convolve.c 27.2 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11
/*
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
 *
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 */

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

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

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

Yaowu Xu's avatar
Yaowu Xu committed
27 28 29
void av1_convolve_horiz_c(const uint8_t *src, int src_stride, uint8_t *dst,
                          int dst_stride, int w, int h,
                          const InterpFilterParams filter_params,
30 31
                          const int subpel_x_q4, int x_step_q4,
                          ConvolveParams *conv_params) {
32
  int x, y;
33
  int filter_size = filter_params.taps;
34 35 36 37 38
  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
39
      const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
40
          filter_params, x_q4 & SUBPEL_MASK);
41 42
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k) sum += src_x[k] * x_filter[k];
43

44
      if (conv_params->round == CONVOLVE_OPT_ROUND) {
45
        sum = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
46 47 48 49 50 51 52 53 54 55
        if (conv_params->ref)
          dst[x] = ROUND_POWER_OF_TWO(dst[x] + sum, 1);
        else
          dst[x] = sum;
      } else {
        int tmp = conv_params->dst[y * conv_params->dst_stride + x];
        if (conv_params->ref)
          tmp = ROUND_POWER_OF_TWO(tmp + sum, 1);
        else
          tmp = sum;
Angie Chiang's avatar
Angie Chiang committed
56 57
        conv_params->dst[y * conv_params->dst_stride + x] =
            clamp(tmp, INT32_MIN, INT32_MAX);
58
      }
59

60 61 62 63 64 65 66
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
67 68 69
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,
70 71
                         const int subpel_y_q4, int y_step_q4,
                         ConvolveParams *conv_params) {
72
  int x, y;
73
  int filter_size = filter_params.taps;
74 75 76 77 78 79
  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
80
      const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
81
          filter_params, y_q4 & SUBPEL_MASK);
82 83 84
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
85

86
      if (conv_params->round == CONVOLVE_OPT_ROUND) {
87
        sum = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
88 89 90 91 92 93 94 95 96 97 98
        if (conv_params->ref)
          dst[y * dst_stride] =
              ROUND_POWER_OF_TWO(dst[y * dst_stride] + sum, 1);
        else
          dst[y * dst_stride] = sum;
      } else {
        int tmp = conv_params->dst[y * conv_params->dst_stride + x];
        if (conv_params->ref)
          tmp = ROUND_POWER_OF_TWO(tmp + sum, 1);
        else
          tmp = sum;
Angie Chiang's avatar
Angie Chiang committed
99 100
        conv_params->dst[y * conv_params->dst_stride + x] =
            clamp(tmp, INT32_MIN, INT32_MAX);
101
      }
102

103 104 105 106 107 108 109
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

Angie Chiang's avatar
Angie Chiang committed
110
static void convolve_copy(const uint8_t *src, int src_stride, uint8_t *dst,
111 112 113 114
                          int dst_stride, int w, int h,
                          ConvolveParams *conv_params) {
  if (conv_params->ref == 0) {
    int r, c;
Angie Chiang's avatar
Angie Chiang committed
115
    for (r = 0; r < h; ++r) {
116 117 118 119 120 121 122
      if (conv_params->round == CONVOLVE_OPT_ROUND) {
        memcpy(dst, src, w);
      } else {
        for (c = 0; c < w; ++c)
          conv_params->dst[r * conv_params->dst_stride + c] = ((uint16_t)src[c])
                                                              << FILTER_BITS;
      }
Angie Chiang's avatar
Angie Chiang committed
123 124 125 126 127 128
      src += src_stride;
      dst += dst_stride;
    }
  } else {
    int r, c;
    for (r = 0; r < h; ++r) {
129 130
      for (c = 0; c < w; ++c) {
        if (conv_params->round == CONVOLVE_OPT_ROUND) {
131
          dst[c] = clip_pixel(ROUND_POWER_OF_TWO(dst[c] + src[c], 1));
132 133
        } else {
          int tmp = conv_params->dst[r * conv_params->dst_stride + c];
Angie Chiang's avatar
Angie Chiang committed
134 135 136
          tmp = ROUND_POWER_OF_TWO(tmp + (((int32_t)src[c]) << FILTER_BITS), 1);
          conv_params->dst[r * conv_params->dst_stride + c] =
              clamp(tmp, INT32_MIN, INT32_MAX);
137 138
        }
      }
Angie Chiang's avatar
Angie Chiang committed
139 140 141 142 143 144
      src += src_stride;
      dst += dst_stride;
    }
  }
}

