summaryrefslogtreecommitdiffstats
path: root/third_party/aom/aom_dsp/x86/variance_impl_avx2.c
blob: 88e27aef3af0a2969edb86dd2d986930c12b9dd4 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
/*
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
 *
 * 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.
 */

#include <immintrin.h>  // AVX2

#include "config/aom_dsp_rtcd.h"

#include "aom_ports/mem.h"

/* clang-format off */
DECLARE_ALIGNED(32, static const uint8_t, bilinear_filters_avx2[512]) = {
  16,  0, 16,  0, 16,  0, 16,  0, 16,  0, 16,  0, 16,  0, 16,  0,
  16,  0, 16,  0, 16,  0, 16,  0, 16,  0, 16,  0, 16,  0, 16,  0,
  14,  2, 14,  2, 14,  2, 14,  2, 14,  2, 14,  2, 14,  2, 14,  2,
  14,  2, 14,  2, 14,  2, 14,  2, 14,  2, 14,  2, 14,  2, 14,  2,
  12,  4, 12,  4, 12,  4, 12,  4, 12,  4, 12,  4, 12,  4, 12,  4,
  12,  4, 12,  4, 12,  4, 12,  4, 12,  4, 12,  4, 12,  4, 12,  4,
  10,  6, 10,  6, 10,  6, 10,  6, 10,  6, 10,  6, 10,  6, 10,  6,
  10,  6, 10,  6, 10,  6, 10,  6, 10,  6, 10,  6, 10,  6, 10,  6,
   8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,
   8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,
   6, 10,  6, 10,  6, 10,  6, 10,  6, 10,  6, 10,  6, 10,  6, 10,
   6, 10,  6, 10,  6, 10,  6, 10,  6, 10,  6, 10,  6, 10,  6, 10,
   4, 12,  4, 12,  4, 12,  4, 12,  4, 12,  4, 12,  4, 12,  4, 12,
   4, 12,  4, 12,  4, 12,  4, 12,  4, 12,  4, 12,  4, 12,  4, 12,
   2, 14,  2, 14,  2, 14,  2, 14,  2, 14,  2, 14,  2, 14,  2, 14,
   2, 14,  2, 14,  2, 14,  2, 14,  2, 14,  2, 14,  2, 14,  2, 14,
};
/* clang-format on */

#define FILTER_SRC(filter)                               \
  /* filter the source */                                \
  exp_src_lo = _mm256_maddubs_epi16(exp_src_lo, filter); \
  exp_src_hi = _mm256_maddubs_epi16(exp_src_hi, filter); \
                                                         \
  /* add 8 to source */                                  \
  exp_src_lo = _mm256_add_epi16(exp_src_lo, pw8);        \
  exp_src_hi = _mm256_add_epi16(exp_src_hi, pw8);        \
                                                         \
  /* divide source by 16 */                              \
  exp_src_lo = _mm256_srai_epi16(exp_src_lo, 4);         \
  exp_src_hi = _mm256_srai_epi16(exp_src_hi, 4);

#define MERGE_WITH_SRC(src_reg, reg)               \
  exp_src_lo = _mm256_unpacklo_epi8(src_reg, reg); \
  exp_src_hi = _mm256_unpackhi_epi8(src_reg, reg);

#define LOAD_SRC_DST                                    \
  /* load source and destination */                     \
  src_reg = _mm256_loadu_si256((__m256i const *)(src)); \
  dst_reg = _mm256_loadu_si256((__m256i const *)(dst));

#define AVG_NEXT_SRC(src_reg, size_stride)                                 \
  src_next_reg = _mm256_loadu_si256((__m256i const *)(src + size_stride)); \
  /* average between current and next stride source */                     \
  src_reg = _mm256_avg_epu8(src_reg, src_next_reg);

#define MERGE_NEXT_SRC(src_reg, size_stride)                               \
  src_next_reg = _mm256_loadu_si256((__m256i const *)(src + size_stride)); \
  MERGE_WITH_SRC(src_reg, src_next_reg)

