summaryrefslogtreecommitdiffstats
path: root/media/libwebp/dsp/lossless_neon.c
blob: 1145d5fad0f0791162a734ffeacea49498e5a993 (plain)
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
// Copyright 2014 Google Inc. All Rights Reserved.
//
// Use of this source code is governed by a BSD-style license
// that can be found in the COPYING 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.
// -----------------------------------------------------------------------------
//
// NEON variant of methods for lossless decoder
//
// Author: Skal (pascal.massimino@gmail.com)

#include "./dsp.h"

#if defined(WEBP_USE_NEON)

#include <arm_neon.h>

#include "./lossless.h"
#include "./neon.h"

//------------------------------------------------------------------------------
// Colorspace conversion functions

#if !defined(WORK_AROUND_GCC)
// gcc 4.6.0 had some trouble (NDK-r9) with this code. We only use it for
// gcc-4.8.x at least.
static void ConvertBGRAToRGBA(const uint32_t* src,
                              int num_pixels, uint8_t* dst) {
  const uint32_t* const end = src + (num_pixels & ~15);
  for (; src < end; src += 16) {
    uint8x16x4_t pixel = vld4q_u8((uint8_t*)src);
    // swap B and R. (VSWP d0,d2 has no intrinsics equivalent!)
    const uint8x16_t tmp = pixel.val[0];
    pixel.val[0] = pixel.val[2];
    pixel.val[2] = tmp;
    vst4q_u8(dst, pixel);
    dst += 64;
  }
  VP8LConvertBGRAToRGBA_C(src, num_pixels & 15, dst);  // left-overs
}

static void ConvertBGRAToBGR(const uint32_t* src,
                             int num_pixels, uint8_t* dst) {
  const uint32_t* const end = src + (num_pixels & ~15);
  for (; src < end; src += 16) {
    const uint8x16x4_t pixel = vld4q_u8((uint8_t*)src);
    const uint8x16x3_t tmp = { { pixel.val[0], pixel.val[1], pixel.val[2] } };
    vst3q_u8(dst, tmp);
    dst += 48;
  }
  VP8LConvertBGRAToBGR_C(src, num_pixels & 15, dst);  // left-overs
}

static void ConvertBGRAToRGB(const uint32_t* src,
                             int num_pixels, uint8_t* dst) {
  const uint32_t* const end = src + (num_pixels & ~15);
  for (; src < end; src += 16) {
    const uint8x16x4_t pixel = vld4q_u8((uint8_t*)src);
    const uint8x16x3_t tmp = { { pixel.val[2], pixel.val[1], pixel.val[0] } };
    vst3q_u8(dst, tmp);
    dst += 48;
  }
  VP8LConvertBGRAToRGB_C(src, num_pixels & 15, dst);  // left-overs
}

#else  // WORK_AROUND_GCC

// gcc-4.6.0 fallback

static const uint8_t kRGBAShuffle[8] = { 2, 1, 0, 3, 6, 5, 4, 7 };

static void ConvertBGRAToRGBA(const uint32_t* src,
                              int num_pixels, uint8_t* dst) {
  const uint32_t* const end = src + (num_pixels & ~1);
  const uint8x8_t shuffle = vld1_u8(kRGBAShuffle);
  for (; src < end; src += 2) {
    const uint8x8_t pixels = vld1_u8((uint8_t*)src);
    vst1_u8(dst, vtbl1_u8(pixels, shuffle));
    dst += 8;
  }
  VP8LConvertBGRAToRGBA_C(src, num_pixels & 1, dst);  // left-overs
}

static const uint8_t kBGRShuffle[3][8] = {
  {  0,  1,  2,  4,  5,  6,  8,  9 },
  { 10, 12, 13, 14, 16, 17, 18, 20 },
  { 21, 22, 24, 25, 26, 28, 29, 30 }
};

