summaryrefslogtreecommitdiffstats
path: root/media/libjpeg/simd/arm/jcphuff-neon.c
blob: b91c5db478a10267e46d6a7df828febcc1e9d283 (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
/*
 * jcphuff-neon.c - prepare data for progressive Huffman encoding (Arm Neon)
 *
 * Copyright (C) 2020-2021, Arm Limited.  All Rights Reserved.
 *
 * This software is provided 'as-is', without any express or implied
 * warranty.  In no event will the authors be held liable for any damages
 * arising from the use of this software.
 *
 * Permission is granted to anyone to use this software for any purpose,
 * including commercial applications, and to alter it and redistribute it
 * freely, subject to the following restrictions:
 *
 * 1. The origin of this software must not be misrepresented; you must not
 *    claim that you wrote the original software. If you use this software
 *    in a product, an acknowledgment in the product documentation would be
 *    appreciated but is not required.
 * 2. Altered source versions must be plainly marked as such, and must not be
 *    misrepresented as being the original software.
 * 3. This notice may not be removed or altered from any source distribution.
 */

#define JPEG_INTERNALS
#include "jconfigint.h"
#include "../../jinclude.h"
#include "../../jpeglib.h"
#include "../../jsimd.h"
#include "../../jdct.h"
#include "../../jsimddct.h"
#include "../jsimd.h"
#include "neon-compat.h"

#include <arm_neon.h>


/* Data preparation for encode_mcu_AC_first().
 *
 * The equivalent scalar C function (encode_mcu_AC_first_prepare()) can be
 * found in jcphuff.c.
 */

void jsimd_encode_mcu_AC_first_prepare_neon
  (const JCOEF *block, const int *jpeg_natural_order_start, int Sl, int Al,
   JCOEF *values, size_t *zerobits)
{
  JCOEF *values_ptr = values;
  JCOEF *diff_values_ptr = values + DCTSIZE2;

  /* Rows of coefficients to zero (since they haven't been processed) */
  int i, rows_to_zero = 8;

  for (i = 0; i < Sl / 16; i++) {
    int16x8_t coefs1 = vld1q_dup_s16(block + jpeg_natural_order_start[0]);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[1], coefs1, 1);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[2], coefs1, 2);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[3], coefs1, 3);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[4], coefs1, 4);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[5], coefs1, 5);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[6], coefs1, 6);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[7], coefs1, 7);
    int16x8_t coefs2 = vld1q_dup_s16(block + jpeg_natural_order_start[8]);
    coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[9], coefs2, 1);
    coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[10], coefs2, 2);
    coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[11], coefs2, 3);
    coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[12], coefs2, 4);
    coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[13], coefs2, 5);
    coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[14], coefs2, 6);
    coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[15], coefs2, 7);

    /* Isolate sign of coefficients. */
    int16x8_t sign_coefs1 = vshrq_n_s16(coefs1, 15);
    int16x8_t sign_coefs2 = vshrq_n_s16(coefs2, 15);
    /* Compute absolute value of coefficients and apply point transform Al. */
    int16x8_t abs_coefs1 = vabsq_s16(coefs1);
    int16x8_t abs_coefs2 = vabsq_s16(coefs2);
    coefs1 = vshlq_s16(abs_coefs1, vdupq_n_s16(-Al));
    coefs2 = vshlq_s16(abs_coefs2, vdupq_n_s16(-Al));

    /* Compute diff values. */
    int16x8_t diff1 = veorq_s16(coefs1, sign_coefs1);
    int16x8_t diff2 = veorq_s16(coefs2, sign_coefs2);

    /* Store transformed coefficients and diff values. */
    vst1q_s16(values_ptr, coefs1);
    vst1q_s16(values_ptr + DCTSIZE, coefs2);
    vst1q_s16(diff_values_ptr, diff1);
    vst1q_s16(diff_values_ptr + DCTSIZE, diff2);
    values_ptr += 16;
    diff_values_ptr += 16;
    jpeg_natural_order_start += 16;
    rows_to_zero -= 2;
  }