#define CALC_SUM_SSE_INSIDE_LOOP                          \
  /* expand each byte to 2 bytes */                       \
  exp_dst_lo = _mm256_unpacklo_epi8(dst_reg, zero_reg);   \
  exp_dst_hi = _mm256_unpackhi_epi8(dst_reg, zero_reg);   \
  /* source - dest */                                     \
  exp_src_lo = _mm256_sub_epi16(exp_src_lo, exp_dst_lo);  \
  exp_src_hi = _mm256_sub_epi16(exp_src_hi, exp_dst_hi);  \
  /* caculate sum */                                      \
  sum_reg = _mm256_add_epi16(sum_reg, exp_src_lo);        \
  exp_src_lo = _mm256_madd_epi16(exp_src_lo, exp_src_lo); \
  sum_reg = _mm256_add_epi16(sum_reg, exp_src_hi);        \
  exp_src_hi = _mm256_madd_epi16(exp_src_hi, exp_src_hi); \
  /* calculate sse */                                     \
  sse_reg = _mm256_add_epi32(sse_reg, exp_src_lo);        \
  sse_reg = _mm256_add_epi32(sse_reg, exp_src_hi);

// final calculation to sum and sse
#define CALC_SUM_AND_SSE                                                   \
  res_cmp = _mm256_cmpgt_epi16(zero_reg, sum_reg);                         \
  sse_reg_hi = _mm256_srli_si256(sse_reg, 8);                              \
  sum_reg_lo = _mm256_unpacklo_epi16(sum_reg, res_cmp);                    \
  sum_reg_hi = _mm256_unpackhi_epi16(sum_reg, res_cmp);                    \
  sse_reg = _mm256_add_epi32(sse_reg, sse_reg_hi);                         \
  sum_reg = _mm256_add_epi32(sum_reg_lo, sum_reg_hi);                      \
                                                                           \
  sse_reg_hi = _mm256_srli_si256(sse_reg, 4);                              \
  sum_reg_hi = _mm256_srli_si256(sum_reg, 8);                              \
                                                                           \
  sse_reg = _mm256_add_epi32(sse_reg, sse_reg_hi);                         \
  sum_reg = _mm256_add_epi32(sum_reg, sum_reg_hi);                         \
  *((int *)sse) = _mm_cvtsi128_si32(_mm256_castsi256_si128(sse_reg)) +     \
                  _mm_cvtsi128_si32(_mm256_extractf128_si256(sse_reg, 1)); \
  sum_reg_hi = _mm256_srli_si256(sum_reg, 4);                              \
  sum_reg = _mm256_add_epi32(sum_reg, sum_reg_hi);                         \
  sum = _mm_cvtsi128_si32(_mm256_castsi256_si128(sum_reg)) +               \
        _mm_cvtsi128_si32(_mm256_extractf128_si256(sum_reg, 1));

unsigned int aom_sub_pixel_variance32xh_avx2(const uint8_t *src, int src_stride,
                                             int x_offset, int y_offset,
                                             const uint8_t *dst, int dst_stride,
                                             int height, unsigned int *sse) {
  __m256i src_reg, dst_reg, exp_src_lo, exp_src_hi, exp_dst_lo, exp_dst_hi;
  __m256i sse_reg, sum_reg, sse_reg_hi, res_cmp, sum_reg_lo, sum_reg_hi;
  __m256i zero_reg;
  int i, sum;
  sum_reg = _mm256_set1_epi16(0);
  sse_reg = _mm256_set1_epi16(0);
  zero_reg = _mm256_set1_epi16(0);

  // x_offset = 0 and y_offset = 0
  if (x_offset == 0) {
    if (y_offset == 0) {
      for (i = 0; i < height; i++) {
        LOAD_SRC_DST
        // expend each byte to 2 bytes
        MERGE_WITH_SRC(src_reg, zero_reg)
        CALC_SUM_SSE_INSIDE_LOOP
        src += src_stride;
        dst += dst_stride;
      }
      // x_offset = 0 and y_offset = 8
    } else if (y_offset == 8) {
      __m256i src_next_reg;
      for (i = 0; i < height; i++) {
        LOAD_SRC_DST
        AVG_NEXT_SRC(src_reg, src_stride)
        // expend each byte to 2 bytes
        MERGE_WITH_SRC(src_reg, zero_reg)
        CALC_SUM_SSE_INSIDE_LOOP
        src += src_stride;
        dst += dst_stride;
      }
      // x_offset = 0 and y_offset = bilin interpolation
    } else {
      __m256i filter, pw8, src_next_reg;

