convolve.c 69 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/blockd.h"
18
#include "av1/common/convolve.h"
19
#include "av1/common/filter.h"
Angie Chiang's avatar
Angie Chiang committed
20
#include "av1/common/onyxc_int.h"
Yaowu Xu's avatar
Yaowu Xu committed
21
#include "aom_dsp/aom_dsp_common.h"
22
#include "aom_ports/mem.h"
23

24 25
#define MAX_BLOCK_WIDTH (MAX_SB_SIZE)
#define MAX_BLOCK_HEIGHT (MAX_SB_SIZE)
26 27
#define MAX_STEP (32)

Yaowu Xu's avatar
Yaowu Xu committed
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,
31 32
                          const int subpel_x_q4, int x_step_q4,
                          ConvolveParams *conv_params) {
33
  int x, y;
34
  int filter_size = filter_params.taps;
Angie Chiang's avatar
Angie Chiang committed
35
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
36 37 38 39 40
  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
41
      const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
42
          filter_params, x_q4 & SUBPEL_MASK);
43 44
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k) sum += src_x[k] * x_filter[k];
45

Angie Chiang's avatar
Angie Chiang committed
46
      sum = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
47
      if (conv_params->do_average)
Angie Chiang's avatar
Angie Chiang committed
48 49 50
        dst[x] = ROUND_POWER_OF_TWO(dst[x] + sum, 1);
      else
        dst[x] = sum;
51

52 53 54 55 56 57 58
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

Fergus Simpson's avatar
Fergus Simpson committed
59 60 61 62 63 64 65
void av1_convolve_horiz_scale(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_qn, int x_step_qn,
                              ConvolveParams *conv_params) {
  int x, y;
  int filter_size = filter_params.taps;
66
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
Fergus Simpson's avatar
Fergus Simpson committed
67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91
  src -= filter_size / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_qn = subpel_x_qn;
    for (x = 0; x < w; ++x) {
      const uint8_t *const src_x = &src[x_qn >> SCALE_SUBPEL_BITS];
      const int x_filter_idx = (x_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
      assert(x_filter_idx < SUBPEL_SHIFTS);
      const int16_t *x_filter =
          av1_get_interp_filter_subpel_kernel(filter_params, x_filter_idx);
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k) sum += src_x[k] * x_filter[k];

      sum = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
      if (conv_params->do_average)
        dst[x] = ROUND_POWER_OF_TWO(dst[x] + sum, 1);
      else
        dst[x] = sum;

      x_qn += x_step_qn;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
92 93 94
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,
95 96
                         const int subpel_y_q4, int y_step_q4,
                         ConvolveParams *conv_params) {
97
  int x, y;
98
  int filter_size = filter_params.taps;
Angie Chiang's avatar
Angie Chiang committed
99
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
100 101 102 103 104
  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
105
      const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
106
          filter_params, y_q4 & SUBPEL_MASK);
107 108 109
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
110

Angie Chiang's avatar
Angie Chiang committed
111
      sum = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
112
      if (conv_params->do_average)
Angie Chiang's avatar
Angie Chiang committed
113 114 115
        dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] + sum, 1);
      else
        dst[y * dst_stride] = sum;
116

117 118 119 120 121 122 123
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

Fergus Simpson's avatar
Fergus Simpson committed
124 125 126 127 128 129 130
void av1_convolve_vert_scale(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_qn, int y_step_qn,
                             ConvolveParams *conv_params) {
  int x, y;
  int filter_size = filter_params.taps;
131
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
Fergus Simpson's avatar
Fergus Simpson committed
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
  src -= src_stride * (filter_size / 2 - 1);
  for (x = 0; x < w; ++x) {
    int y_qn = subpel_y_qn;
    for (y = 0; y < h; ++y) {
      const uint8_t *const src_y =
          &src[(y_qn >> SCALE_SUBPEL_BITS) * src_stride];
      const int y_filter_idx = (y_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
      assert(y_filter_idx < SUBPEL_SHIFTS);
      const int16_t *y_filter =
          av1_get_interp_filter_subpel_kernel(filter_params, y_filter_idx);
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k)
        sum += src_y[k * src_stride] * y_filter[k];

      sum = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
      if (conv_params->do_average)
        dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] + sum, 1);
      else
        dst[y * dst_stride] = sum;

      y_qn += y_step_qn;
    }
    ++src;
    ++dst;
  }
}

Angie Chiang's avatar
Angie Chiang committed
159
static void convolve_copy(const uint8_t *src, int src_stride, uint8_t *dst,
160 161
                          int dst_stride, int w, int h,
                          ConvolveParams *conv_params) {
Angie Chiang's avatar
Angie Chiang committed
162
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
163
  if (conv_params->do_average == 0) {
Angie Chiang's avatar
Angie Chiang committed
164
    int r;
Angie Chiang's avatar
Angie Chiang committed
165
    for (r = 0; r < h; ++r) {
Angie Chiang's avatar
Angie Chiang committed
166
      memcpy(dst, src, w);
Angie Chiang's avatar
Angie Chiang committed
167 168 169 170 171 172
      src += src_stride;
      dst += dst_stride;
    }
  } else {
    int r, c;
    for (r = 0; r < h; ++r) {
173
      for (c = 0; c < w; ++c) {
Angie Chiang's avatar
Angie Chiang committed
174
        dst[c] = clip_pixel(ROUND_POWER_OF_TWO(dst[c] + src[c], 1));
175
      }
Angie Chiang's avatar
Angie Chiang committed
176 177 178 179 180 181
      src += src_stride;
      dst += dst_stride;
    }
  }
}

