warped_motion.c 69.1 KB
Newer Older
1
/*
2
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3
 *
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.
10
11
12
13
14
15
16
17
 */

#include <stdio.h>
#include <stdlib.h>
#include <memory.h>
#include <math.h>
#include <assert.h>

18
#include "./av1_rtcd.h"
19
#include "av1/common/warped_motion.h"
20

21
22
/* clang-format off */
static const int error_measure_lut[512] = {
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
  // pow 0.7
  16384, 16339, 16294, 16249, 16204, 16158, 16113, 16068,
  16022, 15977, 15932, 15886, 15840, 15795, 15749, 15703,
  15657, 15612, 15566, 15520, 15474, 15427, 15381, 15335,
  15289, 15242, 15196, 15149, 15103, 15056, 15010, 14963,
  14916, 14869, 14822, 14775, 14728, 14681, 14634, 14587,
  14539, 14492, 14445, 14397, 14350, 14302, 14254, 14206,
  14159, 14111, 14063, 14015, 13967, 13918, 13870, 13822,
  13773, 13725, 13676, 13628, 13579, 13530, 13481, 13432,
  13383, 13334, 13285, 13236, 13187, 13137, 13088, 13038,
  12988, 12939, 12889, 12839, 12789, 12739, 12689, 12639,
  12588, 12538, 12487, 12437, 12386, 12335, 12285, 12234,
  12183, 12132, 12080, 12029, 11978, 11926, 11875, 11823,
  11771, 11719, 11667, 11615, 11563, 11511, 11458, 11406,
  11353, 11301, 11248, 11195, 11142, 11089, 11036, 10982,
  10929, 10875, 10822, 10768, 10714, 10660, 10606, 10552,
  10497, 10443, 10388, 10333, 10279, 10224, 10168, 10113,
  10058, 10002,  9947,  9891,  9835,  9779,  9723,  9666,
  9610, 9553, 9497, 9440, 9383, 9326, 9268, 9211,
  9153, 9095, 9037, 8979, 8921, 8862, 8804, 8745,
  8686, 8627, 8568, 8508, 8449, 8389, 8329, 8269,
  8208, 8148, 8087, 8026, 7965, 7903, 7842, 7780,
  7718, 7656, 7593, 7531, 7468, 7405, 7341, 7278,
  7214, 7150, 7086, 7021, 6956, 6891, 6826, 6760,
  6695, 6628, 6562, 6495, 6428, 6361, 6293, 6225,
  6157, 6089, 6020, 5950, 5881, 5811, 5741, 5670,
  5599, 5527, 5456, 5383, 5311, 5237, 5164, 5090,
  5015, 4941, 4865, 4789, 4713, 4636, 4558, 4480,
  4401, 4322, 4242, 4162, 4080, 3998, 3916, 3832,
  3748, 3663, 3577, 3490, 3402, 3314, 3224, 3133,
  3041, 2948, 2854, 2758, 2661, 2562, 2461, 2359,
  2255, 2148, 2040, 1929, 1815, 1698, 1577, 1452,
  1323, 1187, 1045,  894,  731,  550,  339,    0,
  339,  550,  731,  894, 1045, 1187, 1323, 1452,
  1577, 1698, 1815, 1929, 2040, 2148, 2255, 2359,
  2461, 2562, 2661, 2758, 2854, 2948, 3041, 3133,
  3224, 3314, 3402, 3490, 3577, 3663, 3748, 3832,
  3916, 3998, 4080, 4162, 4242, 4322, 4401, 4480,
  4558, 4636, 4713, 4789, 4865, 4941, 5015, 5090,
  5164, 5237, 5311, 5383, 5456, 5527, 5599, 5670,
  5741, 5811, 5881, 5950, 6020, 6089, 6157, 6225,
  6293, 6361, 6428, 6495, 6562, 6628, 6695, 6760,
  6826, 6891, 6956, 7021, 7086, 7150, 7214, 7278,
  7341, 7405, 7468, 7531, 7593, 7656, 7718, 7780,
  7842, 7903, 7965, 8026, 8087, 8148, 8208, 8269,
  8329, 8389, 8449, 8508, 8568, 8627, 8686, 8745,
  8804, 8862, 8921, 8979, 9037, 9095, 9153, 9211,
  9268, 9326, 9383, 9440, 9497, 9553, 9610, 9666,
  9723,  9779,  9835,  9891,  9947, 10002, 10058, 10113,
  10168, 10224, 10279, 10333, 10388, 10443, 10497, 10552,
  10606, 10660, 10714, 10768, 10822, 10875, 10929, 10982,
  11036, 11089, 11142, 11195, 11248, 11301, 11353, 11406,
  11458, 11511, 11563, 11615, 11667, 11719, 11771, 11823,
  11875, 11926, 11978, 12029, 12080, 12132, 12183, 12234,
  12285, 12335, 12386, 12437, 12487, 12538, 12588, 12639,
  12689, 12739, 12789, 12839, 12889, 12939, 12988, 13038,
  13088, 13137, 13187, 13236, 13285, 13334, 13383, 13432,
  13481, 13530, 13579, 13628, 13676, 13725, 13773, 13822,
  13870, 13918, 13967, 14015, 14063, 14111, 14159, 14206,
  14254, 14302, 14350, 14397, 14445, 14492, 14539, 14587,
  14634, 14681, 14728, 14775, 14822, 14869, 14916, 14963,
  15010, 15056, 15103, 15149, 15196, 15242, 15289, 15335,
  15381, 15427, 15474, 15520, 15566, 15612, 15657, 15703,
  15749, 15795, 15840, 15886, 15932, 15977, 16022, 16068,
  16113, 16158, 16204, 16249, 16294, 16339, 16384, 16384,
88
89
};
/* clang-format on */
90

91
static ProjectPointsFunc get_project_points_type(TransformationType type) {
92
  switch (type) {
93
94
95
96
    case HOMOGRAPHY: return project_points_homography;
    case AFFINE: return project_points_affine;
    case ROTZOOM: return project_points_rotzoom;
    case TRANSLATION: return project_points_translation;
clang-format's avatar
clang-format committed
97
    default: assert(0); return NULL;
98
99
100
  }
}

101
void project_points_translation(int32_t *mat, int *points, int *proj,
102
103
104
                                const int n, const int stride_points,
                                const int stride_proj, const int subsampling_x,
                                const int subsampling_y) {
105
106
107
108
  int i;
  for (i = 0; i < n; ++i) {
    const int x = *(points++), y = *(points++);
    if (subsampling_x)
109
      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
110
          ((x * (1 << (WARPEDMODEL_PREC_BITS + 1))) + mat[0]),
111
          WARPEDDIFF_PREC_BITS + 1);
112
    else
113
      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
114
          ((x * (1 << WARPEDMODEL_PREC_BITS)) + mat[0]), WARPEDDIFF_PREC_BITS);
115
    if (subsampling_y)
116
      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
117
          ((y * (1 << (WARPEDMODEL_PREC_BITS + 1))) + mat[1]),
118
          WARPEDDIFF_PREC_BITS + 1);
119
    else
120
      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
121
          ((y * (1 << WARPEDMODEL_PREC_BITS))) + mat[1], WARPEDDIFF_PREC_BITS);
122
123
124
125
126
    points += stride_points - 2;
    proj += stride_proj - 2;
  }
}

127
void project_points_rotzoom(int32_t *mat, int *points, int *proj, const int n,
128
129
                            const int stride_points, const int stride_proj,
                            const int subsampling_x, const int subsampling_y) {
130
131
132
133
  int i;
  for (i = 0; i < n; ++i) {
    const int x = *(points++), y = *(points++);
    if (subsampling_x)
134
      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
135
136
          mat[2] * 2 * x + mat[3] * 2 * y + mat[0] +
              (mat[2] + mat[3] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
137
138
          WARPEDDIFF_PREC_BITS + 1);
    else
139
      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(mat[2] * x + mat[3] * y + mat[0],
140
                                            WARPEDDIFF_PREC_BITS);
141
    if (subsampling_y)
142
      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
143
144
          -mat[3] * 2 * x + mat[2] * 2 * y + mat[1] +
              (-mat[3] + mat[2] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
145
146
          WARPEDDIFF_PREC_BITS + 1);
    else
147
      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(-mat[3] * x + mat[2] * y + mat[1],
148
                                            WARPEDDIFF_PREC_BITS);
149
150
151
152
153
    points += stride_points - 2;
    proj += stride_proj - 2;
  }
}

154
void project_points_affine(int32_t *mat, int *points, int *proj, const int n,
155
156
                           const int stride_points, const int stride_proj,
                           const int subsampling_x, const int subsampling_y) {
157
158
159
160
  int i;
  for (i = 0; i < n; ++i) {
    const int x = *(points++), y = *(points++);
    if (subsampling_x)
161
      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
162
163
          mat[2] * 2 * x + mat[3] * 2 * y + mat[0] +
              (mat[2] + mat[3] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
164
165
          WARPEDDIFF_PREC_BITS + 1);
    else
166
      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(mat[2] * x + mat[3] * y + mat[0],
167
                                            WARPEDDIFF_PREC_BITS);
168
    if (subsampling_y)
169
      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
170
          mat[4] * 2 * x + mat[5] * 2 * y + mat[1] +
Sarah Parker's avatar
Sarah Parker committed
171
              (mat[4] + mat[5] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
172
173
          WARPEDDIFF_PREC_BITS + 1);
    else
174
      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(mat[4] * x + mat[5] * y + mat[1],
175
                                            WARPEDDIFF_PREC_BITS);
176
177
178
179
180
    points += stride_points - 2;
    proj += stride_proj - 2;
  }
}

