convolve.c 35 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"
Angie Chiang's avatar
Angie Chiang committed
19
#include "av1/common/onyxc_int.h"
Yaowu Xu's avatar
Yaowu Xu committed
20
#include "aom_dsp/aom_dsp_common.h"
21
#include "aom_ports/mem.h"
22

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

Yaowu Xu's avatar
Yaowu Xu committed
27 28 29
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,
30 31
                          const int subpel_x_q4, int x_step_q4,
                          ConvolveParams *conv_params) {
32
  int x, y;
33
  int filter_size = filter_params.taps;
Angie Chiang's avatar
Angie Chiang committed
34
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
35 36 37 38 39
  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
40
      const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
41
          filter_params, x_q4 & SUBPEL_MASK);
42 43
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k) sum += src_x[k] * x_filter[k];
44

Angie Chiang's avatar
Angie Chiang committed
45 46 47 48 49
      sum = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
      if (conv_params->ref)
        dst[x] = ROUND_POWER_OF_TWO(dst[x] + sum, 1);
      else
        dst[x] = sum;
50

51 52 53 54 55 56 57
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
58 59 60
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,
61 62
                         const int subpel_y_q4, int y_step_q4,
                         ConvolveParams *conv_params) {
63
  int x, y;
64
  int filter_size = filter_params.taps;
Angie Chiang's avatar
Angie Chiang committed
65
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
66 67 68 69 70
  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
71
      const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
72
          filter_params, y_q4 & SUBPEL_MASK);
73 74 75
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
76

Angie Chiang's avatar
Angie Chiang committed
77 78 79 80 81
      sum = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
      if (conv_params->ref)
        dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] + sum, 1);
      else
        dst[y * dst_stride] = sum;
82

83 84 85 86 87 88 89
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

Angie Chiang's avatar
Angie Chiang committed
90
static void convolve_copy(const uint8_t *src, int src_stride, uint8_t *dst,
91 92
                          int dst_stride, int w, int h,
                          ConvolveParams *conv_params) {
Angie Chiang's avatar
Angie Chiang committed
93
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
94
  if (conv_params->ref == 0) {
Angie Chiang's avatar
Angie Chiang committed
95
    int r;
Angie Chiang's avatar
Angie Chiang committed
96
    for (r = 0; r < h; ++r) {
Angie Chiang's avatar
Angie Chiang committed
97
      memcpy(dst, src, w);
Angie Chiang's avatar
Angie Chiang committed
98 99 100 101 102 103
      src += src_stride;
      dst += dst_stride;
    }
  } else {
    int r, c;
    for (r = 0; r < h; ++r) {
104
      for (c = 0; c < w; ++c) {
Angie Chiang's avatar
Angie Chiang committed
105
        dst[c] = clip_pixel(ROUND_POWER_OF_TWO(dst[c] + src[c], 1));
106
      }
Angie Chiang's avatar
Angie Chiang committed
107 108 109 110 111 112
      src += src_stride;
      dst += dst_stride;
    }
  }
}

113 114 115
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,
116 117
                               const int subpel_x_q4, int x_step_q4,
                               ConvolveParams *conv_params) {
Angie Chiang's avatar
Angie Chiang committed
118 119
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
  if (filter_params.taps == SUBPEL_TAPS) {
120 121
    const int16_t *filter_x =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_x_q4);
122
    if (conv_params->ref == 0)
123 124 125 126 127 128
      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 {
Angie Chiang's avatar
Angie Chiang committed
129 130
    av1_convolve_horiz(src, src_stride, dst, dst_stride, w, h, filter_params,
                       subpel_x_q4, x_step_q4, conv_params);
131 132 133
  }
}

134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154
void av1_convolve_horiz_facade_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,
                                 ConvolveParams *conv_params) {
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
  if (filter_params.taps == SUBPEL_TAPS) {
    const int16_t *filter_x =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_x_q4);
    if (conv_params->ref == 0)
      aom_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter_x,
                            x_step_q4, NULL, -1, w, h);
    else
      aom_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter_x,
                                x_step_q4, NULL, -1, w, h);
  } else {
    av1_convolve_horiz_c(src, src_stride, dst, dst_stride, w, h, filter_params,
                         subpel_x_q4, x_step_q4, conv_params);
  }
}