  /* Same operation but for remaining partial vector */
  int remaining_coefs = Sl % 16;
  if (remaining_coefs > 8) {
    int16x8_t coefs1 = vld1q_dup_s16(block + jpeg_natural_order_start[0]);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[1], coefs1, 1);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[2], coefs1, 2);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[3], coefs1, 3);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[4], coefs1, 4);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[5], coefs1, 5);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[6], coefs1, 6);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[7], coefs1, 7);
    int16x8_t coefs2 = vdupq_n_s16(0);
    switch (remaining_coefs) {
    case 15:
      coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[14], coefs2, 6);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 14:
      coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[13], coefs2, 5);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 13:
      coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[12], coefs2, 4);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 12:
      coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[11], coefs2, 3);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 11:
      coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[10], coefs2, 2);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 10:
      coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[9], coefs2, 1);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 9:
      coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[8], coefs2, 0);
      FALLTHROUGH               /*FALLTHROUGH*/
    default:
      break;
    }

    /* Isolate sign of coefficients. */
    int16x8_t sign_coefs1 = vshrq_n_s16(coefs1, 15);
    int16x8_t sign_coefs2 = vshrq_n_s16(coefs2, 15);
    /* Compute absolute value of coefficients and apply point transform Al. */
    int16x8_t abs_coefs1 = vabsq_s16(coefs1);
    int16x8_t abs_coefs2 = vabsq_s16(coefs2);
    coefs1 = vshlq_s16(abs_coefs1, vdupq_n_s16(-Al));
    coefs2 = vshlq_s16(abs_coefs2, vdupq_n_s16(-Al));

    /* Compute diff values. */
    int16x8_t diff1 = veorq_s16(coefs1, sign_coefs1);
    int16x8_t diff2 = veorq_s16(coefs2, sign_coefs2);

    /* Store transformed coefficients and diff values. */
    vst1q_s16(values_ptr, coefs1);
    vst1q_s16(values_ptr + DCTSIZE, coefs2);
    vst1q_s16(diff_values_ptr, diff1);
    vst1q_s16(diff_values_ptr + DCTSIZE, diff2);
    values_ptr += 16;
    diff_values_ptr += 16;
    rows_to_zero -= 2;

  } else if (remaining_coefs > 0) {
    int16x8_t coefs = vdupq_n_s16(0);

    switch (remaining_coefs) {
    case 8:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[7], coefs, 7);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 7:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[6], coefs, 6);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 6:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[5], coefs, 5);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 5:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[4], coefs, 4);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 4:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[3], coefs, 3);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 3:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[2], coefs, 2);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 2:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[1], coefs, 1);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 1:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[0], coefs, 0);
      FALLTHROUGH               /*FALLTHROUGH*/
    default:
      break;
    }

    /* Isolate sign of coefficients. */
    int16x8_t sign_coefs = vshrq_n_s16(coefs, 15);
    /* Compute absolute value of coefficients and apply point transform Al. */
    int16x8_t abs_coefs = vabsq_s16(coefs);
    coefs = vshlq_s16(abs_coefs, vdupq_n_s16(-Al));

    /* Compute diff values. */
    int16x8_t diff = veorq_s16(coefs, sign_coefs);

    /* Store transformed coefficients and diff values. */
    vst1q_s16(values_ptr, coefs);
    vst1q_s16(diff_values_ptr, diff);
    values_ptr += 8;
    diff_values_ptr += 8;
    rows_to_zero--;
  }

  /* Zero remaining memory in the values and diff_values blocks. */
  for (i = 0; i < rows_to_zero; i++) {
    vst1q_s16(values_ptr, vdupq_n_s16(0));
    vst1q_s16(diff_values_ptr, vdupq_n_s16(0));
    values_ptr += 8;
    diff_values_ptr += 8;
  }