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
void project_points_hortrapezoid(int32_t *mat, int *points, int *proj,
                                 const int n, const int stride_points,
                                 const int stride_proj, const int subsampling_x,
                                 const int subsampling_y) {
  int i;
  int64_t x, y, Z;
  int64_t xp, yp;
  for (i = 0; i < n; ++i) {
    x = *(points++), y = *(points++);
    x = (subsampling_x ? 4 * x + 1 : 2 * x);
    y = (subsampling_y ? 4 * y + 1 : 2 * y);

    Z = (mat[7] * y + (1 << (WARPEDMODEL_ROW3HOMO_PREC_BITS + 1)));
    xp = (mat[2] * x + mat[3] * y + 2 * mat[0]) *
         (1 << (WARPEDPIXEL_PREC_BITS + WARPEDMODEL_ROW3HOMO_PREC_BITS -
                WARPEDMODEL_PREC_BITS));
    yp = (mat[5] * y + 2 * mat[1]) *
         (1 << (WARPEDPIXEL_PREC_BITS + WARPEDMODEL_ROW3HOMO_PREC_BITS -
                WARPEDMODEL_PREC_BITS));

    xp = xp > 0 ? (xp + Z / 2) / Z : (xp - Z / 2) / Z;
    yp = yp > 0 ? (yp + Z / 2) / Z : (yp - Z / 2) / Z;

    if (subsampling_x) xp = (xp - (1 << (WARPEDPIXEL_PREC_BITS - 1))) / 2;
    if (subsampling_y) yp = (yp - (1 << (WARPEDPIXEL_PREC_BITS - 1))) / 2;
    *(proj++) = xp;
    *(proj++) = yp;

    points += stride_points - 2;
    proj += stride_proj - 2;
  }
}

void project_points_vertrapezoid(int32_t *mat, int *points, int *proj,
                                 const int n, const int stride_points,
                                 const int stride_proj, const int subsampling_x,
                                 const int subsampling_y) {
  int i;
  int64_t x, y, Z;
  int64_t xp, yp;
  for (i = 0; i < n; ++i) {
    x = *(points++), y = *(points++);
    x = (subsampling_x ? 4 * x + 1 : 2 * x);
    y = (subsampling_y ? 4 * y + 1 : 2 * y);

    Z = (mat[6] * x + (1 << (WARPEDMODEL_ROW3HOMO_PREC_BITS + 1)));
    xp = (mat[2] * x + 2 * mat[0]) *
         (1 << (WARPEDPIXEL_PREC_BITS + WARPEDMODEL_ROW3HOMO_PREC_BITS -
                WARPEDMODEL_PREC_BITS));
    yp = (mat[4] * x + mat[5] * y + 2 * mat[1]) *
         (1 << (WARPEDPIXEL_PREC_BITS + WARPEDMODEL_ROW3HOMO_PREC_BITS -
                WARPEDMODEL_PREC_BITS));

    xp = xp > 0 ? (xp + Z / 2) / Z : (xp - Z / 2) / Z;
    yp = yp > 0 ? (yp + Z / 2) / Z : (yp - Z / 2) / Z;

    if (subsampling_x) xp = (xp - (1 << (WARPEDPIXEL_PREC_BITS - 1))) / 2;
    if (subsampling_y) yp = (yp - (1 << (WARPEDPIXEL_PREC_BITS - 1))) / 2;
    *(proj++) = xp;
    *(proj++) = yp;

    points += stride_points - 2;
    proj += stride_proj - 2;
  }
}

247
void project_points_homography(int32_t *mat, int *points, int *proj,
248
249
250
                               const int n, const int stride_points,
                               const int stride_proj, const int subsampling_x,
                               const int subsampling_y) {
251
  int i;
252
253
  int64_t x, y, Z;
  int64_t xp, yp;
254
255
256
257
258
  for (i = 0; i < n; ++i) {
    x = *(points++), y = *(points++);
    x = (subsampling_x ? 4 * x + 1 : 2 * x);
    y = (subsampling_y ? 4 * y + 1 : 2 * y);

259
260
    Z = (mat[6] * x + mat[7] * y + (1 << (WARPEDMODEL_ROW3HOMO_PREC_BITS + 1)));
    xp = (mat[2] * x + mat[3] * y + 2 * mat[0]) *
261
262
         (1 << (WARPEDPIXEL_PREC_BITS + WARPEDMODEL_ROW3HOMO_PREC_BITS -
                WARPEDMODEL_PREC_BITS));
263
    yp = (mat[4] * x + mat[5] * y + 2 * mat[1]) *
264
265
         (1 << (WARPEDPIXEL_PREC_BITS + WARPEDMODEL_ROW3HOMO_PREC_BITS -
                WARPEDMODEL_PREC_BITS));
266
267
268
269

    xp = xp > 0 ? (xp + Z / 2) / Z : (xp - Z / 2) / Z;
    yp = yp > 0 ? (yp + Z / 2) / Z : (yp - Z / 2) / Z;

clang-format's avatar
clang-format committed
270
271
    if (subsampling_x) xp = (xp - (1 << (WARPEDPIXEL_PREC_BITS - 1))) / 2;
    if (subsampling_y) yp = (yp - (1 << (WARPEDPIXEL_PREC_BITS - 1))) / 2;
272
273
274
275
276
277
278
279
    *(proj++) = xp;
    *(proj++) = yp;

    points += stride_points - 2;
    proj += stride_proj - 2;
  }
}

Yue Chen's avatar
Yue Chen committed
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
// 'points' are at original scale, output 'proj's are scaled up by
// 1 << WARPEDPIXEL_PREC_BITS
void project_points(WarpedMotionParams *wm_params, int *points, int *proj,
                    const int n, const int stride_points, const int stride_proj,
                    const int subsampling_x, const int subsampling_y) {
  switch (wm_params->wmtype) {
    case AFFINE:
      project_points_affine(wm_params->wmmat, points, proj, n, stride_points,
                            stride_proj, subsampling_x, subsampling_y);
      break;
    case ROTZOOM:
      project_points_rotzoom(wm_params->wmmat, points, proj, n, stride_points,
                             stride_proj, subsampling_x, subsampling_y);
      break;
    case HOMOGRAPHY:
      project_points_homography(wm_params->wmmat, points, proj, n,
                                stride_points, stride_proj, subsampling_x,
                                subsampling_y);
      break;
    default: assert(0 && "Invalid warped motion type!"); return;
  }
}

303
static const int16_t
clang-format's avatar
clang-format committed
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
    filter_ntap[WARPEDPIXEL_PREC_SHIFTS][WARPEDPIXEL_FILTER_TAPS] = {
      { 0, 0, 128, 0, 0, 0 },      { 0, -1, 128, 2, -1, 0 },
      { 1, -3, 127, 4, -1, 0 },    { 1, -4, 126, 6, -2, 1 },
      { 1, -5, 126, 8, -3, 1 },    { 1, -6, 125, 11, -4, 1 },
      { 1, -7, 124, 13, -4, 1 },   { 2, -8, 123, 15, -5, 1 },
      { 2, -9, 122, 18, -6, 1 },   { 2, -10, 121, 20, -6, 1 },
      { 2, -11, 120, 22, -7, 2 },  { 2, -12, 119, 25, -8, 2 },
      { 3, -13, 117, 27, -8, 2 },  { 3, -13, 116, 29, -9, 2 },
      { 3, -14, 114, 32, -10, 3 }, { 3, -15, 113, 35, -10, 2 },
      { 3, -15, 111, 37, -11, 3 }, { 3, -16, 109, 40, -11, 3 },
      { 3, -16, 108, 42, -12, 3 }, { 4, -17, 106, 45, -13, 3 },
      { 4, -17, 104, 47, -13, 3 }, { 4, -17, 102, 50, -14, 3 },
      { 4, -17, 100, 52, -14, 3 }, { 4, -18, 98, 55, -15, 4 },
      { 4, -18, 96, 58, -15, 3 },  { 4, -18, 94, 60, -16, 4 },
      { 4, -18, 91, 63, -16, 4 },  { 4, -18, 89, 65, -16, 4 },
      { 4, -18, 87, 68, -17, 4 },  { 4, -18, 85, 70, -17, 4 },
      { 4, -18, 82, 73, -17, 4 },  { 4, -18, 80, 75, -17, 4 },
      { 4, -18, 78, 78, -18, 4 },  { 4, -17, 75, 80, -18, 4 },
      { 4, -17, 73, 82, -18, 4 },  { 4, -17, 70, 85, -18, 4 },
      { 4, -17, 68, 87, -18, 4 },  { 4, -16, 65, 89, -18, 4 },
      { 4, -16, 63, 91, -18, 4 },  { 4, -16, 60, 94, -18, 4 },
      { 3, -15, 58, 96, -18, 4 },  { 4, -15, 55, 98, -18, 4 },
      { 3, -14, 52, 100, -17, 4 }, { 3, -14, 50, 102, -17, 4 },
      { 3, -13, 47, 104, -17, 4 }, { 3, -13, 45, 106, -17, 4 },
      { 3, -12, 42, 108, -16, 3 }, { 3, -11, 40, 109, -16, 3 },
      { 3, -11, 37, 111, -15, 3 }, { 2, -10, 35, 113, -15, 3 },
      { 3, -10, 32, 114, -14, 3 }, { 2, -9, 29, 116, -13, 3 },
      { 2, -8, 27, 117, -13, 3 },  { 2, -8, 25, 119, -12, 2 },
      { 2, -7, 22, 120, -11, 2 },  { 1, -6, 20, 121, -10, 2 },
      { 1, -6, 18, 122, -9, 2 },   { 1, -5, 15, 123, -8, 2 },
      { 1, -4, 13, 124, -7, 1 },   { 1, -4, 11, 125, -6, 1 },
      { 1, -3, 8, 126, -5, 1 },    { 1, -2, 6, 126, -4, 1 },
      { 0, -1, 4, 127, -3, 1 },    { 0, -1, 2, 128, -1, 0 },
    };
338
339
340
341
342
343
344
345
346
347
348