155 156 157
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,
158 159
                              const int subpel_y_q4, int y_step_q4,
                              ConvolveParams *conv_params) {
Angie Chiang's avatar
Angie Chiang committed
160 161
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
  if (filter_params.taps == SUBPEL_TAPS) {
162 163
    const int16_t *filter_y =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_y_q4);
164
    if (conv_params->ref == 0) {
165 166 167 168 169 170 171
      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 {
Angie Chiang's avatar
Angie Chiang committed
172 173
    av1_convolve_vert(src, src_stride, dst, dst_stride, w, h, filter_params,
                      subpel_y_q4, y_step_q4, conv_params);
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
void av1_convolve_vert_facade_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,
                                ConvolveParams *conv_params) {
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
  if (filter_params.taps == SUBPEL_TAPS) {
    const int16_t *filter_y =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_y_q4);
    if (conv_params->ref == 0) {
      aom_convolve8_vert_c(src, src_stride, dst, dst_stride, NULL, -1, filter_y,
                           y_step_q4, w, h);
    } else {
      aom_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, NULL, -1,
                               filter_y, y_step_q4, w, h);
    }
  } else {
    av1_convolve_vert_c(src, src_stride, dst, dst_stride, w, h, filter_params,
                        subpel_y_q4, y_step_q4, conv_params);
  }
}

Angie Chiang's avatar
Angie Chiang committed
199
#if CONFIG_CONVOLVE_ROUND
Angie Chiang's avatar
Angie Chiang committed
200
void av1_convolve_rounding(const int32_t *src, int src_stride, uint8_t *dst,
Angie Chiang's avatar
Angie Chiang committed
201
                           int dst_stride, int w, int h, int bits) {
Angie Chiang's avatar
Angie Chiang committed
202 203 204 205
  int r, c;
  for (r = 0; r < h; ++r) {
    for (c = 0; c < w; ++c) {
      dst[r * dst_stride + c] =
Angie Chiang's avatar
Angie Chiang committed
206
          clip_pixel(ROUND_POWER_OF_TWO_SIGNED(src[r * src_stride + c], bits));
Angie Chiang's avatar
Angie Chiang committed
207 208 209
    }
  }
}
Angie Chiang's avatar
Angie Chiang committed
210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232

void av1_convolve_2d(const uint8_t *src, int src_stride, CONV_BUF_TYPE *dst,
                     int dst_stride, int w, int h,
                     InterpFilterParams *filter_params_x,
                     InterpFilterParams *filter_params_y, const int subpel_x_q4,
                     const int subpel_y_q4, ConvolveParams *conv_params) {
  int x, y, k;
  CONV_BUF_TYPE im_block[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) * MAX_SB_SIZE];
  int im_h = h + filter_params_y->taps - 1;
  int im_stride = w;
  const int fo_vert = filter_params_y->taps / 2 - 1;
  const int fo_horiz = filter_params_x->taps / 2 - 1;
  (void)conv_params;
  // horizontal filter
  const uint8_t *src_horiz = src - fo_vert * src_stride;
  const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
      *filter_params_x, subpel_x_q4 & SUBPEL_MASK);
  for (y = 0; y < im_h; ++y) {
    for (x = 0; x < w; ++x) {
      CONV_BUF_TYPE sum = 0;
      for (k = 0; k < filter_params_x->taps; ++k) {
        sum += x_filter[k] * src_horiz[y * src_stride + x - fo_horiz + k];
      }
Angie Chiang's avatar
Angie Chiang committed
233 234 235 236
#if CONFIG_COMPOUND_ROUND
      im_block[y * im_stride + x] =
          clip_pixel(ROUND_POWER_OF_TWO_SIGNED(sum, conv_params->round_0));
#else
Angie Chiang's avatar
Angie Chiang committed
237 238
      im_block[y * im_stride + x] =
          ROUND_POWER_OF_TWO_SIGNED(sum, conv_params->round_0);
Angie Chiang's avatar
Angie Chiang committed
239
#endif
Angie Chiang's avatar
Angie Chiang committed
240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258
    }
  }

  // vertical filter
  CONV_BUF_TYPE *src_vert = im_block + fo_vert * im_stride;
  const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
      *filter_params_y, subpel_y_q4 & SUBPEL_MASK);
  for (y = 0; y < h; ++y) {
    for (x = 0; x < w; ++x) {
      CONV_BUF_TYPE sum = 0;
      for (k = 0; k < filter_params_y->taps; ++k) {
        sum += y_filter[k] * src_vert[(y - fo_vert + k) * im_stride + x];
      }
      dst[y * dst_stride + x] +=
          ROUND_POWER_OF_TWO_SIGNED(sum, conv_params->round_1);
    }
  }
}

