aom_convolve.c 54.7 KB
Newer Older
John Koleszar's avatar
John Koleszar committed
1
/*
Yaowu Xu's avatar
Yaowu Xu committed
2
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
John Koleszar's avatar
John Koleszar committed
3
 *
Yaowu Xu's avatar
Yaowu Xu committed
4 5 6 7 8 9
 * 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.
John Koleszar's avatar
John Koleszar committed
10
 */
Christian Duvivier's avatar
Christian Duvivier committed
11

John Koleszar's avatar
John Koleszar committed
12
#include <assert.h>
Zoe Liu's avatar
Zoe Liu committed
13
#include <string.h>
John Koleszar's avatar
John Koleszar committed
14

Yaowu Xu's avatar
Yaowu Xu committed
15 16 17 18 19 20
#include "./aom_config.h"
#include "./aom_dsp_rtcd.h"
#include "aom/aom_integer.h"
#include "aom_dsp/aom_convolve.h"
#include "aom_dsp/aom_dsp_common.h"
#include "aom_dsp/aom_filter.h"
21
#include "aom_ports/mem.h"
John Koleszar's avatar
John Koleszar committed
22

Dmitry Kovalev's avatar
Dmitry Kovalev committed
23 24
static void convolve_horiz(const uint8_t *src, ptrdiff_t src_stride,
                           uint8_t *dst, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
25 26
                           const InterpKernel *x_filters, int x0_q4,
                           int x_step_q4, int w, int h) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
27 28
  int x, y;
  src -= SUBPEL_TAPS / 2 - 1;
John Koleszar's avatar
John Koleszar committed
29
  for (y = 0; y < h; ++y) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
30
    int x_q4 = x0_q4;
John Koleszar's avatar
John Koleszar committed
31
    for (x = 0; x < w; ++x) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
32 33 34
      const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
      int k, sum = 0;
clang-format's avatar
clang-format committed
35
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
36
      dst[x] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
John Koleszar's avatar
John Koleszar committed
37 38 39 40 41 42 43
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

Fergus Simpson's avatar
Fergus Simpson committed
44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66
static void convolve_horiz_scale_c(const uint8_t *src, ptrdiff_t src_stride,
                                   uint8_t *dst, ptrdiff_t dst_stride,
                                   const InterpKernel *x_filters, int x0_qn,
                                   int x_step_qn, int w, int h) {
  int x, y;
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_qn = x0_qn;
    for (x = 0; x < w; ++x) {
      const uint8_t *const src_x = &src[x_qn >> SCALE_SUBPEL_BITS];  // q8
      const int x_filter_idx = (x_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
      assert(x_filter_idx < SUBPEL_SHIFTS);
      const int16_t *const x_filter = x_filters[x_filter_idx];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
      dst[x] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
      x_qn += x_step_qn;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

Dmitry Kovalev's avatar
Dmitry Kovalev committed
67 68
static void convolve_avg_horiz(const uint8_t *src, ptrdiff_t src_stride,
                               uint8_t *dst, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
69 70
                               const InterpKernel *x_filters, int x0_q4,
                               int x_step_q4, int w, int h) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
71 72
  int x, y;
  src -= SUBPEL_TAPS / 2 - 1;
John Koleszar's avatar
John Koleszar committed
73
  for (y = 0; y < h; ++y) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
74
    int x_q4 = x0_q4;
John Koleszar's avatar
John Koleszar committed
75
    for (x = 0; x < w; ++x) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
76 77 78
      const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
      int k, sum = 0;
clang-format's avatar
clang-format committed
79 80 81
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
      dst[x] = ROUND_POWER_OF_TWO(
          dst[x] + clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)), 1);
John Koleszar's avatar
John Koleszar committed
82 83 84 85 86 87 88
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

Fergus Simpson's avatar
Fergus Simpson committed
89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112
static void convolve_avg_horiz_scale_c(const uint8_t *src, ptrdiff_t src_stride,
                                       uint8_t *dst, ptrdiff_t dst_stride,
                                       const InterpKernel *x_filters, int x0_qn,
                                       int x_step_qn, int w, int h) {
  int x, y;
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_qn = x0_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 *const x_filter = x_filters[x_filter_idx];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
      dst[x] = ROUND_POWER_OF_TWO(
          dst[x] + clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)), 1);
      x_qn += x_step_qn;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

Dmitry Kovalev's avatar
Dmitry Kovalev committed
113 114
static void convolve_vert(const uint8_t *src, ptrdiff_t src_stride,
                          uint8_t *dst, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
115 116
                          const InterpKernel *y_filters, int y0_q4,
                          int y_step_q4, int w, int h) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
117 118
  int x, y;
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
John Koleszar's avatar
John Koleszar committed
119

120
  for (x = 0; x < w; ++x) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
121
    int y_q4 = y0_q4;
John Koleszar's avatar
John Koleszar committed
122
    for (y = 0; y < h; ++y) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
123 124 125 126 127 128
      const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      dst[y * dst_stride] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
John Koleszar's avatar
John Koleszar committed
129 130 131 132 133 134 135
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

Fergus Simpson's avatar
Fergus Simpson committed
136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160
static void convolve_vert_scale_c(const uint8_t *src, ptrdiff_t src_stride,
                                  uint8_t *dst, ptrdiff_t dst_stride,
                                  const InterpKernel *y_filters, int y0_qn,
                                  int y_step_qn, int w, int h) {
  int x, y;
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);

  for (x = 0; x < w; ++x) {
    int y_qn = y0_qn;
    for (y = 0; y < h; ++y) {
      const unsigned char *src_y =
          &src[(y_qn >> SCALE_SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter =
          y_filters[(y_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      dst[y * dst_stride] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
      y_qn += y_step_qn;
    }
    ++src;
    ++dst;
  }
}

Dmitry Kovalev's avatar
Dmitry Kovalev committed
161 162
static void convolve_avg_vert(const uint8_t *src, ptrdiff_t src_stride,
                              uint8_t *dst, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
163 164
                              const InterpKernel *y_filters, int y0_q4,
                              int y_step_q4, int w, int h) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
165 166
  int x, y;
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
John Koleszar's avatar
John Koleszar committed
167

168
  for (x = 0; x < w; ++x) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
169
    int y_q4 = y0_q4;
John Koleszar's avatar
John Koleszar committed
170
    for (y = 0; y < h; ++y) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
171 172 173 174 175
      const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
clang-format's avatar
clang-format committed
176 177 178 179
      dst[y * dst_stride] = ROUND_POWER_OF_TWO(
          dst[y * dst_stride] +
              clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)),
          1);
John Koleszar's avatar
John Koleszar committed
180 181 182 183 184 185 186
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

