variance.c 21.5 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/*
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

#include "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"

#include "vpx_ports/mem.h"
#include "vpx/vpx_integer.h"

17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
#include "vpx_dsp/variance.h"

static const uint8_t bilinear_filters[8][2] = {
  { 128,   0  },
  { 112,  16  },
  {  96,  32  },
  {  80,  48  },
  {  64,  64  },
  {  48,  80  },
  {  32,  96  },
  {  16, 112  },
};

uint32_t vpx_get4x4sse_cs_c(const uint8_t *a, int  a_stride,
                            const uint8_t *b, int  b_stride) {
32 33 34
  int distortion = 0;
  int r, c;

35 36
  for (r = 0; r < 4; ++r) {
    for (c = 0; c < 4; ++c) {
37 38 39 40 41 42 43 44 45 46 47
      int diff = a[c] - b[c];
      distortion += diff * diff;
    }

    a += a_stride;
    b += b_stride;
  }

  return distortion;
}

48
uint32_t vpx_get_mb_ss_c(const int16_t *a) {
49 50 51 52 53 54 55 56 57
  unsigned int i, sum = 0;

  for (i = 0; i < 256; ++i) {
    sum += a[i] * a[i];
  }

  return sum;
}

58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79
uint32_t vpx_variance_halfpixvar16x16_h_c(const uint8_t *a, int a_stride,
                                          const uint8_t *b, int b_stride,
                                          uint32_t *sse) {
  return vpx_sub_pixel_variance16x16_c(a, a_stride, 4, 0,
                                       b, b_stride, sse);
}


uint32_t vpx_variance_halfpixvar16x16_v_c(const uint8_t *a, int a_stride,
                                          const uint8_t *b, int b_stride,
                                          uint32_t *sse) {
  return vpx_sub_pixel_variance16x16_c(a, a_stride, 0, 4,
                                       b, b_stride, sse);
}

uint32_t vpx_variance_halfpixvar16x16_hv_c(const uint8_t *a, int a_stride,
                                           const uint8_t *b, int b_stride,
                                           uint32_t *sse) {
  return vpx_sub_pixel_variance16x16_c(a, a_stride, 4, 4,
                                       b, b_stride, sse);
}

80 81
static void variance(const uint8_t *a, int  a_stride,
                     const uint8_t *b, int  b_stride,
82
                     int  w, int  h, uint32_t *sse, int *sum) {
83 84 85 86 87
  int i, j;

  *sum = 0;
  *sse = 0;

88 89
  for (i = 0; i < h; ++i) {
    for (j = 0; j < w; ++j) {
90 91 92 93 94 95 96 97 98 99
      const int diff = a[j] - b[j];
      *sum += diff;
      *sse += diff * diff;
    }

    a += a_stride;
    b += b_stride;
  }
}

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
// Applies a 1-D 2-tap bilinear filter to the source block in either horizontal
// or vertical direction to produce the filtered output block. Used to implement
// the first-pass of 2-D separable filter.
//
// Produces int16_t output to retain precision for the next pass. Two filter
// taps should sum to FILTER_WEIGHT. pixel_step defines whether the filter is
// applied horizontally (pixel_step = 1) or vertically (pixel_step = stride).
// It defines the offset required to move from one input to the next.
static void var_filter_block2d_bil_first_pass(const uint8_t *a, uint16_t *b,
                                              unsigned int src_pixels_per_line,
                                              int pixel_step,
                                              unsigned int output_height,
                                              unsigned int output_width,
                                              const uint8_t *filter) {
  unsigned int i, j;

  for (i = 0; i < output_height; ++i) {
    for (j = 0; j < output_width; ++j) {
      b[j] = ROUND_POWER_OF_TWO((int)a[0] * filter[0] +
                          (int)a[pixel_step] * filter[1],
                          FILTER_BITS);

