vpadd_u32(b2, b3);
vst1q_u32(sad_array, vcombine_u32(c0, c1));
#endif
}
// Can handle 4096 pixels' sad sum (such as 64x64)
static INLINE void sad_4096_pel_final_neon(const uint16x8_t sum[8],
uint32_t sad_array[4]) {
#if defined(__aarch64__)
const uint32x4_t a0 = vpaddlq_u16(sum[0]);
const uint32x4_t a1 = vpaddlq_u16(sum[1]);
const uint32x4_t a2 = vpaddlq_u16(sum[2]);
const uint32x4_t a3 = vpaddlq_u16(sum[3]);
const uint32x4_t a4 = vpaddlq_u16(sum[4]);
const uint32x4_t a5 = vpaddlq_u16(sum[5]);
const uint32x4_t a6 = vpaddlq_u16(sum[6]);
const uint32x4_t a7 = vpaddlq_u16(sum[7]);
const uint32x4_t b0 = vaddq_u32(a0, a1);
const uint32x4_t b1 = vaddq_u32(a2, a3);
const uint32x4_t b2 = vaddq_u32(a4, a5);
const uint32x4_t b3 = vaddq_u32(a6, a7);
const uint32x4_t c0 = vpaddq_u32(b0, b1);
const uint32x4_t c1 = vpaddq_u32(b2, b3);
const uint32x4_t r = vpaddq_u32(c0, c1);
vst1q_u32(sad_array, r);
#else
const uint32x4_t a0 = vpaddlq_u16(sum[0]);
const uint32x4_t a1 = vpaddlq_u16(sum[1]);
const uint32x4_t a2 = vpaddlq_u16(sum[2]);
const uint32x4_t a3 = vpaddlq_u16(sum[3]);
const uint32x4_t a4 = vpaddlq_u16(sum[4]);
const uint32x4_t a5 = vpaddlq_u16(sum[5]);
const uint32x4_t a6 = vpaddlq_u16(sum[6]);
const uint32x4_t a7 = vpaddlq_u16(sum[7]);
const uint32x4_t b0 = vaddq_u32(a0, a1);
const uint32x4_t b1 = vaddq_u32(a2, a3);
const uint32x4_t b2 = vaddq_u32(a4, a5);
const uint32x4_t b3 = vaddq_u32(a6, a7);
const uint32x2_t c0 = vadd_u32(vget_low_u32(b0), vget_high_u32(b0));
const uint32x2_t c1 = vadd_u32(vget_low_u32(b1), vget_high_u32(b1));
const uint32x2_t c2 = vadd_u32(vget_low_u32(b2), vget_high_u32(b2));
const uint32x2_t c3 = vadd_u32(vget_low_u32(b3), vget_high_u32(b3));
const uint32x2_t d0 = vpadd_u32(c0, c1);
const uint32x2_t d1 = vpadd_u32(c2, c3);
vst1q_u32(sad_array, vcombine_u32(d0, d1));
#endif
}
#endif
static INLINE void sad8x_4d(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4], const int height) {
int i, j;
const uint8_t *ref_loop[4] = { ref_array[0], ref_array[1], ref_array[2],
ref_array[3] };
uint16x8_t sum[4] = { vdupq_n_u16(0), vdupq_n_u16(0), vdupq_n_u16(0),
vdupq_n_u16(0) };
for (i = 0; i < height; ++i) {
const uint8x8_t s = vld1_u8(src_ptr);
src_ptr += src_stride;
for (j = 0; j < 4; ++j) {
const uint8x8_t b_u8 = vld1_u8(ref_loop[j]);
ref_loop[j] += ref_stride;
sum[j] = vabal_u8(sum[j], s, b_u8);
}
}
sad_512_pel_final_neon(sum, sad_array);
}
void vpx_sad8x4x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
sad8x_4d(src_ptr, src_stride, ref_array, ref_stride, sad_array, 4);
}
void vpx_sad8x8x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
sad8x_4d(src_ptr, src_stride, ref_array, ref_stride, sad_array, 8);
}
void vpx_sad8x16x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
sad8x_4d(src_ptr, src_stride, ref_array, ref_stride, sad_array, 16);
}
////////////////////////////////////////////////////////////////////////////////
#if defined(__aarch64__) && defined(__ARM_FEATURE_DOTPROD)