Fergus Simpson's avatar
Fergus Simpson committed
187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214
static void convolve_avg_vert_scale_c(const uint8_t *src, ptrdiff_t src_stride,
                                      uint8_t *dst, ptrdiff_t dst_stride,
                                      const InterpKernel *y_filters, int y0_qn,
                                      int y_step_qn, int w, int h) {
  int x, y;
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);

  for (x = 0; x < w; ++x) {
    int y_qn = y0_qn;
    for (y = 0; y < h; ++y) {
      const unsigned char *src_y =
          &src[(y_qn >> SCALE_SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter =
          y_filters[(y_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      dst[y * dst_stride] = ROUND_POWER_OF_TWO(
          dst[y * dst_stride] +
              clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)),
          1);
      y_qn += y_step_qn;
    }
    ++src;
    ++dst;
  }
}

clang-format's avatar
clang-format committed
215 216
static void convolve(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                     ptrdiff_t dst_stride, const InterpKernel *const x_filters,
Dmitry Kovalev's avatar
Dmitry Kovalev committed
217
                     int x0_q4, int x_step_q4,
clang-format's avatar
clang-format committed
218 219
                     const InterpKernel *const y_filters, int y0_q4,
                     int y_step_q4, int w, int h) {
220 221 222 223 224 225 226 227 228 229 230 231
  // Note: Fixed size intermediate buffer, temp, places limits on parameters.
  // 2d filtering proceeds in 2 steps:
  //   (1) Interpolate horizontally into an intermediate buffer, temp.
  //   (2) Interpolate temp vertically to derive the sub-pixel result.
  // Deriving the maximum number of rows in the temp buffer (135):
  // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
  // --Largest block size is 64x64 pixels.
  // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
  //   original frame (in 1/16th pixel units).
  // --Must round-up because block may be located at sub-pixel position.
  // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
  // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
232
  uint8_t temp[MAX_EXT_SIZE * MAX_SB_SIZE];
233
  int intermediate_height =
clang-format's avatar
clang-format committed
234
      (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;
235

236 237
  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);
238

239 240
  assert(y_step_q4 <= 32);
  assert(x_step_q4 <= 32);
241

clang-format's avatar
clang-format committed
242 243 244 245 246
  convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride, temp,
                 MAX_SB_SIZE, x_filters, x0_q4, x_step_q4, w,
                 intermediate_height);
  convolve_vert(temp + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1), MAX_SB_SIZE, dst,
                dst_stride, y_filters, y0_q4, y_step_q4, w, h);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
247 248
}

Fergus Simpson's avatar
Fergus Simpson committed
249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283
static void convolve_scale_c(const uint8_t *src, ptrdiff_t src_stride,
                             uint8_t *dst, ptrdiff_t dst_stride,
                             const InterpKernel *const x_filters, int x0_qn,
                             int x_step_qn, const InterpKernel *const y_filters,
                             int y0_qn, int y_step_qn, int w, int h) {
  // TODO(afergs): Update comment here
  // Note: Fixed size intermediate buffer, temp, places limits on parameters.
  // 2d filtering proceeds in 2 steps:
  //   (1) Interpolate horizontally into an intermediate buffer, temp.
  //   (2) Interpolate temp vertically to derive the sub-pixel result.
  // Deriving the maximum number of rows in the temp buffer (135):
  // --Smallest scaling factor is x1/2 ==> y_step_qn = 32 (Normative).
  // --Largest block size is 64x64 pixels.
  // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
  //   original frame (in 1/16th pixel units).
  // --Must round-up because block may be located at sub-pixel position.
  // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
  // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
  uint8_t temp[MAX_EXT_SIZE * MAX_SB_SIZE];
  int intermediate_height =
      (((h - 1) * y_step_qn + y0_qn) >> SCALE_SUBPEL_BITS) + SUBPEL_TAPS;

  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);

  assert(y_step_qn <= SCALE_SUBPEL_BITS * 2);
  assert(x_step_qn <= SCALE_SUBPEL_BITS * 2);

  convolve_horiz_scale_c(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride,
                         temp, MAX_SB_SIZE, x_filters, x0_qn, x_step_qn, w,
                         intermediate_height);
  convolve_vert_scale_c(temp + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1), MAX_SB_SIZE,
                        dst, dst_stride, y_filters, y0_qn, y_step_qn, w, h);
}

284
static const InterpKernel *get_filter_base(const int16_t *filter) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
285 286
  // NOTE: This assumes that the filter table is 256-byte aligned.
  // TODO(agrange) Modify to make independent of table alignment.
287
  return (const InterpKernel *)(((intptr_t)filter) & ~((intptr_t)0xFF));
Dmitry Kovalev's avatar
Dmitry Kovalev committed
288 289
}

290
static int get_filter_offset(const int16_t *f, const InterpKernel *base) {
291
  return (int)((const InterpKernel *)(intptr_t)f - base);
John Koleszar's avatar
John Koleszar committed
292 293
}

Yaowu Xu's avatar
Yaowu Xu committed
294
void aom_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
295
                           uint8_t *dst, ptrdiff_t dst_stride,
John Koleszar's avatar
John Koleszar committed
296
                           const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
297 298
                           const int16_t *filter_y, int y_step_q4, int w,
                           int h) {
299
  const InterpKernel *const filters_x = get_filter_base(filter_x);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
300 301
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

302 303 304
  (void)filter_y;
  (void)y_step_q4;

clang-format's avatar
clang-format committed
305 306
  convolve_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4, x_step_q4,
                 w, h);
John Koleszar's avatar
John Koleszar committed
307 308
}

Fergus Simpson's avatar
Fergus Simpson committed
309 310 311 312 313 314 315 316 317 318 319 320 321 322 323
void aom_convolve8_horiz_scale_c(const uint8_t *src, ptrdiff_t src_stride,
                                 uint8_t *dst, ptrdiff_t dst_stride,
                                 const int16_t *filter_x, int subpel_x,
                                 int x_step_qn, const int16_t *filter_y,
                                 int subpel_y, int y_step_qn, int w, int h) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);

  (void)subpel_y;
  (void)filter_y;
  (void)y_step_qn;

  convolve_horiz_scale_c(src, src_stride, dst, dst_stride, filters_x, subpel_x,
                         x_step_qn, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
324
void aom_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
325
                               uint8_t *dst, ptrdiff_t dst_stride,
John Koleszar's avatar
John Koleszar committed
326
                               const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
327 328
                               const int16_t *filter_y, int y_step_q4, int w,
                               int h) {
329
  const InterpKernel *const filters_x = get_filter_base(filter_x);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
330 331
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

