aom_convolve.c 45.1 KB
Newer Older
John Koleszar's avatar
John Koleszar committed
1
/*
Yaowu Xu's avatar
Yaowu Xu committed
2
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
John Koleszar's avatar
John Koleszar committed
3
 *
Yaowu Xu's avatar
Yaowu Xu committed
4
5
6
7
8
9
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
John Koleszar's avatar
John Koleszar committed
10
 */
Christian Duvivier's avatar
Christian Duvivier committed
11

John Koleszar's avatar
John Koleszar committed
12
#include <assert.h>
Zoe Liu's avatar
Zoe Liu committed
13
#include <string.h>
John Koleszar's avatar
John Koleszar committed
14

Yaowu Xu's avatar
Yaowu Xu committed
15
16
17
18
19
20
#include "./aom_config.h"
#include "./aom_dsp_rtcd.h"
#include "aom/aom_integer.h"
#include "aom_dsp/aom_convolve.h"
#include "aom_dsp/aom_dsp_common.h"
#include "aom_dsp/aom_filter.h"
21
#include "aom_ports/mem.h"
John Koleszar's avatar
John Koleszar committed
22

Dmitry Kovalev's avatar
Dmitry Kovalev committed
23
24
static void convolve_horiz(const uint8_t *src, ptrdiff_t src_stride,
                           uint8_t *dst, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
25
26
                           const InterpKernel *x_filters, int x0_q4,
                           int x_step_q4, int w, int h) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
27
28
  int x, y;
  src -= SUBPEL_TAPS / 2 - 1;
John Koleszar's avatar
John Koleszar committed
29
  for (y = 0; y < h; ++y) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
30
    int x_q4 = x0_q4;
John Koleszar's avatar
John Koleszar committed
31
    for (x = 0; x < w; ++x) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
32
33
34
      const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
      int k, sum = 0;
clang-format's avatar
clang-format committed
35
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
36
      dst[x] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
John Koleszar's avatar
John Koleszar committed
37
38
39
40
41
42
43
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

Dmitry Kovalev's avatar
Dmitry Kovalev committed
44
45
static void convolve_avg_horiz(const uint8_t *src, ptrdiff_t src_stride,
                               uint8_t *dst, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
46
47
                               const InterpKernel *x_filters, int x0_q4,
                               int x_step_q4, int w, int h) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
48
49
  int x, y;
  src -= SUBPEL_TAPS / 2 - 1;
John Koleszar's avatar
John Koleszar committed
50
  for (y = 0; y < h; ++y) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
51
    int x_q4 = x0_q4;
John Koleszar's avatar
John Koleszar committed
52
    for (x = 0; x < w; ++x) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
53
54
55
      const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
      int k, sum = 0;
clang-format's avatar
clang-format committed
56
57
58
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
      dst[x] = ROUND_POWER_OF_TWO(
          dst[x] + clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)), 1);
John Koleszar's avatar
John Koleszar committed
59
60
61
62
63
64
65
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

Dmitry Kovalev's avatar
Dmitry Kovalev committed
66
67
static void convolve_vert(const uint8_t *src, ptrdiff_t src_stride,
                          uint8_t *dst, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
68
69
                          const InterpKernel *y_filters, int y0_q4,
                          int y_step_q4, int w, int h) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
70
71
  int x, y;
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
John Koleszar's avatar
John Koleszar committed
72

73
  for (x = 0; x < w; ++x) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
74
    int y_q4 = y0_q4;
John Koleszar's avatar
John Koleszar committed
75
    for (y = 0; y < h; ++y) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
76
77
78
79
80
81
      const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      dst[y * dst_stride] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
John Koleszar's avatar
John Koleszar committed
82
83
84
85
86
87
88
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

Dmitry Kovalev's avatar
Dmitry Kovalev committed
89
90
static void convolve_avg_vert(const uint8_t *src, ptrdiff_t src_stride,
                              uint8_t *dst, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
91
92
                              const InterpKernel *y_filters, int y0_q4,
                              int y_step_q4, int w, int h) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
93
94
  int x, y;
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
John Koleszar's avatar
John Koleszar committed
95

96
  for (x = 0; x < w; ++x) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
97
    int y_q4 = y0_q4;
John Koleszar's avatar
John Koleszar committed
98
    for (y = 0; y < h; ++y) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
99
100
101
102
103
      const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
clang-format's avatar
clang-format committed
104
105
106
107
      dst[y * dst_stride] = ROUND_POWER_OF_TWO(
          dst[y * dst_stride] +
              clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)),
          1);
John Koleszar's avatar
John Koleszar committed
108
109
110
111
112
113
114
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

clang-format's avatar
clang-format committed
115
116
static void convolve(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                     ptrdiff_t dst_stride, const InterpKernel *const x_filters,
Dmitry Kovalev's avatar
Dmitry Kovalev committed
117
                     int x0_q4, int x_step_q4,
clang-format's avatar
clang-format committed
118
119
                     const InterpKernel *const y_filters, int y0_q4,
                     int y_step_q4, int w, int h) {
120
121
122
123
124
125
126
127
128
129
130
131
  // Note: Fixed size intermediate buffer, temp, places limits on parameters.
  // 2d filtering proceeds in 2 steps:
  //   (1) Interpolate horizontally into an intermediate buffer, temp.
  //   (2) Interpolate temp vertically to derive the sub-pixel result.
  // Deriving the maximum number of rows in the temp buffer (135):
  // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
  // --Largest block size is 64x64 pixels.
  // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
  //   original frame (in 1/16th pixel units).
  // --Must round-up because block may be located at sub-pixel position.
  // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
  // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
132
  uint8_t temp[MAX_EXT_SIZE * MAX_SB_SIZE];
133
  int intermediate_height =
clang-format's avatar
clang-format committed
134
      (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;
135