182 183 184
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,
185 186
                               const int subpel_x_q4, int x_step_q4,
                               ConvolveParams *conv_params) {
Angie Chiang's avatar
Angie Chiang committed
187 188
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
  if (filter_params.taps == SUBPEL_TAPS) {
189 190
    const int16_t *filter_x =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_x_q4);
191
    if (conv_params->do_average == 0)
192 193 194 195 196 197
      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
198 199
    av1_convolve_horiz(src, src_stride, dst, dst_stride, w, h, filter_params,
                       subpel_x_q4, x_step_q4, conv_params);
200 201 202
  }
}

203 204 205 206 207 208 209 210 211
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);
212
    if (conv_params->do_average == 0)
213 214 215 216 217 218 219 220 221 222 223
      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);
  }
}

Fergus Simpson's avatar
Fergus Simpson committed
224 225 226 227 228
void av1_convolve_horiz_facade_scale(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_qn, int x_step_qn,
                                     ConvolveParams *conv_params) {
229
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
Fergus Simpson's avatar
Fergus Simpson committed
230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245
  if (filter_params.taps == SUBPEL_TAPS) {
    const int16_t *filter_x = av1_get_interp_filter_subpel_kernel(
        filter_params, subpel_x_qn >> SCALE_EXTRA_BITS);
    if (conv_params->do_average == 0)
      aom_convolve8_horiz_scale(src, src_stride, dst, dst_stride, filter_x,
                                subpel_x_qn, x_step_qn, NULL, 0, -1, w, h);
    else
      aom_convolve8_avg_horiz_scale(src, src_stride, dst, dst_stride, filter_x,
                                    subpel_x_qn, x_step_qn, NULL, 0, -1, w, h);
  } else {
    av1_convolve_horiz_scale(src, src_stride, dst, dst_stride, w, h,
                             filter_params, subpel_x_qn, x_step_qn,
                             conv_params);
  }
}

246 247 248
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,
249 250
                              const int subpel_y_q4, int y_step_q4,
                              ConvolveParams *conv_params) {
Angie Chiang's avatar
Angie Chiang committed
251 252
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
  if (filter_params.taps == SUBPEL_TAPS) {
253 254
    const int16_t *filter_y =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_y_q4);
255
    if (conv_params->do_average == 0) {
256 257 258 259 260 261 262
      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
263 264
    av1_convolve_vert(src, src_stride, dst, dst_stride, w, h, filter_params,
                      subpel_y_q4, y_step_q4, conv_params);
265 266 267
  }
}

268 269 270 271 272 273 274 275 276
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);
277
    if (conv_params->do_average == 0) {
278 279 280 281 282 283 284 285 286 287 288 289
      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);
  }
}

Fergus Simpson's avatar
Fergus Simpson committed
290 291 292 293 294
void av1_convolve_vert_facade_scale(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_qn, int y_step_qn,
                                    ConvolveParams *conv_params) {
295
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
Fergus Simpson's avatar
Fergus Simpson committed
296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311
  if (filter_params.taps == SUBPEL_TAPS) {
    const int16_t *filter_y = av1_get_interp_filter_subpel_kernel(
        filter_params, subpel_y_qn >> SCALE_EXTRA_BITS);
    if (conv_params->do_average == 0) {
      aom_convolve8_vert_scale(src, src_stride, dst, dst_stride, NULL, 0, -1,
                               filter_y, subpel_y_qn, y_step_qn, w, h);
    } else {
      aom_convolve8_avg_vert_scale(src, src_stride, dst, dst_stride, NULL, 0,
                                   -1, filter_y, subpel_y_qn, y_step_qn, w, h);
    }
  } else {
    av1_convolve_vert_scale(src, src_stride, dst, dst_stride, w, h,
                            filter_params, subpel_y_qn, y_step_qn, conv_params);
  }
}

Angie Chiang's avatar
Angie Chiang committed
312
#if CONFIG_CONVOLVE_ROUND
313 314
void av1_convolve_rounding_c(const int32_t *src, int src_stride, uint8_t *dst,
                             int dst_stride, int w, int h, int bits) {
Angie Chiang's avatar
Angie Chiang committed
315 316 317 318
  int r, c;
  for (r = 0; r < h; ++r) {
    for (c = 0; c < w; ++c) {
      dst[r * dst_stride + c] =
319
          clip_pixel(ROUND_POWER_OF_TWO(src[r * src_stride + c], bits));
Angie Chiang's avatar
Angie Chiang committed
320 321 322
    }
  }
}
Angie Chiang's avatar
Angie Chiang committed
323

324
#if CONFIG_COMPOUND_ROUND
David Barker's avatar
David Barker committed
325 326 327 328 329 330
void av1_convolve_2d_c(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) {
Angie Chiang's avatar
Angie Chiang committed
331
  int x, y, k;
332
  uint8_t im_block[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) * MAX_SB_SIZE];
Angie Chiang's avatar
Angie Chiang committed
333 334 335 336
  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;
337