332 333 334
  (void)filter_y;
  (void)y_step_q4;

clang-format's avatar
clang-format committed
335 336
  convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                     x_step_q4, w, h);
John Koleszar's avatar
John Koleszar committed
337 338
}

Fergus Simpson's avatar
Fergus Simpson committed
339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354
void aom_convolve8_avg_horiz_scale_c(const uint8_t *src, ptrdiff_t src_stride,
                                     uint8_t *dst, ptrdiff_t dst_stride,
                                     const int16_t *filter_x, int subpel_x,
                                     int x_step_qn, const int16_t *filter_y,
                                     int subpel_y, int y_step_qn, int w,
                                     int h) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);

  (void)subpel_y;
  (void)filter_y;
  (void)y_step_qn;

  convolve_avg_horiz_scale_c(src, src_stride, dst, dst_stride, filters_x,
                             subpel_x, x_step_qn, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
355
void aom_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
356
                          uint8_t *dst, ptrdiff_t dst_stride,
John Koleszar's avatar
John Koleszar committed
357
                          const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
358 359
                          const int16_t *filter_y, int y_step_q4, int w,
                          int h) {
360
  const InterpKernel *const filters_y = get_filter_base(filter_y);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
361
  const int y0_q4 = get_filter_offset(filter_y, filters_y);
362 363 364 365

  (void)filter_x;
  (void)x_step_q4;

clang-format's avatar
clang-format committed
366 367
  convolve_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4, y_step_q4,
                w, h);
John Koleszar's avatar
John Koleszar committed
368 369
}

Fergus Simpson's avatar
Fergus Simpson committed
370 371 372 373 374 375 376 377 378 379 380 381 382 383 384
void aom_convolve8_vert_scale_c(const uint8_t *src, ptrdiff_t src_stride,
                                uint8_t *dst, ptrdiff_t dst_stride,
                                const int16_t *filter_x, int subpel_x,
                                int x_step_qn, const int16_t *filter_y,
                                int subpel_y, int y_step_qn, int w, int h) {
  const InterpKernel *const filters_y = get_filter_base(filter_y);

  (void)subpel_x;
  (void)filter_x;
  (void)x_step_qn;

  convolve_vert_scale_c(src, src_stride, dst, dst_stride, filters_y, subpel_y,
                        y_step_qn, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
385
void aom_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
386
                              uint8_t *dst, ptrdiff_t dst_stride,
John Koleszar's avatar
John Koleszar committed
387
                              const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
388 389
                              const int16_t *filter_y, int y_step_q4, int w,
                              int h) {
390
  const InterpKernel *const filters_y = get_filter_base(filter_y);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
391
  const int y0_q4 = get_filter_offset(filter_y, filters_y);
392 393 394 395

  (void)filter_x;
  (void)x_step_q4;

clang-format's avatar
clang-format committed
396 397
  convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
                    y_step_q4, w, h);
John Koleszar's avatar
John Koleszar committed
398 399
}

Fergus Simpson's avatar
Fergus Simpson committed
400 401 402 403 404 405 406 407 408 409 410 411 412 413 414
void aom_convolve8_avg_vert_scale_c(const uint8_t *src, ptrdiff_t src_stride,
                                    uint8_t *dst, ptrdiff_t dst_stride,
                                    const int16_t *filter_x, int subpel_x,
                                    int x_step_qn, const int16_t *filter_y,
                                    int subpel_y, int y_step_qn, int w, int h) {
  const InterpKernel *const filters_y = get_filter_base(filter_y);

  (void)subpel_x;
  (void)filter_x;
  (void)x_step_qn;

  convolve_avg_vert_scale_c(src, src_stride, dst, dst_stride, filters_y,
                            subpel_y, y_step_qn, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
415
void aom_convolve8_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
416 417
                     ptrdiff_t dst_stride, const int16_t *filter_x,
                     int x_step_q4, const int16_t *filter_y, int y_step_q4,
John Koleszar's avatar
John Koleszar committed
418
                     int w, int h) {
419
  const InterpKernel *const filters_x = get_filter_base(filter_x);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
420 421
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

422
  const InterpKernel *const filters_y = get_filter_base(filter_y);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
423 424
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

clang-format's avatar
clang-format committed
425
  convolve(src, src_stride, dst, dst_stride, filters_x, x0_q4, x_step_q4,
Dmitry Kovalev's avatar
Dmitry Kovalev committed
426
           filters_y, y0_q4, y_step_q4, w, h);
John Koleszar's avatar
John Koleszar committed
427 428
}

Fergus Simpson's avatar
Fergus Simpson committed
429 430 431 432 433 434 435 436 437 438 439 440 441
void aom_convolve8_scale_c(const uint8_t *src, ptrdiff_t src_stride,
                           uint8_t *dst, ptrdiff_t dst_stride,
                           const int16_t *filter_x, int subpel_x, int x_step_qn,
                           const int16_t *filter_y, int subpel_y, int y_step_qn,
                           int w, int h) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);

  const InterpKernel *const filters_y = get_filter_base(filter_y);

  convolve_scale_c(src, src_stride, dst, dst_stride, filters_x, subpel_x,
                   x_step_qn, filters_y, subpel_y, y_step_qn, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
442
void aom_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
443 444
                         ptrdiff_t dst_stride, const int16_t *filter_x,
                         int x_step_q4, const int16_t *filter_y, int y_step_q4,
John Koleszar's avatar
John Koleszar committed
445
                         int w, int h) {
Christian Duvivier's avatar
Christian Duvivier committed
446
  /* Fixed size intermediate buffer places limits on parameters. */
447 448 449
  DECLARE_ALIGNED(16, uint8_t, temp[MAX_SB_SIZE * MAX_SB_SIZE]);
  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);
Christian Duvivier's avatar
Christian Duvivier committed
450

Yaowu Xu's avatar
Yaowu Xu committed
451
  aom_convolve8_c(src, src_stride, temp, MAX_SB_SIZE, filter_x, x_step_q4,
clang-format's avatar
clang-format committed
452
                  filter_y, y_step_q4, w, h);
Yaowu Xu's avatar
Yaowu Xu committed
453
  aom_convolve_avg_c(temp, MAX_SB_SIZE, dst, dst_stride, NULL, 0, NULL, 0, w,
clang-format's avatar
clang-format committed
454
                     h);
John Koleszar's avatar
John Koleszar committed
455
}
456

