convolve.c 19.7 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11
/*
 * 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.
 */

12
#include <assert.h>
Angie Chiang's avatar
Angie Chiang committed
13
#include <string.h>
14

15
#include "./aom_dsp_rtcd.h"
Yaowu Xu's avatar
Yaowu Xu committed
16
#include "./av1_rtcd.h"
17
#include "av1/common/convolve.h"
18
#include "av1/common/filter.h"
Yaowu Xu's avatar
Yaowu Xu committed
19
#include "aom_dsp/aom_dsp_common.h"
20
#include "aom_ports/mem.h"
21

22 23
#define MAX_BLOCK_WIDTH (MAX_SB_SIZE)
#define MAX_BLOCK_HEIGHT (MAX_SB_SIZE)
24 25 26
#define MAX_STEP (32)
#define MAX_FILTER_TAP (12)

Yaowu Xu's avatar
Yaowu Xu committed
27 28 29 30
void av1_convolve_horiz_c(const uint8_t *src, int src_stride, uint8_t *dst,
                          int dst_stride, int w, int h,
                          const InterpFilterParams filter_params,
                          const int subpel_x_q4, int x_step_q4, int avg) {
31
  int x, y;
32
  int filter_size = filter_params.taps;
33 34 35 36 37
  src -= filter_size / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = subpel_x_q4;
    for (x = 0; x < w; ++x) {
      const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
Yaowu Xu's avatar
Yaowu Xu committed
38
      const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
39
          filter_params, x_q4 & SUBPEL_MASK);
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k) sum += src_x[k] * x_filter[k];
      if (avg) {
        dst[x] = ROUND_POWER_OF_TWO(
            dst[x] + clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)), 1);
      } else {
        dst[x] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
      }
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
55 56 57 58
void av1_convolve_vert_c(const uint8_t *src, int src_stride, uint8_t *dst,
                         int dst_stride, int w, int h,
                         const InterpFilterParams filter_params,
                         const int subpel_y_q4, int y_step_q4, int avg) {
59
  int x, y;
60
  int filter_size = filter_params.taps;
61 62 63 64 65 66
  src -= src_stride * (filter_size / 2 - 1);

  for (x = 0; x < w; ++x) {
    int y_q4 = subpel_y_q4;
    for (y = 0; y < h; ++y) {
      const uint8_t *const src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
Yaowu Xu's avatar
Yaowu Xu committed
67
      const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
68
          filter_params, y_q4 & SUBPEL_MASK);
69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      if (avg) {
        dst[y * dst_stride] = ROUND_POWER_OF_TWO(
            dst[y * dst_stride] +
                clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)),
            1);
      } else {
        dst[y * dst_stride] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
      }
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

Angie Chiang's avatar
Angie Chiang committed
87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107
static void convolve_copy(const uint8_t *src, int src_stride, uint8_t *dst,
                          int dst_stride, int w, int h, int avg) {
  if (avg == 0) {
    int r;
    for (r = 0; r < h; ++r) {
      memcpy(dst, src, w);
      src += src_stride;
      dst += dst_stride;
    }
  } else {
    int r, c;
    for (r = 0; r < h; ++r) {
      for (c = 0; c < w; ++c) {
        dst[c] = clip_pixel(ROUND_POWER_OF_TWO(dst[c] + src[c], 1));
      }
      src += src_stride;
      dst += dst_stride;
    }
  }
}