  /* Construct zerobits bitmap.  A set bit means that the corresponding
   * coefficient != 0.
   */
  int16x8_t row0 = vld1q_s16(values + 0 * DCTSIZE);
  int16x8_t row1 = vld1q_s16(values + 1 * DCTSIZE);
  int16x8_t row2 = vld1q_s16(values + 2 * DCTSIZE);
  int16x8_t row3 = vld1q_s16(values + 3 * DCTSIZE);
  int16x8_t row4 = vld1q_s16(values + 4 * DCTSIZE);
  int16x8_t row5 = vld1q_s16(values + 5 * DCTSIZE);
  int16x8_t row6 = vld1q_s16(values + 6 * DCTSIZE);
  int16x8_t row7 = vld1q_s16(values + 7 * DCTSIZE);

  uint8x8_t row0_eq0 = vmovn_u16(vceqq_s16(row0, vdupq_n_s16(0)));
  uint8x8_t row1_eq0 = vmovn_u16(vceqq_s16(row1, vdupq_n_s16(0)));
  uint8x8_t row2_eq0 = vmovn_u16(vceqq_s16(row2, vdupq_n_s16(0)));
  uint8x8_t row3_eq0 = vmovn_u16(vceqq_s16(row3, vdupq_n_s16(0)));
  uint8x8_t row4_eq0 = vmovn_u16(vceqq_s16(row4, vdupq_n_s16(0)));
  uint8x8_t row5_eq0 = vmovn_u16(vceqq_s16(row5, vdupq_n_s16(0)));
  uint8x8_t row6_eq0 = vmovn_u16(vceqq_s16(row6, vdupq_n_s16(0)));
  uint8x8_t row7_eq0 = vmovn_u16(vceqq_s16(row7, vdupq_n_s16(0)));

  /* { 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80 } */
  const uint8x8_t bitmap_mask =
    vreinterpret_u8_u64(vmov_n_u64(0x8040201008040201));

  row0_eq0 = vand_u8(row0_eq0, bitmap_mask);
  row1_eq0 = vand_u8(row1_eq0, bitmap_mask);
  row2_eq0 = vand_u8(row2_eq0, bitmap_mask);
  row3_eq0 = vand_u8(row3_eq0, bitmap_mask);
  row4_eq0 = vand_u8(row4_eq0, bitmap_mask);
  row5_eq0 = vand_u8(row5_eq0, bitmap_mask);
  row6_eq0 = vand_u8(row6_eq0, bitmap_mask);
  row7_eq0 = vand_u8(row7_eq0, bitmap_mask);

  uint8x8_t bitmap_rows_01 = vpadd_u8(row0_eq0, row1_eq0);
  uint8x8_t bitmap_rows_23 = vpadd_u8(row2_eq0, row3_eq0);
  uint8x8_t bitmap_rows_45 = vpadd_u8(row4_eq0, row5_eq0);
  uint8x8_t bitmap_rows_67 = vpadd_u8(row6_eq0, row7_eq0);
  uint8x8_t bitmap_rows_0123 = vpadd_u8(bitmap_rows_01, bitmap_rows_23);
  uint8x8_t bitmap_rows_4567 = vpadd_u8(bitmap_rows_45, bitmap_rows_67);
  uint8x8_t bitmap_all = vpadd_u8(bitmap_rows_0123, bitmap_rows_4567);

#if defined(__aarch64__) || defined(_M_ARM64)
  /* Move bitmap to a 64-bit scalar register. */
  uint64_t bitmap = vget_lane_u64(vreinterpret_u64_u8(bitmap_all), 0);
  /* Store zerobits bitmap. */
  *zerobits = ~bitmap;
#else
  /* Move bitmap to two 32-bit scalar registers. */
  uint32_t bitmap0 = vget_lane_u32(vreinterpret_u32_u8(bitmap_all), 0);
  uint32_t bitmap1 = vget_lane_u32(vreinterpret_u32_u8(bitmap_all), 1);
  /* Store zerobits bitmap. */
  zerobits[0] = ~bitmap0;
  zerobits[1] = ~bitmap1;
#endif
}