145 146 147
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,
148 149
                               const int subpel_x_q4, int x_step_q4,
                               ConvolveParams *conv_params) {
150 151
  if (filter_params.taps == SUBPEL_TAPS &&
      conv_params->round == CONVOLVE_OPT_ROUND) {
152 153
    const int16_t *filter_x =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_x_q4);
154
    if (conv_params->ref == 0)
155 156 157 158 159 160
      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 {
161 162 163 164 165 166 167 168
    if (conv_params->round == CONVOLVE_OPT_ROUND) {
      av1_convolve_horiz(src, src_stride, dst, dst_stride, w, h, filter_params,
                         subpel_x_q4, x_step_q4, conv_params);
    } else {
      // TODO(angiebird) need SIMD implementation here
      av1_convolve_horiz_c(src, src_stride, dst, dst_stride, w, h,
                           filter_params, subpel_x_q4, x_step_q4, conv_params);
    }
169 170 171 172 173 174
  }
}

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,
175 176
                              const int subpel_y_q4, int y_step_q4,
                              ConvolveParams *conv_params) {
177 178
  if (filter_params.taps == SUBPEL_TAPS &&
      conv_params->round == CONVOLVE_OPT_ROUND) {
179 180
    const int16_t *filter_y =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_y_q4);
181
    if (conv_params->ref == 0) {
182 183 184 185 186 187 188
      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 {
189 190 191 192 193 194 195 196
    if (conv_params->round == CONVOLVE_OPT_ROUND) {
      av1_convolve_vert(src, src_stride, dst, dst_stride, w, h, filter_params,
                        subpel_y_q4, y_step_q4, conv_params);
    } else {
      // TODO(angiebird) need SIMD implementation here
      av1_convolve_vert_c(src, src_stride, dst, dst_stride, w, h, filter_params,
                          subpel_y_q4, y_step_q4, conv_params);
    }
197 198 199
  }
}

Angie Chiang's avatar
Angie Chiang committed
200
#if CONFIG_CONVOLVE_ROUND
Angie Chiang's avatar
Angie Chiang committed
201
void av1_convolve_rounding(const int32_t *src, int src_stride, uint8_t *dst,
Angie Chiang's avatar
Angie Chiang committed
202
                           int dst_stride, int w, int h, int bits) {
Angie Chiang's avatar
Angie Chiang committed
203 204 205 206
  int r, c;
  for (r = 0; r < h; ++r) {
    for (c = 0; c < w; ++c) {
      dst[r * dst_stride + c] =
Angie Chiang's avatar
Angie Chiang committed
207
          clip_pixel(ROUND_POWER_OF_TWO_SIGNED(src[r * src_stride + c], bits));
Angie Chiang's avatar
Angie Chiang committed
208 209 210
    }
  }
}
Angie Chiang's avatar
Angie Chiang committed
211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254