Fergus Simpson's avatar
Fergus Simpson committed
457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472
void aom_convolve8_avg_scale_c(const uint8_t *src, ptrdiff_t src_stride,
                               uint8_t *dst, ptrdiff_t dst_stride,
                               const int16_t *filter_x, int subpel_x,
                               int x_step_qn, const int16_t *filter_y,
                               int subpel_y, int y_step_qn, int w, int h) {
  /* Fixed size intermediate buffer places limits on parameters. */
  DECLARE_ALIGNED(16, uint8_t, temp[MAX_SB_SIZE * MAX_SB_SIZE]);
  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);

  aom_convolve8_scale_c(src, src_stride, temp, MAX_SB_SIZE, filter_x, subpel_x,
                        x_step_qn, filter_y, subpel_y, y_step_qn, w, h);
  aom_convolve_avg_c(temp, MAX_SB_SIZE, dst, dst_stride, NULL, 0, NULL, 0, w,
                     h);
}

Yaowu Xu's avatar
Yaowu Xu committed
473
void aom_convolve_copy_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
474 475 476
                         ptrdiff_t dst_stride, const int16_t *filter_x,
                         int filter_x_stride, const int16_t *filter_y,
                         int filter_y_stride, int w, int h) {
477 478
  int r;

clang-format's avatar
clang-format committed
479 480 481 482
  (void)filter_x;
  (void)filter_x_stride;
  (void)filter_y;
  (void)filter_y_stride;
483

484
  for (r = h; r > 0; --r) {
James Zern's avatar
James Zern committed
485
    memcpy(dst, src, w);
486 487
    src += src_stride;
    dst += dst_stride;
488 489 490
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
491
void aom_convolve_avg_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
492 493 494
                        ptrdiff_t dst_stride, const int16_t *filter_x,
                        int filter_x_stride, const int16_t *filter_y,
                        int filter_y_stride, int w, int h) {
495 496
  int x, y;

clang-format's avatar
clang-format committed
497 498 499 500
  (void)filter_x;
  (void)filter_x_stride;
  (void)filter_y;
  (void)filter_y_stride;
501

502
  for (y = 0; y < h; ++y) {
clang-format's avatar
clang-format committed
503
    for (x = 0; x < w; ++x) dst[x] = ROUND_POWER_OF_TWO(dst[x] + src[x], 1);
504

505 506 507 508
    src += src_stride;
    dst += dst_stride;
  }
}
509

Yaowu Xu's avatar
Yaowu Xu committed
510
void aom_scaled_horiz_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
511 512
                        ptrdiff_t dst_stride, const int16_t *filter_x,
                        int x_step_q4, const int16_t *filter_y, int y_step_q4,
513
                        int w, int h) {
Yaowu Xu's avatar
Yaowu Xu committed
514
  aom_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
515 516 517
                        filter_y, y_step_q4, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
518
void aom_scaled_vert_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
519 520
                       ptrdiff_t dst_stride, const int16_t *filter_x,
                       int x_step_q4, const int16_t *filter_y, int y_step_q4,
521
                       int w, int h) {
Yaowu Xu's avatar
Yaowu Xu committed
522
  aom_convolve8_vert_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
523 524 525
                       filter_y, y_step_q4, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
526
void aom_scaled_2d_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
527 528
                     ptrdiff_t dst_stride, const int16_t *filter_x,
                     int x_step_q4, const int16_t *filter_y, int y_step_q4,
529
                     int w, int h) {
Yaowu Xu's avatar
Yaowu Xu committed
530
  aom_convolve8_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
531 532 533
                  filter_y, y_step_q4, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
534
void aom_scaled_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
535 536
                            uint8_t *dst, ptrdiff_t dst_stride,
                            const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
537 538
                            const int16_t *filter_y, int y_step_q4, int w,
                            int h) {
Yaowu Xu's avatar
Yaowu Xu committed
539
  aom_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter_x,
540 541 542
                            x_step_q4, filter_y, y_step_q4, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
543
void aom_scaled_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
544 545
                           uint8_t *dst, ptrdiff_t dst_stride,
                           const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
546 547
                           const int16_t *filter_y, int y_step_q4, int w,
                           int h) {
Yaowu Xu's avatar
Yaowu Xu committed
548
  aom_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter_x,
549 550 551
                           x_step_q4, filter_y, y_step_q4, w, h);
}

Yaowu Xu's avatar
Yaowu Xu committed
552
void aom_scaled_avg_2d_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
clang-format's avatar
clang-format committed
553 554 555
                         ptrdiff_t dst_stride, const int16_t *filter_x,
                         int x_step_q4, const int16_t *filter_y, int y_step_q4,
                         int w, int h) {
Yaowu Xu's avatar
Yaowu Xu committed
556
  aom_convolve8_avg_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
557 558 559
                      filter_y, y_step_q4, w, h);
}

Fergus Simpson's avatar
Fergus Simpson committed
560
// TODO(afergs): Make sure this works too
561 562 563 564 565
#if CONFIG_LOOP_RESTORATION
static void convolve_add_src_horiz(const uint8_t *src, ptrdiff_t src_stride,
                                   uint8_t *dst, ptrdiff_t dst_stride,
                                   const InterpKernel *x_filters, int x0_q4,
                                   int x_step_q4, int w, int h) {
566
  int x, y, k;
567 568 569 570 571 572
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = x0_q4;
    for (x = 0; x < w; ++x) {
      const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
573
      int sum = 0;
574 575 576 577 578 579 580 581 582 583 584 585 586 587
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
      dst[x] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS) +
                          src_x[SUBPEL_TAPS / 2 - 1]);
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

static void convolve_add_src_vert(const uint8_t *src, ptrdiff_t src_stride,
                                  uint8_t *dst, ptrdiff_t dst_stride,
                                  const InterpKernel *y_filters, int y0_q4,
                                  int y_step_q4, int w, int h) {
588
  int x, y, k;
589 590 591 592 593 594 595
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);

  for (x = 0; x < w; ++x) {
    int y_q4 = y0_q4;
    for (y = 0; y < h; ++y) {
      const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
596
      int sum = 0;
597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      dst[y * dst_stride] =
          clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS) +
                     src_y[(SUBPEL_TAPS / 2 - 1) * src_stride]);
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

static void convolve_add_src(const uint8_t *src, ptrdiff_t src_stride,
                             uint8_t *dst, ptrdiff_t dst_stride,
                             const InterpKernel *const x_filters, int x0_q4,
                             int x_step_q4, const InterpKernel *const y_filters,
                             int y0_q4, int y_step_q4, int w, int h) {
  uint8_t temp[MAX_EXT_SIZE * MAX_SB_SIZE];
  int intermediate_height =
      (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;