Angie Chiang's avatar
Angie Chiang committed
338 339 340 341 342 343
  // 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) {
344
      int32_t sum = 0;
Angie Chiang's avatar
Angie Chiang committed
345 346 347
      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
348
      im_block[y * im_stride + x] =
349 350 351 352 353 354 355 356 357 358 359 360 361 362 363
          clip_pixel(ROUND_POWER_OF_TWO(sum, conv_params->round_0));
    }
  }

  // vertical filter
  uint8_t *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];
      }
      CONV_BUF_TYPE res = ROUND_POWER_OF_TWO(sum, conv_params->round_1);
364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379
#if CONFIG_JNT_COMP
      if (conv_params->bck_offset == -1) {
        if (conv_params->do_average)
          dst[y * dst_stride + x] += res;
        else
          dst[y * dst_stride + x] = res;
      } else {
        if (conv_params->do_average == 0) {
          dst[y * dst_stride + x] = res * conv_params->fwd_offset;
        } else {
          dst[y * dst_stride + x] += res * conv_params->bck_offset;

          dst[y * dst_stride + x] >>= (DIST_PRECISION_BITS - 1);
        }
      }
#else
380 381 382 383
      if (conv_params->do_average)
        dst[y * dst_stride + x] += res;
      else
        dst[y * dst_stride + x] = res;
384
#endif  // CONFIG_JNT_COMP
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

void av1_convolve_2d_scale_c(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_qn, const int x_step_qn,
                             const int subpel_y_qn, const int y_step_qn,
                             ConvolveParams *conv_params) {
  int x, y, k;
  uint8_t im_block[(2 * MAX_SB_SIZE + MAX_FILTER_TAP) * MAX_SB_SIZE];
  int im_h = (((h - 1) * y_step_qn + subpel_y_qn) >> SCALE_SUBPEL_BITS) +
             filter_params_y->taps;
  int im_stride = w;
  const int fo_vert = filter_params_y->taps / 2 - 1;
  const int fo_horiz = filter_params_x->taps / 2 - 1;

  // horizontal filter
  const uint8_t *src_horiz = src - fo_vert * src_stride;
  for (y = 0; y < im_h; ++y) {
    int x_qn = subpel_x_qn;
    for (x = 0; x < w; ++x, x_qn += x_step_qn) {
      const uint8_t *const src_x = &src_horiz[(x_qn >> SCALE_SUBPEL_BITS)];
      const int x_filter_idx = (x_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
      assert(x_filter_idx < SUBPEL_SHIFTS);
      const int16_t *x_filter =
          av1_get_interp_filter_subpel_kernel(*filter_params_x, x_filter_idx);
      int sum = 0;
      for (k = 0; k < filter_params_x->taps; ++k)
        sum += x_filter[k] * src_x[k - fo_horiz];
      im_block[y * im_stride + x] =
          clip_pixel(ROUND_POWER_OF_TWO(sum, conv_params->round_0));
    }
    src_horiz += src_stride;
  }

  // vertical filter
  const uint8_t *src_vert = im_block + fo_vert * im_stride;
  for (x = 0; x < w; ++x) {
    int y_qn = subpel_y_qn;
    for (y = 0; y < h; ++y, y_qn += y_step_qn) {
      const uint8_t *const src_y =
          &src_vert[(y_qn >> SCALE_SUBPEL_BITS) * im_stride];
      const int y_filter_idx = (y_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
      assert(y_filter_idx < SUBPEL_SHIFTS);
      const int16_t *y_filter =
          av1_get_interp_filter_subpel_kernel(*filter_params_y, y_filter_idx);
      CONV_BUF_TYPE sum = 0;
      for (k = 0; k < filter_params_y->taps; ++k) {
        sum += y_filter[k] * src_y[(k - fo_vert) * im_stride];
      }
      CONV_BUF_TYPE res = ROUND_POWER_OF_TWO(sum, conv_params->round_1);
439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454
#if CONFIG_JNT_COMP
      if (conv_params->bck_offset == -1) {
        if (conv_params->do_average)
          dst[y * dst_stride + x] += res;
        else
          dst[y * dst_stride + x] = res;
      } else {
        if (conv_params->do_average == 0) {
          dst[y * dst_stride + x] = res * conv_params->fwd_offset;
        } else {
          dst[y * dst_stride + x] += res * conv_params->bck_offset;

          dst[y * dst_stride + x] >>= (DIST_PRECISION_BITS - 1);
        }
      }
#else
455 456 457 458
      if (conv_params->do_average)
        dst[y * dst_stride + x] += res;
      else
        dst[y * dst_stride + x] = res;
459
#endif  // CONFIG_JNT_COMP
460 461 462 463 464
    }
    src_vert++;
  }
}

Angie Chiang's avatar
Angie Chiang committed
465
#else
466

467 468 469 470 471 472
/* When convolve-round is enabled and compound-round is disabled, we use a
   high-precision convolve filter.
   Note: For notes on hardware implementations, including the required
   bit widths for various intermediate values, see the comments above
   av1_warp_affine_c.
*/
David Barker's avatar
David Barker committed
473 474 475 476 477 478
void av1_convolve_2d_c(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) {
479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497
  int x, y, k;
  int32_t 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;
  const int bd = 8;