static int32_t do_ntap_filter(int32_t *p, int x) {
  int i;
  int32_t sum = 0;
  for (i = 0; i < WARPEDPIXEL_FILTER_TAPS; ++i) {
    sum += p[i - WARPEDPIXEL_FILTER_TAPS / 2 + 1] * filter_ntap[x][i];
  }
  return sum;
}

static int32_t do_cubic_filter(int32_t *p, int x) {
clang-format's avatar
clang-format committed
349
  if (x == 0) {
350
    return p[0] * (1 << WARPEDPIXEL_FILTER_BITS);
351
  } else if (x == (1 << WARPEDPIXEL_PREC_BITS)) {
352
    return p[1] * (1 << WARPEDPIXEL_FILTER_BITS);
353
  } else {
354
    const int64_t v1 = (int64_t)x * x * x * (3 * (p[0] - p[1]) + p[2] - p[-1]);
355
356
357
    const int64_t v2 = x * x * (2 * p[-1] - 5 * p[0] + 4 * p[1] - p[2]);
    const int64_t v3 = x * (p[1] - p[-1]);
    const int64_t v4 = 2 * p[0];
358
    return (int32_t)ROUND_POWER_OF_TWO_SIGNED(
359
360
361
        (v4 * (1 << (3 * WARPEDPIXEL_PREC_BITS))) +
            (v3 * (1 << (2 * WARPEDPIXEL_PREC_BITS))) +
            (v2 * (1 << WARPEDPIXEL_PREC_BITS)) + v1,
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
        3 * WARPEDPIXEL_PREC_BITS + 1 - WARPEDPIXEL_FILTER_BITS);
  }
}

static INLINE void get_subcolumn(int taps, uint8_t *ref, int32_t *col,
                                 int stride, int x, int y_start) {
  int i;
  for (i = 0; i < taps; ++i) {
    col[i] = ref[(i + y_start) * stride + x];
  }
}

static uint8_t bi_ntap_filter(uint8_t *ref, int x, int y, int stride) {
  int32_t val, arr[WARPEDPIXEL_FILTER_TAPS];
  int k;
clang-format's avatar
clang-format committed
377
378
  int i = (int)x >> WARPEDPIXEL_PREC_BITS;
  int j = (int)y >> WARPEDPIXEL_PREC_BITS;
379
380
381
382
383
384
  for (k = 0; k < WARPEDPIXEL_FILTER_TAPS; ++k) {
    int32_t arr_temp[WARPEDPIXEL_FILTER_TAPS];
    get_subcolumn(WARPEDPIXEL_FILTER_TAPS, ref, arr_temp, stride,
                  i + k + 1 - WARPEDPIXEL_FILTER_TAPS / 2,
                  j + 1 - WARPEDPIXEL_FILTER_TAPS / 2);
    arr[k] = do_ntap_filter(arr_temp + WARPEDPIXEL_FILTER_TAPS / 2 - 1,
385
                            y - (j * (1 << WARPEDPIXEL_PREC_BITS)));
386
387
  }
  val = do_ntap_filter(arr + WARPEDPIXEL_FILTER_TAPS / 2 - 1,
388
                       x - (i * (1 << WARPEDPIXEL_PREC_BITS)));
389
390
391
392
393
394
395
  val = ROUND_POWER_OF_TWO_SIGNED(val, WARPEDPIXEL_FILTER_BITS * 2);
  return (uint8_t)clip_pixel(val);
}

static uint8_t bi_cubic_filter(uint8_t *ref, int x, int y, int stride) {
  int32_t val, arr[4];
  int k;
clang-format's avatar
clang-format committed
396
397
  int i = (int)x >> WARPEDPIXEL_PREC_BITS;
  int j = (int)y >> WARPEDPIXEL_PREC_BITS;
398
399
  for (k = 0; k < 4; ++k) {
    int32_t arr_temp[4];
clang-format's avatar
clang-format committed
400
    get_subcolumn(4, ref, arr_temp, stride, i + k - 1, j - 1);
401
402
    arr[k] =
        do_cubic_filter(arr_temp + 1, y - (j * (1 << WARPEDPIXEL_PREC_BITS)));
403
  }
404
  val = do_cubic_filter(arr + 1, x - (i * (1 << WARPEDPIXEL_PREC_BITS)));
405
406
407
408
409
410
411
  val = ROUND_POWER_OF_TWO_SIGNED(val, WARPEDPIXEL_FILTER_BITS * 2);
  return (uint8_t)clip_pixel(val);
}

static uint8_t bi_linear_filter(uint8_t *ref, int x, int y, int stride) {
  const int ix = x >> WARPEDPIXEL_PREC_BITS;
  const int iy = y >> WARPEDPIXEL_PREC_BITS;
412
413
  const int sx = x - (ix * (1 << WARPEDPIXEL_PREC_BITS));
  const int sy = y - (iy * (1 << WARPEDPIXEL_PREC_BITS));
414
415
416
  int32_t val;
  val = ROUND_POWER_OF_TWO_SIGNED(
      ref[iy * stride + ix] * (WARPEDPIXEL_PREC_SHIFTS - sy) *
clang-format's avatar
clang-format committed
417
418
419
420
              (WARPEDPIXEL_PREC_SHIFTS - sx) +
          ref[iy * stride + ix + 1] * (WARPEDPIXEL_PREC_SHIFTS - sy) * sx +
          ref[(iy + 1) * stride + ix] * sy * (WARPEDPIXEL_PREC_SHIFTS - sx) +
          ref[(iy + 1) * stride + ix + 1] * sy * sx,
421
422
423
424
      WARPEDPIXEL_PREC_BITS * 2);
  return (uint8_t)clip_pixel(val);
}

clang-format's avatar
clang-format committed
425
426
static uint8_t warp_interpolate(uint8_t *ref, int x, int y, int width,
                                int height, int stride) {
427
428
  int ix = x >> WARPEDPIXEL_PREC_BITS;
  int iy = y >> WARPEDPIXEL_PREC_BITS;
429
430
  int sx = x - (ix * (1 << WARPEDPIXEL_PREC_BITS));
  int sy = y - (iy * (1 << WARPEDPIXEL_PREC_BITS));
431
432
  int32_t v;