      y_offset <<= 5;
      filter = _mm256_load_si256(
          (__m256i const *)(bilinear_filters_avx2 + y_offset));
      pw8 = _mm256_set1_epi16(8);
      for (i = 0; i < height; i++) {
        LOAD_SRC_DST
        MERGE_NEXT_SRC(src_reg, src_stride)
        FILTER_SRC(filter)
        CALC_SUM_SSE_INSIDE_LOOP
        src += src_stride;
        dst += dst_stride;
      }
    }
    // x_offset = 8  and y_offset = 0
  } else if (x_offset == 8) {
    if (y_offset == 0) {
      __m256i src_next_reg;
      for (i = 0; i < height; i++) {
        LOAD_SRC_DST
        AVG_NEXT_SRC(src_reg, 1)
        // expand each byte to 2 bytes
        MERGE_WITH_SRC(src_reg, zero_reg)
        CALC_SUM_SSE_INSIDE_LOOP
        src += src_stride;
        dst += dst_stride;
      }
      // x_offset = 8  and y_offset = 8
    } else if (y_offset == 8) {
      __m256i src_next_reg, src_avg;
      // load source and another source starting from the next
      // following byte
      src_reg = _mm256_loadu_si256((__m256i const *)(src));
      AVG_NEXT_SRC(src_reg, 1)
      for (i = 0; i < height; i++) {
        src_avg = src_reg;
        src += src_stride;
        LOAD_SRC_DST
        AVG_NEXT_SRC(src_reg, 1)
        // average between previous average to current average
        src_avg = _mm256_avg_epu8(src_avg, src_reg);
        // expand each byte to 2 bytes
        MERGE_WITH_SRC(src_avg, zero_reg)
        // save current source average
        CALC_SUM_SSE_INSIDE_LOOP
        dst += dst_stride;
      }
      // x_offset = 8  and y_offset = bilin interpolation
    } else {
      __m256i filter, pw8, src_next_reg, src_avg;
      y_offset <<= 5;
      filter = _mm256_load_si256(
          (__m256i const *)(bilinear_filters_avx2 + y_offset));
      pw8 = _mm256_set1_epi16(8);
      // load source and another source starting from the next
      // following byte
      src_reg = _mm256_loadu_si256((__m256i const *)(src));
      AVG_NEXT_SRC(src_reg, 1)
      for (i = 0; i < height; i++) {
        // save current source average
        src_avg = src_reg;
        src += src_stride;
        LOAD_SRC_DST
        AVG_NEXT_SRC(src_reg, 1)
        MERGE_WITH_SRC(src_avg, src_reg)
        FILTER_SRC(filter)
        CALC_SUM_SSE_INSIDE_LOOP
        dst += dst_stride;
      }
    }
    // x_offset = bilin interpolation and y_offset = 0
  } else {
    if (y_offset == 0) {
      __m256i filter, pw8, src_next_reg;
      x_offset <<= 5;
      filter = _mm256_load_si256(
          (__m256i const *)(bilinear_filters_avx2 + x_offset));
      pw8 = _mm256_set1_epi16(8);
      for (i = 0; i < height; i++) {
        LOAD_SRC_DST
        MERGE_NEXT_SRC(src_reg, 1)
        FILTER_SRC(filter)
        CALC_SUM_SSE_INSIDE_LOOP
        src += src_stride;
        dst += dst_stride;
      }
      // x_offset = bilin interpolation and y_offset = 8
    } else if (y_offset == 8) {
      __m256i filter, pw8, src_next_reg, src_pack;
      x_offset <<= 5;
      filter = _mm256_load_si256(
          (__m256i const *)(bilinear_filters_avx2 + x_offset));
      pw8 = _mm256_set1_epi16(8);
      src_reg = _mm256_loadu_si256((__m256i const *)(src));
      MERGE_NEXT_SRC(src_reg, 1)
      FILTER_SRC(filter)
      // convert each 16 bit to 8 bit to each low and high lane source
      src_pack = _mm256_packus_epi16(exp_src_lo, exp_src_hi);
      for (i = 0; i < height; i++) {
        src += src_stride;
        LOAD_SRC_DST
        MERGE_NEXT_SRC(src_reg, 1)
        FILTER_SRC(filter)
        src_reg = _mm256_packus_epi16(exp_src_lo, exp_src_hi);
        // average between previous pack to the current
        src_pack = _mm256_avg_epu8(src_pack, src_reg);
        MERGE_WITH_SRC(src_pack, zero_reg)
        CALC_SUM_SSE_INSIDE_LOOP
        src_pack = src_reg;
        dst += dst_stride;
      }
      // x_offset = bilin interpolation and y_offset = bilin interpolation
    } else {
      __m256i xfilter, yfilter, pw8, src_next_reg, src_pack;
      x_offset <<= 5;
      xfilter = _mm256_load_si256(
          (__m256i const *)(bilinear_filters_avx2 + x_offset));
      y_offset <<= 5;
      yfilter = _mm256_load_si256(
          (__m256i const *)(bilinear_filters_avx2 + y_offset));
      pw8 = _mm256_set1_epi16(8);
      // load source and another source starting from the next
      // following byte
      src_reg = _mm256_loadu_si256((__m256i const *)(src));
      MERGE_NEXT_SRC(src_reg, 1)