/* Data preparation for encode_mcu_AC_refine().
 *
 * The equivalent scalar C function (encode_mcu_AC_refine_prepare()) can be
 * found in jcphuff.c.
 */

int jsimd_encode_mcu_AC_refine_prepare_neon
  (const JCOEF *block, const int *jpeg_natural_order_start, int Sl, int Al,
   JCOEF *absvalues, size_t *bits)
{
  /* Temporary storage buffers for data used to compute the signbits bitmap and
   * the end-of-block (EOB) position
   */
  uint8_t coef_sign_bits[64];
  uint8_t coef_eq1_bits[64];

  JCOEF *absvalues_ptr = absvalues;
  uint8_t *coef_sign_bits_ptr = coef_sign_bits;
  uint8_t *eq1_bits_ptr = coef_eq1_bits;

  /* Rows of coefficients to zero (since they haven't been processed) */
  int i, rows_to_zero = 8;

  for (i = 0; i < Sl / 16; i++) {
    int16x8_t coefs1 = vld1q_dup_s16(block + jpeg_natural_order_start[0]);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[1], coefs1, 1);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[2], coefs1, 2);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[3], coefs1, 3);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[4], coefs1, 4);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[5], coefs1, 5);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[6], coefs1, 6);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[7], coefs1, 7);
    int16x8_t coefs2 = vld1q_dup_s16(block + jpeg_natural_order_start[8]);
    coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[9], coefs2, 1);
    coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[10], coefs2, 2);
    coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[11], coefs2, 3);
    coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[12], coefs2, 4);
    coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[13], coefs2, 5);
    coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[14], coefs2, 6);
    coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[15], coefs2, 7);

    /* Compute and store data for signbits bitmap. */
    uint8x8_t sign_coefs1 =
      vmovn_u16(vreinterpretq_u16_s16(vshrq_n_s16(coefs1, 15)));
    uint8x8_t sign_coefs2 =
      vmovn_u16(vreinterpretq_u16_s16(vshrq_n_s16(coefs2, 15)));
    vst1_u8(coef_sign_bits_ptr, sign_coefs1);
    vst1_u8(coef_sign_bits_ptr + DCTSIZE, sign_coefs2);

    /* Compute absolute value of coefficients and apply point transform Al. */
    int16x8_t abs_coefs1 = vabsq_s16(coefs1);
    int16x8_t abs_coefs2 = vabsq_s16(coefs2);
    coefs1 = vshlq_s16(abs_coefs1, vdupq_n_s16(-Al));
    coefs2 = vshlq_s16(abs_coefs2, vdupq_n_s16(-Al));
    vst1q_s16(absvalues_ptr, coefs1);
    vst1q_s16(absvalues_ptr + DCTSIZE, coefs2);

    /* Test whether transformed coefficient values == 1 (used to find EOB
     * position.)
     */
    uint8x8_t coefs_eq11 = vmovn_u16(vceqq_s16(coefs1, vdupq_n_s16(1)));
    uint8x8_t coefs_eq12 = vmovn_u16(vceqq_s16(coefs2, vdupq_n_s16(1)));
    vst1_u8(eq1_bits_ptr, coefs_eq11);
    vst1_u8(eq1_bits_ptr + DCTSIZE, coefs_eq12);

    absvalues_ptr += 16;
    coef_sign_bits_ptr += 16;
    eq1_bits_ptr += 16;
    jpeg_natural_order_start += 16;
    rows_to_zero -= 2;
  }