clang-format's avatar
clang-format committed
433
434
  if (ix < 0 && iy < 0)
    return ref[0];
Yue Chen's avatar
Yue Chen committed
435
  else if (ix < 0 && iy >= height - 1)
436
    return ref[(height - 1) * stride];
Yue Chen's avatar
Yue Chen committed
437
  else if (ix >= width - 1 && iy < 0)
438
    return ref[width - 1];
Yue Chen's avatar
Yue Chen committed
439
  else if (ix >= width - 1 && iy >= height - 1)
440
441
442
443
    return ref[(height - 1) * stride + (width - 1)];
  else if (ix < 0) {
    v = ROUND_POWER_OF_TWO_SIGNED(
        ref[iy * stride] * (WARPEDPIXEL_PREC_SHIFTS - sy) +
clang-format's avatar
clang-format committed
444
            ref[(iy + 1) * stride] * sy,
445
446
447
448
        WARPEDPIXEL_PREC_BITS);
    return clip_pixel(v);
  } else if (iy < 0) {
    v = ROUND_POWER_OF_TWO_SIGNED(
clang-format's avatar
clang-format committed
449
        ref[ix] * (WARPEDPIXEL_PREC_SHIFTS - sx) + ref[ix + 1] * sx,
450
451
        WARPEDPIXEL_PREC_BITS);
    return clip_pixel(v);
Yue Chen's avatar
Yue Chen committed
452
  } else if (ix >= width - 1) {
453
454
    v = ROUND_POWER_OF_TWO_SIGNED(
        ref[iy * stride + width - 1] * (WARPEDPIXEL_PREC_SHIFTS - sy) +
clang-format's avatar
clang-format committed
455
            ref[(iy + 1) * stride + width - 1] * sy,
456
457
        WARPEDPIXEL_PREC_BITS);
    return clip_pixel(v);
Yue Chen's avatar
Yue Chen committed
458
  } else if (iy >= height - 1) {
459
460
    v = ROUND_POWER_OF_TWO_SIGNED(
        ref[(height - 1) * stride + ix] * (WARPEDPIXEL_PREC_SHIFTS - sx) +
clang-format's avatar
clang-format committed
461
            ref[(height - 1) * stride + ix + 1] * sx,
462
463
464
465
466
467
468
        WARPEDPIXEL_PREC_BITS);
    return clip_pixel(v);
  } else if (ix >= WARPEDPIXEL_FILTER_TAPS / 2 - 1 &&
             iy >= WARPEDPIXEL_FILTER_TAPS / 2 - 1 &&
             ix < width - WARPEDPIXEL_FILTER_TAPS / 2 &&
             iy < height - WARPEDPIXEL_FILTER_TAPS / 2) {
    return bi_ntap_filter(ref, x, y, stride);
clang-format's avatar
clang-format committed
469
  } else if (ix >= 1 && iy >= 1 && ix < width - 2 && iy < height - 2) {
470
471
472
473
474
475
    return bi_cubic_filter(ref, x, y, stride);
  } else {
    return bi_linear_filter(ref, x, y, stride);
  }
}

476
477
478
479
480
481
// For warping, we really use a 6-tap filter, but we do blocks of 8 pixels
// at a time. The zoom/rotation/shear in the model are applied to the
// "fractional" position of each pixel, which therefore varies within
// [-1, 2) * WARPEDPIXEL_PREC_SHIFTS.
// We need an extra 2 taps to fit this in, for a total of 8 taps.
/* clang-format off */
482
const int16_t warped_filter[WARPEDPIXEL_PREC_SHIFTS * 3 + 1][8] = {
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
  // [-1, 0)
  { 0,   0, 128,   0,   0, 0, 0, 0 }, { 0, - 1, 128,   2, - 1, 0, 0, 0 },
  { 1, - 3, 127,   4, - 1, 0, 0, 0 }, { 1, - 4, 126,   6, - 2, 1, 0, 0 },
  { 1, - 5, 126,   8, - 3, 1, 0, 0 }, { 1, - 6, 125,  11, - 4, 1, 0, 0 },
  { 1, - 7, 124,  13, - 4, 1, 0, 0 }, { 2, - 8, 123,  15, - 5, 1, 0, 0 },
  { 2, - 9, 122,  18, - 6, 1, 0, 0 }, { 2, -10, 121,  20, - 6, 1, 0, 0 },
  { 2, -11, 120,  22, - 7, 2, 0, 0 }, { 2, -12, 119,  25, - 8, 2, 0, 0 },
  { 3, -13, 117,  27, - 8, 2, 0, 0 }, { 3, -13, 116,  29, - 9, 2, 0, 0 },
  { 3, -14, 114,  32, -10, 3, 0, 0 }, { 3, -15, 113,  35, -10, 2, 0, 0 },
  { 3, -15, 111,  37, -11, 3, 0, 0 }, { 3, -16, 109,  40, -11, 3, 0, 0 },
  { 3, -16, 108,  42, -12, 3, 0, 0 }, { 4, -17, 106,  45, -13, 3, 0, 0 },
  { 4, -17, 104,  47, -13, 3, 0, 0 }, { 4, -17, 102,  50, -14, 3, 0, 0 },
  { 4, -17, 100,  52, -14, 3, 0, 0 }, { 4, -18,  98,  55, -15, 4, 0, 0 },
  { 4, -18,  96,  58, -15, 3, 0, 0 }, { 4, -18,  94,  60, -16, 4, 0, 0 },
  { 4, -18,  91,  63, -16, 4, 0, 0 }, { 4, -18,  89,  65, -16, 4, 0, 0 },
  { 4, -18,  87,  68, -17, 4, 0, 0 }, { 4, -18,  85,  70, -17, 4, 0, 0 },
  { 4, -18,  82,  73, -17, 4, 0, 0 }, { 4, -18,  80,  75, -17, 4, 0, 0 },
  { 4, -18,  78,  78, -18, 4, 0, 0 }, { 4, -17,  75,  80, -18, 4, 0, 0 },
  { 4, -17,  73,  82, -18, 4, 0, 0 }, { 4, -17,  70,  85, -18, 4, 0, 0 },
  { 4, -17,  68,  87, -18, 4, 0, 0 }, { 4, -16,  65,  89, -18, 4, 0, 0 },
  { 4, -16,  63,  91, -18, 4, 0, 0 }, { 4, -16,  60,  94, -18, 4, 0, 0 },
  { 3, -15,  58,  96, -18, 4, 0, 0 }, { 4, -15,  55,  98, -18, 4, 0, 0 },
  { 3, -14,  52, 100, -17, 4, 0, 0 }, { 3, -14,  50, 102, -17, 4, 0, 0 },
  { 3, -13,  47, 104, -17, 4, 0, 0 }, { 3, -13,  45, 106, -17, 4, 0, 0 },
  { 3, -12,  42, 108, -16, 3, 0, 0 }, { 3, -11,  40, 109, -16, 3, 0, 0 },
  { 3, -11,  37, 111, -15, 3, 0, 0 }, { 2, -10,  35, 113, -15, 3, 0, 0 },
  { 3, -10,  32, 114, -14, 3, 0, 0 }, { 2, - 9,  29, 116, -13, 3, 0, 0 },
  { 2, - 8,  27, 117, -13, 3, 0, 0 }, { 2, - 8,  25, 119, -12, 2, 0, 0 },
  { 2, - 7,  22, 120, -11, 2, 0, 0 }, { 1, - 6,  20, 121, -10, 2, 0, 0 },
  { 1, - 6,  18, 122, - 9, 2, 0, 0 }, { 1, - 5,  15, 123, - 8, 2, 0, 0 },
  { 1, - 4,  13, 124, - 7, 1, 0, 0 }, { 1, - 4,  11, 125, - 6, 1, 0, 0 },
  { 1, - 3,   8, 126, - 5, 1, 0, 0 }, { 1, - 2,   6, 126, - 4, 1, 0, 0 },
  { 0, - 1,   4, 127, - 3, 1, 0, 0 }, { 0, - 1,   2, 128, - 1, 0, 0, 0 },

  // [0, 1)
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
  { 0,  0,   0, 128,   0,   0,  0,  0}, { 0,  1,  -2, 128,   2,  -1,  0,  0},
  { 0,  1,  -3, 127,   4,  -2,  1,  0}, { 0,  1,  -5, 127,   6,  -2,  1,  0},
  { 0,  2,  -6, 126,   8,  -3,  1,  0}, {-1,  2,  -7, 126,  11,  -4,  2, -1},
  {-1,  3,  -8, 125,  13,  -5,  2, -1}, {-1,  3, -10, 124,  16,  -6,  3, -1},
  {-1,  4, -11, 123,  18,  -7,  3, -1}, {-1,  4, -12, 122,  20,  -7,  3, -1},
  {-1,  4, -13, 121,  23,  -8,  3, -1}, {-2,  5, -14, 120,  25,  -9,  4, -1},
  {-1,  5, -15, 119,  27, -10,  4, -1}, {-1,  5, -16, 118,  30, -11,  4, -1},
  {-2,  6, -17, 116,  33, -12,  5, -1}, {-2,  6, -17, 114,  35, -12,  5, -1},
  {-2,  6, -18, 113,  38, -13,  5, -1}, {-2,  7, -19, 111,  41, -14,  6, -2},
  {-2,  7, -19, 110,  43, -15,  6, -2}, {-2,  7, -20, 108,  46, -15,  6, -2},
  {-2,  7, -20, 106,  49, -16,  6, -2}, {-2,  7, -21, 104,  51, -16,  7, -2},
  {-2,  7, -21, 102,  54, -17,  7, -2}, {-2,  8, -21, 100,  56, -18,  7, -2},
  {-2,  8, -22,  98,  59, -18,  7, -2}, {-2,  8, -22,  96,  62, -19,  7, -2},
  {-2,  8, -22,  94,  64, -19,  7, -2}, {-2,  8, -22,  91,  67, -20,  8, -2},
  {-2,  8, -22,  89,  69, -20,  8, -2}, {-2,  8, -22,  87,  72, -21,  8, -2},
  {-2,  8, -21,  84,  74, -21,  8, -2}, {-2,  8, -22,  82,  77, -21,  8, -2},
  {-2,  8, -21,  79,  79, -21,  8, -2}, {-2,  8, -21,  77,  82, -22,  8, -2},
  {-2,  8, -21,  74,  84, -21,  8, -2}, {-2,  8, -21,  72,  87, -22,  8, -2},
  {-2,  8, -20,  69,  89, -22,  8, -2}, {-2,  8, -20,  67,  91, -22,  8, -2},
  {-2,  7, -19,  64,  94, -22,  8, -2}, {-2,  7, -19,  62,  96, -22,  8, -2},
  {-2,  7, -18,  59,  98, -22,  8, -2}, {-2,  7, -18,  56, 100, -21,  8, -2},
  {-2,  7, -17,  54, 102, -21,  7, -2}, {-2,  7, -16,  51, 104, -21,  7, -2},
  {-2,  6, -16,  49, 106, -20,  7, -2}, {-2,  6, -15,  46, 108, -20,  7, -2},
  {-2,  6, -15,  43, 110, -19,  7, -2}, {-2,  6, -14,  41, 111, -19,  7, -2},
  {-1,  5, -13,  38, 113, -18,  6, -2}, {-1,  5, -12,  35, 114, -17,  6, -2},
  {-1,  5, -12,  33, 116, -17,  6, -2}, {-1,  4, -11,  30, 118, -16,  5, -1},
  {-1,  4, -10,  27, 119, -15,  5, -1}, {-1,  4,  -9,  25, 120, -14,  5, -2},
  {-1,  3,  -8,  23, 121, -13,  4, -1}, {-1,  3,  -7,  20, 122, -12,  4, -1},
  {-1,  3,  -7,  18, 123, -11,  4, -1}, {-1,  3,  -6,  16, 124, -10,  3, -1},
  {-1,  2,  -5,  13, 125,  -8,  3, -1}, {-1,  2,  -4,  11, 126,  -7,  2, -1},
  { 0,  1,  -3,   8, 126,  -6,  2,  0}, { 0,  1,  -2,   6, 127,  -5,  1,  0},
  { 0,  1,  -2,   4, 127,  -3,  1,  0}, { 0,  0,  -1,   2, 128,  -2,  1,  0},
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