259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276
static INLINE void transpose_uint8(uint8_t *dst, int dst_stride,
                                   const uint8_t *src, int src_stride, int w,
                                   int h) {
  int r, c;
  for (r = 0; r < h; ++r)
    for (c = 0; c < w; ++c)
      dst[c * (dst_stride) + r] = src[r * (src_stride) + c];
}

static INLINE void transpose_int32(int32_t *dst, int dst_stride,
                                   const int32_t *src, int src_stride, int w,
                                   int h) {
  int r, c;
  for (r = 0; r < h; ++r)
    for (c = 0; c < w; ++c)
      dst[c * (dst_stride) + r] = src[r * (src_stride) + c];
}

Angie Chiang's avatar
Angie Chiang committed
277 278 279 280 281 282 283 284 285 286
void av1_convolve_2d_facade(const uint8_t *src, int src_stride, uint8_t *dst,
                            int dst_stride, int w, int h,
                            const InterpFilter *interp_filter,
                            const int subpel_x_q4, int x_step_q4,
                            const int subpel_y_q4, int y_step_q4,
                            ConvolveParams *conv_params) {
  (void)x_step_q4;
  (void)y_step_q4;
  (void)dst;
  (void)dst_stride;
287
#if CONFIG_DUAL_FILTER
Angie Chiang's avatar
Angie Chiang committed
288 289 290 291
  InterpFilterParams filter_params_x =
      av1_get_interp_filter_params(interp_filter[1 + 2 * conv_params->ref]);
  InterpFilterParams filter_params_y =
      av1_get_interp_filter_params(interp_filter[0 + 2 * conv_params->ref]);
292

293 294 295 296 297 298
  if (filter_params_x.interp_filter == MULTITAP_SHARP &&
      filter_params_y.interp_filter == 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);
  }
299 300 301 302 303
#else
  InterpFilterParams filter_params_x =
      av1_get_interp_filter_params(*interp_filter);
  InterpFilterParams filter_params_y =
      av1_get_interp_filter_params(*interp_filter);
304
#endif
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

  if (filter_params_y.taps < filter_params_x.taps) {
    uint8_t tr_src[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) *
                   (MAX_SB_SIZE + MAX_FILTER_TAP - 1)];
    int tr_src_stride = MAX_SB_SIZE + MAX_FILTER_TAP - 1;
    CONV_BUF_TYPE tr_dst[MAX_SB_SIZE * MAX_SB_SIZE];
    int tr_dst_stride = MAX_SB_SIZE;
    int fo_vert = filter_params_y.taps / 2 - 1;
    int fo_horiz = filter_params_x.taps / 2 - 1;

    transpose_uint8(tr_src, tr_src_stride,
                    src - fo_vert * src_stride - fo_horiz, src_stride,
                    w + filter_params_x.taps - 1, h + filter_params_y.taps - 1);
    transpose_int32(tr_dst, tr_dst_stride, conv_params->dst,
                    conv_params->dst_stride, w, h);

    // horizontal and vertical parameters are swapped because of the transpose
    av1_convolve_2d(tr_src + fo_horiz * tr_src_stride + fo_vert, tr_src_stride,
                    tr_dst, tr_dst_stride, h, w, &filter_params_y,
                    &filter_params_x, subpel_y_q4, subpel_x_q4, conv_params);
    transpose_int32(conv_params->dst, conv_params->dst_stride, tr_dst,
                    tr_dst_stride, h, w);
  } else {
    av1_convolve_2d(src, src_stride, conv_params->dst, conv_params->dst_stride,
                    w, h, &filter_params_x, &filter_params_y, subpel_x_q4,
                    subpel_y_q4, conv_params);
  }
Angie Chiang's avatar
Angie Chiang committed
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
#if CONFIG_HIGHBITDEPTH
static INLINE void transpose_uint16(uint16_t *dst, int dst_stride,
                                    const uint16_t *src, int src_stride, int w,
                                    int h) {
  int r, c;
  for (r = 0; r < h; ++r)
    for (c = 0; c < w; ++c) dst[c * dst_stride + r] = src[r * src_stride + c];
}