  // 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) {
      int32_t sum = (1 << (bd + FILTER_BITS - 1));
      for (k = 0; k < filter_params_x->taps; ++k) {
        sum += x_filter[k] * src_horiz[y * src_stride + x - fo_horiz + k];
      }
      assert(0 <= sum && sum < (1 << (bd + FILTER_BITS + 1)));
Angie Chiang's avatar
Angie Chiang committed
498
      im_block[y * im_stride + x] =
499
          ROUND_POWER_OF_TWO(sum, conv_params->round_0);
Angie Chiang's avatar
Angie Chiang committed
500 501 502 503
    }
  }

  // vertical filter
504
  int32_t *src_vert = im_block + fo_vert * im_stride;
Angie Chiang's avatar
Angie Chiang committed
505 506
  const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
      *filter_params_y, subpel_y_q4 & SUBPEL_MASK);
507
  const int offset_bits = bd + 2 * FILTER_BITS - conv_params->round_0;
Angie Chiang's avatar
Angie Chiang committed
508 509
  for (y = 0; y < h; ++y) {
    for (x = 0; x < w; ++x) {
510
      CONV_BUF_TYPE sum = 1 << offset_bits;
Angie Chiang's avatar
Angie Chiang committed
511 512 513
      for (k = 0; k < filter_params_y->taps; ++k) {
        sum += y_filter[k] * src_vert[(y - fo_vert + k) * im_stride + x];
      }
514 515 516 517
      assert(0 <= sum && sum < (1 << (offset_bits + 2)));
      CONV_BUF_TYPE res = ROUND_POWER_OF_TWO(sum, conv_params->round_1) -
                          ((1 << (offset_bits - conv_params->round_1)) +
                           (1 << (offset_bits - conv_params->round_1 - 1)));
518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533
#if CONFIG_JNT_COMP
      if (conv_params->fwd_offset == -1) {
        if (conv_params->do_average)
          dst[y * dst_stride + x] += res;
        else
          dst[y * dst_stride + x] = res;
      } else {
        if (conv_params->do_average) {
          dst[y * dst_stride + x] += res * conv_params->bck_offset;

          dst[y * dst_stride + x] >>= (DIST_PRECISION_BITS - 1);
        } else {
          dst[y * dst_stride + x] = res * conv_params->fwd_offset;
        }
      }
#else
534 535 536 537
      if (conv_params->do_average)
        dst[y * dst_stride + x] += res;
      else
        dst[y * dst_stride + x] = res;
538
#endif  // CONFIG_JNT_COMP
Angie Chiang's avatar
Angie Chiang committed
539 540 541
    }
  }
}
542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598

void av1_convolve_2d_scale_c(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_qn, const int x_step_qn,
                             const int subpel_y_qn, const int y_step_qn,
                             ConvolveParams *conv_params) {
  int x, y, k;
  int32_t im_block[(2 * MAX_SB_SIZE + MAX_FILTER_TAP) * MAX_SB_SIZE];
  int im_h = (((h - 1) * y_step_qn + subpel_y_qn) >> SCALE_SUBPEL_BITS) +
             filter_params_y->taps;
  int im_stride = w;
  const int fo_vert = filter_params_y->taps / 2 - 1;
  const int fo_horiz = filter_params_x->taps / 2 - 1;
  const int bd = 8;

  // horizontal filter
  const uint8_t *src_horiz = src - fo_vert * src_stride;
  for (y = 0; y < im_h; ++y) {
    int x_qn = subpel_x_qn;
    for (x = 0; x < w; ++x, x_qn += x_step_qn) {
      const uint8_t *const src_x = &src_horiz[(x_qn >> SCALE_SUBPEL_BITS)];
      const int x_filter_idx = (x_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
      assert(x_filter_idx < SUBPEL_SHIFTS);
      const int16_t *x_filter =
          av1_get_interp_filter_subpel_kernel(*filter_params_x, x_filter_idx);
      int32_t sum = (1 << (bd + FILTER_BITS - 1));
      for (k = 0; k < filter_params_x->taps; ++k) {
        sum += x_filter[k] * src_x[k - fo_horiz];
      }
      assert(0 <= sum && sum < (1 << (bd + FILTER_BITS + 1)));
      im_block[y * im_stride + x] =
          ROUND_POWER_OF_TWO(sum, conv_params->round_0);
    }
    src_horiz += src_stride;
  }

  // vertical filter
  int32_t *src_vert = im_block + fo_vert * im_stride;
  const int offset_bits = bd + 2 * FILTER_BITS - conv_params->round_0;
  for (x = 0; x < w; ++x) {
    int y_qn = subpel_y_qn;
    for (y = 0; y < h; ++y, y_qn += y_step_qn) {
      const int32_t *src_y = &src_vert[(y_qn >> SCALE_SUBPEL_BITS) * im_stride];
      const int y_filter_idx = (y_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
      assert(y_filter_idx < SUBPEL_SHIFTS);
      const int16_t *y_filter =
          av1_get_interp_filter_subpel_kernel(*filter_params_y, y_filter_idx);
      CONV_BUF_TYPE sum = 1 << offset_bits;
      for (k = 0; k < filter_params_y->taps; ++k) {
        sum += y_filter[k] * src_y[(k - fo_vert) * im_stride];
      }
      assert(0 <= sum && sum < (1 << (offset_bits + 2)));
      CONV_BUF_TYPE res = ROUND_POWER_OF_TWO(sum, conv_params->round_1) -
                          ((1 << (offset_bits - conv_params->round_1)) +
                           (1 << (offset_bits - conv_params->round_1 - 1)));
599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614
#if CONFIG_JNT_COMP
      if (conv_params->fwd_offset == -1) {
        if (conv_params->do_average)
          dst[y * dst_stride + x] += res;
        else
          dst[y * dst_stride + x] = res;
      } else {
        if (conv_params->do_average) {
          dst[y * dst_stride + x] += res * conv_params->bck_offset;