  // [1, 2)
  { 0, 0, 0,   0, 128,   0,   0, 0 }, { 0, 0, 0, - 1, 128,   2, - 1, 0 },
  { 0, 0, 1, - 3, 127,   4, - 1, 0 }, { 0, 0, 1, - 4, 126,   6, - 2, 1 },
  { 0, 0, 1, - 5, 126,   8, - 3, 1 }, { 0, 0, 1, - 6, 125,  11, - 4, 1 },
  { 0, 0, 1, - 7, 124,  13, - 4, 1 }, { 0, 0, 2, - 8, 123,  15, - 5, 1 },
  { 0, 0, 2, - 9, 122,  18, - 6, 1 }, { 0, 0, 2, -10, 121,  20, - 6, 1 },
  { 0, 0, 2, -11, 120,  22, - 7, 2 }, { 0, 0, 2, -12, 119,  25, - 8, 2 },
  { 0, 0, 3, -13, 117,  27, - 8, 2 }, { 0, 0, 3, -13, 116,  29, - 9, 2 },
  { 0, 0, 3, -14, 114,  32, -10, 3 }, { 0, 0, 3, -15, 113,  35, -10, 2 },
  { 0, 0, 3, -15, 111,  37, -11, 3 }, { 0, 0, 3, -16, 109,  40, -11, 3 },
  { 0, 0, 3, -16, 108,  42, -12, 3 }, { 0, 0, 4, -17, 106,  45, -13, 3 },
  { 0, 0, 4, -17, 104,  47, -13, 3 }, { 0, 0, 4, -17, 102,  50, -14, 3 },
  { 0, 0, 4, -17, 100,  52, -14, 3 }, { 0, 0, 4, -18,  98,  55, -15, 4 },
  { 0, 0, 4, -18,  96,  58, -15, 3 }, { 0, 0, 4, -18,  94,  60, -16, 4 },
  { 0, 0, 4, -18,  91,  63, -16, 4 }, { 0, 0, 4, -18,  89,  65, -16, 4 },
  { 0, 0, 4, -18,  87,  68, -17, 4 }, { 0, 0, 4, -18,  85,  70, -17, 4 },
  { 0, 0, 4, -18,  82,  73, -17, 4 }, { 0, 0, 4, -18,  80,  75, -17, 4 },
  { 0, 0, 4, -18,  78,  78, -18, 4 }, { 0, 0, 4, -17,  75,  80, -18, 4 },
  { 0, 0, 4, -17,  73,  82, -18, 4 }, { 0, 0, 4, -17,  70,  85, -18, 4 },
  { 0, 0, 4, -17,  68,  87, -18, 4 }, { 0, 0, 4, -16,  65,  89, -18, 4 },
  { 0, 0, 4, -16,  63,  91, -18, 4 }, { 0, 0, 4, -16,  60,  94, -18, 4 },
  { 0, 0, 3, -15,  58,  96, -18, 4 }, { 0, 0, 4, -15,  55,  98, -18, 4 },
  { 0, 0, 3, -14,  52, 100, -17, 4 }, { 0, 0, 3, -14,  50, 102, -17, 4 },
  { 0, 0, 3, -13,  47, 104, -17, 4 }, { 0, 0, 3, -13,  45, 106, -17, 4 },
  { 0, 0, 3, -12,  42, 108, -16, 3 }, { 0, 0, 3, -11,  40, 109, -16, 3 },
  { 0, 0, 3, -11,  37, 111, -15, 3 }, { 0, 0, 2, -10,  35, 113, -15, 3 },
  { 0, 0, 3, -10,  32, 114, -14, 3 }, { 0, 0, 2, - 9,  29, 116, -13, 3 },
  { 0, 0, 2, - 8,  27, 117, -13, 3 }, { 0, 0, 2, - 8,  25, 119, -12, 2 },
  { 0, 0, 2, - 7,  22, 120, -11, 2 }, { 0, 0, 1, - 6,  20, 121, -10, 2 },
  { 0, 0, 1, - 6,  18, 122, - 9, 2 }, { 0, 0, 1, - 5,  15, 123, - 8, 2 },
  { 0, 0, 1, - 4,  13, 124, - 7, 1 }, { 0, 0, 1, - 4,  11, 125, - 6, 1 },
  { 0, 0, 1, - 3,   8, 126, - 5, 1 }, { 0, 0, 1, - 2,   6, 126, - 4, 1 },
  { 0, 0, 0, - 1,   4, 127, - 3, 1 }, { 0, 0, 0, - 1,   2, 128, - 1, 0 },
584
585
586

  // dummy
  { 0, 0, 0, 0,   0, 128, 0, 0 },
587
588
589
};
/* clang-format on */

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
#define DIV_LUT_PREC_BITS 14
#define DIV_LUT_BITS 8
#define DIV_LUT_NUM (1 << DIV_LUT_BITS)

static const uint16_t div_lut[DIV_LUT_NUM + 1] = {
  16384, 16320, 16257, 16194, 16132, 16070, 16009, 15948, 15888, 15828, 15768,
  15709, 15650, 15592, 15534, 15477, 15420, 15364, 15308, 15252, 15197, 15142,
  15087, 15033, 14980, 14926, 14873, 14821, 14769, 14717, 14665, 14614, 14564,
  14513, 14463, 14413, 14364, 14315, 14266, 14218, 14170, 14122, 14075, 14028,
  13981, 13935, 13888, 13843, 13797, 13752, 13707, 13662, 13618, 13574, 13530,
  13487, 13443, 13400, 13358, 13315, 13273, 13231, 13190, 13148, 13107, 13066,
  13026, 12985, 12945, 12906, 12866, 12827, 12788, 12749, 12710, 12672, 12633,
  12596, 12558, 12520, 12483, 12446, 12409, 12373, 12336, 12300, 12264, 12228,
  12193, 12157, 12122, 12087, 12053, 12018, 11984, 11950, 11916, 11882, 11848,
  11815, 11782, 11749, 11716, 11683, 11651, 11619, 11586, 11555, 11523, 11491,
  11460, 11429, 11398, 11367, 11336, 11305, 11275, 11245, 11215, 11185, 11155,
  11125, 11096, 11067, 11038, 11009, 10980, 10951, 10923, 10894, 10866, 10838,
  10810, 10782, 10755, 10727, 10700, 10673, 10645, 10618, 10592, 10565, 10538,
  10512, 10486, 10460, 10434, 10408, 10382, 10356, 10331, 10305, 10280, 10255,
  10230, 10205, 10180, 10156, 10131, 10107, 10082, 10058, 10034, 10010, 9986,
  9963,  9939,  9916,  9892,  9869,  9846,  9823,  9800,  9777,  9754,  9732,
  9709,  9687,  9664,  9642,  9620,  9598,  9576,  9554,  9533,  9511,  9489,
  9468,  9447,  9425,  9404,  9383,  9362,  9341,  9321,  9300,  9279,  9259,
  9239,  9218,  9198,  9178,  9158,  9138,  9118,  9098,  9079,  9059,  9039,
  9020,  9001,  8981,  8962,  8943,  8924,  8905,  8886,  8867,  8849,  8830,
  8812,  8793,  8775,  8756,  8738,  8720,  8702,  8684,  8666,  8648,  8630,
  8613,  8595,  8577,  8560,  8542,  8525,  8508,  8490,  8473,  8456,  8439,
  8422,  8405,  8389,  8372,  8355,  8339,  8322,  8306,  8289,  8273,  8257,
  8240,  8224,  8208,  8192,
};