void av1_convolve_2d(const uint8_t *src, int src_stride, CONV_BUF_TYPE *dst,
                     int dst_stride, int w, int h,
                     InterpFilterParams *filter_params_x,
                     InterpFilterParams *filter_params_y, const int subpel_x_q4,
                     const int subpel_y_q4, ConvolveParams *conv_params) {
  int x, y, k;
  CONV_BUF_TYPE im_block[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) * MAX_SB_SIZE];
  int im_h = h + filter_params_y->taps - 1;
  int im_stride = w;
  const int fo_vert = filter_params_y->taps / 2 - 1;
  const int fo_horiz = filter_params_x->taps / 2 - 1;
  (void)conv_params;
  // horizontal filter
  const uint8_t *src_horiz = src - fo_vert * src_stride;
  const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
      *filter_params_x, subpel_x_q4 & SUBPEL_MASK);
  for (y = 0; y < im_h; ++y) {
    for (x = 0; x < w; ++x) {
      CONV_BUF_TYPE sum = 0;
      for (k = 0; k < filter_params_x->taps; ++k) {
        sum += x_filter[k] * src_horiz[y * src_stride + x - fo_horiz + k];
      }
      im_block[y * im_stride + x] =
          ROUND_POWER_OF_TWO_SIGNED(sum, conv_params->round_0);
    }
  }

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

255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272
static INLINE void transpose_uint8(uint8_t *dst, int dst_stride,
                                   const uint8_t *src, int src_stride, int w,
                                   int h) {
  int r, c;
  for (r = 0; r < h; ++r)
    for (c = 0; c < w; ++c)
      dst[c * (dst_stride) + r] = src[r * (src_stride) + c];
}

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

Angie Chiang's avatar
Angie Chiang committed
273 274 275 276 277 278 279 280 281 282 283 284 285 286
void av1_convolve_2d_facade(const uint8_t *src, int src_stride, uint8_t *dst,
                            int dst_stride, int w, int h,
                            const InterpFilter *interp_filter,
                            const int subpel_x_q4, int x_step_q4,
                            const int subpel_y_q4, int y_step_q4,
                            ConvolveParams *conv_params) {
  (void)x_step_q4;
  (void)y_step_q4;
  (void)dst;
  (void)dst_stride;
  InterpFilterParams filter_params_x =
      av1_get_interp_filter_params(interp_filter[1 + 2 * conv_params->ref]);
  InterpFilterParams filter_params_y =
      av1_get_interp_filter_params(interp_filter[0 + 2 * conv_params->ref]);
287 288 289 290 291 292
  if (filter_params_x.interp_filter == MULTITAP_SHARP &&
      filter_params_y.interp_filter == MULTITAP_SHARP) {
    // Avoid two directions both using 12-tap filter.
    // This will reduce hardware implementation cost.
    filter_params_y = av1_get_interp_filter_params(EIGHTTAP_SHARP);
  }
293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319

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

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

    // horizontal and vertical parameters are swapped because of the transpose
    av1_convolve_2d(tr_src + fo_horiz * tr_src_stride + fo_vert, tr_src_stride,
                    tr_dst, tr_dst_stride, h, w, &filter_params_y,
                    &filter_params_x, subpel_y_q4, subpel_x_q4, conv_params);
    transpose_int32(conv_params->dst, conv_params->dst_stride, tr_dst,
                    tr_dst_stride, h, w);
  } else {
    av1_convolve_2d(src, src_stride, conv_params->dst, conv_params->dst_stride,
                    w, h, &filter_params_x, &filter_params_y, subpel_x_q4,
                    subpel_y_q4, conv_params);
  }
Angie Chiang's avatar
Angie Chiang committed
320 321
}

Angie Chiang's avatar
Angie Chiang committed
322
#endif  // CONFIG_CONVOLVE_ROUND
Angie Chiang's avatar
Angie Chiang committed
323