static void ConvertBGRAToBGR(const uint32_t* src,
                             int num_pixels, uint8_t* dst) {
  const uint32_t* const end = src + (num_pixels & ~7);
  const uint8x8_t shuffle0 = vld1_u8(kBGRShuffle[0]);
  const uint8x8_t shuffle1 = vld1_u8(kBGRShuffle[1]);
  const uint8x8_t shuffle2 = vld1_u8(kBGRShuffle[2]);
  for (; src < end; src += 8) {
    uint8x8x4_t pixels;
    INIT_VECTOR4(pixels,
                 vld1_u8((const uint8_t*)(src + 0)),
                 vld1_u8((const uint8_t*)(src + 2)),
                 vld1_u8((const uint8_t*)(src + 4)),
                 vld1_u8((const uint8_t*)(src + 6)));
    vst1_u8(dst +  0, vtbl4_u8(pixels, shuffle0));
    vst1_u8(dst +  8, vtbl4_u8(pixels, shuffle1));
    vst1_u8(dst + 16, vtbl4_u8(pixels, shuffle2));
    dst += 8 * 3;
  }
  VP8LConvertBGRAToBGR_C(src, num_pixels & 7, dst);  // left-overs
}

static const uint8_t kRGBShuffle[3][8] = {
  {  2,  1,  0,  6,  5,  4, 10,  9 },
  {  8, 14, 13, 12, 18, 17, 16, 22 },
  { 21, 20, 26, 25, 24, 30, 29, 28 }
};

static void ConvertBGRAToRGB(const uint32_t* src,
                             int num_pixels, uint8_t* dst) {
  const uint32_t* const end = src + (num_pixels & ~7);
  const uint8x8_t shuffle0 = vld1_u8(kRGBShuffle[0]);
  const uint8x8_t shuffle1 = vld1_u8(kRGBShuffle[1]);
  const uint8x8_t shuffle2 = vld1_u8(kRGBShuffle[2]);
  for (; src < end; src += 8) {
    uint8x8x4_t pixels;
    INIT_VECTOR4(pixels,
                 vld1_u8((const uint8_t*)(src + 0)),
                 vld1_u8((const uint8_t*)(src + 2)),
                 vld1_u8((const uint8_t*)(src + 4)),
                 vld1_u8((const uint8_t*)(src + 6)));
    vst1_u8(dst +  0, vtbl4_u8(pixels, shuffle0));
    vst1_u8(dst +  8, vtbl4_u8(pixels, shuffle1));
    vst1_u8(dst + 16, vtbl4_u8(pixels, shuffle2));
    dst += 8 * 3;
  }
  VP8LConvertBGRAToRGB_C(src, num_pixels & 7, dst);  // left-overs
}

#endif   // !WORK_AROUND_GCC


//------------------------------------------------------------------------------
// Predictor Transform

#define LOAD_U32_AS_U8(IN) vreinterpret_u8_u32(vdup_n_u32((IN)))
#define LOAD_U32P_AS_U8(IN) vreinterpret_u8_u32(vld1_u32((IN)))
#define LOADQ_U32_AS_U8(IN) vreinterpretq_u8_u32(vdupq_n_u32((IN)))
#define LOADQ_U32P_AS_U8(IN) vreinterpretq_u8_u32(vld1q_u32((IN)))
#define GET_U8_AS_U32(IN) vget_lane_u32(vreinterpret_u32_u8((IN)), 0);
#define GETQ_U8_AS_U32(IN) vgetq_lane_u32(vreinterpretq_u32_u8((IN)), 0);
#define STOREQ_U8_AS_U32P(OUT, IN) vst1q_u32((OUT), vreinterpretq_u32_u8((IN)));
#define ROTATE32_LEFT(L) vextq_u8((L), (L), 12)    // D|C|B|A -> C|B|A|D

static WEBP_INLINE uint8x8_t Average2_u8_NEON(uint32_t a0, uint32_t a1) {
  const uint8x8_t A0 = LOAD_U32_AS_U8(a0);
  const uint8x8_t A1 = LOAD_U32_AS_U8(a1);
  return vhadd_u8(A0, A1);
}