136
137
  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);
138

139
140
  assert(y_step_q4 <= 32);
  assert(x_step_q4 <= 32);
141

clang-format's avatar
clang-format committed
142
143
144
145
146
  convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride, temp,
                 MAX_SB_SIZE, x_filters, x0_q4, x_step_q4, w,
                 intermediate_height);
  convolve_vert(temp + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1), MAX_SB_SIZE, dst,
                dst_stride, y_filters, y0_q4, y_step_q4, w, h);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
147
148
}

149
static const InterpKernel *get_filter_base(const int16_t *filter) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
150
151
  // NOTE: This assumes that the filter table is 256-byte aligned.
  // TODO(agrange) Modify to make independent of table alignment.
152
  return (const InterpKernel *)(((intptr_t)filter) & ~((intptr_t)0xFF));
Dmitry Kovalev's avatar
Dmitry Kovalev committed
153
154
}

155
static int get_filter_offset(const int16_t *f, const InterpKernel *base) {
156
  return (int)((const InterpKernel *)(intptr_t)f - base);
John Koleszar's avatar
John Koleszar committed
157
158
}

Yaowu Xu's avatar
Yaowu Xu committed
159
void aom_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
160
                           uint8_t *dst, ptrdiff_t dst_stride,
John Koleszar's avatar
John Koleszar committed
161
                           const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
162
163
                           const int16_t *filter_y, int y_step_q4, int w,
                           int h) {
164
  const InterpKernel *const filters_x = get_filter_base(filter_x);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
165
166
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

167
168
169
  (void)filter_y;
  (void)y_step_q4;

clang-format's avatar
clang-format committed
170
171
  convolve_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4, x_step_q4,
                 w, h);
John Koleszar's avatar
John Koleszar committed
172
173
}

Yaowu Xu's avatar
Yaowu Xu committed
174
void aom_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
175
                               uint8_t *dst, ptrdiff_t dst_stride,
John Koleszar's avatar
John Koleszar committed
176
                               const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
177
178
                               const int16_t *filter_y, int y_step_q4, int w,
                               int h) {
179
  const InterpKernel *const filters_x = get_filter_base(filter_x);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
180
181
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

182
183
184
  (void)filter_y;
  (void)y_step_q4;

clang-format's avatar
clang-format committed
185
186
  convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                     x_step_q4, w, h);
John Koleszar's avatar
John Koleszar committed
187
188
}

Yaowu Xu's avatar
Yaowu Xu committed
189
void aom_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
190
                          uint8_t *dst, ptrdiff_t dst_stride,
John Koleszar's avatar
John Koleszar committed
191
                          const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
192
193
                          const int16_t *filter_y, int y_step_q4, int w,
                          int h) {
194
  const InterpKernel *const filters_y = get_filter_base(filter_y);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
195
  const int y0_q4 = get_filter_offset(filter_y, filters_y);
196
197
198
199

  (void)filter_x;
  (void)x_step_q4;

clang-format's avatar
clang-format committed
200
201
  convolve_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4, y_step_q4,
                w, h);
John Koleszar's avatar
John Koleszar committed
202
203
}

Yaowu Xu's avatar
Yaowu Xu committed
204
void aom_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
205
                              uint8_t *dst, ptrdiff_t dst_stride,
John Koleszar's avatar
John Koleszar committed
206
                              const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
207
208
                              const int16_t *filter_y, int y_step_q4, int w,
                              int h) {
209
  const InterpKernel *const filters_y = get_filter_base(filter_y);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
210
  const int y0_q4 = get_filter_offset(filter_y, filters_y);
211
212
213
214

  (void)filter_x;
  (void)x_step_q4;

clang-format's avatar
clang-format committed
215
216
  convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
                    y_step_q4, w, h);
John Koleszar's avatar
John Koleszar committed
217
218
}

Yaowu Xu's avatar
Yaowu Xu committed
219
void aom_convolve8_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
220
221
                     ptrdiff_t dst_stride, const int16_t *filter_x,
                     int x_step_q4, const int16_t *filter_y, int y_step_q4,
John Koleszar's avatar
John Koleszar committed
222
                     int w, int h) {
223
  const InterpKernel *const filters_x = get_filter_base(filter_x);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
224
225
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

226
  const InterpKernel *const filters_y = get_filter_base(filter_y);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
227
228
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

clang-format's avatar
clang-format committed
229
  convolve(src, src_stride, dst, dst_stride, filters_x, x0_q4, x_step_q4,
Dmitry Kovalev's avatar
Dmitry Kovalev committed
230
           filters_y, y0_q4, y_step_q4, w, h);
John Koleszar's avatar
John Koleszar committed
231
232
}