621
622
623
624
625
626
627
628
static inline int16_t saturate_int16(int32_t v) {
  if (v > 32767)
    return 32767;
  else if (v < -32768)
    return -32768;
  return v;
}

629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
#if CONFIG_WARPED_MOTION
// Decomposes a divisor D such that 1/D = y/2^shift, where y is returned
// at precision of DIV_LUT_PREC_BITS along with the shift.
static int16_t resolve_divisor_64(uint64_t D, int16_t *shift) {
  int64_t e, f;
  *shift = (D >> 32) ? get_msb(D >> 32) + 32 : get_msb(D);
  // e is obtained from D after resetting the most significant 1 bit.
  e = D - ((uint64_t)1 << *shift);
  // Get the most significant DIV_LUT_BITS (8) bits of e into f
  if (*shift > DIV_LUT_BITS)
    f = ROUND_POWER_OF_TWO_64(e, *shift - DIV_LUT_BITS);
  else
    f = e << (DIV_LUT_BITS - *shift);
  assert(f <= DIV_LUT_NUM);
  *shift += DIV_LUT_PREC_BITS;
  // Use f as lookup into the precomputed table of multipliers
  return div_lut[f];
}
#endif  // CONFIG_WARPED_MOTION

static int16_t resolve_divisor_32(uint32_t D, int16_t *shift) {
  int32_t e, f;
  *shift = get_msb(D);
  // e is obtained from D after resetting the most significant 1 bit.
  e = D - ((uint32_t)1 << *shift);
  // Get the most significant DIV_LUT_BITS (8) bits of e into f
  if (*shift > DIV_LUT_BITS)
    f = ROUND_POWER_OF_TWO(e, *shift - DIV_LUT_BITS);
  else
    f = e << (DIV_LUT_BITS - *shift);
  assert(f <= DIV_LUT_NUM);
  *shift += DIV_LUT_PREC_BITS;
  // Use f as lookup into the precomputed table of multipliers
  return div_lut[f];
}

static int is_affine_valid(WarpedMotionParams *wm) {
  const int32_t *mat = wm->wmmat;
  return (mat[2] > 0);
}

static int is_affine_shear_allowed(int32_t alpha, int32_t beta, int32_t gamma,
                                   int32_t delta) {
672
673
  if ((4 * abs(alpha) + 7 * abs(beta) > (1 << WARPEDMODEL_PREC_BITS)) ||
      (4 * abs(gamma) + 4 * abs(delta) > (1 << WARPEDMODEL_PREC_BITS)))
674
675
676
677
678
679
    return 0;
  else
    return 1;
}

// Returns 1 on success or 0 on an invalid affine set
680
int get_shear_params(WarpedMotionParams *wm) {
681
682
  const int32_t *mat = wm->wmmat;
  if (!is_affine_valid(wm)) return 0;
683
684
  wm->alpha = mat[2] - (1 << WARPEDMODEL_PREC_BITS);
  wm->beta = mat[3];
685
686
687
688
  int16_t shift;
  int16_t y = resolve_divisor_32(abs(mat[2]), &shift) * (mat[2] < 0 ? -1 : 1);
  int64_t v;
  v = ((int64_t)mat[4] << WARPEDMODEL_PREC_BITS) * y;
689
  wm->gamma = ROUND_POWER_OF_TWO_SIGNED_64(v, shift);
690
  v = ((int64_t)mat[3] * mat[4]) * y;
691
692
693
694
  wm->delta = mat[5] - ROUND_POWER_OF_TWO_SIGNED_64(v, shift) -
              (1 << WARPEDMODEL_PREC_BITS);
  if (!is_affine_shear_allowed(wm->alpha, wm->beta, wm->gamma, wm->delta))
    return 0;
695
696
697
  return 1;
}

Yaowu Xu's avatar
Yaowu Xu committed
698
#if CONFIG_AOM_HIGHBITDEPTH
699
700
701
702
703
704
705
706
static INLINE void highbd_get_subcolumn(int taps, uint16_t *ref, int32_t *col,
                                        int stride, int x, int y_start) {
  int i;
  for (i = 0; i < taps; ++i) {
    col[i] = ref[(i + y_start) * stride + x];
  }
}

clang-format's avatar
clang-format committed
707
static uint16_t highbd_bi_ntap_filter(uint16_t *ref, int x, int y, int stride,
708
709
710
                                      int bd) {
  int32_t val, arr[WARPEDPIXEL_FILTER_TAPS];
  int k;
clang-format's avatar
clang-format committed
711
712
  int i = (int)x >> WARPEDPIXEL_PREC_BITS;
  int j = (int)y >> WARPEDPIXEL_PREC_BITS;
713
714
715
716
717
718
  for (k = 0; k < WARPEDPIXEL_FILTER_TAPS; ++k) {
    int32_t arr_temp[WARPEDPIXEL_FILTER_TAPS];
    highbd_get_subcolumn(WARPEDPIXEL_FILTER_TAPS, ref, arr_temp, stride,
                         i + k + 1 - WARPEDPIXEL_FILTER_TAPS / 2,
                         j + 1 - WARPEDPIXEL_FILTER_TAPS / 2);
    arr[k] = do_ntap_filter(arr_temp + WARPEDPIXEL_FILTER_TAPS / 2 - 1,
719
                            y - (j * (1 << WARPEDPIXEL_PREC_BITS)));
720
721
  }
  val = do_ntap_filter(arr + WARPEDPIXEL_FILTER_TAPS / 2 - 1,
722
                       x - (i * (1 << WARPEDPIXEL_PREC_BITS)));
723
  val = ROUND_POWER_OF_TWO_SIGNED(val, WARPEDPIXEL_FILTER_BITS * 2);
724
  return (uint16_t)clip_pixel_highbd(val, bd);
725
726
}

clang-format's avatar
clang-format committed
727
static uint16_t highbd_bi_cubic_filter(uint16_t *ref, int x, int y, int stride,
728
729
730
                                       int bd) {
  int32_t val, arr[4];
  int k;
clang-format's avatar
clang-format committed
731
732
  int i = (int)x >> WARPEDPIXEL_PREC_BITS;
  int j = (int)y >> WARPEDPIXEL_PREC_BITS;
733
734
  for (k = 0; k < 4; ++k) {
    int32_t arr_temp[4];
clang-format's avatar
clang-format committed
735
    highbd_get_subcolumn(4, ref, arr_temp, stride, i + k - 1, j - 1);
736
737
    arr[k] =
        do_cubic_filter(arr_temp + 1, y - (j * (1 << WARPEDPIXEL_PREC_BITS)));
738
  }
739
  val = do_cubic_filter(arr + 1, x - (i * (1 << WARPEDPIXEL_PREC_BITS)));
740
  val = ROUND_POWER_OF_TWO_SIGNED(val, WARPEDPIXEL_FILTER_BITS * 2);
741
  return (uint16_t)clip_pixel_highbd(val, bd);
742
743
}

clang-format's avatar
clang-format committed
744
static uint16_t highbd_bi_linear_filter(uint16_t *ref, int x, int y, int stride,
745
746
747
                                        int bd) {
  const int ix = x >> WARPEDPIXEL_PREC_BITS;
  const int iy = y >> WARPEDPIXEL_PREC_BITS;
748
749
  const int sx = x - (ix * (1 << WARPEDPIXEL_PREC_BITS));
  const int sy = y - (iy * (1 << WARPEDPIXEL_PREC_BITS));
750
751
752
  int32_t val;
  val = ROUND_POWER_OF_TWO_SIGNED(
      ref[iy * stride + ix] * (WARPEDPIXEL_PREC_SHIFTS - sy) *
clang-format's avatar
clang-format committed
753
754
755
756
              (WARPEDPIXEL_PREC_SHIFTS - sx) +
          ref[iy * stride + ix + 1] * (WARPEDPIXEL_PREC_SHIFTS - sy) * sx +
          ref[(iy + 1) * stride + ix] * sy * (WARPEDPIXEL_PREC_SHIFTS - sx) +
          ref[(iy + 1) * stride + ix + 1] * sy * sx,
757
      WARPEDPIXEL_PREC_BITS * 2);
758
  return (uint16_t)clip_pixel_highbd(val, bd);
759
760
}

clang-format's avatar
clang-format committed
761
762
static uint16_t highbd_warp_interpolate(uint16_t *ref, int x, int y, int width,
                                        int height, int stride, int bd) {
763
764
  int ix = x >> WARPEDPIXEL_PREC_BITS;
  int iy = y >> WARPEDPIXEL_PREC_BITS;
765
766
  int sx = x - (ix * (1 << WARPEDPIXEL_PREC_BITS));
  int sy = y - (iy * (1 << WARPEDPIXEL_PREC_BITS));
767
768
  int32_t v;