static INLINE void sad16_neon(const uint8_t *ref_ptr, const uint8x16_t src_ptr,
uint32x4_t *const sum) {
const uint8x16_t r = vld1q_u8(ref_ptr);
const uint8x16_t diff = vabdq_u8(src_ptr, r);
*sum = vdotq_u32(*sum, diff, vdupq_n_u8(1));
}
static INLINE void sad16x_4d(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4], const int height) {
int i;
uint32x4_t r0, r1;
const uint8_t *ref_loop[4] = { ref_array[0], ref_array[1], ref_array[2],
ref_array[3] };
uint32x4_t sum[4] = { vdupq_n_u32(0), vdupq_n_u32(0), vdupq_n_u32(0),
vdupq_n_u32(0) };
for (i = 0; i < height; ++i) {
const uint8x16_t s = vld1q_u8(src_ptr + i * src_stride);
sad16_neon(ref_loop[0] + i * ref_stride, s, &sum[0]);
sad16_neon(ref_loop[1] + i * ref_stride, s, &sum[1]);
sad16_neon(ref_loop[2] + i * ref_stride, s, &sum[2]);
sad16_neon(ref_loop[3] + i * ref_stride, s, &sum[3]);
}
r0 = vpaddq_u32(sum[0], sum[1]);
r1 = vpaddq_u32(sum[2], sum[3]);
vst1q_u32(sad_array, vpaddq_u32(r0, r1));
}
#else // !(defined(__aarch64__) && defined(__ARM_FEATURE_DOTPROD))
static INLINE void sad16_neon(const uint8_t *ref_ptr, const uint8x16_t src_ptr,
uint16x8_t *const sum) {
const uint8x16_t r = vld1q_u8(ref_ptr);
*sum = vabal_u8(*sum, vget_low_u8(src_ptr), vget_low_u8(r));
*sum = vabal_u8(*sum, vget_high_u8(src_ptr), vget_high_u8(r));
}
static INLINE void sad16x_4d(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4], const int height) {
int i;
const uint8_t *ref_loop[4] = { ref_array[0], ref_array[1], ref_array[2],
ref_array[3] };
uint16x8_t sum[4] = { vdupq_n_u16(0), vdupq_n_u16(0), vdupq_n_u16(0),
vdupq_n_u16(0) };
for (i = 0; i < height; ++i) {
const uint8x16_t s = vld1q_u8(src_ptr);
src_ptr += src_stride;
/* Manual unrolling here stops the compiler from getting confused. */
sad16_neon(ref_loop[0], s, &sum[0]);
ref_loop[0] += ref_stride;
sad16_neon(ref_loop[1], s, &sum[1]);
ref_loop[1] += ref_stride;
sad16_neon(ref_loop[2], s, &sum[2]);
ref_loop[2] += ref_stride;
sad16_neon(ref_loop[3], s, &sum[3]);
ref_loop[3] += ref_stride;
}
sad_512_pel_final_neon(sum, sad_array);
}
#endif // defined(__aarch64__) && defined(__ARM_FEATURE_DOTPROD)
void vpx_sad16x8x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
sad16x_4d(src_ptr, src_stride, ref_array, ref_stride, sad_array, 8);
}
void vpx_sad16x16x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
sad16x_4d(src_ptr, src_stride, ref_array, ref_stride, sad_array, 16);
}
void vpx_sad16x32x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
sad16x_4d(src_ptr, src_stride, ref_array, ref_stride, sad_array, 32);
}
////////////////////////////////////////////////////////////////////////////////
#if defined(__aarch64__) && defined(__ARM_FEATURE_DOTPROD)
static INLINE void sad32x_4d(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4], const int height) {
int i;
uint32x4_t r0, r1;
const uint8_t *ref_loop[4] = { ref_array[0], ref_array[1], ref_array[2],
ref_array[3] };
uint32x4_t sum[4] = { vdupq_n_u32(0), vdupq_n_u32(0), vdupq_n_u32(0),
vdupq_n_u32(0) };
for (i = 0; i < height; ++i) {