  /* Same operation but for remaining partial vector */
  int remaining_coefs = Sl % 16;
  if (remaining_coefs > 8) {
    int16x8_t coefs1 = vld1q_dup_s16(block + jpeg_natural_order_start[0]);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[1], coefs1, 1);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[2], coefs1, 2);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[3], coefs1, 3);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[4], coefs1, 4);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[5], coefs1, 5);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[6], coefs1, 6);
    coefs1 = vld1q_lane_s16(block + jpeg_natural_order_start[7], coefs1, 7);
    int16x8_t coefs2 = vdupq_n_s16(0);
    switch (remaining_coefs) {
    case 15:
      coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[14], coefs2, 6);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 14:
      coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[13], coefs2, 5);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 13:
      coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[12], coefs2, 4);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 12:
      coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[11], coefs2, 3);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 11:
      coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[10], coefs2, 2);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 10:
      coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[9], coefs2, 1);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 9:
      coefs2 = vld1q_lane_s16(block + jpeg_natural_order_start[8], coefs2, 0);
      FALLTHROUGH               /*FALLTHROUGH*/
    default:
      break;
    }

    /* Compute and store data for signbits bitmap. */
    uint8x8_t sign_coefs1 =
      vmovn_u16(vreinterpretq_u16_s16(vshrq_n_s16(coefs1, 15)));
    uint8x8_t sign_coefs2 =
      vmovn_u16(vreinterpretq_u16_s16(vshrq_n_s16(coefs2, 15)));
    vst1_u8(coef_sign_bits_ptr, sign_coefs1);
    vst1_u8(coef_sign_bits_ptr + DCTSIZE, sign_coefs2);

    /* Compute absolute value of coefficients and apply point transform Al. */
    int16x8_t abs_coefs1 = vabsq_s16(coefs1);
    int16x8_t abs_coefs2 = vabsq_s16(coefs2);
    coefs1 = vshlq_s16(abs_coefs1, vdupq_n_s16(-Al));
    coefs2 = vshlq_s16(abs_coefs2, vdupq_n_s16(-Al));
    vst1q_s16(absvalues_ptr, coefs1);
    vst1q_s16(absvalues_ptr + DCTSIZE, coefs2);

    /* Test whether transformed coefficient values == 1 (used to find EOB
     * position.)
     */
    uint8x8_t coefs_eq11 = vmovn_u16(vceqq_s16(coefs1, vdupq_n_s16(1)));
    uint8x8_t coefs_eq12 = vmovn_u16(vceqq_s16(coefs2, vdupq_n_s16(1)));
    vst1_u8(eq1_bits_ptr, coefs_eq11);
    vst1_u8(eq1_bits_ptr + DCTSIZE, coefs_eq12);

    absvalues_ptr += 16;
    coef_sign_bits_ptr += 16;
    eq1_bits_ptr += 16;
    jpeg_natural_order_start += 16;
    rows_to_zero -= 2;

  } else if (remaining_coefs > 0) {
    int16x8_t coefs = vdupq_n_s16(0);

    switch (remaining_coefs) {
    case 8:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[7], coefs, 7);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 7:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[6], coefs, 6);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 6:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[5], coefs, 5);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 5:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[4], coefs, 4);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 4:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[3], coefs, 3);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 3:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[2], coefs, 2);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 2:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[1], coefs, 1);
      FALLTHROUGH               /*FALLTHROUGH*/
    case 1:
      coefs = vld1q_lane_s16(block + jpeg_natural_order_start[0], coefs, 0);
      FALLTHROUGH               /*FALLTHROUGH*/
    default:
      break;
    }

    /* Compute and store data for signbits bitmap. */
    uint8x8_t sign_coefs =
      vmovn_u16(vreinterpretq_u16_s16(vshrq_n_s16(coefs, 15)));
    vst1_u8(coef_sign_bits_ptr, sign_coefs);

    /* Compute absolute value of coefficients and apply point transform Al. */
    int16x8_t abs_coefs = vabsq_s16(coefs);
    coefs = vshlq_s16(abs_coefs, vdupq_n_s16(-Al));
    vst1q_s16(absvalues_ptr, coefs);