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
void av1_convolve_horiz_facade(const uint8_t *src, int src_stride, uint8_t *dst,
                               int dst_stride, int w, int h,
                               const InterpFilterParams filter_params,
                               const int subpel_x_q4, int x_step_q4, int avg) {
  if (filter_params.taps == SUBPEL_TAPS) {
    const int16_t *filter_x =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_x_q4);
    if (avg == 0)
      aom_convolve8_horiz(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
                          NULL, -1, w, h);
    else
      aom_convolve8_avg_horiz(src, src_stride, dst, dst_stride, filter_x,
                              x_step_q4, NULL, -1, w, h);
  } else {
    av1_convolve_horiz(src, src_stride, dst, dst_stride, w, h, filter_params,
                       subpel_x_q4, x_step_q4, avg);
  }
}

void av1_convolve_vert_facade(const uint8_t *src, int src_stride, uint8_t *dst,
                              int dst_stride, int w, int h,
                              const InterpFilterParams filter_params,
                              const int subpel_y_q4, int y_step_q4, int avg) {
  if (filter_params.taps == SUBPEL_TAPS) {
    const int16_t *filter_y =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_y_q4);
    if (avg == 0) {
      aom_convolve8_vert(src, src_stride, dst, dst_stride, NULL, -1, filter_y,
                         y_step_q4, w, h);
    } else {
      aom_convolve8_avg_vert(src, src_stride, dst, dst_stride, NULL, -1,
                             filter_y, y_step_q4, w, h);
    }
  } else {
    av1_convolve_vert(src, src_stride, dst, dst_stride, w, h, filter_params,
                      subpel_y_q4, y_step_q4, avg);
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
147 148
void av1_convolve(const uint8_t *src, int src_stride, uint8_t *dst,
                  int dst_stride, int w, int h,
149
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
150
                  const InterpFilter *interp_filter,
151
#else
James Zern's avatar
James Zern committed
152
                  const InterpFilter interp_filter,
153
#endif
Yaowu Xu's avatar
Yaowu Xu committed
154 155
                  const int subpel_x_q4, int x_step_q4, const int subpel_y_q4,
                  int y_step_q4, int ref_idx) {
Angie Chiang's avatar
Angie Chiang committed
156 157
  int ignore_horiz = x_step_q4 == 16 && subpel_x_q4 == 0;
  int ignore_vert = y_step_q4 == 16 && subpel_y_q4 == 0;
158 159 160 161 162 163

  assert(w <= MAX_BLOCK_WIDTH);
  assert(h <= MAX_BLOCK_HEIGHT);
  assert(y_step_q4 <= MAX_STEP);
  assert(x_step_q4 <= MAX_STEP);

Angie Chiang's avatar
Angie Chiang committed
164
  if (ignore_horiz && ignore_vert) {
165
    convolve_copy(src, src_stride, dst, dst_stride, w, h, ref_idx);
Angie Chiang's avatar
Angie Chiang committed
166
  } else if (ignore_vert) {
167 168
#if CONFIG_DUAL_FILTER
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
169
        av1_get_interp_filter_params(interp_filter[1 + 2 * ref_idx]);
170 171
#else
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
172
        av1_get_interp_filter_params(interp_filter);
173 174
#endif
    assert(filter_params.taps <= MAX_FILTER_TAP);
175 176
    av1_convolve_horiz_facade(src, src_stride, dst, dst_stride, w, h,
                              filter_params, subpel_x_q4, x_step_q4, ref_idx);
Angie Chiang's avatar
Angie Chiang committed
177
  } else if (ignore_horiz) {
178 179
#if CONFIG_DUAL_FILTER
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
180
        av1_get_interp_filter_params(interp_filter[2 * ref_idx]);
181 182
#else
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
183
        av1_get_interp_filter_params(interp_filter);
184 185
#endif
    assert(filter_params.taps <= MAX_FILTER_TAP);
186 187
    av1_convolve_vert_facade(src, src_stride, dst, dst_stride, w, h,
                             filter_params, subpel_y_q4, y_step_q4, ref_idx);
Angie Chiang's avatar
Angie Chiang committed
188
  } else {
189 190 191 192 193 194
    // temp's size is set to a 256 aligned value to facilitate SIMD
    // implementation. The value is greater than (maximum possible intermediate
    // height or width) * MAX_SB_SIZE
    DECLARE_ALIGNED(16, uint8_t,
                    temp[((MAX_SB_SIZE * 2 + 16) + 16) * MAX_SB_SIZE]);
    int max_intermediate_size = ((MAX_SB_SIZE * 2 + 16) + 16);
195 196
    int filter_size;
    InterpFilterParams filter_params;
197
#if CONFIG_DUAL_FILTER
198
    InterpFilterParams filter_params_x =
Yaowu Xu's avatar
Yaowu Xu committed
199
        av1_get_interp_filter_params(interp_filter[1 + 2 * ref_idx]);
200
    InterpFilterParams filter_params_y =
Yaowu Xu's avatar
Yaowu Xu committed
201
        av1_get_interp_filter_params(interp_filter[0 + 2 * ref_idx]);
202 203 204 205 206 207
    if (interp_filter[0 + 2 * ref_idx] == MULTITAP_SHARP &&
        interp_filter[1 + 2 * ref_idx] == MULTITAP_SHARP) {
      // Avoid two directions both using 12-tap filter.
      // This will reduce hardware implementation cost.
      filter_params_y = av1_get_interp_filter_params(EIGHTTAP_SHARP);
    }
208
#endif
209

