convolve.c 20 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

209 210 211 212
    // we do filter with fewer taps first to reduce hardware implementation
    // complexity
    if (filter_params_y.taps < filter_params_x.taps) {
      int intermediate_width;
213
      int temp_stride = max_intermediate_size;
214 215 216 217
      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;
218
      assert(intermediate_width <= max_intermediate_size);
Angie Chiang's avatar
Angie Chiang committed
219

220
      assert(filter_params.taps <= MAX_FILTER_TAP);
221

222 223 224
      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);
225

226 227
      filter_params = filter_params_x;
      assert(filter_params.taps <= MAX_FILTER_TAP);
228

229 230 231
      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);
232
    } else
233
#endif  // CONFIG_DUAL_FILTER
234 235
    {
      int intermediate_height;
236
      int temp_stride = MAX_SB_SIZE;
237 238 239 240 241 242 243 244 245
#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;
246 247
      assert(intermediate_height <= max_intermediate_size);
      (void)max_intermediate_size;
248 249 250

      assert(filter_params.taps <= MAX_FILTER_TAP);

251 252 253 254
      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);
255 256 257 258 259 260

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

261 262 263
      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);
264
    }
Angie Chiang's avatar
Angie Chiang committed
265
  }
266 267
}

Angie Chiang's avatar
Angie Chiang committed
268 269 270 271 272
void av1_convolve_init_c(void) {
  // A placeholder for SIMD initialization
  return;
}

Yaowu Xu's avatar
Yaowu Xu committed
273 274 275 276 277 278
#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) {
279
  int x, y;
280
  int filter_size = filter_params.taps;
281 282 283 284 285
  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
286
      const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
287
          filter_params, x_q4 & SUBPEL_MASK);
288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303
      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
304 305 306 307 308
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) {
309
  int x, y;
310
  int filter_size = filter_params.taps;
311 312 313 314 315 316
  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
317
      const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
318
          filter_params, y_q4 & SUBPEL_MASK);
319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337
      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
338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359
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;
    }
  }
}

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
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);
391

392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407
  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
408 409
void av1_highbd_convolve(const uint8_t *src8, int src_stride, uint8_t *dst8,
                         int dst_stride, int w, int h,
410
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
411
                         const InterpFilter *interp_filter,
412
#else
James Zern's avatar
James Zern committed
413
                         const InterpFilter interp_filter,
414
#endif
Yaowu Xu's avatar
Yaowu Xu committed
415 416 417
                         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
418 419 420 421
  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;
422 423 424 425 426 427

  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
428
  if (ignore_horiz && ignore_vert) {
429
    highbd_convolve_copy(src, src_stride, dst, dst_stride, w, h, ref_idx, bd);
Angie Chiang's avatar
Angie Chiang committed
430
  } else if (ignore_vert) {
431 432
#if CONFIG_DUAL_FILTER
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
433
        av1_get_interp_filter_params(interp_filter[1 + 2 * ref_idx]);
434 435
#else
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
436
        av1_get_interp_filter_params(interp_filter);
437
#endif
438 439 440
    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
441
  } else if (ignore_horiz) {
442 443
#if CONFIG_DUAL_FILTER
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
444
        av1_get_interp_filter_params(interp_filter[0 + 2 * ref_idx]);
445 446
#else
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
447
        av1_get_interp_filter_params(interp_filter);
448
#endif
449 450 451
    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
452
  } else {
453 454 455 456 457
    // 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
458
    uint8_t *temp8 = CONVERT_TO_BYTEPTR(temp);
459 460 461
    int max_intermediate_size = ((MAX_SB_SIZE * 2 + 16) + 16);
    int filter_size;
    InterpFilterParams filter_params;
462
#if CONFIG_DUAL_FILTER
Yaowu Xu's avatar
Yaowu Xu committed
463 464 465 466
    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]);
467 468 469 470 471 472
    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);
    }
473
#endif
474

475 476 477 478 479 480 481 482 483 484 485
#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
486

487 488 489
      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);
490

491 492 493 494 495 496 497 498 499 500 501
      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;
502
#if CONFIG_DUAL_FILTER
503 504 505 506 507
      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;
508
#endif
509 510 511 512 513 514 515 516 517 518 519 520 521 522 523
      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);
524

525 526 527 528
      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
529
  }
530
}
Yaowu Xu's avatar
Yaowu Xu committed
531
#endif  // CONFIG_AOM_HIGHBITDEPTH