uint8x16_t s;
s = vld1q_u8(src_ptr + 0 * 16);
sad16_neon(ref_loop[0] + 0 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 0 * 16, s, &sum[1]);
sad16_neon(ref_loop[2] + 0 * 16, s, &sum[2]);
sad16_neon(ref_loop[3] + 0 * 16, s, &sum[3]);
s = vld1q_u8(src_ptr + 1 * 16);
sad16_neon(ref_loop[0] + 1 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 1 * 16, s, &sum[1]);
sad16_neon(ref_loop[2] + 1 * 16, s, &sum[2]);
sad16_neon(ref_loop[3] + 1 * 16, s, &sum[3]);
src_ptr += src_stride;
ref_loop[0] += ref_stride;
ref_loop[1] += ref_stride;
ref_loop[2] += ref_stride;
ref_loop[3] += ref_stride;
}
r0 = vpaddq_u32(sum[0], sum[1]);
r1 = vpaddq_u32(sum[2], sum[3]);
vst1q_u32(sad_array, vpaddq_u32(r0, r1));
}
void vpx_sad32x16x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
sad32x_4d(src_ptr, src_stride, ref_array, ref_stride, sad_array, 16);
}
void vpx_sad32x32x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
sad32x_4d(src_ptr, src_stride, ref_array, ref_stride, sad_array, 32);
}
void vpx_sad32x64x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
sad32x_4d(src_ptr, src_stride, ref_array, ref_stride, sad_array, 64);
}
#else // !(defined(__aarch64__) && defined(__ARM_FEATURE_DOTPROD))
static INLINE void sad32x_4d(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
const int height, uint16x8_t *const sum) {
int i;
const uint8_t *ref_loop[4] = { ref_array[0], ref_array[1], ref_array[2],
ref_array[3] };
sum[0] = sum[1] = sum[2] = sum[3] = vdupq_n_u16(0);
for (i = 0; i < height; ++i) {
uint8x16_t s;
s = vld1q_u8(src_ptr + 0 * 16);
sad16_neon(ref_loop[0] + 0 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 0 * 16, s, &sum[1]);
sad16_neon(ref_loop[2] + 0 * 16, s, &sum[2]);
sad16_neon(ref_loop[3] + 0 * 16, s, &sum[3]);
s = vld1q_u8(src_ptr + 1 * 16);
sad16_neon(ref_loop[0] + 1 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 1 * 16, s, &sum[1]);
sad16_neon(ref_loop[2] + 1 * 16, s, &sum[2]);
sad16_neon(ref_loop[3] + 1 * 16, s, &sum[3]);
src_ptr += src_stride;
ref_loop[0] += ref_stride;
ref_loop[1] += ref_stride;
ref_loop[2] += ref_stride;
ref_loop[3] += ref_stride;
}
}
void vpx_sad32x16x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
uint16x8_t sum[4];
sad32x_4d(src_ptr, src_stride, ref_array, ref_stride, 16, sum);
sad_512_pel_final_neon(sum, sad_array);
}
void vpx_sad32x32x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
uint16x8_t sum[4];
sad32x_4d(src_ptr, src_stride, ref_array, ref_stride, 32, sum);
sad_1024_pel_final_neon(sum, sad_array);
}
void vpx_sad32x64x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
uint16x8_t sum[4];
sad32x_4d(src_ptr, src_stride, ref_array, ref_stride, 64, sum);
sad_2048_pel_final_neon(sum, sad_array);
}
#endif // defined(__aarch64__) && defined(__ARM_FEATURE_DOTPROD)
////////////////////////////////////////////////////////////////////////////////
#if defined(__aarch64__) && defined(__ARM_FEATURE_DOTPROD)
void vpx_sad64x32x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
int i;
uint32x4_t r0, r1;