      FILTER_SRC(xfilter)
      // convert each 16 bit to 8 bit to each low and high lane source
      src_pack = _mm256_packus_epi16(exp_src_lo, exp_src_hi);
      for (i = 0; i < height; i++) {
        src += src_stride;
        LOAD_SRC_DST
        MERGE_NEXT_SRC(src_reg, 1)
        FILTER_SRC(xfilter)
        src_reg = _mm256_packus_epi16(exp_src_lo, exp_src_hi);
        // merge previous pack to current pack source
        MERGE_WITH_SRC(src_pack, src_reg)
        // filter the source
        FILTER_SRC(yfilter)
        src_pack = src_reg;
        CALC_SUM_SSE_INSIDE_LOOP
        dst += dst_stride;
      }
    }
  }
  CALC_SUM_AND_SSE
  _mm256_zeroupper();
  return sum;
}

unsigned int aom_sub_pixel_avg_variance32xh_avx2(
    const uint8_t *src, int src_stride, int x_offset, int y_offset,
    const uint8_t *dst, int dst_stride, const uint8_t *sec, int sec_stride,
    int height, unsigned int *sse) {
  __m256i sec_reg;
  __m256i src_reg, dst_reg, exp_src_lo, exp_src_hi, exp_dst_lo, exp_dst_hi;
  __m256i sse_reg, sum_reg, sse_reg_hi, res_cmp, sum_reg_lo, sum_reg_hi;
  __m256i zero_reg;
  int i, sum;
  sum_reg = _mm256_set1_epi16(0);
  sse_reg = _mm256_set1_epi16(0);
  zero_reg = _mm256_set1_epi16(0);

  // x_offset = 0 and y_offset = 0
  if (x_offset == 0) {
    if (y_offset == 0) {
      for (i = 0; i < height; i++) {
        LOAD_SRC_DST
        sec_reg = _mm256_loadu_si256((__m256i const *)(sec));
        src_reg = _mm256_avg_epu8(src_reg, sec_reg);
        sec += sec_stride;
        // expend each byte to 2 bytes
        MERGE_WITH_SRC(src_reg, zero_reg)
        CALC_SUM_SSE_INSIDE_LOOP
        src += src_stride;
        dst += dst_stride;
      }
    } else if (y_offset == 8) {
      __m256i src_next_reg;
      for (i = 0; i < height; i++) {
        LOAD_SRC_DST
        AVG_NEXT_SRC(src_reg, src_stride)
        sec_reg = _mm256_loadu_si256((__m256i const *)(sec));
        src_reg = _mm256_avg_epu8(src_reg, sec_reg);
        sec += sec_stride;
        // expend each byte to 2 bytes
        MERGE_WITH_SRC(src_reg, zero_reg)
        CALC_SUM_SSE_INSIDE_LOOP
        src += src_stride;
        dst += dst_stride;
      }
      // x_offset = 0 and y_offset = bilin interpolation
    } else {
      __m256i filter, pw8, src_next_reg;