210 211 212 213 214
#if CONFIG_DUAL_FILTER
    // we do filter with fewer taps first to reduce hardware implementation
    // complexity
    if (filter_params_y.taps < filter_params_x.taps) {
      int intermediate_width;
215
      int temp_stride = max_intermediate_size;
216 217 218 219
      filter_params = filter_params_y;
      filter_size = filter_params_x.taps;
      intermediate_width =
          (((w - 1) * x_step_q4 + subpel_x_q4) >> SUBPEL_BITS) + filter_size;
220
      assert(intermediate_width <= max_intermediate_size);
Angie Chiang's avatar
Angie Chiang committed
221

222
      assert(filter_params.taps <= MAX_FILTER_TAP);
223

224 225 226
      av1_convolve_vert_facade(src - (filter_size / 2 - 1), src_stride, temp,
                               temp_stride, intermediate_width, h,
                               filter_params, subpel_y_q4, y_step_q4, 0);
227

228 229
      filter_params = filter_params_x;
      assert(filter_params.taps <= MAX_FILTER_TAP);
230

231 232 233
      av1_convolve_horiz_facade(temp + (filter_size / 2 - 1), temp_stride, dst,
                                dst_stride, w, h, filter_params, subpel_x_q4,
                                x_step_q4, ref_idx);
234
    } else
235
#endif  // CONFIG_DUAL_FILTER
236 237
    {
      int intermediate_height;
238
      int temp_stride = MAX_SB_SIZE;
239 240 241 242 243 244 245 246 247
#if CONFIG_DUAL_FILTER
      filter_params = filter_params_x;
      filter_size = filter_params_y.taps;
#else
      filter_params = av1_get_interp_filter_params(interp_filter);
      filter_size = filter_params.taps;
#endif
      intermediate_height =
          (((h - 1) * y_step_q4 + subpel_y_q4) >> SUBPEL_BITS) + filter_size;
248 249
      assert(intermediate_height <= max_intermediate_size);
      (void)max_intermediate_size;
250 251 252

      assert(filter_params.taps <= MAX_FILTER_TAP);

253 254 255 256
      av1_convolve_horiz_facade(src - src_stride * (filter_size / 2 - 1),
                                src_stride, temp, temp_stride, w,
                                intermediate_height, filter_params, subpel_x_q4,
                                x_step_q4, 0);
257 258 259 260 261 262

#if CONFIG_DUAL_FILTER
      filter_params = filter_params_y;
#endif
      assert(filter_params.taps <= MAX_FILTER_TAP);