          dst[y * dst_stride + x] >>= (DIST_PRECISION_BITS - 1);
        } else {
          dst[y * dst_stride + x] = res * conv_params->fwd_offset;
        }
      }
#else
615 616 617 618
      if (conv_params->do_average)
        dst[y * dst_stride + x] += res;
      else
        dst[y * dst_stride + x] = res;
619
#endif  // CONFIG_JNT_COMP
620 621 622 623 624
    }
    src_vert++;
  }
}
#endif  // CONFIG_COMPOUND_ROUND
Angie Chiang's avatar
Angie Chiang committed
625 626 627

void av1_convolve_2d_facade(const uint8_t *src, int src_stride, uint8_t *dst,
                            int dst_stride, int w, int h,
628 629 630
                            InterpFilters interp_filters, const int subpel_x_q4,
                            int x_step_q4, const int subpel_y_q4, int y_step_q4,
                            int scaled, ConvolveParams *conv_params) {
Angie Chiang's avatar
Angie Chiang committed
631 632 633 634
  (void)x_step_q4;
  (void)y_step_q4;
  (void)dst;
  (void)dst_stride;
635 636

  InterpFilterParams filter_params_x, filter_params_y;
Zhijie Yang's avatar
Zhijie Yang committed
637 638 639 640
#if CONFIG_SHORT_FILTER
  av1_get_convolve_filter_params(interp_filters, 1, &filter_params_x,
                                 &filter_params_y, w, h);
#else
641 642
  av1_get_convolve_filter_params(interp_filters, 1, &filter_params_x,
                                 &filter_params_y);
Zhijie Yang's avatar
Zhijie Yang committed
643
#endif
644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659

  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);

660 661 662 663 664 665 666 667 668 669 670 671 672
// horizontal and vertical parameters are swapped because of the transpose
#if CONFIG_JNT_COMP
    if (scaled)
      av1_convolve_2d_scale_c(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,
                              y_step_q4, subpel_x_q4, x_step_q4, conv_params);
    else
      av1_convolve_2d_c(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);
#else
673 674 675 676 677 678 679 680 681 682
    if (scaled)
      av1_convolve_2d_scale(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,
                            y_step_q4, subpel_x_q4, x_step_q4, conv_params);
    else
      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);
683
#endif  // CONFIG_JNT_COMP
684 685 686
    transpose_int32(conv_params->dst, conv_params->dst_stride, tr_dst,
                    tr_dst_stride, h, w);
  } else {
687 688 689 690 691 692 693 694 695 696 697 698
#if CONFIG_JNT_COMP
    if (scaled)
      av1_convolve_2d_scale_c(src, src_stride, conv_params->dst,
                              conv_params->dst_stride, w, h, &filter_params_x,
                              &filter_params_y, subpel_x_q4, x_step_q4,
                              subpel_y_q4, y_step_q4, conv_params);
    else
      av1_convolve_2d_c(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);
#else
699 700 701 702 703 704 705 706 707
    if (scaled)
      av1_convolve_2d_scale(src, src_stride, conv_params->dst,
                            conv_params->dst_stride, w, h, &filter_params_x,
                            &filter_params_y, subpel_x_q4, x_step_q4,
                            subpel_y_q4, y_step_q4, conv_params);
    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);
708
#endif  // CONFIG_JNT_COMP
709
  }
Angie Chiang's avatar
Angie Chiang committed
710 711
}

712
#if CONFIG_HIGHBITDEPTH
713 714 715
void av1_highbd_convolve_rounding_c(const int32_t *src, int src_stride,
                                    uint8_t *dst8, int dst_stride, int w, int h,
                                    int bits, int bd) {
716 717 718 719 720
  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(
721
          ROUND_POWER_OF_TWO(src[r * src_stride + c], bits), bd);
722 723 724 725
    }
  }
}

726
#if CONFIG_COMPOUND_ROUND
David Barker's avatar
David Barker committed
727 728 729 730 731 732
void av1_highbd_convolve_2d_c(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) {
733
  int x, y, k;
734
  uint16_t im_block[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) * MAX_SB_SIZE];
735 736 737 738
  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;
739

740 741 742 743 744 745
  // 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) {
746
      int32_t sum = 0;
747 748 749
      for (k = 0; k < filter_params_x->taps; ++k) {
        sum += x_filter[k] * src_horiz[y * src_stride + x - fo_horiz + k];
      }
750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765
      im_block[y * im_stride + x] =
          clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, conv_params->round_0), bd);
    }
  }