Yaowu Xu's avatar
Yaowu Xu committed
233
void aom_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
234
235
                         ptrdiff_t dst_stride, const int16_t *filter_x,
                         int x_step_q4, const int16_t *filter_y, int y_step_q4,
John Koleszar's avatar
John Koleszar committed
236
                         int w, int h) {
Christian Duvivier's avatar
Christian Duvivier committed
237
  /* Fixed size intermediate buffer places limits on parameters. */
238
239
240
  DECLARE_ALIGNED(16, uint8_t, temp[MAX_SB_SIZE * MAX_SB_SIZE]);
  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);
Christian Duvivier's avatar
Christian Duvivier committed
241

Yaowu Xu's avatar
Yaowu Xu committed
242
  aom_convolve8_c(src, src_stride, temp, MAX_SB_SIZE, filter_x, x_step_q4,
clang-format's avatar
clang-format committed
243
                  filter_y, y_step_q4, w, h);
Yaowu Xu's avatar
Yaowu Xu committed
244
  aom_convolve_avg_c(temp, MAX_SB_SIZE, dst, dst_stride, NULL, 0, NULL, 0, w,
clang-format's avatar
clang-format committed
245
                     h);
John Koleszar's avatar
John Koleszar committed
246
}
247

Yaowu Xu's avatar
Yaowu Xu committed
248
void aom_convolve_copy_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
249
250
251
                         ptrdiff_t dst_stride, const int16_t *filter_x,
                         int filter_x_stride, const int16_t *filter_y,
                         int filter_y_stride, int w, int h) {
252
253
  int r;

clang-format's avatar
clang-format committed
254
255
256
257
  (void)filter_x;
  (void)filter_x_stride;
  (void)filter_y;
  (void)filter_y_stride;
258

259
  for (r = h; r > 0; --r) {
James Zern's avatar
James Zern committed
260
    memcpy(dst, src, w);
261
262
    src += src_stride;
    dst += dst_stride;
263
264
265
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
266
void aom_convolve_avg_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
267
268
269
                        ptrdiff_t dst_stride, const int16_t *filter_x,
                        int filter_x_stride, const int16_t *filter_y,
                        int filter_y_stride, int w, int h) {
270
271
  int x, y;

clang-format's avatar
clang-format committed
272
273
274
275
  (void)filter_x;
  (void)filter_x_stride;
  (void)filter_y;
  (void)filter_y_stride;
276

277
  for (y = 0; y < h; ++y) {
clang-format's avatar
clang-format committed
278
    for (x = 0; x < w; ++x) dst[x] = ROUND_POWER_OF_TWO(dst[x] + src[x], 1);
279

280
281
282
283
    src += src_stride;
    dst += dst_stride;
  }
}
284

Yaowu Xu's avatar
Yaowu Xu committed
285
void aom_scaled_horiz_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
286
287
                        ptrdiff_t dst_stride, const int16_t *filter_x,
                        int x_step_q4, const int16_t *filter_y, int y_step_q4,
288
                        int w, int h) {
Yaowu Xu's avatar
Yaowu Xu committed
289
  aom_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
290
291
292
                        filter_y, y_step_q4, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
293
void aom_scaled_vert_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
294
295
                       ptrdiff_t dst_stride, const int16_t *filter_x,
                       int x_step_q4, const int16_t *filter_y, int y_step_q4,
296
                       int w, int h) {
Yaowu Xu's avatar
Yaowu Xu committed
297
  aom_convolve8_vert_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
298
299
300
                       filter_y, y_step_q4, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
301
void aom_scaled_2d_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
302
303
                     ptrdiff_t dst_stride, const int16_t *filter_x,
                     int x_step_q4, const int16_t *filter_y, int y_step_q4,
304
                     int w, int h) {
Yaowu Xu's avatar
Yaowu Xu committed
305
  aom_convolve8_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
306
307
308
                  filter_y, y_step_q4, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
309
void aom_scaled_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
310
311
                            uint8_t *dst, ptrdiff_t dst_stride,
                            const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
312
313
                            const int16_t *filter_y, int y_step_q4, int w,
                            int h) {
Yaowu Xu's avatar
Yaowu Xu committed
314
  aom_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter_x,
315
316
317
                            x_step_q4, filter_y, y_step_q4, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
318
void aom_scaled_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
319
320
                           uint8_t *dst, ptrdiff_t dst_stride,
                           const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
321
322
                           const int16_t *filter_y, int y_step_q4, int w,
                           int h) {
Yaowu Xu's avatar
Yaowu Xu committed
323
  aom_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter_x,
324
325
326
                           x_step_q4, filter_y, y_step_q4, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
327
void aom_scaled_avg_2d_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
328
329
330
                         ptrdiff_t dst_stride, const int16_t *filter_x,
                         int x_step_q4, const int16_t *filter_y, int y_step_q4,
                         int w, int h) {
Yaowu Xu's avatar
Yaowu Xu committed
331
  aom_convolve8_avg_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
332
333
334
                      filter_y, y_step_q4, w, h);
}