      ++a;
    }

    a += src_pixels_per_line - output_width;
    b += output_width;
  }
}

// Applies a 1-D 2-tap bilinear filter to the source block in either horizontal
// or vertical direction to produce the filtered output block. Used to implement
// the second-pass of 2-D separable filter.
//
// Requires 16-bit input as produced by filter_block2d_bil_first_pass. Two
// filter taps should sum to FILTER_WEIGHT. pixel_step defines whether the
// filter is applied horizontally (pixel_step = 1) or vertically
// (pixel_step = stride). It defines the offset required to move from one input
// to the next. Output is 8-bit.
static void var_filter_block2d_bil_second_pass(const uint16_t *a, uint8_t *b,
                                               unsigned int src_pixels_per_line,
                                               unsigned int pixel_step,
                                               unsigned int output_height,
                                               unsigned int output_width,
                                               const uint8_t *filter) {
  unsigned int  i, j;

  for (i = 0; i < output_height; ++i) {
    for (j = 0; j < output_width; ++j) {
      b[j] = ROUND_POWER_OF_TWO((int)a[0] * filter[0] +
                          (int)a[pixel_step] * filter[1],
                          FILTER_BITS);
      ++a;
    }

    a += src_pixels_per_line - output_width;
    b += output_width;
  }
}

160
#define VAR(W, H) \
161 162 163
uint32_t vpx_variance##W##x##H##_c(const uint8_t *a, int a_stride, \
                                   const uint8_t *b, int b_stride, \
                                   uint32_t *sse) { \
164 165 166 167 168
  int sum; \
  variance(a, a_stride, b, b_stride, W, H, sse, &sum); \
  return *sse - (((int64_t)sum * sum) / (W * H)); \
}

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
#define SUBPIX_VAR(W, H) \
uint32_t vpx_sub_pixel_variance##W##x##H##_c(const uint8_t *a, int a_stride, \
                                             int xoffset, int  yoffset, \
                                             const uint8_t *b, int b_stride, \
                                             uint32_t *sse) { \
  uint16_t fdata3[(H + 1) * W]; \
  uint8_t temp2[H * W]; \
\
  var_filter_block2d_bil_first_pass(a, fdata3, a_stride, 1, H + 1, W, \
                                    bilinear_filters[xoffset]); \
  var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
                                     bilinear_filters[yoffset]); \
\
  return vpx_variance##W##x##H##_c(temp2, W, b, b_stride, sse); \
}

#define SUBPIX_AVG_VAR(W, H) \
uint32_t vpx_sub_pixel_avg_variance##W##x##H##_c(const uint8_t *a, \
                                                 int  a_stride, \
                                                 int xoffset, int  yoffset, \
                                                 const uint8_t *b, \
                                                 int b_stride, \
                                                 uint32_t *sse, \
                                                 const uint8_t *second_pred) { \
  uint16_t fdata3[(H + 1) * W]; \
  uint8_t temp2[H * W]; \
  DECLARE_ALIGNED(16, uint8_t, temp3[H * W]); \
\
  var_filter_block2d_bil_first_pass(a, fdata3, a_stride, 1, H + 1, W, \
                                    bilinear_filters[xoffset]); \
  var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
                                     bilinear_filters[yoffset]); \
\
  vpx_comp_avg_pred(temp3, second_pred, W, H, temp2, W); \
\
  return vpx_variance##W##x##H##_c(temp3, W, b, b_stride, sse); \
}

207 208 209 210 211 212 213
/* Identical to the variance call except it takes an additional parameter, sum,
 * and returns that value using pass-by-reference instead of returning
 * sse - sum^2 / w*h
 */
#define GET_VAR(W, H) \
void vpx_get##W##x##H##var_c(const uint8_t *a, int a_stride, \
                             const uint8_t *b, int b_stride, \
214
                             uint32_t *sse, int *sum) { \
215 216 217 218 219 220 221 222
  variance(a, a_stride, b, b_stride, W, H, sse, sum); \
}