  // vertical filter
  uint16_t *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];
      }
      CONV_BUF_TYPE res = ROUND_POWER_OF_TWO(sum, conv_params->round_1);
766 767 768 769
      if (conv_params->do_average)
        dst[y * dst_stride + x] += res;
      else
        dst[y * dst_stride + x] = res;
770 771 772
    }
  }
}
773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824

void av1_highbd_convolve_2d_scale_c(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_qn, const int x_step_qn,
                                    const int subpel_y_qn, const int y_step_qn,
                                    ConvolveParams *conv_params, int bd) {
  int x, y, k;
  uint16_t im_block[(2 * MAX_SB_SIZE + MAX_FILTER_TAP) * MAX_SB_SIZE];
  int im_h = (((h - 1) * y_step_qn + subpel_y_qn) >> SCALE_SUBPEL_BITS) +
             filter_params_y->taps;
  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)bd;

  // horizontal filter
  const uint16_t *src_horiz = src - fo_vert * src_stride;
  for (y = 0; y < im_h; ++y) {
    int x_qn = subpel_x_qn;
    for (x = 0; x < w; ++x, x_qn += x_step_qn) {
      const uint16_t *const src_x = &src_horiz[(x_qn >> SCALE_SUBPEL_BITS)];
      const int x_filter_idx = (x_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
      assert(x_filter_idx < SUBPEL_SHIFTS);
      const int16_t *x_filter =
          av1_get_interp_filter_subpel_kernel(*filter_params_x, x_filter_idx);
      int sum = 0;
      for (k = 0; k < filter_params_x->taps; ++k)
        sum += x_filter[k] * src_x[k - fo_horiz];
      im_block[y * im_stride + x] =
          clip_pixel(ROUND_POWER_OF_TWO(sum, conv_params->round_0));
    }
    src_horiz += src_stride;
  }

  // vertical filter
  uint16_t *src_vert = im_block + fo_vert * im_stride;
  for (x = 0; x < w; ++x) {
    int y_qn = subpel_y_qn;
    for (y = 0; y < h; ++y, y_qn += y_step_qn) {
      const uint16_t *const src_y =
          &src_vert[(y_qn >> SCALE_SUBPEL_BITS) * im_stride];
      const int y_filter_idx = (y_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
      assert(y_filter_idx < SUBPEL_SHIFTS);
      const int16_t *y_filter =
          av1_get_interp_filter_subpel_kernel(*filter_params_y, y_filter_idx);
      CONV_BUF_TYPE sum = 0;
      for (k = 0; k < filter_params_y->taps; ++k) {
        sum += y_filter[k] * src_y[(k - fo_vert) * im_stride];
      }
      CONV_BUF_TYPE res = ROUND_POWER_OF_TWO(sum, conv_params->round_1);
825 826 827 828
      if (conv_params->do_average)
        dst[y * dst_stride + x] += res;
      else
        dst[y * dst_stride + x] = res;
829 830 831 832 833
    }
    src_vert++;
  }
}

834
#else
835

David Barker's avatar
David Barker committed
836 837 838 839 840 841
void av1_highbd_convolve_2d_c(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) {
842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859
  int x, y, k;
  int32_t 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;

  // 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) {
      int32_t sum = (1 << (bd + FILTER_BITS - 1));
      for (k = 0; k < filter_params_x->taps; ++k) {
        sum += x_filter[k] * src_horiz[y * src_stride + x - fo_horiz + k];
      }
      assert(0 <= sum && sum < (1 << (bd + FILTER_BITS + 1)));
860 861
      (void)bd;
      im_block[y * im_stride + x] =
862
          ROUND_POWER_OF_TWO(sum, conv_params->round_0);
863 864 865 866
    }
  }

  // vertical filter
867 868
  int32_t *src_vert = im_block + fo_vert * im_stride;
  const int offset_bits = bd + 2 * FILTER_BITS - conv_params->round_0;
869 870 871 872
  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) {
873
      CONV_BUF_TYPE sum = 1 << offset_bits;
874 875 876
      for (k = 0; k < filter_params_y->taps; ++k) {
        sum += y_filter[k] * src_vert[(y - fo_vert + k) * im_stride + x];
      }
877 878 879 880
      assert(0 <= sum && sum < (1 << (offset_bits + 2)));
      CONV_BUF_TYPE res = ROUND_POWER_OF_TWO(sum, conv_params->round_1) -
                          ((1 << (offset_bits - conv_params->round_1)) +
                           (1 << (offset_bits - conv_params->round_1 - 1)));
881 882 883 884
      if (conv_params->do_average)
        dst[y * dst_stride + x] += res;
      else
        dst[y * dst_stride + x] = res;
885 886 887
    }
  }
}
888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943

void av1_highbd_convolve_2d_scale_c(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_qn, const int x_step_qn,
                                    const int subpel_y_qn, const int y_step_qn,
                                    ConvolveParams *conv_params, int bd) {
  int x, y, k;
  int32_t im_block[(2 * MAX_SB_SIZE + MAX_FILTER_TAP) * MAX_SB_SIZE];
  int im_h = (((h - 1) * y_step_qn + subpel_y_qn) >> SCALE_SUBPEL_BITS) +
             filter_params_y->taps;
  int im_stride = w;
  const int fo_vert = filter_params_y->taps / 2 - 1;
  const int fo_horiz = filter_params_x->taps / 2 - 1;