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
#if CONFIG_LOOP_RESTORATION
static void convolve_add_src_horiz(const uint8_t *src, ptrdiff_t src_stride,
                                   uint8_t *dst, ptrdiff_t dst_stride,
                                   const InterpKernel *x_filters, int x0_q4,
                                   int x_step_q4, int w, int h) {
  int x, y;
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = x0_q4;
    for (x = 0; x < w; ++x) {
      const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
      dst[x] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS) +
                          src_x[SUBPEL_TAPS / 2 - 1]);
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

static void convolve_add_src_vert(const uint8_t *src, ptrdiff_t src_stride,
                                  uint8_t *dst, ptrdiff_t dst_stride,
                                  const InterpKernel *y_filters, int y0_q4,
                                  int y_step_q4, int w, int h) {
  int x, y;
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);

  for (x = 0; x < w; ++x) {
    int y_q4 = y0_q4;
    for (y = 0; y < h; ++y) {
      const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      dst[y * dst_stride] =
          clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS) +
                     src_y[(SUBPEL_TAPS / 2 - 1) * src_stride]);
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

static void convolve_add_src(const uint8_t *src, ptrdiff_t src_stride,
                             uint8_t *dst, ptrdiff_t dst_stride,
                             const InterpKernel *const x_filters, int x0_q4,
                             int x_step_q4, const InterpKernel *const y_filters,
                             int y0_q4, int y_step_q4, int w, int h) {
  uint8_t temp[MAX_EXT_SIZE * MAX_SB_SIZE];
  int intermediate_height =
      (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;

  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);

  assert(y_step_q4 <= 32);
  assert(x_step_q4 <= 32);

  convolve_add_src_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride,
                         temp, MAX_SB_SIZE, x_filters, x0_q4, x_step_q4, w,
                         intermediate_height);
  convolve_add_src_vert(temp + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1), MAX_SB_SIZE,
                        dst, dst_stride, y_filters, y0_q4, y_step_q4, w, h);
}

void aom_convolve8_add_src_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
                                   uint8_t *dst, ptrdiff_t dst_stride,
                                   const int16_t *filter_x, int x_step_q4,
                                   const int16_t *filter_y, int y_step_q4,
                                   int w, int h) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

  (void)filter_y;
  (void)y_step_q4;

  convolve_add_src_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                         x_step_q4, w, h);
}

void aom_convolve8_add_src_vert_c(const uint8_t *src, ptrdiff_t src_stride,
                                  uint8_t *dst, ptrdiff_t dst_stride,
                                  const int16_t *filter_x, int x_step_q4,
                                  const int16_t *filter_y, int y_step_q4, int w,
                                  int h) {
  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

  (void)filter_x;
  (void)x_step_q4;

  convolve_add_src_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
                        y_step_q4, w, h);
}

void aom_convolve8_add_src_c(const uint8_t *src, ptrdiff_t src_stride,
                             uint8_t *dst, ptrdiff_t dst_stride,
                             const int16_t *filter_x, int x_step_q4,
                             const int16_t *filter_y, int y_step_q4, int w,
                             int h) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

  convolve_add_src(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                   x_step_q4, filters_y, y0_q4, y_step_q4, w, h);
}
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

#define EXTRAPREC_BITS 2
#define EXTRAPREC_CLAMP_LIMIT (512 << EXTRAPREC_BITS)
static void convolve_add_src_horiz_hip(const uint8_t *src, ptrdiff_t src_stride,
                                       uint16_t *dst, ptrdiff_t dst_stride,
                                       const InterpKernel *x_filters, int x0_q4,
                                       int x_step_q4, int w, int h) {
  int x, y;
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = x0_q4;
    for (x = 0; x < w; ++x) {
      const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
      int k, sum = ((int)src_x[SUBPEL_TAPS / 2 - 1] << FILTER_BITS);
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
      dst[x] =
          (uint16_t)clamp(ROUND_POWER_OF_TWO(sum, FILTER_BITS - EXTRAPREC_BITS),
                          0, EXTRAPREC_CLAMP_LIMIT - 1);
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

static void convolve_add_src_vert_hip(const uint16_t *src, ptrdiff_t src_stride,
                                      uint8_t *dst, ptrdiff_t dst_stride,
                                      const InterpKernel *y_filters, int y0_q4,
                                      int y_step_q4, int w, int h) {
  int x, y;
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);

  for (x = 0; x < w; ++x) {
    int y_q4 = y0_q4;
    for (y = 0; y < h; ++y) {
      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
      int k,
          sum = ((int)src_y[(SUBPEL_TAPS / 2 - 1) * src_stride] << FILTER_BITS);
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      dst[y * dst_stride] =
          clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS + EXTRAPREC_BITS));
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

static void convolve_add_src_hip(const uint8_t *src, ptrdiff_t src_stride,
                                 uint8_t *dst, ptrdiff_t dst_stride,
                                 const InterpKernel *const x_filters, int x0_q4,
                                 int x_step_q4,
                                 const InterpKernel *const y_filters, int y0_q4,
                                 int y_step_q4, int w, int h) {
  uint16_t temp[MAX_EXT_SIZE * MAX_SB_SIZE];
  int intermediate_height =
      (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;