/* Identical to the variance call except it does not calculate the
 * sse - sum^2 / w*h and returns sse in addtion to modifying the passed in
 * variable.
 */
#define MSE(W, H) \
223 224 225
uint32_t vpx_mse##W##x##H##_c(const uint8_t *a, int a_stride, \
                              const uint8_t *b, int b_stride, \
                              uint32_t *sse) { \
226 227 228 229 230
  int sum; \
  variance(a, a_stride, b, b_stride, W, H, sse, &sum); \
  return *sse; \
}

231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249
/* All three forms of the variance are available in the same sizes. */
#define VARIANCES(W, H) \
    VAR(W, H) \
    SUBPIX_VAR(W, H) \
    SUBPIX_AVG_VAR(W, H)

VARIANCES(64, 64)
VARIANCES(64, 32)
VARIANCES(32, 64)
VARIANCES(32, 32)
VARIANCES(32, 16)
VARIANCES(16, 32)
VARIANCES(16, 16)
VARIANCES(16, 8)
VARIANCES(8, 16)
VARIANCES(8, 8)
VARIANCES(8, 4)
VARIANCES(4, 8)
VARIANCES(4, 4)
250 251 252 253 254 255 256 257 258

GET_VAR(16, 16)
GET_VAR(8, 8)

MSE(16, 16)
MSE(16, 8)
MSE(8, 16)
MSE(8, 8)

259 260 261
void vpx_comp_avg_pred_c(uint8_t *comp_pred, const uint8_t *pred,
                         int width, int height,
                         const uint8_t *ref, int ref_stride) {
262 263
  int i, j;

264 265
  for (i = 0; i < height; ++i) {
    for (j = 0; j < width; ++j) {
266 267 268 269 270 271 272 273 274 275 276 277
      const int tmp = pred[j] + ref[j];
      comp_pred[j] = ROUND_POWER_OF_TWO(tmp, 1);
    }
    comp_pred += width;
    pred += width;
    ref += ref_stride;
  }
}

#if CONFIG_VP9_HIGHBITDEPTH
static void highbd_variance64(const uint8_t *a8, int  a_stride,
                              const uint8_t *b8, int  b_stride,
278
                              int w, int h, uint64_t *sse, int64_t *sum) {
279 280 281 282 283 284 285
  int i, j;

  uint16_t *a = CONVERT_TO_SHORTPTR(a8);
  uint16_t *b = CONVERT_TO_SHORTPTR(b8);
  *sum = 0;
  *sse = 0;

286 287
  for (i = 0; i < h; ++i) {
    for (j = 0; j < w; ++j) {
288 289 290 291 292 293 294 295 296 297 298
      const int diff = a[j] - b[j];
      *sum += diff;
      *sse += diff * diff;
    }
    a += a_stride;
    b += b_stride;
  }
}

static void highbd_8_variance(const uint8_t *a8, int  a_stride,
                              const uint8_t *b8, int  b_stride,
299
                              int w, int h, uint32_t *sse, int *sum) {
300
  uint64_t sse_long = 0;
301
  int64_t sum_long = 0;
302
  highbd_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long, &sum_long);
303
  *sse = (uint32_t)sse_long;
304
  *sum = (int)sum_long;
305 306 307 308
}

static void highbd_10_variance(const uint8_t *a8, int  a_stride,
                               const uint8_t *b8, int  b_stride,
309
                               int w, int h, uint32_t *sse, int *sum) {
310
  uint64_t sse_long = 0;
311
  int64_t sum_long = 0;
312 313 314
  highbd_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long, &sum_long);
  *sse = (uint32_t)ROUND_POWER_OF_TWO(sse_long, 4);
  *sum = (int)ROUND_POWER_OF_TWO(sum_long, 2);
315 316 317 318
}

static void highbd_12_variance(const uint8_t *a8, int  a_stride,
                               const uint8_t *b8, int  b_stride,
319
                               int w, int h, uint32_t *sse, int *sum) {
320
  uint64_t sse_long = 0;
321
  int64_t sum_long = 0;
322 323 324
  highbd_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long, &sum_long);
  *sse = (uint32_t)ROUND_POWER_OF_TWO(sse_long, 8);
  *sum = (int)ROUND_POWER_OF_TWO(sum_long, 4);
325 326 327
}