  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);

  assert(y_step_q4 <= 32);
  assert(x_step_q4 <= 32);

  convolve_add_src_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride,
                         temp, MAX_SB_SIZE, x_filters, x0_q4, x_step_q4, w,
                         intermediate_height);
  convolve_add_src_vert(temp + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1), MAX_SB_SIZE,
                        dst, dst_stride, y_filters, y0_q4, y_step_q4, w, h);
}

void aom_convolve8_add_src_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
                                   uint8_t *dst, ptrdiff_t dst_stride,
                                   const int16_t *filter_x, int x_step_q4,
                                   const int16_t *filter_y, int y_step_q4,
                                   int w, int h) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

  (void)filter_y;
  (void)y_step_q4;

  convolve_add_src_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                         x_step_q4, w, h);
}

void aom_convolve8_add_src_vert_c(const uint8_t *src, ptrdiff_t src_stride,
                                  uint8_t *dst, ptrdiff_t dst_stride,
                                  const int16_t *filter_x, int x_step_q4,
                                  const int16_t *filter_y, int y_step_q4, int w,
                                  int h) {
  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

  (void)filter_x;
  (void)x_step_q4;

  convolve_add_src_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
                        y_step_q4, w, h);
}

void aom_convolve8_add_src_c(const uint8_t *src, ptrdiff_t src_stride,
                             uint8_t *dst, ptrdiff_t dst_stride,
                             const int16_t *filter_x, int x_step_q4,
                             const int16_t *filter_y, int y_step_q4, int w,
                             int h) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

  convolve_add_src(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                   x_step_q4, filters_y, y0_q4, y_step_q4, w, h);
}
675 676 677 678 679

static void convolve_add_src_horiz_hip(const uint8_t *src, ptrdiff_t src_stride,
                                       uint16_t *dst, ptrdiff_t dst_stride,
                                       const InterpKernel *x_filters, int x0_q4,
                                       int x_step_q4, int w, int h) {
680 681
  const int bd = 8;
  int x, y, k;
682 683 684 685 686 687
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = x0_q4;
    for (x = 0; x < w; ++x) {
      const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
688 689
      int sum = ((int)src_x[SUBPEL_TAPS / 2 - 1] << FILTER_BITS) +
                (1 << (bd + FILTER_BITS - 1));
690 691 692
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
      dst[x] =
          (uint16_t)clamp(ROUND_POWER_OF_TWO(sum, FILTER_BITS - EXTRAPREC_BITS),
693
                          0, EXTRAPREC_CLAMP_LIMIT(bd) - 1);
694 695 696 697 698 699 700 701 702 703 704
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

static void convolve_add_src_vert_hip(const uint16_t *src, ptrdiff_t src_stride,
                                      uint8_t *dst, ptrdiff_t dst_stride,
                                      const InterpKernel *y_filters, int y0_q4,
                                      int y_step_q4, int w, int h) {
705 706
  const int bd = 8;
  int x, y, k;
707 708 709 710 711 712 713
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);

  for (x = 0; x < w; ++x) {
    int y_q4 = y0_q4;
    for (y = 0; y < h; ++y) {
      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
714 715 716
      int sum =
          ((int)src_y[(SUBPEL_TAPS / 2 - 1) * src_stride] << FILTER_BITS) -
          (1 << (bd + FILTER_BITS + EXTRAPREC_BITS - 1));
717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      dst[y * dst_stride] =
          clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS + EXTRAPREC_BITS));
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

static void convolve_add_src_hip(const uint8_t *src, ptrdiff_t src_stride,
                                 uint8_t *dst, ptrdiff_t dst_stride,
                                 const InterpKernel *const x_filters, int x0_q4,
                                 int x_step_q4,
                                 const InterpKernel *const y_filters, int y0_q4,
                                 int y_step_q4, int w, int h) {
  uint16_t temp[MAX_EXT_SIZE * MAX_SB_SIZE];
  int intermediate_height =
      (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;

  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);

  assert(y_step_q4 <= 32);
  assert(x_step_q4 <= 32);

  convolve_add_src_horiz_hip(src - src_stride * (SUBPEL_TAPS / 2 - 1),
                             src_stride, temp, MAX_SB_SIZE, x_filters, x0_q4,
                             x_step_q4, w, intermediate_height);
  convolve_add_src_vert_hip(temp + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1),
                            MAX_SB_SIZE, dst, dst_stride, y_filters, y0_q4,
                            y_step_q4, w, h);
}

void aom_convolve8_add_src_horiz_hip_c(const uint8_t *src, ptrdiff_t src_stride,
                                       uint16_t *dst, ptrdiff_t dst_stride,
                                       const int16_t *filter_x, int x_step_q4,
                                       const int16_t *filter_y, int y_step_q4,
                                       int w, int h) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

  (void)filter_y;
  (void)y_step_q4;

  convolve_add_src_horiz_hip(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                             x_step_q4, w, h);
}

void aom_convolve8_add_src_vert_hip_c(const uint16_t *src, ptrdiff_t src_stride,
                                      uint8_t *dst, ptrdiff_t dst_stride,
                                      const int16_t *filter_x, int x_step_q4,
                                      const int16_t *filter_y, int y_step_q4,
                                      int w, int h) {
  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

  (void)filter_x;
  (void)x_step_q4;

  convolve_add_src_vert_hip(src, src_stride, dst, dst_stride, filters_y, y0_q4,
                            y_step_q4, w, h);
}

void aom_convolve8_add_src_hip_c(const uint8_t *src, ptrdiff_t src_stride,
                                 uint8_t *dst, ptrdiff_t dst_stride,
                                 const int16_t *filter_x, int x_step_q4,
                                 const int16_t *filter_y, int y_step_q4, int w,
                                 int h) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

  convolve_add_src_hip(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                       x_step_q4, filters_y, y0_q4, y_step_q4, w, h);
}
796 797
#endif  // CONFIG_LOOP_RESTORATION

Fergus Simpson's avatar
Fergus Simpson committed
798
// TODO(afergs): Make sure this works too
799
#if CONFIG_HIGHBITDEPTH
800 801
static void highbd_convolve_horiz(const uint8_t *src8, ptrdiff_t src_stride,
                                  uint8_t *dst8, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
802 803
                                  const InterpKernel *x_filters, int x0_q4,
                                  int x_step_q4, int w, int h, int bd) {
804 805 806 807 808 809 810 811 812 813
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = x0_q4;
    for (x = 0; x < w; ++x) {
      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
      int k, sum = 0;
clang-format's avatar
clang-format committed
814
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
815
      dst[x] = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
816 817 818 819 820 821 822
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