  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);

  assert(y_step_q4 <= 32);
  assert(x_step_q4 <= 32);

  convolve_add_src_horiz_hip(src - src_stride * (SUBPEL_TAPS / 2 - 1),
                             src_stride, temp, MAX_SB_SIZE, x_filters, x0_q4,
                             x_step_q4, w, intermediate_height);
  convolve_add_src_vert_hip(temp + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1),
                            MAX_SB_SIZE, dst, dst_stride, y_filters, y0_q4,
                            y_step_q4, w, h);
}

void aom_convolve8_add_src_horiz_hip_c(const uint8_t *src, ptrdiff_t src_stride,
                                       uint16_t *dst, ptrdiff_t dst_stride,
                                       const int16_t *filter_x, int x_step_q4,
                                       const int16_t *filter_y, int y_step_q4,
                                       int w, int h) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

  (void)filter_y;
  (void)y_step_q4;

  convolve_add_src_horiz_hip(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                             x_step_q4, w, h);
}

void aom_convolve8_add_src_vert_hip_c(const uint16_t *src, ptrdiff_t src_stride,
                                      uint8_t *dst, ptrdiff_t dst_stride,
                                      const int16_t *filter_x, int x_step_q4,
                                      const int16_t *filter_y, int y_step_q4,
                                      int w, int h) {
  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

  (void)filter_x;
  (void)x_step_q4;

  convolve_add_src_vert_hip(src, src_stride, dst, dst_stride, filters_y, y0_q4,
                            y_step_q4, w, h);
}

void aom_convolve8_add_src_hip_c(const uint8_t *src, ptrdiff_t src_stride,
                                 uint8_t *dst, ptrdiff_t dst_stride,
                                 const int16_t *filter_x, int x_step_q4,
                                 const int16_t *filter_y, int y_step_q4, int w,
                                 int h) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

  convolve_add_src_hip(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                       x_step_q4, filters_y, y0_q4, y_step_q4, w, h);
}
568
569
#endif  // CONFIG_LOOP_RESTORATION

570
#if CONFIG_HIGHBITDEPTH
571
572
static void highbd_convolve_horiz(const uint8_t *src8, ptrdiff_t src_stride,
                                  uint8_t *dst8, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
573
574
                                  const InterpKernel *x_filters, int x0_q4,
                                  int x_step_q4, int w, int h, int bd) {
575
576
577
578
579
580
581
582
583
584
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = x0_q4;
    for (x = 0; x < w; ++x) {
      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
      int k, sum = 0;
clang-format's avatar
clang-format committed
585
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
586
      dst[x] = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
587
588
589
590
591
592
593
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

594
595
static void highbd_convolve_avg_horiz(const uint8_t *src8, ptrdiff_t src_stride,
                                      uint8_t *dst8, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
596
597
                                      const InterpKernel *x_filters, int x0_q4,
                                      int x_step_q4, int w, int h, int bd) {
598
599
600
601
602
603
604
605
606
607
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = x0_q4;
    for (x = 0; x < w; ++x) {
      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
      int k, sum = 0;
clang-format's avatar
clang-format committed
608
609
610
611
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
      dst[x] = ROUND_POWER_OF_TWO(
          dst[x] + clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd),
          1);
612
613
614
615
616
617
618
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

619
620
static void highbd_convolve_vert(const uint8_t *src8, ptrdiff_t src_stride,
                                 uint8_t *dst8, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
621
622
                                 const InterpKernel *y_filters, int y0_q4,
                                 int y_step_q4, int w, int h, int bd) {
623
624
625
626
627
628
629
630
631
632
633
634
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
  for (x = 0; x < w; ++x) {
    int y_q4 = y0_q4;
    for (y = 0; y < h; ++y) {
      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
clang-format's avatar
clang-format committed
635
636
      dst[y * dst_stride] =
          clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
637
638
639
640
641
642
643
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

644
645
static void highbd_convolve_avg_vert(const uint8_t *src8, ptrdiff_t src_stride,
                                     uint8_t *dst8, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
646
647
                                     const InterpKernel *y_filters, int y0_q4,
                                     int y_step_q4, int w, int h, int bd) {
648
649
650
651
652
653
654
655
656
657
658
659
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
  for (x = 0; x < w; ++x) {
    int y_q4 = y0_q4;
    for (y = 0; y < h; ++y) {
      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
clang-format's avatar
clang-format committed
660
661
662
663
      dst[y * dst_stride] = ROUND_POWER_OF_TWO(
          dst[y * dst_stride] +
              clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd),
          1);
664
665
666
667
668
669
670
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

671
672
static void highbd_convolve(const uint8_t *src, ptrdiff_t src_stride,
                            uint8_t *dst, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
673
674
675
                            const InterpKernel *const x_filters, int x0_q4,
                            int x_step_q4, const InterpKernel *const y_filters,
                            int y0_q4, int y_step_q4, int w, int h, int bd) {
676
677
678
679
680
681
682
683
684
685
686
687
  // Note: Fixed size intermediate buffer, temp, places limits on parameters.
  // 2d filtering proceeds in 2 steps:
  //   (1) Interpolate horizontally into an intermediate buffer, temp.
  //   (2) Interpolate temp vertically to derive the sub-pixel result.
  // Deriving the maximum number of rows in the temp buffer (135):
  // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
  // --Largest block size is 64x64 pixels.
  // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
  //   original frame (in 1/16th pixel units).
  // --Must round-up because block may be located at sub-pixel position.
  // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
  // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
688
  uint16_t temp[MAX_EXT_SIZE * MAX_SB_SIZE];
689
  int intermediate_height =
clang-format's avatar
clang-format committed
690
      (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;
691