static WEBP_INLINE uint32_t ClampedAddSubtractHalf_NEON(uint32_t c0,
                                                        uint32_t c1,
                                                        uint32_t c2) {
  const uint8x8_t avg = Average2_u8_NEON(c0, c1);
  // Remove one to c2 when bigger than avg.
  const uint8x8_t C2 = LOAD_U32_AS_U8(c2);
  const uint8x8_t cmp = vcgt_u8(C2, avg);
  const uint8x8_t C2_1 = vadd_u8(C2, cmp);
  // Compute half of the difference between avg and c2.
  const int8x8_t diff_avg = vreinterpret_s8_u8(vhsub_u8(avg, C2_1));
  // Compute the sum with avg and saturate.
  const int16x8_t avg_16 = vreinterpretq_s16_u16(vmovl_u8(avg));
  const uint8x8_t res = vqmovun_s16(vaddw_s8(avg_16, diff_avg));
  const uint32_t output = GET_U8_AS_U32(res);
  return output;
}

static WEBP_INLINE uint32_t Average2_NEON(uint32_t a0, uint32_t a1) {
  const uint8x8_t avg_u8x8 = Average2_u8_NEON(a0, a1);
  const uint32_t avg = GET_U8_AS_U32(avg_u8x8);
  return avg;
}

static WEBP_INLINE uint32_t Average3_NEON(uint32_t a0, uint32_t a1,
                                          uint32_t a2) {
  const uint8x8_t avg0 = Average2_u8_NEON(a0, a2);
  const uint8x8_t A1 = LOAD_U32_AS_U8(a1);
  const uint32_t avg = GET_U8_AS_U32(vhadd_u8(avg0, A1));
  return avg;
}

static uint32_t Predictor5_NEON(uint32_t left, const uint32_t* const top) {
  return Average3_NEON(left, top[0], top[1]);
}
static uint32_t Predictor6_NEON(uint32_t left, const uint32_t* const top) {
  return Average2_NEON(left, top[-1]);
}
static uint32_t Predictor7_NEON(uint32_t left, const uint32_t* const top) {
  return Average2_NEON(left, top[0]);
}
static uint32_t Predictor13_NEON(uint32_t left, const uint32_t* const top) {
  return ClampedAddSubtractHalf_NEON(left, top[0], top[-1]);
}

// Batch versions of those functions.

// Predictor0: ARGB_BLACK.
static void PredictorAdd0_NEON(const uint32_t* in, const uint32_t* upper,
                               int num_pixels, uint32_t* out) {
  int i;
  const uint8x16_t black = vreinterpretq_u8_u32(vdupq_n_u32(ARGB_BLACK));
  for (i = 0; i + 4 <= num_pixels; i += 4) {
    const uint8x16_t src = LOADQ_U32P_AS_U8(&in[i]);
    const uint8x16_t res = vaddq_u8(src, black);
    STOREQ_U8_AS_U32P(&out[i], res);
  }
  VP8LPredictorsAdd_C[0](in + i, upper + i, num_pixels - i, out + i);
}

// Predictor1: left.
static void PredictorAdd1_NEON(const uint32_t* in, const uint32_t* upper,
                               int num_pixels, uint32_t* out) {
  int i;
  const uint8x16_t zero = LOADQ_U32_AS_U8(0);
  for (i = 0; i + 4 <= num_pixels; i += 4) {
    // a | b | c | d
    const uint8x16_t src = LOADQ_U32P_AS_U8(&in[i]);
    // 0 | a | b | c
    const uint8x16_t shift0 = vextq_u8(zero, src, 12);
    // a | a + b | b + c | c + d
    const uint8x16_t sum0 = vaddq_u8(src, shift0);
    // 0 | 0 | a | a + b
    const uint8x16_t shift1 = vextq_u8(zero, sum0, 8);
    // a | a + b | a + b + c | a + b + c + d
    const uint8x16_t sum1 = vaddq_u8(sum0, shift1);
    const uint8x16_t prev = LOADQ_U32_AS_U8(out[i - 1]);
    const uint8x16_t res = vaddq_u8(sum1, prev);
    STOREQ_U8_AS_U32P(&out[i], res);
  }
  VP8LPredictorsAdd_C[1](in + i, upper + i, num_pixels - i, out + i);
}