      y_offset <<= 5;
      filter = _mm256_load_si256(
          (__m256i const *)(bilinear_filters_avx2 + y_offset));
      pw8 = _mm256_set1_epi16(8);
      for (i = 0; i < height; i++) {
        LOAD_SRC_DST
        MERGE_NEXT_SRC(src_reg, src_stride)
        FILTER_SRC(filter)
        src_reg = _mm256_packus_epi16(exp_src_lo, exp_src_hi);
        sec_reg = _mm256_loadu_si256((__m256i const *)(sec));
        src_reg = _mm256_avg_epu8(src_reg, sec_reg);
        sec += sec_stride;
        MERGE_WITH_SRC(src_reg, zero_reg)
        CALC_SUM_SSE_INSIDE_LOOP
        src += src_stride;
        dst += dst_stride;
      }
    }
    // x_offset = 8  and y_offset = 0
  } else if (x_offset == 8) {
    if (y_offset == 0) {
      __m256i src_next_reg;
      for (i = 0; i < height; i++) {
        LOAD_SRC_DST
        AVG_NEXT_SRC(src_reg, 1)
        sec_reg = _mm256_loadu_si256((__m256i const *)(sec));
        src_reg = _mm256_avg_epu8(src_reg, sec_reg);
        sec += sec_stride;
        // expand each byte to 2 bytes
        MERGE_WITH_SRC(src_reg, zero_reg)
        CALC_SUM_SSE_INSIDE_LOOP
        src += src_stride;
        dst += dst_stride;
      }
      // x_offset = 8  and y_offset = 8
    } else if (y_offset == 8) {
      __m256i src_next_reg, src_avg;
      // load source and another source starting from the next
      // following byte
      src_reg = _mm256_loadu_si256((__m256i const *)(src));
      AVG_NEXT_SRC(src_reg, 1)
      for (i = 0; i < height; i++) {
        // save current source average
        src_avg = src_reg;
        src += src_stride;
        LOAD_SRC_DST
        AVG_NEXT_SRC(src_reg, 1)
        // average between previous average to current average
        src_avg = _mm256_avg_epu8(src_avg, src_reg);
        sec_reg = _mm256_loadu_si256((__m256i const *)(sec));
        src_avg = _mm256_avg_epu8(src_avg, sec_reg);
        sec += sec_stride;
        // expand each byte to 2 bytes
        MERGE_WITH_SRC(src_avg, zero_reg)
        CALC_SUM_SSE_INSIDE_LOOP
        dst += dst_stride;
      }
      // x_offset = 8  and y_offset = bilin interpolation
    } else {
      __m256i filter, pw8, src_next_reg, src_avg;
      y_offset <<= 5;
      filter = _mm256_load_si256(
          (__m256i const *)(bilinear_filters_avx2 + y_offset));
      pw8 = _mm256_set1_epi16(8);
      // load source and another source starting from the next
      // following byte
      src_reg = _mm256_loadu_si256((__m256i const *)(src));
      AVG_NEXT_SRC(src_reg, 1)
      for (i = 0; i < height; i++) {
        // save current source average
        src_avg = src_reg;
        src += src_stride;
        LOAD_SRC_DST
        AVG_NEXT_SRC(src_reg, 1)
        MERGE_WITH_SRC(src_avg, src_reg)
        FILTER_SRC(filter)
        src_avg = _mm256_packus_epi16(exp_src_lo, exp_src_hi);
        sec_reg = _mm256_loadu_si256((__m256i const *)(sec));
        src_avg = _mm256_avg_epu8(src_avg, sec_reg);
        // expand each byte to 2 bytes
        MERGE_WITH_SRC(src_avg, zero_reg)
        sec += sec_stride;
        CALC_SUM_SSE_INSIDE_LOOP
        dst += dst_stride;
      }
    }
    // x_offset = bilin interpolation and y_offset = 0
  } else {
    if (y_offset == 0) {
      __m256i filter, pw8, src_next_reg;
      x_offset <<= 5;