void av1_highbd_convolve_rounding(const int32_t *src, int src_stride,
                                  uint8_t *dst8, int dst_stride, int w, int h,
                                  int bits, int bd) {
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  int r, c;
  for (r = 0; r < h; ++r) {
    for (c = 0; c < w; ++c) {
      dst[r * dst_stride + c] = clip_pixel_highbd(
          ROUND_POWER_OF_TWO_SIGNED(src[r * src_stride + c], bits), bd);
    }
  }
}

void av1_highbd_convolve_2d(const uint16_t *src, int src_stride,
                            CONV_BUF_TYPE *dst, int dst_stride, int w, int h,
                            InterpFilterParams *filter_params_x,
                            InterpFilterParams *filter_params_y,
                            const int subpel_x_q4, const int subpel_y_q4,
                            ConvolveParams *conv_params, int bd) {
  int x, y, k;
  CONV_BUF_TYPE im_block[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) * MAX_SB_SIZE];
  int im_h = h + filter_params_y->taps - 1;
  int im_stride = w;
  const int fo_vert = filter_params_y->taps / 2 - 1;
  const int fo_horiz = filter_params_x->taps / 2 - 1;
  (void)conv_params;
  // horizontal filter
  const uint16_t *src_horiz = src - fo_vert * src_stride;
  const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
      *filter_params_x, subpel_x_q4 & SUBPEL_MASK);
  for (y = 0; y < im_h; ++y) {
    for (x = 0; x < w; ++x) {
      CONV_BUF_TYPE sum = 0;
      for (k = 0; k < filter_params_x->taps; ++k) {
        sum += x_filter[k] * src_horiz[y * src_stride + x - fo_horiz + k];
      }
#if CONFIG_COMPOUND_ROUND
      im_block[y * im_stride + x] = clip_pixel_highbd(
          ROUND_POWER_OF_TWO_SIGNED(sum, conv_params->round_0), bd);
#else
      (void)bd;
      im_block[y * im_stride + x] =
          ROUND_POWER_OF_TWO_SIGNED(sum, conv_params->round_0);
#endif
    }
  }

  // vertical filter
  CONV_BUF_TYPE *src_vert = im_block + fo_vert * im_stride;
  const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
      *filter_params_y, subpel_y_q4 & SUBPEL_MASK);
  for (y = 0; y < h; ++y) {
    for (x = 0; x < w; ++x) {
      CONV_BUF_TYPE sum = 0;
      for (k = 0; k < filter_params_y->taps; ++k) {
        sum += y_filter[k] * src_vert[(y - fo_vert + k) * im_stride + x];
      }
      dst[y * dst_stride + x] +=
          ROUND_POWER_OF_TWO_SIGNED(sum, conv_params->round_1);
    }
  }
}

void av1_highbd_convolve_2d_facade(const uint8_t *src8, int src_stride,
                                   uint8_t *dst, int dst_stride, int w, int h,
                                   const InterpFilter *interp_filter,
                                   const int subpel_x_q4, int x_step_q4,
                                   const int subpel_y_q4, int y_step_q4,
                                   ConvolveParams *conv_params, int bd) {
  (void)x_step_q4;
  (void)y_step_q4;
  (void)dst;
  (void)dst_stride;
#if CONFIG_DUAL_FILTER
  InterpFilterParams filter_params_x =
      av1_get_interp_filter_params(interp_filter[1 + 2 * conv_params->ref]);
  InterpFilterParams filter_params_y =
      av1_get_interp_filter_params(interp_filter[0 + 2 * conv_params->ref]);

  if (filter_params_x.interp_filter == MULTITAP_SHARP &&
      filter_params_y.interp_filter == 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);
  }
#else
  InterpFilterParams filter_params_x =
      av1_get_interp_filter_params(*interp_filter);
  InterpFilterParams filter_params_y =
      av1_get_interp_filter_params(*interp_filter);
#endif
  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  if (filter_params_y.taps < filter_params_x.taps) {
    uint16_t tr_src[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) *
                    (MAX_SB_SIZE + MAX_FILTER_TAP - 1)];
    int tr_src_stride = MAX_SB_SIZE + MAX_FILTER_TAP - 1;
    CONV_BUF_TYPE tr_dst[MAX_SB_SIZE * MAX_SB_SIZE];
    int tr_dst_stride = MAX_SB_SIZE;
    int fo_vert = filter_params_y.taps / 2 - 1;
    int fo_horiz = filter_params_x.taps / 2 - 1;

    transpose_uint16(
        tr_src, tr_src_stride, src - fo_vert * src_stride - fo_horiz,
        src_stride, w + filter_params_x.taps - 1, h + filter_params_y.taps - 1);
    transpose_int32(tr_dst, tr_dst_stride, conv_params->dst,
                    conv_params->dst_stride, w, h);