// Macro that adds 32-bit integers from IN using mod 256 arithmetic
// per 8 bit channel.
#define GENERATE_PREDICTOR_1(X, IN)                                       \
static void PredictorAdd##X##_NEON(const uint32_t* in,                    \
                                   const uint32_t* upper, int num_pixels, \
                                   uint32_t* out) {                       \
  int i;                                                                  \
  for (i = 0; i + 4 <= num_pixels; i += 4) {                              \
    const uint8x16_t src = LOADQ_U32P_AS_U8(&in[i]);                      \
    const uint8x16_t other = LOADQ_U32P_AS_U8(&(IN));                     \
    const uint8x16_t res = vaddq_u8(src, other);                          \
    STOREQ_U8_AS_U32P(&out[i], res);                                      \
  }                                                                       \
  VP8LPredictorsAdd_C[(X)](in + i, upper + i, num_pixels - i, out + i);   \
}
// Predictor2: Top.
GENERATE_PREDICTOR_1(2, upper[i])
// Predictor3: Top-right.
GENERATE_PREDICTOR_1(3, upper[i + 1])
// Predictor4: Top-left.
GENERATE_PREDICTOR_1(4, upper[i - 1])
#undef GENERATE_PREDICTOR_1

// Predictor5: average(average(left, TR), T)
#define DO_PRED5(LANE) do {                                              \
  const uint8x16_t avgLTR = vhaddq_u8(L, TR);                            \
  const uint8x16_t avg = vhaddq_u8(avgLTR, T);                           \
  const uint8x16_t res = vaddq_u8(avg, src);                             \
  vst1q_lane_u32(&out[i + (LANE)], vreinterpretq_u32_u8(res), (LANE));   \
  L = ROTATE32_LEFT(res);                                                \
} while (0)

static void PredictorAdd5_NEON(const uint32_t* in, const uint32_t* upper,
                               int num_pixels, uint32_t* out) {
  int i;
  uint8x16_t L = LOADQ_U32_AS_U8(out[-1]);
  for (i = 0; i + 4 <= num_pixels; i += 4) {
    const uint8x16_t src = LOADQ_U32P_AS_U8(&in[i]);
    const uint8x16_t T = LOADQ_U32P_AS_U8(&upper[i + 0]);
    const uint8x16_t TR = LOADQ_U32P_AS_U8(&upper[i + 1]);
    DO_PRED5(0);
    DO_PRED5(1);
    DO_PRED5(2);
    DO_PRED5(3);
  }
  VP8LPredictorsAdd_C[5](in + i, upper + i, num_pixels - i, out + i);
}
#undef DO_PRED5

#define DO_PRED67(LANE) do {                                             \
  const uint8x16_t avg = vhaddq_u8(L, top);                              \
  const uint8x16_t res = vaddq_u8(avg, src);                             \
  vst1q_lane_u32(&out[i + (LANE)], vreinterpretq_u32_u8(res), (LANE));   \
  L = ROTATE32_LEFT(res);                                                \
} while (0)

// Predictor6: average(left, TL)
static void PredictorAdd6_NEON(const uint32_t* in, const uint32_t* upper,
                               int num_pixels, uint32_t* out) {
  int i;
  uint8x16_t L = LOADQ_U32_AS_U8(out[-1]);
  for (i = 0; i + 4 <= num_pixels; i += 4) {
    const uint8x16_t src = LOADQ_U32P_AS_U8(&in[i]);
    const uint8x16_t top = LOADQ_U32P_AS_U8(&upper[i - 1]);
    DO_PRED67(0);
    DO_PRED67(1);
    DO_PRED67(2);
    DO_PRED67(3);
  }
  VP8LPredictorsAdd_C[6](in + i, upper + i, num_pixels - i, out + i);
}

// Predictor7: average(left, T)
static void PredictorAdd7_NEON(const uint32_t* in, const uint32_t* upper,
                               int num_pixels, uint32_t* out) {
  int i;
  uint8x16_t L = LOADQ_U32_AS_U8(out[-1]);
  for (i = 0; i + 4 <= num_pixels; i += 4) {
    const uint8x16_t src = LOADQ_U32P_AS_U8(&in[i]);
    const uint8x16_t top = LOADQ_U32P_AS_U8(&upper[i]);
    DO_PRED67(0);
    DO_PRED67(1);
    DO_PRED67(2);
    DO_PRED67(3);
  }
  VP8LPredictorsAdd_C[7](in + i, upper + i, num_pixels - i, out + i);
}
#undef DO_PRED67