692
693
  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);
694
695
696
  assert(y_step_q4 <= 32);
  assert(x_step_q4 <= 32);

697
  highbd_convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride,
clang-format's avatar
clang-format committed
698
699
                        CONVERT_TO_BYTEPTR(temp), MAX_SB_SIZE, x_filters, x0_q4,
                        x_step_q4, w, intermediate_height, bd);
700
  highbd_convolve_vert(
clang-format's avatar
clang-format committed
701
702
      CONVERT_TO_BYTEPTR(temp) + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1),
      MAX_SB_SIZE, dst, dst_stride, y_filters, y0_q4, y_step_q4, w, h, bd);
703
704
}

Yaowu Xu's avatar
Yaowu Xu committed
705
void aom_highbd_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
706
707
                                  uint8_t *dst, ptrdiff_t dst_stride,
                                  const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
708
709
                                  const int16_t *filter_y, int y_step_q4, int w,
                                  int h, int bd) {
710
711
712
713
714
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);
  (void)filter_y;
  (void)y_step_q4;

clang-format's avatar
clang-format committed
715
716
  highbd_convolve_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                        x_step_q4, w, h, bd);
717
718
}

Yaowu Xu's avatar
Yaowu Xu committed
719
void aom_highbd_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
720
721
722
723
                                      uint8_t *dst, ptrdiff_t dst_stride,
                                      const int16_t *filter_x, int x_step_q4,
                                      const int16_t *filter_y, int y_step_q4,
                                      int w, int h, int bd) {
724
725
726
727
728
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);
  (void)filter_y;
  (void)y_step_q4;

clang-format's avatar
clang-format committed
729
730
  highbd_convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                            x_step_q4, w, h, bd);
731
732
}

Yaowu Xu's avatar
Yaowu Xu committed
733
void aom_highbd_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
734
735
                                 uint8_t *dst, ptrdiff_t dst_stride,
                                 const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
736
737
                                 const int16_t *filter_y, int y_step_q4, int w,
                                 int h, int bd) {
738
739
740
741
742
  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);
  (void)filter_x;
  (void)x_step_q4;

clang-format's avatar
clang-format committed
743
744
  highbd_convolve_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
                       y_step_q4, w, h, bd);
745
746
}

Yaowu Xu's avatar
Yaowu Xu committed
747
void aom_highbd_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
748
749
750
751
                                     uint8_t *dst, ptrdiff_t dst_stride,
                                     const int16_t *filter_x, int x_step_q4,
                                     const int16_t *filter_y, int y_step_q4,
                                     int w, int h, int bd) {
752
753
754
755
756
  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);
  (void)filter_x;
  (void)x_step_q4;

clang-format's avatar
clang-format committed
757
758
  highbd_convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
                           y_step_q4, w, h, bd);
759
760
}

Yaowu Xu's avatar
Yaowu Xu committed
761
void aom_highbd_convolve8_c(const uint8_t *src, ptrdiff_t src_stride,
762
763
                            uint8_t *dst, ptrdiff_t dst_stride,
                            const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
764
765
                            const int16_t *filter_y, int y_step_q4, int w,
                            int h, int bd) {
766
767
768
769
770
771
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

clang-format's avatar
clang-format committed
772
  highbd_convolve(src, src_stride, dst, dst_stride, filters_x, x0_q4, x_step_q4,
773
                  filters_y, y0_q4, y_step_q4, w, h, bd);
774
775
}

Yaowu Xu's avatar
Yaowu Xu committed
776
void aom_highbd_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride,
777
778
                                uint8_t *dst, ptrdiff_t dst_stride,
                                const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
779
780
                                const int16_t *filter_y, int y_step_q4, int w,
                                int h, int bd) {
781
  // Fixed size intermediate buffer places limits on parameters.
782
783
784
  DECLARE_ALIGNED(16, uint16_t, temp[MAX_SB_SIZE * MAX_SB_SIZE]);
  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);
785

Yaowu Xu's avatar
Yaowu Xu committed
786
  aom_highbd_convolve8_c(src, src_stride, CONVERT_TO_BYTEPTR(temp), MAX_SB_SIZE,
787
                         filter_x, x_step_q4, filter_y, y_step_q4, w, h, bd);
Yaowu Xu's avatar
Yaowu Xu committed
788
  aom_highbd_convolve_avg_c(CONVERT_TO_BYTEPTR(temp), MAX_SB_SIZE, dst,
clang-format's avatar
clang-format committed
789
                            dst_stride, NULL, 0, NULL, 0, w, h, bd);
790
791
}