823 824
static void highbd_convolve_avg_horiz(const uint8_t *src8, ptrdiff_t src_stride,
                                      uint8_t *dst8, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
825 826
                                      const InterpKernel *x_filters, int x0_q4,
                                      int x_step_q4, int w, int h, int bd) {
827 828 829 830 831 832 833 834 835 836
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = x0_q4;
    for (x = 0; x < w; ++x) {
      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
      int k, sum = 0;
clang-format's avatar
clang-format committed
837 838 839 840
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
      dst[x] = ROUND_POWER_OF_TWO(
          dst[x] + clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd),
          1);
841 842 843 844 845 846 847
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

848 849
static void highbd_convolve_vert(const uint8_t *src8, ptrdiff_t src_stride,
                                 uint8_t *dst8, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
850 851
                                 const InterpKernel *y_filters, int y0_q4,
                                 int y_step_q4, int w, int h, int bd) {
852 853 854 855 856 857 858 859 860 861 862 863
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
  for (x = 0; x < w; ++x) {
    int y_q4 = y0_q4;
    for (y = 0; y < h; ++y) {
      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
clang-format's avatar
clang-format committed
864 865
      dst[y * dst_stride] =
          clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
866 867 868 869 870 871 872
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

873 874
static void highbd_convolve_avg_vert(const uint8_t *src8, ptrdiff_t src_stride,
                                     uint8_t *dst8, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
875 876
                                     const InterpKernel *y_filters, int y0_q4,
                                     int y_step_q4, int w, int h, int bd) {
877 878 879 880 881 882 883 884 885 886 887 888
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
  for (x = 0; x < w; ++x) {
    int y_q4 = y0_q4;
    for (y = 0; y < h; ++y) {
      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
clang-format's avatar
clang-format committed
889 890 891 892
      dst[y * dst_stride] = ROUND_POWER_OF_TWO(
          dst[y * dst_stride] +
              clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd),
          1);
893 894 895 896 897 898 899
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

900 901
static void highbd_convolve(const uint8_t *src, ptrdiff_t src_stride,
                            uint8_t *dst, ptrdiff_t dst_stride,
clang-format's avatar
clang-format committed
902 903 904
                            const InterpKernel *const x_filters, int x0_q4,
                            int x_step_q4, const InterpKernel *const y_filters,
                            int y0_q4, int y_step_q4, int w, int h, int bd) {
905 906 907 908 909 910 911 912 913 914 915 916
  // Note: Fixed size intermediate buffer, temp, places limits on parameters.
  // 2d filtering proceeds in 2 steps:
  //   (1) Interpolate horizontally into an intermediate buffer, temp.
  //   (2) Interpolate temp vertically to derive the sub-pixel result.
  // Deriving the maximum number of rows in the temp buffer (135):
  // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
  // --Largest block size is 64x64 pixels.
  // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
  //   original frame (in 1/16th pixel units).
  // --Must round-up because block may be located at sub-pixel position.
  // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
  // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
917
  uint16_t temp[MAX_EXT_SIZE * MAX_SB_SIZE];
918
  int intermediate_height =
clang-format's avatar
clang-format committed
919
      (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;
920

921 922
  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);
923 924 925
  assert(y_step_q4 <= 32);
  assert(x_step_q4 <= 32);

926
  highbd_convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride,
clang-format's avatar
clang-format committed
927 928
                        CONVERT_TO_BYTEPTR(temp), MAX_SB_SIZE, x_filters, x0_q4,
                        x_step_q4, w, intermediate_height, bd);
929
  highbd_convolve_vert(
clang-format's avatar
clang-format committed
930 931
      CONVERT_TO_BYTEPTR(temp) + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1),
      MAX_SB_SIZE, dst, dst_stride, y_filters, y0_q4, y_step_q4, w, h, bd);
932 933
}

Yaowu Xu's avatar
Yaowu Xu committed
934
void aom_highbd_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
935 936
                                  uint8_t *dst, ptrdiff_t dst_stride,
                                  const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
937 938
                                  const int16_t *filter_y, int y_step_q4, int w,
                                  int h, int bd) {
939 940 941 942 943
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);
  (void)filter_y;
  (void)y_step_q4;

clang-format's avatar
clang-format committed
944 945
  highbd_convolve_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                        x_step_q4, w, h, bd);
946 947
}

Yaowu Xu's avatar
Yaowu Xu committed
948
void aom_highbd_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
949 950 951 952
                                      uint8_t *dst, ptrdiff_t dst_stride,
                                      const int16_t *filter_x, int x_step_q4,
                                      const int16_t *filter_y, int y_step_q4,
                                      int w, int h, int bd) {
953 954 955 956 957
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);
  (void)filter_y;
  (void)y_step_q4;

clang-format's avatar
clang-format committed
958 959
  highbd_convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                            x_step_q4, w, h, bd);
960 961
}

Yaowu Xu's avatar
Yaowu Xu committed
962
void aom_highbd_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
963 964
                                 uint8_t *dst, ptrdiff_t dst_stride,
                                 const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
965 966
                                 const int16_t *filter_y, int y_step_q4, int w,
                                 int h, int bd) {
967 968 969 970 971
  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);
  (void)filter_x;
  (void)x_step_q4;

clang-format's avatar
clang-format committed
972 973
  highbd_convolve_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
                       y_step_q4, w, h, bd);
974 975
}

Yaowu Xu's avatar
Yaowu Xu committed
976
void aom_highbd_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
977 978 979 980
                                     uint8_t *dst, ptrdiff_t dst_stride,
                                     const int16_t *filter_x, int x_step_q4,
                                     const int16_t *filter_y, int y_step_q4,
                                     int w, int h, int bd) {
981 982 983 984 985
  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);
  (void)filter_x;
  (void)x_step_q4;

clang-format's avatar
clang-format committed
986 987
  highbd_convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
                           y_step_q4, w, h, bd);
988 989
}

Yaowu Xu's avatar
Yaowu Xu committed
990
void aom_highbd_convolve8_c(const uint8_t *src, ptrdiff_t src_stride,
991 992
                            uint8_t *dst, ptrdiff_t dst_stride,
                            const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
993 994
                            const int16_t *filter_y, int y_step_q4, int w,
                            int h, int bd) {
995 996 997 998 999 1000
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

clang-format's avatar
clang-format committed
1001
  highbd_convolve(src, src_stride, dst, dst_stride, filters_x, x0_q4, x_step_q4,
1002
                  filters_y, y0_q4, y_step_q4, w, h, bd);
1003 1004
}