#define GENERATE_PREDICTOR_2(X, IN)                                       \
static void PredictorAdd##X##_NEON(const uint32_t* in,                    \
                                   const uint32_t* upper, int num_pixels, \
                                   uint32_t* out) {                       \
  int i;                                                                  \
  for (i = 0; i + 4 <= num_pixels; i += 4) {                              \
    const uint8x16_t src = LOADQ_U32P_AS_U8(&in[i]);                      \
    const uint8x16_t Tother = LOADQ_U32P_AS_U8(&(IN));                    \
    const uint8x16_t T = LOADQ_U32P_AS_U8(&upper[i]);                     \
    const uint8x16_t avg = vhaddq_u8(T, Tother);                          \
    const uint8x16_t res = vaddq_u8(avg, src);                            \
    STOREQ_U8_AS_U32P(&out[i], res);                                      \
  }                                                                       \
  VP8LPredictorsAdd_C[(X)](in + i, upper + i, num_pixels - i, out + i);   \
}
// Predictor8: average TL T.
GENERATE_PREDICTOR_2(8, upper[i - 1])
// Predictor9: average T TR.
GENERATE_PREDICTOR_2(9, upper[i + 1])
#undef GENERATE_PREDICTOR_2

// Predictor10: average of (average of (L,TL), average of (T, TR)).
#define DO_PRED10(LANE) do {                                             \
  const uint8x16_t avgLTL = vhaddq_u8(L, TL);                            \
  const uint8x16_t avg = vhaddq_u8(avgTTR, avgLTL);                      \
  const uint8x16_t res = vaddq_u8(avg, src);                             \
  vst1q_lane_u32(&out[i + (LANE)], vreinterpretq_u32_u8(res), (LANE));   \
  L = ROTATE32_LEFT(res);                                                \
} while (0)

static void PredictorAdd10_NEON(const uint32_t* in, const uint32_t* upper,
                                int num_pixels, uint32_t* out) {
  int i;
  uint8x16_t L = LOADQ_U32_AS_U8(out[-1]);
  for (i = 0; i + 4 <= num_pixels; i += 4) {
    const uint8x16_t src = LOADQ_U32P_AS_U8(&in[i]);
    const uint8x16_t TL = LOADQ_U32P_AS_U8(&upper[i - 1]);
    const uint8x16_t T = LOADQ_U32P_AS_U8(&upper[i]);
    const uint8x16_t TR = LOADQ_U32P_AS_U8(&upper[i + 1]);
    const uint8x16_t avgTTR = vhaddq_u8(T, TR);
    DO_PRED10(0);
    DO_PRED10(1);
    DO_PRED10(2);
    DO_PRED10(3);
  }
  VP8LPredictorsAdd_C[10](in + i, upper + i, num_pixels - i, out + i);
}
#undef DO_PRED10

// Predictor11: select.
#define DO_PRED11(LANE) do {                                                   \
  const uint8x16_t sumLin = vaddq_u8(L, src);  /* in + L */                    \
  const uint8x16_t pLTL = vabdq_u8(L, TL);  /* |L - TL| */                     \
  const uint16x8_t sum_LTL = vpaddlq_u8(pLTL);                                 \
  const uint32x4_t pa = vpaddlq_u16(sum_LTL);                                  \
  const uint32x4_t mask = vcleq_u32(pa, pb);                                   \
  const uint8x16_t res = vbslq_u8(vreinterpretq_u8_u32(mask), sumTin, sumLin); \
  vst1q_lane_u32(&out[i + (LANE)], vreinterpretq_u32_u8(res), (LANE));         \
  L = ROTATE32_LEFT(res);                                                      \
} while (0)

static void PredictorAdd11_NEON(const uint32_t* in, const uint32_t* upper,
                                int num_pixels, uint32_t* out) {
  int i;
  uint8x16_t L = LOADQ_U32_AS_U8(out[-1]);
  for (i = 0; i + 4 <= num_pixels; i += 4) {
    const uint8x16_t T = LOADQ_U32P_AS_U8(&upper[i]);
    const uint8x16_t TL = LOADQ_U32P_AS_U8(&upper[i - 1]);
    const uint8x16_t pTTL = vabdq_u8(T, TL);   // |T - TL|
    const uint16x8_t sum_TTL = vpaddlq_u8(pTTL);
    const uint32x4_t pb = vpaddlq_u16(sum_TTL);
    const uint8x16_t src = LOADQ_U32P_AS_U8(&in[i]);
    const uint8x16_t sumTin = vaddq_u8(T, src);   // in + T
    DO_PRED11(0);
    DO_PRED11(1);
    DO_PRED11(2);
    DO_PRED11(3);
  }
  VP8LPredictorsAdd_C[11](in + i, upper + i, num_pixels - i, out + i);
}
#undef DO_PRED11