    /* Test whether transformed coefficient values == 1 (used to find EOB
     * position.)
     */
    uint8x8_t coefs_eq1 = vmovn_u16(vceqq_s16(coefs, vdupq_n_s16(1)));
    vst1_u8(eq1_bits_ptr, coefs_eq1);

    absvalues_ptr += 8;
    coef_sign_bits_ptr += 8;
    eq1_bits_ptr += 8;
    rows_to_zero--;
  }

  /* Zero remaining memory in blocks. */
  for (i = 0; i < rows_to_zero; i++) {
    vst1q_s16(absvalues_ptr, vdupq_n_s16(0));
    vst1_u8(coef_sign_bits_ptr, vdup_n_u8(0));
    vst1_u8(eq1_bits_ptr, vdup_n_u8(0));
    absvalues_ptr += 8;
    coef_sign_bits_ptr += 8;
    eq1_bits_ptr += 8;
  }

  /* Construct zerobits bitmap. */
  int16x8_t abs_row0 = vld1q_s16(absvalues + 0 * DCTSIZE);
  int16x8_t abs_row1 = vld1q_s16(absvalues + 1 * DCTSIZE);
  int16x8_t abs_row2 = vld1q_s16(absvalues + 2 * DCTSIZE);
  int16x8_t abs_row3 = vld1q_s16(absvalues + 3 * DCTSIZE);
  int16x8_t abs_row4 = vld1q_s16(absvalues + 4 * DCTSIZE);
  int16x8_t abs_row5 = vld1q_s16(absvalues + 5 * DCTSIZE);
  int16x8_t abs_row6 = vld1q_s16(absvalues + 6 * DCTSIZE);
  int16x8_t abs_row7 = vld1q_s16(absvalues + 7 * DCTSIZE);

  uint8x8_t abs_row0_eq0 = vmovn_u16(vceqq_s16(abs_row0, vdupq_n_s16(0)));
  uint8x8_t abs_row1_eq0 = vmovn_u16(vceqq_s16(abs_row1, vdupq_n_s16(0)));
  uint8x8_t abs_row2_eq0 = vmovn_u16(vceqq_s16(abs_row2, vdupq_n_s16(0)));
  uint8x8_t abs_row3_eq0 = vmovn_u16(vceqq_s16(abs_row3, vdupq_n_s16(0)));
  uint8x8_t abs_row4_eq0 = vmovn_u16(vceqq_s16(abs_row4, vdupq_n_s16(0)));
  uint8x8_t abs_row5_eq0 = vmovn_u16(vceqq_s16(abs_row5, vdupq_n_s16(0)));
  uint8x8_t abs_row6_eq0 = vmovn_u16(vceqq_s16(abs_row6, vdupq_n_s16(0)));
  uint8x8_t abs_row7_eq0 = vmovn_u16(vceqq_s16(abs_row7, vdupq_n_s16(0)));

  /* { 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80 } */
  const uint8x8_t bitmap_mask =
    vreinterpret_u8_u64(vmov_n_u64(0x8040201008040201));

  abs_row0_eq0 = vand_u8(abs_row0_eq0, bitmap_mask);
  abs_row1_eq0 = vand_u8(abs_row1_eq0, bitmap_mask);
  abs_row2_eq0 = vand_u8(abs_row2_eq0, bitmap_mask);
  abs_row3_eq0 = vand_u8(abs_row3_eq0, bitmap_mask);
  abs_row4_eq0 = vand_u8(abs_row4_eq0, bitmap_mask);
  abs_row5_eq0 = vand_u8(abs_row5_eq0, bitmap_mask);
  abs_row6_eq0 = vand_u8(abs_row6_eq0, bitmap_mask);
  abs_row7_eq0 = vand_u8(abs_row7_eq0, bitmap_mask);