      filter = _mm256_load_si256(
          (__m256i const *)(bilinear_filters_avx2 + x_offset));
      pw8 = _mm256_set1_epi16(8);
      for (i = 0; i < height; i++) {
        LOAD_SRC_DST
        MERGE_NEXT_SRC(src_reg, 1)
        FILTER_SRC(filter)
        src_reg = _mm256_packus_epi16(exp_src_lo, exp_src_hi);
        sec_reg = _mm256_loadu_si256((__m256i const *)(sec));
        src_reg = _mm256_avg_epu8(src_reg, sec_reg);
        MERGE_WITH_SRC(src_reg, zero_reg)
        sec += sec_stride;
        CALC_SUM_SSE_INSIDE_LOOP
        src += src_stride;
        dst += dst_stride;
      }
      // x_offset = bilin interpolation and y_offset = 8
    } else if (y_offset == 8) {
      __m256i filter, pw8, src_next_reg, src_pack;
      x_offset <<= 5;
      filter = _mm256_load_si256(
          (__m256i const *)(bilinear_filters_avx2 + x_offset));
      pw8 = _mm256_set1_epi16(8);
      src_reg = _mm256_loadu_si256((__m256i const *)(src));
      MERGE_NEXT_SRC(src_reg, 1)
      FILTER_SRC(filter)
      // convert each 16 bit to 8 bit to each low and high lane source
      src_pack = _mm256_packus_epi16(exp_src_lo, exp_src_hi);
      for (i = 0; i < height; i++) {
        src += src_stride;
        LOAD_SRC_DST
        MERGE_NEXT_SRC(src_reg, 1)
        FILTER_SRC(filter)
        src_reg = _mm256_packus_epi16(exp_src_lo, exp_src_hi);
        // average between previous pack to the current
        src_pack = _mm256_avg_epu8(src_pack, src_reg);
        sec_reg = _mm256_loadu_si256((__m256i const *)(sec));
        src_pack = _mm256_avg_epu8(src_pack, sec_reg);
        sec += sec_stride;
        MERGE_WITH_SRC(src_pack, zero_reg)
        src_pack = src_reg;
        CALC_SUM_SSE_INSIDE_LOOP
        dst += dst_stride;
      }
      // x_offset = bilin interpolation and y_offset = bilin interpolation
    } else {
      __m256i xfilter, yfilter, pw8, src_next_reg, src_pack;
      x_offset <<= 5;
      xfilter = _mm256_load_si256(
          (__m256i const *)(bilinear_filters_avx2 + x_offset));
      y_offset <<= 5;
      yfilter = _mm256_load_si256(
          (__m256i const *)(bilinear_filters_avx2 + y_offset));
      pw8 = _mm256_set1_epi16(8);
      // load source and another source starting from the next
      // following byte
      src_reg = _mm256_loadu_si256((__m256i const *)(src));
      MERGE_NEXT_SRC(src_reg, 1)

      FILTER_SRC(xfilter)
      // convert each 16 bit to 8 bit to each low and high lane source
      src_pack = _mm256_packus_epi16(exp_src_lo, exp_src_hi);
      for (i = 0; i < height; i++) {
        src += src_stride;
        LOAD_SRC_DST
        MERGE_NEXT_SRC(src_reg, 1)
        FILTER_SRC(xfilter)
        src_reg = _mm256_packus_epi16(exp_src_lo, exp_src_hi);
        // merge previous pack to current pack source
        MERGE_WITH_SRC(src_pack, src_reg)
        // filter the source
        FILTER_SRC(yfilter)
        src_pack = _mm256_packus_epi16(exp_src_lo, exp_src_hi);
        sec_reg = _mm256_loadu_si256((__m256i const *)(sec));
        src_pack = _mm256_avg_epu8(src_pack, sec_reg);
        MERGE_WITH_SRC(src_pack, zero_reg)
        src_pack = src_reg;
        sec += sec_stride;
        CALC_SUM_SSE_INSIDE_LOOP
        dst += dst_stride;
      }
    }
  }
  CALC_SUM_AND_SSE
  _mm256_zeroupper();
  return sum;
}