const uint8_t *ref_loop[4] = { ref_array[0], ref_array[1], ref_array[2],
ref_array[3] };
uint32x4_t sum[4] = { vdupq_n_u32(0), vdupq_n_u32(0), vdupq_n_u32(0),
vdupq_n_u32(0) };
for (i = 0; i < 32; ++i) {
uint8x16_t s;
s = vld1q_u8(src_ptr + 0 * 16);
sad16_neon(ref_loop[0] + 0 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 0 * 16, s, &sum[1]);
sad16_neon(ref_loop[2] + 0 * 16, s, &sum[2]);
sad16_neon(ref_loop[3] + 0 * 16, s, &sum[3]);
s = vld1q_u8(src_ptr + 1 * 16);
sad16_neon(ref_loop[0] + 1 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 1 * 16, s, &sum[1]);
sad16_neon(ref_loop[2] + 1 * 16, s, &sum[2]);
sad16_neon(ref_loop[3] + 1 * 16, s, &sum[3]);
s = vld1q_u8(src_ptr + 2 * 16);
sad16_neon(ref_loop[0] + 2 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 2 * 16, s, &sum[1]);
sad16_neon(ref_loop[2] + 2 * 16, s, &sum[2]);
sad16_neon(ref_loop[3] + 2 * 16, s, &sum[3]);
s = vld1q_u8(src_ptr + 3 * 16);
sad16_neon(ref_loop[0] + 3 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 3 * 16, s, &sum[1]);
sad16_neon(ref_loop[2] + 3 * 16, s, &sum[2]);
sad16_neon(ref_loop[3] + 3 * 16, s, &sum[3]);
src_ptr += src_stride;
ref_loop[0] += ref_stride;
ref_loop[1] += ref_stride;
ref_loop[2] += ref_stride;
ref_loop[3] += ref_stride;
}
r0 = vpaddq_u32(sum[0], sum[1]);
r1 = vpaddq_u32(sum[2], sum[3]);
vst1q_u32(sad_array, vpaddq_u32(r0, r1));
}
void vpx_sad64x64x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
int i;
uint32x4_t r0, r1, r2, r3;
const uint8_t *ref_loop[4] = { ref_array[0], ref_array[1], ref_array[2],
ref_array[3] };
uint32x4_t sum[8] = { vdupq_n_u32(0), vdupq_n_u32(0), vdupq_n_u32(0),
vdupq_n_u32(0), vdupq_n_u32(0), vdupq_n_u32(0),
vdupq_n_u32(0), vdupq_n_u32(0) };
for (i = 0; i < 64; ++i) {
uint8x16_t s;
s = vld1q_u8(src_ptr + 0 * 16);
sad16_neon(ref_loop[0] + 0 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 0 * 16, s, &sum[2]);
sad16_neon(ref_loop[2] + 0 * 16, s, &sum[4]);
sad16_neon(ref_loop[3] + 0 * 16, s, &sum[6]);
s = vld1q_u8(src_ptr + 1 * 16);
sad16_neon(ref_loop[0] + 1 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 1 * 16, s, &sum[2]);
sad16_neon(ref_loop[2] + 1 * 16, s, &sum[4]);
sad16_neon(ref_loop[3] + 1 * 16, s, &sum[6]);
s = vld1q_u8(src_ptr + 2 * 16);
sad16_neon(ref_loop[0] + 2 * 16, s, &sum[1]);
sad16_neon(ref_loop[1] + 2 * 16, s, &sum[3]);
sad16_neon(ref_loop[2] + 2 * 16, s, &sum[5]);
sad16_neon(ref_loop[3] + 2 * 16, s, &sum[7]);
s = vld1q_u8(src_ptr + 3 * 16);
sad16_neon(ref_loop[0] + 3 * 16, s, &sum[1]);
sad16_neon(ref_loop[1] + 3 * 16, s, &sum[3]);
sad16_neon(ref_loop[2] + 3 * 16, s, &sum[5]);
sad16_neon(ref_loop[3] + 3 * 16, s, &sum[7]);
src_ptr += src_stride;
ref_loop[0] += ref_stride;
ref_loop[1] += ref_stride;
ref_loop[2] += ref_stride;
ref_loop[3] += ref_stride;
}
r0 = vpaddq_u32(sum[0], sum[1]);
r1 = vpaddq_u32(sum[2], sum[3]);
r2 = vpaddq_u32(sum[4], sum[5]);
r3 = vpaddq_u32(sum[6], sum[7]);
r0 = vpaddq_u32(r0, r1);
r1 = vpaddq_u32(r2, r3);
vst1q_u32(sad_array, vpaddq_u32(r0, r1));