    // horizontal and vertical parameters are swapped because of the transpose
    av1_highbd_convolve_2d(tr_src + fo_horiz * tr_src_stride + fo_vert,
                           tr_src_stride, tr_dst, tr_dst_stride, h, w,
                           &filter_params_y, &filter_params_x, subpel_y_q4,
                           subpel_x_q4, conv_params, bd);
    transpose_int32(conv_params->dst, conv_params->dst_stride, tr_dst,
                    tr_dst_stride, h, w);
  } else {
    av1_highbd_convolve_2d(src, src_stride, conv_params->dst,
                           conv_params->dst_stride, w, h, &filter_params_x,
                           &filter_params_y, subpel_x_q4, subpel_y_q4,
                           conv_params, bd);
  }
}
#endif  // CONFIG_HIGHBITDEPTH

Angie Chiang's avatar
Angie Chiang committed
466
#endif  // CONFIG_CONVOLVE_ROUND
Angie Chiang's avatar
Angie Chiang committed
467

468 469 470 471 472 473 474 475
typedef void (*ConvolveFunc)(const uint8_t *src, int src_stride, uint8_t *dst,
                             int dst_stride, int w, int h,
                             const InterpFilterParams filter_params,
                             const int subpel_q4, int step_q4,
                             ConvolveParams *conv_params);

static void convolve_helper(const uint8_t *src, int src_stride, uint8_t *dst,
                            int dst_stride, int w, int h,
476
#if CONFIG_DUAL_FILTER
477
                            const InterpFilter *interp_filter,
478
#else
479
                            const InterpFilter interp_filter,
480
#endif
481 482 483 484 485
                            const int subpel_x_q4, int x_step_q4,
                            const int subpel_y_q4, int y_step_q4,
                            ConvolveParams *conv_params,
                            ConvolveFunc convolve_horiz,
                            ConvolveFunc convolve_vert) {
Angie Chiang's avatar
Angie Chiang committed
486 487
  int ignore_horiz = x_step_q4 == 16 && subpel_x_q4 == 0;
  int ignore_vert = y_step_q4 == 16 && subpel_y_q4 == 0;
Angie Chiang's avatar
Angie Chiang committed
488 489 490 491 492 493 494 495 496 497
#if CONFIG_DUAL_FILTER
  InterpFilterParams filter_params_x =
      av1_get_interp_filter_params(interp_filter[1 + 2 * conv_params->ref]);
  InterpFilterParams filter_params_y =
      av1_get_interp_filter_params(interp_filter[0 + 2 * conv_params->ref]);
  InterpFilterParams filter_params;
#else
  InterpFilterParams filter_params =
      av1_get_interp_filter_params(interp_filter);
#endif
Angie Chiang's avatar
Angie Chiang committed
498
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
499 500 501 502 503 504

  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
505
  if (ignore_horiz && ignore_vert) {
506
    convolve_copy(src, src_stride, dst, dst_stride, w, h, conv_params);
Angie Chiang's avatar
Angie Chiang committed
507
  } else if (ignore_vert) {
508
#if CONFIG_DUAL_FILTER
Angie Chiang's avatar
Angie Chiang committed
509
    filter_params = filter_params_x;
510 511
#endif
    assert(filter_params.taps <= MAX_FILTER_TAP);
512 513
    convolve_horiz(src, src_stride, dst, dst_stride, w, h, filter_params,
                   subpel_x_q4, x_step_q4, conv_params);
Angie Chiang's avatar
Angie Chiang committed
514
  } else if (ignore_horiz) {
515
#if CONFIG_DUAL_FILTER
Angie Chiang's avatar
Angie Chiang committed
516
    filter_params = filter_params_y;
517 518
#endif
    assert(filter_params.taps <= MAX_FILTER_TAP);
519 520
    convolve_vert(src, src_stride, dst, dst_stride, w, h, filter_params,
                  subpel_y_q4, y_step_q4, conv_params);
Angie Chiang's avatar
Angie Chiang committed
521
  } else {
522 523 524 525 526 527
    // 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);
528
    int filter_size;
529
#if CONFIG_DUAL_FILTER
530 531
    if (interp_filter[0 + 2 * conv_params->ref] == MULTITAP_SHARP &&
        interp_filter[1 + 2 * conv_params->ref] == MULTITAP_SHARP) {
532 533 534 535
      // Avoid two directions both using 12-tap filter.
      // This will reduce hardware implementation cost.
      filter_params_y = av1_get_interp_filter_params(EIGHTTAP_SHARP);
    }
536