clang-format's avatar
clang-format committed
769
770
  if (ix < 0 && iy < 0)
    return ref[0];
771
772
773
774
775
776
777
778
779
  else if (ix < 0 && iy > height - 1)
    return ref[(height - 1) * stride];
  else if (ix > width - 1 && iy < 0)
    return ref[width - 1];
  else if (ix > width - 1 && iy > height - 1)
    return ref[(height - 1) * stride + (width - 1)];
  else if (ix < 0) {
    v = ROUND_POWER_OF_TWO_SIGNED(
        ref[iy * stride] * (WARPEDPIXEL_PREC_SHIFTS - sy) +
clang-format's avatar
clang-format committed
780
            ref[(iy + 1) * stride] * sy,
781
        WARPEDPIXEL_PREC_BITS);
782
    return clip_pixel_highbd(v, bd);
783
784
  } else if (iy < 0) {
    v = ROUND_POWER_OF_TWO_SIGNED(
clang-format's avatar
clang-format committed
785
        ref[ix] * (WARPEDPIXEL_PREC_SHIFTS - sx) + ref[ix + 1] * sx,
786
        WARPEDPIXEL_PREC_BITS);
787
    return clip_pixel_highbd(v, bd);
788
789
790
  } else if (ix > width - 1) {
    v = ROUND_POWER_OF_TWO_SIGNED(
        ref[iy * stride + width - 1] * (WARPEDPIXEL_PREC_SHIFTS - sy) +
clang-format's avatar
clang-format committed
791
            ref[(iy + 1) * stride + width - 1] * sy,
792
        WARPEDPIXEL_PREC_BITS);
793
    return clip_pixel_highbd(v, bd);
794
795
796
  } else if (iy > height - 1) {
    v = ROUND_POWER_OF_TWO_SIGNED(
        ref[(height - 1) * stride + ix] * (WARPEDPIXEL_PREC_SHIFTS - sx) +
clang-format's avatar
clang-format committed
797
            ref[(height - 1) * stride + ix + 1] * sx,
798
        WARPEDPIXEL_PREC_BITS);
799
    return clip_pixel_highbd(v, bd);
800
801
802
803
804
  } else if (ix >= WARPEDPIXEL_FILTER_TAPS / 2 - 1 &&
             iy >= WARPEDPIXEL_FILTER_TAPS / 2 - 1 &&
             ix < width - WARPEDPIXEL_FILTER_TAPS / 2 &&
             iy < height - WARPEDPIXEL_FILTER_TAPS / 2) {
    return highbd_bi_ntap_filter(ref, x, y, stride, bd);
clang-format's avatar
clang-format committed
805
  } else if (ix >= 1 && iy >= 1 && ix < width - 2 && iy < height - 2) {
806
807
808
809
810
811
    return highbd_bi_cubic_filter(ref, x, y, stride, bd);
  } else {
    return highbd_bi_linear_filter(ref, x, y, stride, bd);
  }
}

812
813
814
static INLINE int highbd_error_measure(int err, int bd) {
  const int b = bd - 8;
  const int bmask = (1 << b) - 1;
815
  const int v = (1 << b);
816
817
818
819
820
821
822
823
  int e1, e2;
  err = abs(err);
  e1 = err >> b;
  e2 = err & bmask;
  return error_measure_lut[255 + e1] * (v - e2) +
         error_measure_lut[256 + e1] * e2;
}

824
825
826
827
828
829
830
static void highbd_warp_plane_old(WarpedMotionParams *wm, uint8_t *ref8,
                                  int width, int height, int stride,
                                  uint8_t *pred8, int p_col, int p_row,
                                  int p_width, int p_height, int p_stride,
                                  int subsampling_x, int subsampling_y,
                                  int x_scale, int y_scale, int bd,
                                  int ref_frm) {
831
  int i, j;
832
  ProjectPointsFunc projectpoints = get_project_points_type(wm->wmtype);
833
  uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
834
  uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
835
  if (projectpoints == NULL) return;
836
837
838
839
840
  for (i = p_row; i < p_row + p_height; ++i) {
    for (j = p_col; j < p_col + p_width; ++j) {
      int in[2], out[2];
      in[0] = j;
      in[1] = i;
841
      projectpoints(wm->wmmat, in, out, 1, 2, 2, subsampling_x, subsampling_y);
842
843
      out[0] = ROUND_POWER_OF_TWO_SIGNED(out[0] * x_scale, 4);
      out[1] = ROUND_POWER_OF_TWO_SIGNED(out[1] * y_scale, 4);
844
845
846
847
848
849
850
851
852
      if (ref_frm)
        pred[(j - p_col) + (i - p_row) * p_stride] = ROUND_POWER_OF_TWO(
            pred[(j - p_col) + (i - p_row) * p_stride] +
                highbd_warp_interpolate(ref, out[0], out[1], width, height,
                                        stride, bd),
            1);
      else
        pred[(j - p_col) + (i - p_row) * p_stride] = highbd_warp_interpolate(
            ref, out[0], out[1], width, height, stride, bd);
853
854
855
856
    }
  }
}

857
858
// Note: For an explanation of the warp algorithm, see the comment
// above warp_plane()
859
860
861
862
863
864
865
866
867
868
869
870
871
//
// Note also: The "worst case" in terms of modulus of the data stored into 'tmp'
// (ie, the result of 'sum' in the horizontal filter) occurs when:
// coeffs = { -2,   8, -22,  87,  72, -21,   8, -2}, and
// ref =    {  0, 255,   0, 255, 255,   0, 255,  0}
// Before rounding, this gives sum = 716625. After rounding,
// HORSHEAR_REDUCE_PREC_BITS = 4 => sum = 44789 > 2^15
// HORSHEAR_REDUCE_PREC_BITS = 5 => sum = 22395 < 2^15
//
// So, as long as HORSHEAR_REDUCE_PREC_BITS >= 5, we can safely use a 16-bit
// intermediate array.
void av1_highbd_warp_affine_c(int32_t *mat, uint16_t *ref, int width,
                              int height, int stride, uint16_t *pred, int p_col,
clang-format's avatar
clang-format committed
872
873
                              int p_row, int p_width, int p_height,
                              int p_stride, int subsampling_x,
874
875
876
877
878
879
880
881
882
                              int subsampling_y, int bd, int ref_frm,
                              int32_t alpha, int32_t beta, int32_t gamma,
                              int32_t delta) {
#if HORSHEAR_REDUCE_PREC_BITS >= 5
  int16_t tmp[15 * 8];
#else
  int32_t tmp[15 * 8];
#endif
  int i, j, k, l, m;
883

884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
  /* Note: For this code to work, the left/right frame borders need to be
     extended by at least 13 pixels each. By the time we get here, other
     code will have set up this border, but we allow an explicit check
     for debugging purposes.
  */
  /*for (i = 0; i < height; ++i) {
    for (j = 0; j < 13; ++j) {
      assert(ref[i * stride - 13 + j] == ref[i * stride]);
      assert(ref[i * stride + width + j] == ref[i * stride + (width - 1)]);
    }
  }*/

  for (i = p_row; i < p_row + p_height; i += 8) {
    for (j = p_col; j < p_col + p_width; j += 8) {
      int32_t x4, y4, ix4, sx4, iy4, sy4;
      if (subsampling_x)
        x4 = ROUND_POWER_OF_TWO_SIGNED(
            mat[2] * 2 * (j + 4) + mat[3] * 2 * (i + 4) + mat[0] +
                (mat[2] + mat[3] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
            1);
      else
        x4 = mat[2] * (j + 4) + mat[3] * (i + 4) + mat[0];
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
      if (subsampling_y)
        y4 = ROUND_POWER_OF_TWO_SIGNED(
            mat[4] * 2 * (j + 4) + mat[5] * 2 * (i + 4) + mat[1] +
                (mat[4] + mat[5] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
            1);
      else
        y4 = mat[4] * (j + 4) + mat[5] * (i + 4) + mat[1];

      ix4 = x4 >> WARPEDMODEL_PREC_BITS;
      sx4 = x4 & ((1 << WARPEDMODEL_PREC_BITS) - 1);
      iy4 = y4 >> WARPEDMODEL_PREC_BITS;
      sy4 = y4 & ((1 << WARPEDMODEL_PREC_BITS) - 1);