#define HIGHBD_VAR(W, H) \
328 329 330 331 332
uint32_t vpx_highbd_8_variance##W##x##H##_c(const uint8_t *a, \
                                            int a_stride, \
                                            const uint8_t *b, \
                                            int b_stride, \
                                            uint32_t *sse) { \
333 334 335 336 337
  int sum; \
  highbd_8_variance(a, a_stride, b, b_stride, W, H, sse, &sum); \
  return *sse - (((int64_t)sum * sum) / (W * H)); \
} \
\
338 339 340 341 342
uint32_t vpx_highbd_10_variance##W##x##H##_c(const uint8_t *a, \
                                             int a_stride, \
                                             const uint8_t *b, \
                                             int b_stride, \
                                             uint32_t *sse) { \
343 344 345 346 347
  int sum; \
  highbd_10_variance(a, a_stride, b, b_stride, W, H, sse, &sum); \
  return *sse - (((int64_t)sum * sum) / (W * H)); \
} \
\
348 349 350 351 352
uint32_t vpx_highbd_12_variance##W##x##H##_c(const uint8_t *a, \
                                             int a_stride, \
                                             const uint8_t *b, \
                                             int b_stride, \
                                             uint32_t *sse) { \
353 354 355 356 357 358 359
  int sum; \
  highbd_12_variance(a, a_stride, b, b_stride, W, H, sse, &sum); \
  return *sse - (((int64_t)sum * sum) / (W * H)); \
}

#define HIGHBD_GET_VAR(S) \
void vpx_highbd_8_get##S##x##S##var_c(const uint8_t *src, int src_stride, \
360 361
                                      const uint8_t *ref, int ref_stride, \
                                      uint32_t *sse, int *sum) { \
362 363 364 365 366
  highbd_8_variance(src, src_stride, ref, ref_stride, S, S, sse, sum); \
} \
\
void vpx_highbd_10_get##S##x##S##var_c(const uint8_t *src, int src_stride, \
                                       const uint8_t *ref, int ref_stride, \
367
                                       uint32_t *sse, int *sum) { \
368 369 370 371 372
  highbd_10_variance(src, src_stride, ref, ref_stride, S, S, sse, sum); \
} \
\
void vpx_highbd_12_get##S##x##S##var_c(const uint8_t *src, int src_stride, \
                                       const uint8_t *ref, int ref_stride, \
373
                                       uint32_t *sse, int *sum) { \
374 375 376 377
  highbd_12_variance(src, src_stride, ref, ref_stride, S, S, sse, sum); \
}

#define HIGHBD_MSE(W, H) \
378 379 380 381 382
uint32_t vpx_highbd_8_mse##W##x##H##_c(const uint8_t *src, \
                                       int src_stride, \
                                       const uint8_t *ref, \
                                       int ref_stride, \
                                       uint32_t *sse) { \
383 384 385 386 387
  int sum; \
  highbd_8_variance(src, src_stride, ref, ref_stride, W, H, sse, &sum); \
  return *sse; \
} \
\
388 389 390 391 392
uint32_t vpx_highbd_10_mse##W##x##H##_c(const uint8_t *src, \
                                        int src_stride, \
                                        const uint8_t *ref, \
                                        int ref_stride, \
                                        uint32_t *sse) { \
393 394 395 396 397
  int sum; \
  highbd_10_variance(src, src_stride, ref, ref_stride, W, H, sse, &sum); \
  return *sse; \
} \
\
398 399 400 401 402
uint32_t vpx_highbd_12_mse##W##x##H##_c(const uint8_t *src, \
                                        int src_stride, \
                                        const uint8_t *ref, \
                                        int ref_stride, \
                                        uint32_t *sse) { \
403 404 405 406 407
  int sum; \
  highbd_12_variance(src, src_stride, ref, ref_stride, W, H, sse, &sum); \
  return *sse; \
}