Yaowu Xu's avatar
Yaowu Xu committed
792
void aom_highbd_convolve_copy_c(const uint8_t *src8, ptrdiff_t src_stride,
793
794
795
796
                                uint8_t *dst8, ptrdiff_t dst_stride,
                                const int16_t *filter_x, int filter_x_stride,
                                const int16_t *filter_y, int filter_y_stride,
                                int w, int h, int bd) {
797
798
799
800
801
802
803
804
805
806
  int r;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  (void)filter_x;
  (void)filter_y;
  (void)filter_x_stride;
  (void)filter_y_stride;
  (void)bd;

  for (r = h; r > 0; --r) {
James Zern's avatar
James Zern committed
807
    memcpy(dst, src, w * sizeof(uint16_t));
808
809
810
811
812
    src += src_stride;
    dst += dst_stride;
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
813
void aom_highbd_convolve_avg_c(const uint8_t *src8, ptrdiff_t src_stride,
814
815
816
817
                               uint8_t *dst8, ptrdiff_t dst_stride,
                               const int16_t *filter_x, int filter_x_stride,
                               const int16_t *filter_y, int filter_y_stride,
                               int w, int h, int bd) {
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  (void)filter_x;
  (void)filter_y;
  (void)filter_x_stride;
  (void)filter_y_stride;
  (void)bd;

  for (y = 0; y < h; ++y) {
    for (x = 0; x < w; ++x) {
      dst[x] = ROUND_POWER_OF_TWO(dst[x] + src[x], 1);
    }
    src += src_stride;
    dst += dst_stride;
  }
}
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971

#if CONFIG_LOOP_RESTORATION
static void highbd_convolve_add_src_horiz(const uint8_t *src8,
                                          ptrdiff_t src_stride, uint8_t *dst8,
                                          ptrdiff_t dst_stride,
                                          const InterpKernel *x_filters,
                                          int x0_q4, int x_step_q4, int w,
                                          int h, int bd) {
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = x0_q4;
    for (x = 0; x < w; ++x) {
      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
      dst[x] = clip_pixel_highbd(
          ROUND_POWER_OF_TWO(sum, FILTER_BITS) + src_x[SUBPEL_TAPS / 2 - 1],
          bd);
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

static void highbd_convolve_add_src_vert(const uint8_t *src8,
                                         ptrdiff_t src_stride, uint8_t *dst8,
                                         ptrdiff_t dst_stride,
                                         const InterpKernel *y_filters,
                                         int y0_q4, int y_step_q4, int w, int h,
                                         int bd) {
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
  for (x = 0; x < w; ++x) {
    int y_q4 = y0_q4;
    for (y = 0; y < h; ++y) {
      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      dst[y * dst_stride] =
          clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS) +
                                src_y[(SUBPEL_TAPS / 2 - 1) * src_stride],
                            bd);
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

static void highbd_convolve_add_src(const uint8_t *src, ptrdiff_t src_stride,
                                    uint8_t *dst, ptrdiff_t dst_stride,
                                    const InterpKernel *const x_filters,
                                    int x0_q4, int x_step_q4,
                                    const InterpKernel *const y_filters,
                                    int y0_q4, int y_step_q4, int w, int h,
                                    int bd) {
  // Note: Fixed size intermediate buffer, temp, places limits on parameters.
  // 2d filtering proceeds in 2 steps:
  //   (1) Interpolate horizontally into an intermediate buffer, temp.
  //   (2) Interpolate temp vertically to derive the sub-pixel result.
  // Deriving the maximum number of rows in the temp buffer (135):
  // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
  // --Largest block size is 64x64 pixels.
  // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
  //   original frame (in 1/16th pixel units).
  // --Must round-up because block may be located at sub-pixel position.
  // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
  // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
  uint16_t temp[MAX_EXT_SIZE * MAX_SB_SIZE];
  int intermediate_height =
      (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;

  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);
  assert(y_step_q4 <= 32);
  assert(x_step_q4 <= 32);

  highbd_convolve_add_src_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1),
                                src_stride, CONVERT_TO_BYTEPTR(temp),
                                MAX_SB_SIZE, x_filters, x0_q4, x_step_q4, w,
                                intermediate_height, bd);
  highbd_convolve_add_src_vert(
      CONVERT_TO_BYTEPTR(temp) + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1),
      MAX_SB_SIZE, dst, dst_stride, y_filters, y0_q4, y_step_q4, w, h, bd);
}

void aom_highbd_convolve8_add_src_horiz_c(
    const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
    ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4,
    const int16_t *filter_y, int y_step_q4, int w, int h, int bd) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);
  (void)filter_y;
  (void)y_step_q4;

  highbd_convolve_add_src_horiz(src, src_stride, dst, dst_stride, filters_x,
                                x0_q4, x_step_q4, w, h, bd);
}

void aom_highbd_convolve8_add_src_vert_c(const uint8_t *src,
                                         ptrdiff_t src_stride, uint8_t *dst,
                                         ptrdiff_t dst_stride,
                                         const int16_t *filter_x, int x_step_q4,
                                         const int16_t *filter_y, int y_step_q4,
                                         int w, int h, int bd) {
  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);
  (void)filter_x;
  (void)x_step_q4;