537 538 539 540
    // we do filter with fewer taps first to reduce hardware implementation
    // complexity
    if (filter_params_y.taps < filter_params_x.taps) {
      int intermediate_width;
541
      int temp_stride = max_intermediate_size;
542 543
      ConvolveParams temp_conv_params;
      temp_conv_params.ref = 0;
544
      temp_conv_params.round = CONVOLVE_OPT_ROUND;
545 546 547 548
      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;
549
      assert(intermediate_width <= max_intermediate_size);
Angie Chiang's avatar
Angie Chiang committed
550

551
      assert(filter_params.taps <= MAX_FILTER_TAP);
552

553 554 555
      convolve_vert(src - (filter_size / 2 - 1), src_stride, temp, temp_stride,
                    intermediate_width, h, filter_params, subpel_y_q4,
                    y_step_q4, &temp_conv_params);
556

557 558
      filter_params = filter_params_x;
      assert(filter_params.taps <= MAX_FILTER_TAP);
559 560
      convolve_horiz(temp + (filter_size / 2 - 1), temp_stride, dst, dst_stride,
                     w, h, filter_params, subpel_x_q4, x_step_q4, conv_params);
561
    } else
562
#endif  // CONFIG_DUAL_FILTER
563 564
    {
      int intermediate_height;
565
      int temp_stride = MAX_SB_SIZE;
566 567
      ConvolveParams temp_conv_params;
      temp_conv_params.ref = 0;
568
      temp_conv_params.round = CONVOLVE_OPT_ROUND;
569 570 571 572 573 574 575 576
#if CONFIG_DUAL_FILTER
      filter_params = filter_params_x;
      filter_size = filter_params_y.taps;
#else
      filter_size = filter_params.taps;
#endif
      intermediate_height =
          (((h - 1) * y_step_q4 + subpel_y_q4) >> SUBPEL_BITS) + filter_size;
577 578
      assert(intermediate_height <= max_intermediate_size);
      (void)max_intermediate_size;
579 580 581

      assert(filter_params.taps <= MAX_FILTER_TAP);

582 583 584
      convolve_horiz(src - src_stride * (filter_size / 2 - 1), src_stride, temp,
                     temp_stride, w, intermediate_height, filter_params,
                     subpel_x_q4, x_step_q4, &temp_conv_params);
585 586 587 588 589 590

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

591 592 593
      convolve_vert(temp + temp_stride * (filter_size / 2 - 1), temp_stride,
                    dst, dst_stride, w, h, filter_params, subpel_y_q4,
                    y_step_q4, conv_params);
594
    }
Angie Chiang's avatar
Angie Chiang committed
595
  }
596 597
}

598 599 600 601 602 603 604 605 606 607 608 609 610 611
void av1_convolve(const uint8_t *src, int src_stride, uint8_t *dst,
                  int dst_stride, int w, int h,
#if CONFIG_DUAL_FILTER
                  const InterpFilter *interp_filter,
#else
                  const InterpFilter interp_filter,
#endif
                  const int subpel_x_q4, int x_step_q4, const int subpel_y_q4,
                  int y_step_q4, ConvolveParams *conv_params) {
  convolve_helper(src, src_stride, dst, dst_stride, w, h, interp_filter,
                  subpel_x_q4, x_step_q4, subpel_y_q4, y_step_q4, conv_params,
                  av1_convolve_horiz_facade, av1_convolve_vert_facade);
}