408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596
static void highbd_var_filter_block2d_bil_first_pass(
    const uint8_t *src_ptr8,
    uint16_t *output_ptr,
    unsigned int src_pixels_per_line,
    int pixel_step,
    unsigned int output_height,
    unsigned int output_width,
    const uint8_t *filter) {
  unsigned int i, j;
  uint16_t *src_ptr = CONVERT_TO_SHORTPTR(src_ptr8);
  for (i = 0; i < output_height; ++i) {
    for (j = 0; j < output_width; ++j) {
      output_ptr[j] =
          ROUND_POWER_OF_TWO((int)src_ptr[0] * filter[0] +
                             (int)src_ptr[pixel_step] * filter[1],
                             FILTER_BITS);

      ++src_ptr;
    }

    // Next row...
    src_ptr += src_pixels_per_line - output_width;
    output_ptr += output_width;
  }
}

static void highbd_var_filter_block2d_bil_second_pass(
    const uint16_t *src_ptr,
    uint16_t *output_ptr,
    unsigned int src_pixels_per_line,
    unsigned int pixel_step,
    unsigned int output_height,
    unsigned int output_width,
    const uint8_t *filter) {
  unsigned int  i, j;

  for (i = 0; i < output_height; ++i) {
    for (j = 0; j < output_width; ++j) {
      output_ptr[j] =
          ROUND_POWER_OF_TWO((int)src_ptr[0] * filter[0] +
                             (int)src_ptr[pixel_step] * filter[1],
                             FILTER_BITS);
      ++src_ptr;
    }

    src_ptr += src_pixels_per_line - output_width;
    output_ptr += output_width;
  }
}

#define HIGHBD_SUBPIX_VAR(W, H) \
uint32_t vpx_highbd_8_sub_pixel_variance##W##x##H##_c( \
  const uint8_t *src, int  src_stride, \
  int xoffset, int  yoffset, \
  const uint8_t *dst, int dst_stride, \
  uint32_t *sse) { \
  uint16_t fdata3[(H + 1) * W]; \
  uint16_t temp2[H * W]; \
\
  highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
                                           W, bilinear_filters[xoffset]); \
  highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
                                            bilinear_filters[yoffset]); \
\
  return vpx_highbd_8_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), W, dst, \
                                          dst_stride, sse); \
} \
\
uint32_t vpx_highbd_10_sub_pixel_variance##W##x##H##_c( \
  const uint8_t *src, int  src_stride, \
  int xoffset, int  yoffset, \
  const uint8_t *dst, int dst_stride, \
  uint32_t *sse) { \
  uint16_t fdata3[(H + 1) * W]; \
  uint16_t temp2[H * W]; \
\
  highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
                                           W, bilinear_filters[xoffset]); \
  highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
                                            bilinear_filters[yoffset]); \
\
  return vpx_highbd_10_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), \
                                             W, dst, dst_stride, sse); \
} \
\
uint32_t vpx_highbd_12_sub_pixel_variance##W##x##H##_c( \
  const uint8_t *src, int  src_stride, \
  int xoffset, int  yoffset, \
  const uint8_t *dst, int dst_stride, \
  uint32_t *sse) { \
  uint16_t fdata3[(H + 1) * W]; \
  uint16_t temp2[H * W]; \
\
  highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
                                           W, bilinear_filters[xoffset]); \
  highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
                                            bilinear_filters[yoffset]); \
\
  return vpx_highbd_12_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), \
                                             W, dst, dst_stride, sse); \
}