Yaowu Xu's avatar
Yaowu Xu committed
1005
void aom_highbd_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride,
1006 1007
                                uint8_t *dst, ptrdiff_t dst_stride,
                                const int16_t *filter_x, int x_step_q4,
clang-format's avatar
clang-format committed
1008 1009
                                const int16_t *filter_y, int y_step_q4, int w,
                                int h, int bd) {
1010
  // Fixed size intermediate buffer places limits on parameters.
1011 1012 1013
  DECLARE_ALIGNED(16, uint16_t, temp[MAX_SB_SIZE * MAX_SB_SIZE]);
  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);
1014

Yaowu Xu's avatar
Yaowu Xu committed
1015
  aom_highbd_convolve8_c(src, src_stride, CONVERT_TO_BYTEPTR(temp), MAX_SB_SIZE,
1016
                         filter_x, x_step_q4, filter_y, y_step_q4, w, h, bd);
Yaowu Xu's avatar
Yaowu Xu committed
1017
  aom_highbd_convolve_avg_c(CONVERT_TO_BYTEPTR(temp), MAX_SB_SIZE, dst,
clang-format's avatar
clang-format committed
1018
                            dst_stride, NULL, 0, NULL, 0, w, h, bd);
1019 1020
}

Yaowu Xu's avatar
Yaowu Xu committed
1021
void aom_highbd_convolve_copy_c(const uint8_t *src8, ptrdiff_t src_stride,
1022 1023 1024 1025
                                uint8_t *dst8, ptrdiff_t dst_stride,
                                const int16_t *filter_x, int filter_x_stride,
                                const int16_t *filter_y, int filter_y_stride,
                                int w, int h, int bd) {
1026 1027 1028 1029 1030 1031 1032 1033 1034 1035
  int r;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  (void)filter_x;
  (void)filter_y;
  (void)filter_x_stride;
  (void)filter_y_stride;
  (void)bd;

  for (r = h; r > 0; --r) {
James Zern's avatar
James Zern committed
1036
    memcpy(dst, src, w * sizeof(uint16_t));
1037 1038 1039 1040 1041
    src += src_stride;
    dst += dst_stride;
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
1042
void aom_highbd_convolve_avg_c(const uint8_t *src8, ptrdiff_t src_stride,
1043 1044 1045 1046
                               uint8_t *dst8, ptrdiff_t dst_stride,
                               const int16_t *filter_x, int filter_x_stride,
                               const int16_t *filter_y, int filter_y_stride,
                               int w, int h, int bd) {
1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  (void)filter_x;
  (void)filter_y;
  (void)filter_x_stride;
  (void)filter_y_stride;
  (void)bd;

  for (y = 0; y < h; ++y) {
    for (x = 0; x < w; ++x) {
      dst[x] = ROUND_POWER_OF_TWO(dst[x] + src[x], 1);
    }
    src += src_stride;
    dst += dst_stride;
  }
}
1064 1065 1066 1067 1068 1069 1070 1071

#if CONFIG_LOOP_RESTORATION
static void highbd_convolve_add_src_horiz(const uint8_t *src8,
                                          ptrdiff_t src_stride, uint8_t *dst8,
                                          ptrdiff_t dst_stride,
                                          const InterpKernel *x_filters,
                                          int x0_q4, int x_step_q4, int w,
                                          int h, int bd) {
1072
  int x, y, k;
1073 1074 1075 1076 1077 1078 1079 1080
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = x0_q4;
    for (x = 0; x < w; ++x) {
      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
1081
      int sum = 0;
1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
      dst[x] = clip_pixel_highbd(
          ROUND_POWER_OF_TWO(sum, FILTER_BITS) + src_x[SUBPEL_TAPS / 2 - 1],
          bd);
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

static void highbd_convolve_add_src_vert(const uint8_t *src8,
                                         ptrdiff_t src_stride, uint8_t *dst8,
                                         ptrdiff_t dst_stride,
                                         const InterpKernel *y_filters,
                                         int y0_q4, int y_step_q4, int w, int h,
                                         int bd) {
1099
  int x, y, k;
1100 1101 1102 1103 1104 1105 1106 1107
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
  for (x = 0; x < w; ++x) {
    int y_q4 = y0_q4;
    for (y = 0; y < h; ++y) {
      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
1108
      int sum = 0;
1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      dst[y * dst_stride] =
          clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS) +
                                src_y[(SUBPEL_TAPS / 2 - 1) * src_stride],
                            bd);
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

static void highbd_convolve_add_src(const uint8_t *src, ptrdiff_t src_stride,
                                    uint8_t *dst, ptrdiff_t dst_stride,
                                    const InterpKernel *const x_filters,
                                    int x0_q4, int x_step_q4,
                                    const InterpKernel *const y_filters,
                                    int y0_q4, int y_step_q4, int w, int h,
                                    int bd) {
  // Note: Fixed size intermediate buffer, temp, places limits on parameters.
  // 2d filtering proceeds in 2 steps:
  //   (1) Interpolate horizontally into an intermediate buffer, temp.
  //   (2) Interpolate temp vertically to derive the sub-pixel result.
  // Deriving the maximum number of rows in the temp buffer (135):
  // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
  // --Largest block size is 64x64 pixels.
  // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
  //   original frame (in 1/16th pixel units).
  // --Must round-up because block may be located at sub-pixel position.
  // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
  // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
  uint16_t temp[MAX_EXT_SIZE * MAX_SB_SIZE];
  int intermediate_height =
      (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;