263 264 265
      av1_convolve_vert_facade(temp + temp_stride * (filter_size / 2 - 1),
                               temp_stride, dst, dst_stride, w, h,
                               filter_params, subpel_y_q4, y_step_q4, ref_idx);
266
    }
Angie Chiang's avatar
Angie Chiang committed
267
  }
268 269
}

Angie Chiang's avatar
Angie Chiang committed
270 271 272 273 274
void av1_convolve_init_c(void) {
  // A placeholder for SIMD initialization
  return;
}

Yaowu Xu's avatar
Yaowu Xu committed
275 276 277 278 279 280
#if CONFIG_AOM_HIGHBITDEPTH
void av1_highbd_convolve_horiz_c(const uint16_t *src, int src_stride,
                                 uint16_t *dst, int dst_stride, int w, int h,
                                 const InterpFilterParams filter_params,
                                 const int subpel_x_q4, int x_step_q4, int avg,
                                 int bd) {
281
  int x, y;
282
  int filter_size = filter_params.taps;
283 284 285 286 287
  src -= filter_size / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = subpel_x_q4;
    for (x = 0; x < w; ++x) {
      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
Yaowu Xu's avatar
Yaowu Xu committed
288
      const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
289
          filter_params, x_q4 & SUBPEL_MASK);
290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k) sum += src_x[k] * x_filter[k];
      if (avg)
        dst[x] = ROUND_POWER_OF_TWO(
            dst[x] +
                clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd),
            1);
      else
        dst[x] = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
306 307 308 309 310
void av1_highbd_convolve_vert_c(const uint16_t *src, int src_stride,
                                uint16_t *dst, int dst_stride, int w, int h,
                                const InterpFilterParams filter_params,
                                const int subpel_y_q4, int y_step_q4, int avg,
                                int bd) {
311
  int x, y;
312
  int filter_size = filter_params.taps;
313 314 315 316 317 318
  src -= src_stride * (filter_size / 2 - 1);

  for (x = 0; x < w; ++x) {
    int y_q4 = subpel_y_q4;
    for (y = 0; y < h; ++y) {
      const uint16_t *const src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
Yaowu Xu's avatar
Yaowu Xu committed
319
      const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
320
          filter_params, y_q4 & SUBPEL_MASK);
321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      if (avg) {
        dst[y * dst_stride] = ROUND_POWER_OF_TWO(
            dst[y * dst_stride] +
                clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd),
            1);
      } else {
        dst[y * dst_stride] =
            clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
      }
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

Angie Chiang's avatar
Angie Chiang committed
340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361
static void highbd_convolve_copy(const uint16_t *src, int src_stride,
                                 uint16_t *dst, int dst_stride, int w, int h,
                                 int avg, int bd) {
  if (avg == 0) {
    int r;
    for (r = 0; r < h; ++r) {
      memcpy(dst, src, w * sizeof(*src));
      src += src_stride;
      dst += dst_stride;
    }
  } else {
    int r, c;
    for (r = 0; r < h; ++r) {
      for (c = 0; c < w; ++c) {
        dst[c] = clip_pixel_highbd(ROUND_POWER_OF_TWO(dst[c] + src[c], 1), bd);
      }
      src += src_stride;
      dst += dst_stride;
    }
  }
}

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
void av1_highbd_convolve_horiz_facade(const uint8_t *src8, int src_stride,
                                      uint8_t *dst8, int dst_stride, int w,
                                      int h,
                                      const InterpFilterParams filter_params,
                                      const int subpel_x_q4, int x_step_q4,
                                      int avg, int bd) {
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  if (filter_params.taps == SUBPEL_TAPS) {
    const int16_t *filter_x =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_x_q4);
    if (avg == 0)
      aom_highbd_convolve8_horiz(src8, src_stride, dst8, dst_stride, filter_x,
                                 x_step_q4, NULL, -1, w, h, bd);
    else
      aom_highbd_convolve8_avg_horiz(src8, src_stride, dst8, dst_stride,
                                     filter_x, x_step_q4, NULL, -1, w, h, bd);
  } else {
    av1_highbd_convolve_horiz(src, src_stride, dst, dst_stride, w, h,
                              filter_params, subpel_x_q4, x_step_q4, avg, bd);
  }
}

void av1_highbd_convolve_vert_facade(const uint8_t *src8, int src_stride,
                                     uint8_t *dst8, int dst_stride, int w,
                                     int h,
                                     const InterpFilterParams filter_params,
                                     const int subpel_y_q4, int y_step_q4,
                                     int avg, int bd) {
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
393