  uint8x8_t bitmap_rows_01 = vpadd_u8(abs_row0_eq0, abs_row1_eq0);
  uint8x8_t bitmap_rows_23 = vpadd_u8(abs_row2_eq0, abs_row3_eq0);
  uint8x8_t bitmap_rows_45 = vpadd_u8(abs_row4_eq0, abs_row5_eq0);
  uint8x8_t bitmap_rows_67 = vpadd_u8(abs_row6_eq0, abs_row7_eq0);
  uint8x8_t bitmap_rows_0123 = vpadd_u8(bitmap_rows_01, bitmap_rows_23);
  uint8x8_t bitmap_rows_4567 = vpadd_u8(bitmap_rows_45, bitmap_rows_67);
  uint8x8_t bitmap_all = vpadd_u8(bitmap_rows_0123, bitmap_rows_4567);

#if defined(__aarch64__) || defined(_M_ARM64)
  /* Move bitmap to a 64-bit scalar register. */
  uint64_t bitmap = vget_lane_u64(vreinterpret_u64_u8(bitmap_all), 0);
  /* Store zerobits bitmap. */
  bits[0] = ~bitmap;
#else
  /* Move bitmap to two 32-bit scalar registers. */
  uint32_t bitmap0 = vget_lane_u32(vreinterpret_u32_u8(bitmap_all), 0);
  uint32_t bitmap1 = vget_lane_u32(vreinterpret_u32_u8(bitmap_all), 1);
  /* Store zerobits bitmap. */
  bits[0] = ~bitmap0;
  bits[1] = ~bitmap1;
#endif

  /* Construct signbits bitmap. */
  uint8x8_t signbits_row0 = vld1_u8(coef_sign_bits + 0 * DCTSIZE);
  uint8x8_t signbits_row1 = vld1_u8(coef_sign_bits + 1 * DCTSIZE);
  uint8x8_t signbits_row2 = vld1_u8(coef_sign_bits + 2 * DCTSIZE);
  uint8x8_t signbits_row3 = vld1_u8(coef_sign_bits + 3 * DCTSIZE);
  uint8x8_t signbits_row4 = vld1_u8(coef_sign_bits + 4 * DCTSIZE);
  uint8x8_t signbits_row5 = vld1_u8(coef_sign_bits + 5 * DCTSIZE);
  uint8x8_t signbits_row6 = vld1_u8(coef_sign_bits + 6 * DCTSIZE);
  uint8x8_t signbits_row7 = vld1_u8(coef_sign_bits + 7 * DCTSIZE);

  signbits_row0 = vand_u8(signbits_row0, bitmap_mask);
  signbits_row1 = vand_u8(signbits_row1, bitmap_mask);
  signbits_row2 = vand_u8(signbits_row2, bitmap_mask);
  signbits_row3 = vand_u8(signbits_row3, bitmap_mask);
  signbits_row4 = vand_u8(signbits_row4, bitmap_mask);
  signbits_row5 = vand_u8(signbits_row5, bitmap_mask);
  signbits_row6 = vand_u8(signbits_row6, bitmap_mask);
  signbits_row7 = vand_u8(signbits_row7, bitmap_mask);

  bitmap_rows_01 = vpadd_u8(signbits_row0, signbits_row1);
  bitmap_rows_23 = vpadd_u8(signbits_row2, signbits_row3);
  bitmap_rows_45 = vpadd_u8(signbits_row4, signbits_row5);
  bitmap_rows_67 = vpadd_u8(signbits_row6, signbits_row7);
  bitmap_rows_0123 = vpadd_u8(bitmap_rows_01, bitmap_rows_23);
  bitmap_rows_4567 = vpadd_u8(bitmap_rows_45, bitmap_rows_67);
  bitmap_all = vpadd_u8(bitmap_rows_0123, bitmap_rows_4567);