#define HIGHBD_SUBPIX_AVG_VAR(W, H) \
uint32_t vpx_highbd_8_sub_pixel_avg_variance##W##x##H##_c( \
  const uint8_t *src, int  src_stride, \
  int xoffset, int  yoffset, \
  const uint8_t *dst, int dst_stride, \
  uint32_t *sse, \
  const uint8_t *second_pred) { \
  uint16_t fdata3[(H + 1) * W]; \
  uint16_t temp2[H * W]; \
  DECLARE_ALIGNED(16, uint16_t, temp3[H * W]); \
\
  highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
                                           W, bilinear_filters[xoffset]); \
  highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
                                            bilinear_filters[yoffset]); \
\
  vpx_highbd_comp_avg_pred(temp3, second_pred, W, H, \
                           CONVERT_TO_BYTEPTR(temp2), W); \
\
  return vpx_highbd_8_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W, dst, \
                                          dst_stride, sse); \
} \
\
uint32_t vpx_highbd_10_sub_pixel_avg_variance##W##x##H##_c( \
  const uint8_t *src, int  src_stride, \
  int xoffset, int  yoffset, \
  const uint8_t *dst, int dst_stride, \
  uint32_t *sse, \
  const uint8_t *second_pred) { \
  uint16_t fdata3[(H + 1) * W]; \
  uint16_t temp2[H * W]; \
  DECLARE_ALIGNED(16, uint16_t, temp3[H * W]); \
\
  highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
                                           W, bilinear_filters[xoffset]); \
  highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
                                            bilinear_filters[yoffset]); \
\
  vpx_highbd_comp_avg_pred(temp3, second_pred, W, H, \
                           CONVERT_TO_BYTEPTR(temp2), W); \
\
  return vpx_highbd_10_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), \
                                             W, dst, dst_stride, sse); \
} \
\
uint32_t vpx_highbd_12_sub_pixel_avg_variance##W##x##H##_c( \
  const uint8_t *src, int  src_stride, \
  int xoffset, int  yoffset, \
  const uint8_t *dst, int dst_stride, \
  uint32_t *sse, \
  const uint8_t *second_pred) { \
  uint16_t fdata3[(H + 1) * W]; \
  uint16_t temp2[H * W]; \
  DECLARE_ALIGNED(16, uint16_t, temp3[H * W]); \
\
  highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
                                           W, bilinear_filters[xoffset]); \
  highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
                                            bilinear_filters[yoffset]); \
\
  vpx_highbd_comp_avg_pred(temp3, second_pred, W, H, \
                           CONVERT_TO_BYTEPTR(temp2), W); \
\
  return vpx_highbd_12_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), \
                                             W, dst, dst_stride, sse); \
}

/* All three forms of the variance are available in the same sizes. */
#define HIGHBD_VARIANCES(W, H) \
    HIGHBD_VAR(W, H) \
    HIGHBD_SUBPIX_VAR(W, H) \
    HIGHBD_SUBPIX_AVG_VAR(W, H)

HIGHBD_VARIANCES(64, 64)
HIGHBD_VARIANCES(64, 32)
HIGHBD_VARIANCES(32, 64)
HIGHBD_VARIANCES(32, 32)
HIGHBD_VARIANCES(32, 16)
HIGHBD_VARIANCES(16, 32)
HIGHBD_VARIANCES(16, 16)
HIGHBD_VARIANCES(16, 8)
HIGHBD_VARIANCES(8, 16)
HIGHBD_VARIANCES(8, 8)
HIGHBD_VARIANCES(8, 4)
HIGHBD_VARIANCES(4, 8)
HIGHBD_VARIANCES(4, 4)

597 598 599 600 601 602 603 604 605 606 607 608 609 610
HIGHBD_GET_VAR(8)
HIGHBD_GET_VAR(16)

HIGHBD_MSE(16, 16)
HIGHBD_MSE(16, 8)
HIGHBD_MSE(8, 16)
HIGHBD_MSE(8, 8)

void vpx_highbd_comp_avg_pred(uint16_t *comp_pred, const uint8_t *pred8,
                              int width, int height, const uint8_t *ref8,
                              int ref_stride) {
  int i, j;
  uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
  uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
611 612
  for (i = 0; i < height; ++i) {
    for (j = 0; j < width; ++j) {
613 614 615 616 617 618 619 620 621
      const int tmp = pred[j] + ref[j];
      comp_pred[j] = ROUND_POWER_OF_TWO(tmp, 1);
    }
    comp_pred += width;
    pred += width;
    ref += ref_stride;
  }
}
#endif  // CONFIG_VP9_HIGHBITDEPTH