  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);
  assert(y_step_q4 <= 32);
  assert(x_step_q4 <= 32);

  highbd_convolve_add_src_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1),
                                src_stride, CONVERT_TO_BYTEPTR(temp),
                                MAX_SB_SIZE, x_filters, x0_q4, x_step_q4, w,
                                intermediate_height, bd);
  highbd_convolve_add_src_vert(
      CONVERT_TO_BYTEPTR(temp) + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1),
      MAX_SB_SIZE, dst, dst_stride, y_filters, y0_q4, y_step_q4, w, h, bd);
}

void aom_highbd_convolve8_add_src_horiz_c(
    const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
    ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4,
    const int16_t *filter_y, int y_step_q4, int w, int h, int bd) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);
  (void)filter_y;
  (void)y_step_q4;

  highbd_convolve_add_src_horiz(src, src_stride, dst, dst_stride, filters_x,
                                x0_q4, x_step_q4, w, h, bd);
}

void aom_highbd_convolve8_add_src_vert_c(const uint8_t *src,
                                         ptrdiff_t src_stride, uint8_t *dst,
                                         ptrdiff_t dst_stride,
                                         const int16_t *filter_x, int x_step_q4,
                                         const int16_t *filter_y, int y_step_q4,
                                         int w, int h, int bd) {
  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);
  (void)filter_x;
  (void)x_step_q4;

  highbd_convolve_add_src_vert(src, src_stride, dst, dst_stride, filters_y,
                               y0_q4, y_step_q4, w, h, bd);
}

void aom_highbd_convolve8_add_src_c(const uint8_t *src, ptrdiff_t src_stride,
                                    uint8_t *dst, ptrdiff_t dst_stride,
                                    const int16_t *filter_x, int x_step_q4,
                                    const int16_t *filter_y, int y_step_q4,
                                    int w, int h, int bd) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

  highbd_convolve_add_src(src, src_stride, dst, dst_stride, filters_x, x0_q4,
                          x_step_q4, filters_y, y0_q4, y_step_q4, w, h, bd);
}
1201 1202 1203 1204 1205

static void highbd_convolve_add_src_horiz_hip(
    const uint8_t *src8, ptrdiff_t src_stride, uint16_t *dst,
    ptrdiff_t dst_stride, const InterpKernel *x_filters, int x0_q4,
    int x_step_q4, int w, int h, int bd) {
1206 1207
  const int extraprec_clamp_limit = EXTRAPREC_CLAMP_LIMIT(bd);
  int x, y, k;
1208 1209 1210 1211 1212 1213 1214
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = x0_q4;
    for (x = 0; x < w; ++x) {
      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
1215 1216
      int sum = ((int)src_x[SUBPEL_TAPS / 2 - 1] << FILTER_BITS) +
                (1 << (bd + FILTER_BITS - 1));
1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231
      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
      dst[x] =
          (uint16_t)clamp(ROUND_POWER_OF_TWO(sum, FILTER_BITS - EXTRAPREC_BITS),
                          0, extraprec_clamp_limit - 1);
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

static void highbd_convolve_add_src_vert_hip(
    const uint16_t *src, ptrdiff_t src_stride, uint8_t *dst8,
    ptrdiff_t dst_stride, const InterpKernel *y_filters, int y0_q4,
    int y_step_q4, int w, int h, int bd) {
1232
  int x, y, k;
1233 1234 1235 1236 1237 1238 1239
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
  for (x = 0; x < w; ++x) {
    int y_q4 = y0_q4;
    for (y = 0; y < h; ++y) {
      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
1240 1241 1242
      int sum =
          ((int)src_y[(SUBPEL_TAPS / 2 - 1) * src_stride] << FILTER_BITS) -
          (1 << (bd + FILTER_BITS + EXTRAPREC_BITS - 1));
1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      dst[y * dst_stride] = clip_pixel_highbd(
          ROUND_POWER_OF_TWO(sum, FILTER_BITS + EXTRAPREC_BITS), bd);
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

static void highbd_convolve_add_src_hip(
    const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
    ptrdiff_t dst_stride, const InterpKernel *const x_filters, int x0_q4,
    int x_step_q4, const InterpKernel *const y_filters, int y0_q4,
    int y_step_q4, int w, int h, int bd) {
  // Note: Fixed size intermediate buffer, temp, places limits on parameters.
  // 2d filtering proceeds in 2 steps:
  //   (1) Interpolate horizontally into an intermediate buffer, temp.
  //   (2) Interpolate temp vertically to derive the sub-pixel result.
  // Deriving the maximum number of rows in the temp buffer (135):
  // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
  // --Largest block size is 64x64 pixels.
  // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
  //   original frame (in 1/16th pixel units).
  // --Must round-up because block may be located at sub-pixel position.
  // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
  // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
  uint16_t temp[MAX_EXT_SIZE * MAX_SB_SIZE];
  int intermediate_height =
      (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;

  assert(w <= MAX_SB_SIZE);
  assert(h <= MAX_SB_SIZE);
  assert(y_step_q4 <= 32);
  assert(x_step_q4 <= 32);

  highbd_convolve_add_src_horiz_hip(
      src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride, temp, MAX_SB_SIZE,
      x_filters, x0_q4, x_step_q4, w, intermediate_height, bd);
  highbd_convolve_add_src_vert_hip(temp + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1),
                                   MAX_SB_SIZE, dst, dst_stride, y_filters,
                                   y0_q4, y_step_q4, w, h, bd);
}

void aom_highbd_convolve8_add_src_horiz_hip_c(
    const uint8_t *src, ptrdiff_t src_stride, uint16_t *dst,
    ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4,
    const int16_t *filter_y, int y_step_q4, int w, int h, int bd) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);
  (void)filter_y;
  (void)y_step_q4;

  highbd_convolve_add_src_horiz_hip(src, src_stride, dst, dst_stride, filters_x,
                                    x0_q4, x_step_q4, w, h, bd);
}

void aom_highbd_convolve8_add_src_vert_hip_c(
    const uint16_t *src, ptrdiff_t src_stride, uint8_t *dst,
    ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4,
    const int16_t *filter_y, int y_step_q4, int w, int h, int bd) {
  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);
  (void)filter_x;
  (void)x_step_q4;

  highbd_convolve_add_src_vert_hip(src, src_stride, dst, dst_stride, filters_y,
                                   y0_q4, y_step_q4, w, h, bd);
}

void aom_highbd_convolve8_add_src_hip_c(const uint8_t *src,
                                        ptrdiff_t src_stride, uint8_t *dst,
                                        ptrdiff_t dst_stride,
                                        const int16_t *filter_x, int x_step_q4,
                                        const int16_t *filter_y, int y_step_q4,
                                        int w, int h, int bd) {
  const InterpKernel *const filters_x = get_filter_base(filter_x);
  const int x0_q4 = get_filter_offset(filter_x, filters_x);

  const InterpKernel *const filters_y = get_filter_base(filter_y);
  const int y0_q4 = get_filter_offset(filter_y, filters_y);

  highbd_convolve_add_src_hip(src, src_stride, dst, dst_stride, filters_x,
                              x0_q4, x_step_q4, filters_y, y0_q4, y_step_q4, w,
                              h, bd);
}

1331
#endif  // CONFIG_LOOP_RESTORATION
1332
#endif  // CONFIG_HIGHBITDEPTH