// Predictor12: ClampedAddSubtractFull.
#define DO_PRED12(DIFF, LANE) do {                                       \
  const uint8x8_t pred =                                                 \
      vqmovun_s16(vaddq_s16(vreinterpretq_s16_u16(L), (DIFF)));          \
  const uint8x8_t res =                                                  \
      vadd_u8(pred, (LANE <= 1) ? vget_low_u8(src) : vget_high_u8(src)); \
  const uint16x8_t res16 = vmovl_u8(res);                                \
  vst1_lane_u32(&out[i + (LANE)], vreinterpret_u32_u8(res), (LANE) & 1); \
  /* rotate in the left predictor for next iteration */                  \
  L = vextq_u16(res16, res16, 4);                                        \
} while (0)

static void PredictorAdd12_NEON(const uint32_t* in, const uint32_t* upper,
                                int num_pixels, uint32_t* out) {
  int i;
  uint16x8_t L = vmovl_u8(LOAD_U32_AS_U8(out[-1]));
  for (i = 0; i + 4 <= num_pixels; i += 4) {
    // load four pixels of source
    const uint8x16_t src = LOADQ_U32P_AS_U8(&in[i]);
    // precompute the difference T - TL once for all, stored as s16
    const uint8x16_t TL = LOADQ_U32P_AS_U8(&upper[i - 1]);
    const uint8x16_t T = LOADQ_U32P_AS_U8(&upper[i]);
    const int16x8_t diff_lo =
        vreinterpretq_s16_u16(vsubl_u8(vget_low_u8(T), vget_low_u8(TL)));
    const int16x8_t diff_hi =
        vreinterpretq_s16_u16(vsubl_u8(vget_high_u8(T), vget_high_u8(TL)));
    // loop over the four reconstructed pixels
    DO_PRED12(diff_lo, 0);
    DO_PRED12(diff_lo, 1);
    DO_PRED12(diff_hi, 2);
    DO_PRED12(diff_hi, 3);
  }
  VP8LPredictorsAdd_C[12](in + i, upper + i, num_pixels - i, out + i);
}
#undef DO_PRED12

// Predictor13: ClampedAddSubtractHalf
#define DO_PRED13(LANE, LOW_OR_HI) do {                                        \
  const uint8x16_t avg = vhaddq_u8(L, T);                                      \
  const uint8x16_t cmp = vcgtq_u8(TL, avg);                                    \
  const uint8x16_t TL_1 = vaddq_u8(TL, cmp);                                   \
  /* Compute half of the difference between avg and TL'. */                    \
  const int8x8_t diff_avg =                                                    \
      vreinterpret_s8_u8(LOW_OR_HI(vhsubq_u8(avg, TL_1)));                     \
  /* Compute the sum with avg and saturate. */                                 \
  const int16x8_t avg_16 = vreinterpretq_s16_u16(vmovl_u8(LOW_OR_HI(avg)));    \
  const uint8x8_t delta = vqmovun_s16(vaddw_s8(avg_16, diff_avg));             \
  const uint8x8_t res = vadd_u8(LOW_OR_HI(src), delta);                        \
  const uint8x16_t res2 = vcombine_u8(res, res);                               \
  vst1_lane_u32(&out[i + (LANE)], vreinterpret_u32_u8(res), (LANE) & 1);       \
  L = ROTATE32_LEFT(res2);                                                     \
} while (0)

static void PredictorAdd13_NEON(const uint32_t* in, const uint32_t* upper,
                                int num_pixels, uint32_t* out) {
  int i;
  uint8x16_t L = LOADQ_U32_AS_U8(out[-1]);
  for (i = 0; i + 4 <= num_pixels; i += 4) {
    const uint8x16_t src = LOADQ_U32P_AS_U8(&in[i]);
    const uint8x16_t T = LOADQ_U32P_AS_U8(&upper[i]);
    const uint8x16_t TL = LOADQ_U32P_AS_U8(&upper[i - 1]);
    DO_PRED13(0, vget_low_u8);
    DO_PRED13(1, vget_low_u8);
    DO_PRED13(2, vget_high_u8);
    DO_PRED13(3, vget_high_u8);
  }
  VP8LPredictorsAdd_C[13](in + i, upper + i, num_pixels - i, out + i);
}
#undef DO_PRED13