      // Horizontal filter
      for (k = -7; k < 8; ++k) {
        int iy = iy4 + k;
        if (iy < 0)
          iy = 0;
        else if (iy > height - 1)
          iy = height - 1;

        if (ix4 <= -7) {
          for (l = 0; l < 8; ++l) {
            tmp[(k + 7) * 8 + l] =
                ref[iy * stride] *
                (1 << (WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS));
          }
        } else if (ix4 >= width + 6) {
          for (l = 0; l < 8; ++l) {
            tmp[(k + 7) * 8 + l] =
                ref[iy * stride + (width - 1)] *
                (1 << (WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS));
          }
        } else {
          int sx = sx4 + alpha * (-4) + beta * k;
942
943

          for (l = -4; l < 4; ++l) {
944
945
946
947
            int ix = ix4 + l - 3;
            const int offs = ROUND_POWER_OF_TWO(sx, WARPEDDIFF_PREC_BITS) +
                             WARPEDPIXEL_PREC_SHIFTS;
            const int16_t *coeffs = warped_filter[offs];
948
            int32_t sum = 0;
949
            // assert(offs >= 0 && offs <= WARPEDPIXEL_PREC_SHIFTS * 3);
950
            for (m = 0; m < 8; ++m) {
951
              sum += ref[iy * stride + ix + m] * coeffs[m];
952
            }
953
            sum = ROUND_POWER_OF_TWO(sum, HORSHEAR_REDUCE_PREC_BITS);
954
955
956
#if HORSHEAR_REDUCE_PREC_BITS >= 5
            tmp[(k + 7) * 8 + (l + 4)] = saturate_int16(sum);
#else
957
            tmp[(k + 7) * 8 + (l + 4)] = sum;
958
959
#endif
            sx += alpha;
960
961
          }
        }
962
      }
963

964
965
966
967
968
969
970
971
972
973
974
975
976
      // Vertical filter
      for (k = -4; k < AOMMIN(4, p_row + p_height - i - 4); ++k) {
        int sy = sy4 + gamma * (-4) + delta * k;
        for (l = -4; l < 4; ++l) {
          uint16_t *p =
              &pred[(i - p_row + k + 4) * p_stride + (j - p_col + l + 4)];
          const int offs = ROUND_POWER_OF_TWO(sy, WARPEDDIFF_PREC_BITS) +
                           WARPEDPIXEL_PREC_SHIFTS;
          const int16_t *coeffs = warped_filter[offs];
          int32_t sum = 0;
          // assert(offs >= 0 && offs <= WARPEDPIXEL_PREC_SHIFTS * 3);
          for (m = 0; m < 8; ++m) {
            sum += tmp[(k + m + 4) * 8 + (l + 4)] * coeffs[m];
977
          }
978
979
980
981
982
983
984
          sum = clip_pixel_highbd(
              ROUND_POWER_OF_TWO(sum, VERSHEAR_REDUCE_PREC_BITS), bd);
          if (ref_frm)
            *p = ROUND_POWER_OF_TWO(*p + sum, 1);
          else
            *p = sum;
          sy += gamma;
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
  }
}

static void highbd_warp_plane(WarpedMotionParams *wm, uint8_t *ref8, int width,
                              int height, int stride, uint8_t *pred8, int p_col,
                              int p_row, int p_width, int p_height,
                              int p_stride, int subsampling_x,
                              int subsampling_y, int x_scale, int y_scale,
                              int bd, int ref_frm) {
  if (wm->wmtype == ROTZOOM) {
    wm->wmmat[5] = wm->wmmat[2];
    wm->wmmat[4] = -wm->wmmat[3];
  }
  if ((wm->wmtype == ROTZOOM || wm->wmtype == AFFINE) && x_scale == 16 &&
      y_scale == 16) {
    int32_t *mat = wm->wmmat;
    const int32_t alpha = wm->alpha;
    const int32_t beta = wm->beta;
    const int32_t gamma = wm->gamma;
    const int32_t delta = wm->delta;

    uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
    uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
    av1_highbd_warp_affine(mat, ref, width, height, stride, pred, p_col, p_row,
                           p_width, p_height, p_stride, subsampling_x,
                           subsampling_y, bd, ref_frm, alpha, beta, gamma,
                           delta);
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
  } else {
    highbd_warp_plane_old(wm, ref8, width, height, stride, pred8, p_col, p_row,
                          p_width, p_height, p_stride, subsampling_x,
                          subsampling_y, x_scale, y_scale, bd, ref_frm);
  }
}

static double highbd_warp_erroradv(WarpedMotionParams *wm, uint8_t *ref8,
                                   int width, int height, int stride,
                                   uint8_t *dst8, int p_col, int p_row,
                                   int p_width, int p_height, int p_stride,
                                   int subsampling_x, int subsampling_y,
                                   int x_scale, int y_scale, int bd) {
  int gm_err = 0, no_gm_err = 0;
  int64_t gm_sumerr = 0, no_gm_sumerr = 0;
1030
  int i, j;
1031
1032
  uint16_t *tmp = aom_malloc(p_width * p_height * sizeof(*tmp));
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
1033
  uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
  highbd_warp_plane(wm, ref8, width, height, stride, CONVERT_TO_BYTEPTR(tmp),
                    p_col, p_row, p_width, p_height, p_width, subsampling_x,
                    subsampling_y, x_scale, y_scale, bd, 0);
  for (i = 0; i < p_height; ++i) {
    for (j = 0; j < p_width; ++j) {
      gm_err = dst[j + i * p_stride] - tmp[j + i * p_width];
      no_gm_err =
          dst[j + i * p_stride] - ref[(j + p_col) + (i + p_row) * stride];
      gm_sumerr += highbd_error_measure(gm_err, bd);
      no_gm_sumerr += highbd_error_measure(no_gm_err, bd);
1044
1045
    }
  }
1046
1047
  aom_free(tmp);
  return (double)gm_sumerr / no_gm_sumerr;
1048
}
Yaowu Xu's avatar
Yaowu Xu committed
1049
#endif  // CONFIG_AOM_HIGHBITDEPTH
1050

1051
1052
1053
1054
static INLINE int error_measure(int err) {
  return error_measure_lut[255 + err];
}

1055
1056
1057
1058
1059
static void warp_plane_old(WarpedMotionParams *wm, uint8_t *ref, int width,
                           int height, int stride, uint8_t *pred, int p_col,
                           int p_row, int p_width, int p_height, int p_stride,
                           int subsampling_x, int subsampling_y, int x_scale,
                           int y_scale, int ref_frm) {
1060
  int i, j;
1061
  ProjectPointsFunc projectpoints = get_project_points_type(wm->wmtype);
1062
  if (projectpoints == NULL) return;
1063
1064
1065
1066
1067
  for (i = p_row; i < p_row + p_height; ++i) {
    for (j = p_col; j < p_col + p_width; ++j) {
      int in[2], out[2];
      in[0] = j;
      in[1] = i;
1068
      projectpoints(wm->wmmat, in, out, 1, 2, 2, subsampling_x, subsampling_y);
1069
1070
      out[0] = ROUND_POWER_OF_TWO_SIGNED(out[0] * x_scale, 4);
      out[1] = ROUND_POWER_OF_TWO_SIGNED(out[1] * y_scale, 4);
1071
1072
1073
1074
1075
1076
1077
1078
      if (ref_frm)
        pred[(j - p_col) + (i - p_row) * p_stride] = ROUND_POWER_OF_TWO(
            pred[(j - p_col) + (i - p_row) * p_stride] +
                warp_interpolate(ref, out[0], out[1], width, height, stride),
            1);
      else
        pred[(j - p_col) + (i - p_row) * p_stride] =
            warp_interpolate(ref, out[0], out[1], width, height, stride);
1079
1080
1081
1082
    }
  }
}

1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
/* The warp filter for ROTZOOM and AFFINE models works as follows:
   * Split the input into 8x8 blocks
   * For each block, project the point (4, 4) within the block, to get the
     overall block position. Split into integer and fractional coordinates,
     maintaining full WARPEDMODEL precision
   * Filter horizontally: Generate 15 rows of 8 pixels each. Each pixel gets a
     variable horizontal offset. This means that, while the rows of the
     intermediate buffer align with the rows of the *reference* image, the
     columns align with the columns of the *destination* image.
   * Filter vertically: Generate the output block (up to 8x8 pixels, but if the
     destination is too small we crop the output at this stage). Each pixel has
     a variable vertical offset, so that the resulting rows are aligned with
     the rows of the destination image.

   To accomplish these alignments, we factor the warp matrix as a
   product of two shear / asymmetric zoom matrices:
   / a b \  = /   1       0    \ * / 1+alpha  beta \
   \ c d /    \ gamma  1+delta /   \    0      1   /
   where a, b, c, d are wmmat[2], wmmat[3], wmmat[4], wmmat[5] respectively.
   The second shear (with alpha and beta) is applied by the horizontal filter,
   then the first shear (with gamma and delta) is applied by the vertical
   filter.

   The only limitation is that, to fit this in a fixed 8-tap filter size,
   the fractional pixel offsets must be at most +-1. Since the horizontal filter
   generates 15 rows of 8 columns, and the initial point we project is at (4, 4)
   within the block, the parameters must satisfy
   4 * |alpha| + 7 * |beta| <= 1   and   4 * |gamma| + 7 * |delta| <= 1
   for this filter to be applicable.
1112
1113
1114
1115
1116
1117
1118

   Note: warp_affine() assumes that the caller has done all of the relevant
   checks, ie. that we have a ROTZOOM or AFFINE model, that wm[4] and wm[5]
   are set appropriately (if using a ROTZOOM model), and that alpha, beta,
   gamma, delta are all in range.

   TODO(david.barker): Maybe support scaled references?
1119
*/