394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409
  if (filter_params.taps == SUBPEL_TAPS) {
    const int16_t *filter_y =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_y_q4);
    if (avg == 0) {
      aom_highbd_convolve8_vert(src8, src_stride, dst8, dst_stride, NULL, -1,
                                filter_y, y_step_q4, w, h, bd);
    } else {
      aom_highbd_convolve8_avg_vert(src8, src_stride, dst8, dst_stride, NULL,
                                    -1, filter_y, y_step_q4, w, h, bd);
    }
  } else {
    av1_highbd_convolve_vert(src, src_stride, dst, dst_stride, w, h,
                             filter_params, subpel_y_q4, y_step_q4, avg, bd);
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
410 411
void av1_highbd_convolve(const uint8_t *src8, int src_stride, uint8_t *dst8,
                         int dst_stride, int w, int h,
412
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
413
                         const InterpFilter *interp_filter,
414
#else
James Zern's avatar
James Zern committed
415
                         const InterpFilter interp_filter,
416
#endif
Yaowu Xu's avatar
Yaowu Xu committed
417 418 419
                         const int subpel_x_q4, int x_step_q4,
                         const int subpel_y_q4, int y_step_q4, int ref_idx,
                         int bd) {
Angie Chiang's avatar
Angie Chiang committed
420 421 422 423
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  int ignore_horiz = x_step_q4 == 16 && subpel_x_q4 == 0;
  int ignore_vert = y_step_q4 == 16 && subpel_y_q4 == 0;
424 425 426 427 428 429

  assert(w <= MAX_BLOCK_WIDTH);
  assert(h <= MAX_BLOCK_HEIGHT);
  assert(y_step_q4 <= MAX_STEP);
  assert(x_step_q4 <= MAX_STEP);

Angie Chiang's avatar
Angie Chiang committed
430
  if (ignore_horiz && ignore_vert) {
431
    highbd_convolve_copy(src, src_stride, dst, dst_stride, w, h, ref_idx, bd);
Angie Chiang's avatar
Angie Chiang committed
432
  } else if (ignore_vert) {
433 434
#if CONFIG_DUAL_FILTER
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
435
        av1_get_interp_filter_params(interp_filter[1 + 2 * ref_idx]);
436 437
#else
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
438
        av1_get_interp_filter_params(interp_filter);
439
#endif
440 441 442
    av1_highbd_convolve_horiz_facade(src8, src_stride, dst8, dst_stride, w, h,
                                     filter_params, subpel_x_q4, x_step_q4,
                                     ref_idx, bd);
Angie Chiang's avatar
Angie Chiang committed
443
  } else if (ignore_horiz) {
444 445
#if CONFIG_DUAL_FILTER
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
446
        av1_get_interp_filter_params(interp_filter[0 + 2 * ref_idx]);
447 448
#else
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
449
        av1_get_interp_filter_params(interp_filter);
450
#endif
451 452 453
    av1_highbd_convolve_vert_facade(src8, src_stride, dst8, dst_stride, w, h,
                                    filter_params, subpel_y_q4, y_step_q4,
                                    ref_idx, bd);
Angie Chiang's avatar
Angie Chiang committed
454
  } else {
455 456 457 458 459
    // temp's size is set to a 256 aligned value to facilitate SIMD
    // implementation. The value is greater than (maximum possible intermediate
    // height or width) * MAX_SB_SIZE
    DECLARE_ALIGNED(16, uint16_t,
                    temp[((MAX_SB_SIZE * 2 + 16) + 16) * MAX_SB_SIZE]);
Yaowu Xu's avatar
Yaowu Xu committed
460
    uint8_t *temp8 = CONVERT_TO_BYTEPTR(temp);
461 462 463
    int max_intermediate_size = ((MAX_SB_SIZE * 2 + 16) + 16);
    int filter_size;
    InterpFilterParams filter_params;
464
#if CONFIG_DUAL_FILTER
Yaowu Xu's avatar
Yaowu Xu committed
465 466 467 468
    InterpFilterParams filter_params_x =
        av1_get_interp_filter_params(interp_filter[1 + 2 * ref_idx]);
    InterpFilterParams filter_params_y =
        av1_get_interp_filter_params(interp_filter[0 + 2 * ref_idx]);
469
#endif
470