#undef LOAD_U32_AS_U8
#undef LOAD_U32P_AS_U8
#undef LOADQ_U32_AS_U8
#undef LOADQ_U32P_AS_U8
#undef GET_U8_AS_U32
#undef GETQ_U8_AS_U32
#undef STOREQ_U8_AS_U32P
#undef ROTATE32_LEFT

//------------------------------------------------------------------------------
// Subtract-Green Transform

// vtbl?_u8 are marked unavailable for iOS arm64 with Xcode < 6.3, use
// non-standard versions there.
#if defined(__APPLE__) && defined(__aarch64__) && \
    defined(__apple_build_version__) && (__apple_build_version__< 6020037)
#define USE_VTBLQ
#endif

#ifdef USE_VTBLQ
// 255 = byte will be zeroed
static const uint8_t kGreenShuffle[16] = {
  1, 255, 1, 255, 5, 255, 5, 255, 9, 255, 9, 255, 13, 255, 13, 255
};

static WEBP_INLINE uint8x16_t DoGreenShuffle(const uint8x16_t argb,
                                             const uint8x16_t shuffle) {
  return vcombine_u8(vtbl1q_u8(argb, vget_low_u8(shuffle)),
                     vtbl1q_u8(argb, vget_high_u8(shuffle)));
}
#else  // !USE_VTBLQ
// 255 = byte will be zeroed
static const uint8_t kGreenShuffle[8] = { 1, 255, 1, 255, 5, 255, 5, 255  };

static WEBP_INLINE uint8x16_t DoGreenShuffle(const uint8x16_t argb,
                                             const uint8x8_t shuffle) {
  return vcombine_u8(vtbl1_u8(vget_low_u8(argb), shuffle),
                     vtbl1_u8(vget_high_u8(argb), shuffle));
}
#endif  // USE_VTBLQ

static void AddGreenToBlueAndRed(const uint32_t* src, int num_pixels,
                                 uint32_t* dst) {
  const uint32_t* const end = src + (num_pixels & ~3);
#ifdef USE_VTBLQ
  const uint8x16_t shuffle = vld1q_u8(kGreenShuffle);
#else
  const uint8x8_t shuffle = vld1_u8(kGreenShuffle);
#endif
  for (; src < end; src += 4, dst += 4) {
    const uint8x16_t argb = vld1q_u8((const uint8_t*)src);
    const uint8x16_t greens = DoGreenShuffle(argb, shuffle);
    vst1q_u8((uint8_t*)dst, vaddq_u8(argb, greens));
  }
  // fallthrough and finish off with plain-C
  VP8LAddGreenToBlueAndRed_C(src, num_pixels & 3, dst);
}

//------------------------------------------------------------------------------
// Color Transform