}
#else // !(defined(__aarch64__) && defined(__ARM_FEATURE_DOTPROD))
void vpx_sad64x32x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
int i;
const uint8_t *ref_loop[4] = { ref_array[0], ref_array[1], ref_array[2],
ref_array[3] };
uint16x8_t sum[4] = { vdupq_n_u16(0), vdupq_n_u16(0), vdupq_n_u16(0),
vdupq_n_u16(0) };
for (i = 0; i < 32; ++i) {
uint8x16_t s;
s = vld1q_u8(src_ptr + 0 * 16);
sad16_neon(ref_loop[0] + 0 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 0 * 16, s, &sum[1]);
sad16_neon(ref_loop[2] + 0 * 16, s, &sum[2]);
sad16_neon(ref_loop[3] + 0 * 16, s, &sum[3]);
s = vld1q_u8(src_ptr + 1 * 16);
sad16_neon(ref_loop[0] + 1 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 1 * 16, s, &sum[1]);
sad16_neon(ref_loop[2] + 1 * 16, s, &sum[2]);
sad16_neon(ref_loop[3] + 1 * 16, s, &sum[3]);
s = vld1q_u8(src_ptr + 2 * 16);
sad16_neon(ref_loop[0] + 2 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 2 * 16, s, &sum[1]);
sad16_neon(ref_loop[2] + 2 * 16, s, &sum[2]);
sad16_neon(ref_loop[3] + 2 * 16, s, &sum[3]);
s = vld1q_u8(src_ptr + 3 * 16);
sad16_neon(ref_loop[0] + 3 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 3 * 16, s, &sum[1]);
sad16_neon(ref_loop[2] + 3 * 16, s, &sum[2]);
sad16_neon(ref_loop[3] + 3 * 16, s, &sum[3]);
src_ptr += src_stride;
ref_loop[0] += ref_stride;
ref_loop[1] += ref_stride;
ref_loop[2] += ref_stride;
ref_loop[3] += ref_stride;
}
sad_2048_pel_final_neon(sum, sad_array);
}
void vpx_sad64x64x4d_neon(const uint8_t *src_ptr, int src_stride,
const uint8_t *const ref_array[4], int ref_stride,
uint32_t sad_array[4]) {
int i;
const uint8_t *ref_loop[4] = { ref_array[0], ref_array[1], ref_array[2],
ref_array[3] };
uint16x8_t sum[8] = { vdupq_n_u16(0), vdupq_n_u16(0), vdupq_n_u16(0),
vdupq_n_u16(0), vdupq_n_u16(0), vdupq_n_u16(0),
vdupq_n_u16(0), vdupq_n_u16(0) };
for (i = 0; i < 64; ++i) {
uint8x16_t s;
s = vld1q_u8(src_ptr + 0 * 16);
sad16_neon(ref_loop[0] + 0 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 0 * 16, s, &sum[2]);
sad16_neon(ref_loop[2] + 0 * 16, s, &sum[4]);
sad16_neon(ref_loop[3] + 0 * 16, s, &sum[6]);
s = vld1q_u8(src_ptr + 1 * 16);
sad16_neon(ref_loop[0] + 1 * 16, s, &sum[0]);
sad16_neon(ref_loop[1] + 1 * 16, s, &sum[2]);
sad16_neon(ref_loop[2] + 1 * 16, s, &sum[4]);
sad16_neon(ref_loop[3] + 1 * 16, s, &sum[6]);
s = vld1q_u8(src_ptr + 2 * 16);
sad16_neon(ref_loop[0] + 2 * 16, s, &sum[1]);
sad16_neon(ref_loop[1] + 2 * 16, s, &sum[3]);
sad16_neon(ref_loop[2] + 2 * 16, s, &sum[5]);
sad16_neon(ref_loop[3] + 2 * 16, s, &sum[7]);
s = vld1q_u8(src_ptr + 3 * 16);
sad16_neon(ref_loop[0] + 3 * 16, s, &sum[1]);
sad16_neon(ref_loop[1] + 3 * 16, s, &sum[3]);
sad16_neon(ref_loop[2] + 3 * 16, s, &sum[5]);
sad16_neon(ref_loop[3] + 3 * 16, s, &sum[7]);
src_ptr += src_stride;
ref_loop[0] += ref_stride;
ref_loop[1] += ref_stride;
ref_loop[2] += ref_stride;
ref_loop[3] += ref_stride;
}
sad_4096_pel_final_neon(sum, sad_array);
}
#endif // defined(__aarch64__) && defined(__ARM_FEATURE_DOTPROD)
|