471 472 473 474 475 476 477 478 479 480 481
#if CONFIG_DUAL_FILTER
    if (filter_params_y.taps < filter_params_x.taps) {
      int intermediate_width;
      int temp_stride = max_intermediate_size;
      filter_params = filter_params_y;
      filter_size = filter_params_x.taps;
      intermediate_width =
          (((w - 1) * x_step_q4 + subpel_x_q4) >> SUBPEL_BITS) + filter_size;
      assert(intermediate_width <= max_intermediate_size);

      assert(filter_params.taps <= MAX_FILTER_TAP);
Angie Chiang's avatar
Angie Chiang committed
482

483 484 485
      av1_highbd_convolve_vert_facade(
          src8 - (filter_size / 2 - 1), src_stride, temp8, temp_stride,
          intermediate_width, h, filter_params, subpel_y_q4, y_step_q4, 0, bd);
486

487 488 489 490 491 492 493 494 495 496 497
      filter_params = filter_params_x;
      assert(filter_params.taps <= MAX_FILTER_TAP);

      av1_highbd_convolve_horiz_facade(
          temp8 + (filter_size / 2 - 1), temp_stride, dst8, dst_stride, w, h,
          filter_params, subpel_x_q4, x_step_q4, ref_idx, bd);
    } else
#endif  // CONFIG_DUAL_FILTER
    {
      int intermediate_height;
      int temp_stride = MAX_SB_SIZE;
498
#if CONFIG_DUAL_FILTER
499 500 501 502 503
      filter_params = filter_params_x;
      filter_size = filter_params_y.taps;
#else
      filter_params = av1_get_interp_filter_params(interp_filter);
      filter_size = filter_params.taps;
504
#endif
505 506 507 508 509 510 511 512 513 514 515 516 517 518 519
      intermediate_height =
          (((h - 1) * y_step_q4 + subpel_y_q4) >> SUBPEL_BITS) + filter_size;
      assert(intermediate_height <= max_intermediate_size);
      (void)max_intermediate_size;

      av1_highbd_convolve_horiz_facade(
          src8 - src_stride * (filter_size / 2 - 1), src_stride, temp8,
          temp_stride, w, intermediate_height, filter_params, subpel_x_q4,
          x_step_q4, 0, bd);

#if CONFIG_DUAL_FILTER
      filter_params = filter_params_y;
#endif
      filter_size = filter_params.taps;
      assert(filter_params.taps <= MAX_FILTER_TAP);
520

521 522 523 524
      av1_highbd_convolve_vert_facade(
          temp8 + temp_stride * (filter_size / 2 - 1), temp_stride, dst8,
          dst_stride, w, h, filter_params, subpel_y_q4, y_step_q4, ref_idx, bd);
    }
Angie Chiang's avatar
Angie Chiang committed
525
  }
526
}
Yaowu Xu's avatar
Yaowu Xu committed
527
#endif  // CONFIG_AOM_HIGHBITDEPTH