  highbd_convolve_add_src_vert(src, src_stride, dst, dst_stride, filters_y,
                               y0_q4, y_step_q4, w, h, bd);
}

void aom_highbd_convolve8_add_src_c(const uint8_t *src, ptrdiff_t src_stride,
                                    uint8_t *dst, ptrdiff_t dst_stride,
                                    const int16_t *filter_x, int x_step_q4,
                                    const int16_t *filter_y, int y_step_q4,
                                    int w, int h, int bd) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

  highbd_convolve_add_src(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                          x_step_q4, filters_y, y0_q4, y_step_q4, w, h, bd);
}
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099

static void highbd_convolve_add_src_horiz_hip(
    const uint8_t *src8, ptrdiff_t src_stride, uint16_t *dst,
    ptrdiff_t dst_stride, const InterpKernel *x_filters, int x0_q4,
    int x_step_q4, int w, int h, int bd) {
  const int extraprec_clamp_limit = (EXTRAPREC_CLAMP_LIMIT << (bd - 8));
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = x0_q4;
    for (x = 0; x < w; ++x) {
      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
      int k, sum = ((int)src_x[SUBPEL_TAPS / 2 - 1] << FILTER_BITS);
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
      dst[x] =
          (uint16_t)clamp(ROUND_POWER_OF_TWO(sum, FILTER_BITS - EXTRAPREC_BITS),
                          0, extraprec_clamp_limit - 1);
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

static void highbd_convolve_add_src_vert_hip(
    const uint16_t *src, ptrdiff_t src_stride, uint8_t *dst8,
    ptrdiff_t dst_stride, const InterpKernel *y_filters, int y0_q4,
    int y_step_q4, int w, int h, int bd) {
  int x, y;
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
  for (x = 0; x < w; ++x) {
    int y_q4 = y0_q4;
    for (y = 0; y < h; ++y) {
      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
      int k,
          sum = ((int)src_y[(SUBPEL_TAPS / 2 - 1) * src_stride] << FILTER_BITS);
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      dst[y * dst_stride] = clip_pixel_highbd(
          ROUND_POWER_OF_TWO(sum, FILTER_BITS + EXTRAPREC_BITS), bd);
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

static void highbd_convolve_add_src_hip(
    const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
    ptrdiff_t dst_stride, const InterpKernel *const x_filters, int x0_q4,
    int x_step_q4, const InterpKernel *const y_filters, int y0_q4,
    int y_step_q4, int w, int h, int bd) {
  // Note: Fixed size intermediate buffer, temp, places limits on parameters.
  // 2d filtering proceeds in 2 steps:
  //   (1) Interpolate horizontally into an intermediate buffer, temp.
  //   (2) Interpolate temp vertically to derive the sub-pixel result.
  // Deriving the maximum number of rows in the temp buffer (135):
  // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
  // --Largest block size is 64x64 pixels.
  // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
  //   original frame (in 1/16th pixel units).
  // --Must round-up because block may be located at sub-pixel position.
  // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
  // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
  uint16_t temp[MAX_EXT_SIZE * MAX_SB_SIZE];
  int intermediate_height =
      (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;

  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);
  assert(y_step_q4 <= 32);
  assert(x_step_q4 <= 32);

  highbd_convolve_add_src_horiz_hip(
      src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride, temp, MAX_SB_SIZE,
      x_filters, x0_q4, x_step_q4, w, intermediate_height, bd);
  highbd_convolve_add_src_vert_hip(temp + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1),
                                   MAX_SB_SIZE, dst, dst_stride, y_filters,
                                   y0_q4, y_step_q4, w, h, bd);
}

void aom_highbd_convolve8_add_src_horiz_hip_c(
    const uint8_t *src, ptrdiff_t src_stride, uint16_t *dst,
    ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4,
    const int16_t *filter_y, int y_step_q4, int w, int h, int bd) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);
  (void)filter_y;
  (void)y_step_q4;

  highbd_convolve_add_src_horiz_hip(src, src_stride, dst, dst_stride, filters_x,
                                    x0_q4, x_step_q4, w, h, bd);
}

void aom_highbd_convolve8_add_src_vert_hip_c(
    const uint16_t *src, ptrdiff_t src_stride, uint8_t *dst,
    ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4,
    const int16_t *filter_y, int y_step_q4, int w, int h, int bd) {
  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);
  (void)filter_x;
  (void)x_step_q4;

  highbd_convolve_add_src_vert_hip(src, src_stride, dst, dst_stride, filters_y,
                                   y0_q4, y_step_q4, w, h, bd);
}

void aom_highbd_convolve8_add_src_hip_c(const uint8_t *src,
                                        ptrdiff_t src_stride, uint8_t *dst,
                                        ptrdiff_t dst_stride,
                                        const int16_t *filter_x, int x_step_q4,
                                        const int16_t *filter_y, int y_step_q4,
                                        int w, int h, int bd) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

  highbd_convolve_add_src_hip(src, src_stride, dst, dst_stride, filters_x,
                              x0_q4, x_step_q4, filters_y, y0_q4, y_step_q4, w,
                              h, bd);
}

1100
#endif  // CONFIG_LOOP_RESTORATION
1101
#endif  // CONFIG_HIGHBITDEPTH