  // horizontal filter
  const uint16_t *src_horiz = src - fo_vert * src_stride;
  for (y = 0; y < im_h; ++y) {
    int x_qn = subpel_x_qn;
    for (x = 0; x < w; ++x, x_qn += x_step_qn) {
      const uint16_t *const src_x = &src_horiz[(x_qn >> SCALE_SUBPEL_BITS)];
      const int x_filter_idx = (x_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
      assert(x_filter_idx < SUBPEL_SHIFTS);
      const int16_t *x_filter =
          av1_get_interp_filter_subpel_kernel(*filter_params_x, x_filter_idx);
      int32_t sum = (1 << (bd + FILTER_BITS - 1));
      for (k = 0; k < filter_params_x->taps; ++k) {
        sum += x_filter[k] * src_x[k - fo_horiz];
      }
      assert(0 <= sum && sum < (1 << (bd + FILTER_BITS + 1)));
      im_block[y * im_stride + x] =
          ROUND_POWER_OF_TWO(sum, conv_params->round_0);
    }
    src_horiz += src_stride;
  }

  // vertical filter
  int32_t *src_vert = im_block + fo_vert * im_stride;
  const int offset_bits = bd + 2 * FILTER_BITS - conv_params->round_0;
  for (x = 0; x < w; ++x) {
    int y_qn = subpel_y_qn;
    for (y = 0; y < h; ++y, y_qn += y_step_qn) {
      const int32_t *src_y = &src_vert[(y_qn >> SCALE_SUBPEL_BITS) * im_stride];
      const int y_filter_idx = (y_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
      assert(y_filter_idx < SUBPEL_SHIFTS);
      const int16_t *y_filter =
          av1_get_interp_filter_subpel_kernel(*filter_params_y, y_filter_idx);
      CONV_BUF_TYPE sum = 1 << offset_bits;
      for (k = 0; k < filter_params_y->taps; ++k) {
        sum += y_filter[k] * src_y[(k - fo_vert) * im_stride];
      }
      assert(0 <= sum && sum < (1 << (offset_bits + 2)));
      CONV_BUF_TYPE res = ROUND_POWER_OF_TWO(sum, conv_params->round_1) -
                          ((1 << (offset_bits - conv_params->round_1)) +
                           (1 << (offset_bits - conv_params->round_1 - 1)));
944 945 946 947
      if (conv_params->do_average)
        dst[y * dst_stride + x] += res;
      else
        dst[y * dst_stride + x] = res;
948 949 950 951 952
    }
    src_vert++;
  }
}
#endif  // CONFIG_COMPOUND_ROUND
953 954 955

void av1_highbd_convolve_2d_facade(const uint8_t *src8, int src_stride,
                                   uint8_t *dst, int dst_stride, int w, int h,
956
                                   InterpFilters interp_filters,
957 958
                                   const int subpel_x_q4, int x_step_q4,
                                   const int subpel_y_q4, int y_step_q4,
959 960
                                   int scaled, ConvolveParams *conv_params,
                                   int bd) {
961 962 963 964
  (void)x_step_q4;
  (void)y_step_q4;
  (void)dst;
  (void)dst_stride;
965 966

  InterpFilterParams filter_params_x, filter_params_y;
Zhijie Yang's avatar
Zhijie Yang committed
967 968 969 970
#if CONFIG_SHORT_FILTER
  av1_get_convolve_filter_params(interp_filters, 1, &filter_params_x,
                                 &filter_params_y, w, h);
#else
971 972
  av1_get_convolve_filter_params(interp_filters, 1, &filter_params_x,
                                 &filter_params_y);
Zhijie Yang's avatar
Zhijie Yang committed
973
#endif
974

975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991
  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
992 993 994 995 996 997 998 999 1000 1001
    if (scaled)
      av1_highbd_convolve_2d_scale(
          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,
          y_step_q4, subpel_x_q4, x_step_q4, conv_params, bd);
    else
      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);
1002 1003 1004
    transpose_int32(conv_params->dst, conv_params->dst_stride, tr_dst,
                    tr_dst_stride, h, w);
  } else {
1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016
    if (scaled)
      av1_highbd_convolve_2d_scale(
          src, src_stride, conv_params->dst, conv_params->dst_stride, w, h,
          &filter_params_x, &filter_params_y, subpel_x_q4, x_step_q4,
          subpel_y_q4, y_step_q4, conv_params, bd);
    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);
  }
}
1017 1018
#endif  // CONFIG_HIGHBITDEPTH

Angie Chiang's avatar
Angie Chiang committed
1019
#endif  // CONFIG_CONVOLVE_ROUND
Angie Chiang's avatar
Angie Chiang committed
1020

1021 1022 1023 1024 1025 1026 1027 1028
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,
1029
                            const InterpFilters interp_filters,
1030 1031 1032 1033 1034
                            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) {
Fergus Simpson's avatar
Fergus Simpson committed
1035 1036
  int ignore_horiz = x_step_q4 == SUBPEL_SHIFTS && subpel_x_q4 == 0;
  int ignore_vert = y_step_q4 == SUBPEL_SHIFTS && subpel_y_q4 == 0;
1037 1038

  InterpFilterParams filter_params_x, filter_params_y;
Zhijie Yang's avatar
Zhijie Yang committed
1039 1040 1041 1042
#if CONFIG_SHORT_FILTER
  av1_get_convolve_filter_params(interp_filters, 0, &filter_params_x,
                                 &filter_params_y, w, h);
#else
1043 1044
  av1_get_convolve_filter_params(interp_filters, 0, &filter_params_x,
                                 &filter_params_y);
Zhijie Yang's avatar
Zhijie Yang committed
1045
#endif
1046