612 613 614 615 616 617 618 619 620
void av1_convolve_c(const uint8_t *src, int src_stride, uint8_t *dst,
                    int dst_stride, int w, int h,
#if CONFIG_DUAL_FILTER
                    const InterpFilter *interp_filter,
#else
                    const InterpFilter interp_filter,
#endif
                    const int subpel_x_q4, int x_step_q4, const int subpel_y_q4,
                    int y_step_q4, ConvolveParams *conv_params) {
621 622 623
  convolve_helper(src, src_stride, dst, dst_stride, w, h, interp_filter,
                  subpel_x_q4, x_step_q4, subpel_y_q4, y_step_q4, conv_params,
                  av1_convolve_horiz_facade_c, av1_convolve_vert_facade_c);
624 625
}

Angie Chiang's avatar
Angie Chiang committed
626
void av1_lowbd_convolve_init_c(void) {
Angie Chiang's avatar
Angie Chiang committed
627 628 629 630
  // A placeholder for SIMD initialization
  return;
}

Angie Chiang's avatar
Angie Chiang committed
631 632 633 634 635 636
void av1_highbd_convolve_init_c(void) {
  // A placeholder for SIMD initialization
  return;
}

void av1_convolve_init(AV1_COMMON *cm) {
637
#if CONFIG_HIGHBITDEPTH
Angie Chiang's avatar
Angie Chiang committed
638 639 640 641 642 643 644 645 646 647 648
  if (cm->use_highbitdepth)
    av1_highbd_convolve_init();
  else
    av1_lowbd_convolve_init();
#else
  (void)cm;
  av1_lowbd_convolve_init();
#endif
  return;
}

649
#if CONFIG_HIGHBITDEPTH
Yaowu Xu's avatar
Yaowu Xu committed
650 651 652 653 654
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) {
655
  int x, y;
656
  int filter_size = filter_params.taps;
657 658 659 660 661
  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
662
      const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
663
          filter_params, x_q4 & SUBPEL_MASK);
664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679
      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
680 681 682 683 684
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) {
685
  int x, y;
686
  int filter_size = filter_params.taps;
687 688 689 690 691 692
  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
693
      const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
694
          filter_params, y_q4 & SUBPEL_MASK);
695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713
      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
714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735
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;
    }
  }
}

736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766
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);
767

768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783
  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
784 785
void av1_highbd_convolve(const uint8_t *src8, int src_stride, uint8_t *dst8,
                         int dst_stride, int w, int h,
786
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
787
                         const InterpFilter *interp_filter,
788
#else
James Zern's avatar
James Zern committed
789
                         const InterpFilter interp_filter,
790
#endif
Yaowu Xu's avatar
Yaowu Xu committed
791 792 793
                         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
794 795 796 797
  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;
798 799 800 801 802 803

  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
804
  if (ignore_horiz && ignore_vert) {
805
    highbd_convolve_copy(src, src_stride, dst, dst_stride, w, h, ref_idx, bd);
Angie Chiang's avatar
Angie Chiang committed
806
  } else if (ignore_vert) {
807 808
#if CONFIG_DUAL_FILTER
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
809
        av1_get_interp_filter_params(interp_filter[1 + 2 * ref_idx]);
810 811
#else
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
812
        av1_get_interp_filter_params(interp_filter);
813
#endif
814 815 816
    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
817
  } else if (ignore_horiz) {
818 819
#if CONFIG_DUAL_FILTER
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
820
        av1_get_interp_filter_params(interp_filter[0 + 2 * ref_idx]);
821 822
#else
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
823
        av1_get_interp_filter_params(interp_filter);
824
#endif
825 826 827
    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
828
  } else {
829 830 831 832 833
    // 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
834
    uint8_t *temp8 = CONVERT_TO_BYTEPTR(temp);
835 836 837
    int max_intermediate_size = ((MAX_SB_SIZE * 2 + 16) + 16);
    int filter_size;
    InterpFilterParams filter_params;
838
#if CONFIG_DUAL_FILTER
Yaowu Xu's avatar
Yaowu Xu committed
839 840 841 842
    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]);
843 844 845 846 847 848
    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);
    }
849
#endif
850

851 852 853 854 855 856 857 858 859 860 861
#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
862

863 864 865
      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);
866

867 868 869 870 871 872 873 874 875 876 877
      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;
878
#if CONFIG_DUAL_FILTER
879 880 881 882 883
      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;
884
#endif
885 886 887 888 889 890 891 892 893 894 895 896 897 898 899
      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);
900

901 902 903 904
      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
905
  }
906
}
907
#endif  // CONFIG_HIGHBITDEPTH