static void TransformColorInverse(const VP8LMultipliers* const m,
                                  const uint32_t* const src, int num_pixels,
                                  uint32_t* dst) {
// sign-extended multiplying constants, pre-shifted by 6.
#define CST(X)  (((int16_t)(m->X << 8)) >> 6)
  const int16_t rb[8] = {
    CST(green_to_blue_), CST(green_to_red_),
    CST(green_to_blue_), CST(green_to_red_),
    CST(green_to_blue_), CST(green_to_red_),
    CST(green_to_blue_), CST(green_to_red_)
  };
  const int16x8_t mults_rb = vld1q_s16(rb);
  const int16_t b2[8] = {
    0, CST(red_to_blue_), 0, CST(red_to_blue_),
    0, CST(red_to_blue_), 0, CST(red_to_blue_),
  };
  const int16x8_t mults_b2 = vld1q_s16(b2);
#undef CST
#ifdef USE_VTBLQ
  static const uint8_t kg0g0[16] = {
    255, 1, 255, 1, 255, 5, 255, 5, 255, 9, 255, 9, 255, 13, 255, 13
  };
  const uint8x16_t shuffle = vld1q_u8(kg0g0);
#else
  static const uint8_t k0g0g[8] = { 255, 1, 255, 1, 255, 5, 255, 5 };
  const uint8x8_t shuffle = vld1_u8(k0g0g);
#endif
  const uint32x4_t mask_ag = vdupq_n_u32(0xff00ff00u);
  int i;
  for (i = 0; i + 4 <= num_pixels; i += 4) {
    const uint8x16_t in = vld1q_u8((const uint8_t*)(src + i));
    const uint32x4_t a0g0 = vandq_u32(vreinterpretq_u32_u8(in), mask_ag);
    // 0 g 0 g
    const uint8x16_t greens = DoGreenShuffle(in, shuffle);
    // x dr  x db1
    const int16x8_t A = vqdmulhq_s16(vreinterpretq_s16_u8(greens), mults_rb);
    // x r'  x   b'
    const int8x16_t B = vaddq_s8(vreinterpretq_s8_u8(in),
                                 vreinterpretq_s8_s16(A));
    // r' 0   b' 0
    const int16x8_t C = vshlq_n_s16(vreinterpretq_s16_s8(B), 8);
    // x db2  0  0
    const int16x8_t D = vqdmulhq_s16(C, mults_b2);
    // 0  x db2  0
    const uint32x4_t E = vshrq_n_u32(vreinterpretq_u32_s16(D), 8);
    // r' x  b'' 0
    const int8x16_t F = vaddq_s8(vreinterpretq_s8_u32(E),
                                 vreinterpretq_s8_s16(C));
    // 0  r'  0  b''
    const uint16x8_t G = vshrq_n_u16(vreinterpretq_u16_s8(F), 8);
    const uint32x4_t out = vorrq_u32(vreinterpretq_u32_u16(G), a0g0);
    vst1q_u32(dst + i, out);
  }
  // Fall-back to C-version for left-overs.
  VP8LTransformColorInverse_C(m, src + i, num_pixels - i, dst + i);
}

#undef USE_VTBLQ

//------------------------------------------------------------------------------
// Entry point

extern void VP8LDspInitNEON(void);

WEBP_TSAN_IGNORE_FUNCTION void VP8LDspInitNEON(void) {
  VP8LPredictors[5] = Predictor5_NEON;
  VP8LPredictors[6] = Predictor6_NEON;
  VP8LPredictors[7] = Predictor7_NEON;
  VP8LPredictors[13] = Predictor13_NEON;

  VP8LPredictorsAdd[0] = PredictorAdd0_NEON;
  VP8LPredictorsAdd[1] = PredictorAdd1_NEON;
  VP8LPredictorsAdd[2] = PredictorAdd2_NEON;
  VP8LPredictorsAdd[3] = PredictorAdd3_NEON;
  VP8LPredictorsAdd[4] = PredictorAdd4_NEON;
  VP8LPredictorsAdd[5] = PredictorAdd5_NEON;
  VP8LPredictorsAdd[6] = PredictorAdd6_NEON;
  VP8LPredictorsAdd[7] = PredictorAdd7_NEON;
  VP8LPredictorsAdd[8] = PredictorAdd8_NEON;
  VP8LPredictorsAdd[9] = PredictorAdd9_NEON;
  VP8LPredictorsAdd[10] = PredictorAdd10_NEON;
  VP8LPredictorsAdd[11] = PredictorAdd11_NEON;
  VP8LPredictorsAdd[12] = PredictorAdd12_NEON;
  VP8LPredictorsAdd[13] = PredictorAdd13_NEON;

  VP8LConvertBGRAToRGBA = ConvertBGRAToRGBA;
  VP8LConvertBGRAToBGR = ConvertBGRAToBGR;
  VP8LConvertBGRAToRGB = ConvertBGRAToRGB;

  VP8LAddGreenToBlueAndRed = AddGreenToBlueAndRed;
  VP8LTransformColorInverse = TransformColorInverse;
}

#else  // !WEBP_USE_NEON

WEBP_DSP_INIT_STUB(VP8LDspInitNEON)

#endif  // WEBP_USE_NEON