#if defined(__aarch64__) || defined(_M_ARM64)
  /* Move bitmap to a 64-bit scalar register. */
  bitmap = vget_lane_u64(vreinterpret_u64_u8(bitmap_all), 0);
  /* Store signbits bitmap. */
  bits[1] = ~bitmap;
#else
  /* Move bitmap to two 32-bit scalar registers. */
  bitmap0 = vget_lane_u32(vreinterpret_u32_u8(bitmap_all), 0);
  bitmap1 = vget_lane_u32(vreinterpret_u32_u8(bitmap_all), 1);
  /* Store signbits bitmap. */
  bits[2] = ~bitmap0;
  bits[3] = ~bitmap1;
#endif

  /* Construct bitmap to find EOB position (the index of the last coefficient
   * equal to 1.)
   */
  uint8x8_t row0_eq1 = vld1_u8(coef_eq1_bits + 0 * DCTSIZE);
  uint8x8_t row1_eq1 = vld1_u8(coef_eq1_bits + 1 * DCTSIZE);
  uint8x8_t row2_eq1 = vld1_u8(coef_eq1_bits + 2 * DCTSIZE);
  uint8x8_t row3_eq1 = vld1_u8(coef_eq1_bits + 3 * DCTSIZE);
  uint8x8_t row4_eq1 = vld1_u8(coef_eq1_bits + 4 * DCTSIZE);
  uint8x8_t row5_eq1 = vld1_u8(coef_eq1_bits + 5 * DCTSIZE);
  uint8x8_t row6_eq1 = vld1_u8(coef_eq1_bits + 6 * DCTSIZE);
  uint8x8_t row7_eq1 = vld1_u8(coef_eq1_bits + 7 * DCTSIZE);

  row0_eq1 = vand_u8(row0_eq1, bitmap_mask);
  row1_eq1 = vand_u8(row1_eq1, bitmap_mask);
  row2_eq1 = vand_u8(row2_eq1, bitmap_mask);
  row3_eq1 = vand_u8(row3_eq1, bitmap_mask);
  row4_eq1 = vand_u8(row4_eq1, bitmap_mask);
  row5_eq1 = vand_u8(row5_eq1, bitmap_mask);
  row6_eq1 = vand_u8(row6_eq1, bitmap_mask);
  row7_eq1 = vand_u8(row7_eq1, bitmap_mask);

  bitmap_rows_01 = vpadd_u8(row0_eq1, row1_eq1);
  bitmap_rows_23 = vpadd_u8(row2_eq1, row3_eq1);
  bitmap_rows_45 = vpadd_u8(row4_eq1, row5_eq1);
  bitmap_rows_67 = vpadd_u8(row6_eq1, row7_eq1);
  bitmap_rows_0123 = vpadd_u8(bitmap_rows_01, bitmap_rows_23);
  bitmap_rows_4567 = vpadd_u8(bitmap_rows_45, bitmap_rows_67);
  bitmap_all = vpadd_u8(bitmap_rows_0123, bitmap_rows_4567);

#if defined(__aarch64__) || defined(_M_ARM64)
  /* Move bitmap to a 64-bit scalar register. */
  bitmap = vget_lane_u64(vreinterpret_u64_u8(bitmap_all), 0);

  /* Return EOB position. */
  if (bitmap == 0) {
    /* EOB position is defined to be 0 if all coefficients != 1. */
    return 0;
  } else {
    return 63 - BUILTIN_CLZLL(bitmap);
  }
#else
  /* Move bitmap to two 32-bit scalar registers. */
  bitmap0 = vget_lane_u32(vreinterpret_u32_u8(bitmap_all), 0);
  bitmap1 = vget_lane_u32(vreinterpret_u32_u8(bitmap_all), 1);

  /* Return EOB position. */
  if (bitmap0 == 0 && bitmap1 == 0) {
    return 0;
  } else if (bitmap1 != 0) {
    return 63 - BUILTIN_CLZ(bitmap1);
  } else {
    return 31 - BUILTIN_CLZ(bitmap0);
  }
#endif
}