Angie Chiang's avatar
Angie Chiang committed
1047
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
1048 1049 1050 1051 1052 1053

  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
1054
  if (ignore_horiz && ignore_vert) {
1055
    convolve_copy(src, src_stride, dst, dst_stride, w, h, conv_params);
Angie Chiang's avatar
Angie Chiang committed
1056
  } else if (ignore_vert) {
1057 1058
    assert(filter_params_x.taps <= MAX_FILTER_TAP);
    convolve_horiz(src, src_stride, dst, dst_stride, w, h, filter_params_x,
1059
                   subpel_x_q4, x_step_q4, conv_params);
Angie Chiang's avatar
Angie Chiang committed
1060
  } else if (ignore_horiz) {
1061 1062
    assert(filter_params_y.taps <= MAX_FILTER_TAP);
    convolve_vert(src, src_stride, dst, dst_stride, w, h, filter_params_y,
1063
                  subpel_y_q4, y_step_q4, conv_params);
Angie Chiang's avatar
Angie Chiang committed
1064
  } else {
1065 1066 1067 1068 1069 1070
    // 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);
1071
    int filter_size;
1072
#if CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
1073
    av1_convolve_filter_params_fixup_1212(&filter_params_x, &filter_params_y);
1074

1075 1076 1077 1078
    // we do filter with fewer taps first to reduce hardware implementation
    // complexity
    if (filter_params_y.taps < filter_params_x.taps) {
      int intermediate_width;
1079
      int temp_stride = max_intermediate_size;
1080 1081
      ConvolveParams temp_conv_params;
      temp_conv_params.ref = 0;
1082
      temp_conv_params.do_average = 0;
1083
      temp_conv_params.round = CONVOLVE_OPT_ROUND;
1084 1085 1086
      filter_size = filter_params_x.taps;
      intermediate_width =
          (((w - 1) * x_step_q4 + subpel_x_q4) >> SUBPEL_BITS) + filter_size;
1087
      assert(intermediate_width <= max_intermediate_size);
Angie Chiang's avatar
Angie Chiang committed
1088

1089
      assert(filter_params_y.taps <= MAX_FILTER_TAP);
1090

1091
      convolve_vert(src - (filter_size / 2 - 1), src_stride, temp, temp_stride,
1092
                    intermediate_width, h, filter_params_y, subpel_y_q4,
1093
                    y_step_q4, &temp_conv_params);
1094

1095
      assert(filter_params_x.taps <= MAX_FILTER_TAP);
1096
      convolve_horiz(temp + (filter_size / 2 - 1), temp_stride, dst, dst_stride,
1097 1098
                     w, h, filter_params_x, subpel_x_q4, x_step_q4,
                     conv_params);
1099
    } else
1100
#endif  // CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
1101 1102
    {
      int intermediate_height;
1103
      int temp_stride = MAX_SB_SIZE;
1104 1105
      ConvolveParams temp_conv_params;
      temp_conv_params.ref = 0;
1106
      temp_conv_params.do_average = 0;
1107
      temp_conv_params.round = CONVOLVE_OPT_ROUND;
1108 1109 1110
      filter_size = filter_params_y.taps;
      intermediate_height =
          (((h - 1) * y_step_q4 + subpel_y_q4) >> SUBPEL_BITS) + filter_size;
1111 1112
      assert(intermediate_height <= max_intermediate_size);
      (void)max_intermediate_size;
1113

1114
      assert(filter_params_x.taps <= MAX_FILTER_TAP);
1115

1116
      convolve_horiz(src - src_stride * (filter_size / 2 - 1), src_stride, temp,
1117
                     temp_stride, w, intermediate_height, filter_params_x,
1118
                     subpel_x_q4, x_step_q4, &temp_conv_params);
1119

1120
      assert(filter_params_y.taps <= MAX_FILTER_TAP);
1121

1122
      convolve_vert(temp + temp_stride * (filter_size / 2 - 1), temp_stride,
1123
                    dst, dst_stride, w, h, filter_params_y, subpel_y_q4,
1124
                    y_step_q4, conv_params);
1125
    }
Angie Chiang's avatar
Angie Chiang committed
1126
  }
1127 1128
}

Fergus Simpson's avatar
Fergus Simpson committed
1129 1130
static void convolve_scale_helper(const uint8_t *src, int src_stride,
                                  uint8_t *dst, int dst_stride, int w, int h,
1131
                                  const InterpFilters interp_filters,
Fergus Simpson's avatar
Fergus Simpson committed
1132 1133 1134 1135 1136 1137 1138
                                  const int subpel_x_qn, int x_step_qn,
                                  const int subpel_y_qn, int y_step_qn,
                                  ConvolveParams *conv_params,
                                  ConvolveFunc convolve_horiz,
                                  ConvolveFunc convolve_vert) {
  int ignore_horiz = x_step_qn == SCALE_SUBPEL_SHIFTS && subpel_x_qn == 0;
  int ignore_vert = y_step_qn == SCALE_SUBPEL_SHIFTS && subpel_y_qn == 0