Yaowu Xu's avatar
Yaowu Xu committed
324 325
void av1_convolve(const uint8_t *src, int src_stride, uint8_t *dst,
                  int dst_stride, int w, int h,
326
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
327
                  const InterpFilter *interp_filter,
328
#else
James Zern's avatar
James Zern committed
329
                  const InterpFilter interp_filter,
330
#endif
Yaowu Xu's avatar
Yaowu Xu committed
331
                  const int subpel_x_q4, int x_step_q4, const int subpel_y_q4,
332
                  int y_step_q4, ConvolveParams *conv_params) {
Angie Chiang's avatar
Angie Chiang committed
333 334
  int ignore_horiz = x_step_q4 == 16 && subpel_x_q4 == 0;
  int ignore_vert = y_step_q4 == 16 && subpel_y_q4 == 0;
Angie Chiang's avatar
Angie Chiang committed
335 336 337 338 339 340 341 342 343 344
#if CONFIG_DUAL_FILTER
  InterpFilterParams filter_params_x =
      av1_get_interp_filter_params(interp_filter[1 + 2 * conv_params->ref]);
  InterpFilterParams filter_params_y =
      av1_get_interp_filter_params(interp_filter[0 + 2 * conv_params->ref]);
  InterpFilterParams filter_params;
#else
  InterpFilterParams filter_params =
      av1_get_interp_filter_params(interp_filter);
#endif
345 346 347 348 349 350

  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
351
  if (ignore_horiz && ignore_vert) {
352
    convolve_copy(src, src_stride, dst, dst_stride, w, h, conv_params);
Angie Chiang's avatar
Angie Chiang committed
353
  } else if (ignore_vert) {
354
#if CONFIG_DUAL_FILTER
Angie Chiang's avatar
Angie Chiang committed
355
    filter_params = filter_params_x;
356 357
#endif
    assert(filter_params.taps <= MAX_FILTER_TAP);
358
    av1_convolve_horiz_facade(src, src_stride, dst, dst_stride, w, h,
359 360
                              filter_params, subpel_x_q4, x_step_q4,
                              conv_params);
Angie Chiang's avatar
Angie Chiang committed
361
  } else if (ignore_horiz) {
362
#if CONFIG_DUAL_FILTER
Angie Chiang's avatar
Angie Chiang committed
363
    filter_params = filter_params_y;
364 365
#endif
    assert(filter_params.taps <= MAX_FILTER_TAP);
366
    av1_convolve_vert_facade(src, src_stride, dst, dst_stride, w, h,
367 368
                             filter_params, subpel_y_q4, y_step_q4,
                             conv_params);
Angie Chiang's avatar
Angie Chiang committed
369
  } else {
370 371 372 373 374 375
    // 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);
376
    int filter_size;
377
#if CONFIG_DUAL_FILTER
378 379
    if (interp_filter[0 + 2 * conv_params->ref] == MULTITAP_SHARP &&
        interp_filter[1 + 2 * conv_params->ref] == MULTITAP_SHARP) {
380 381 382 383
      // Avoid two directions both using 12-tap filter.
      // This will reduce hardware implementation cost.
      filter_params_y = av1_get_interp_filter_params(EIGHTTAP_SHARP);
    }
384

385 386 387 388
    // we do filter with fewer taps first to reduce hardware implementation
    // complexity
    if (filter_params_y.taps < filter_params_x.taps) {
      int intermediate_width;
389
      int temp_stride = max_intermediate_size;
390 391
      ConvolveParams temp_conv_params;
      temp_conv_params.ref = 0;
392
      temp_conv_params.round = CONVOLVE_OPT_ROUND;
393 394 395 396
      filter_params = filter_params_y;
      filter_size = filter_params_x.taps;
      intermediate_width =
          (((w - 1) * x_step_q4 + subpel_x_q4) >> SUBPEL_BITS) + filter_size;
397
      assert(intermediate_width <= max_intermediate_size);
Angie Chiang's avatar
Angie Chiang committed
398

399
      assert(filter_params.taps <= MAX_FILTER_TAP);
400

401 402
      av1_convolve_vert_facade(src - (filter_size / 2 - 1), src_stride, temp,
                               temp_stride, intermediate_width, h,
403 404
                               filter_params, subpel_y_q4, y_step_q4,
                               &temp_conv_params);
405

406 407
      filter_params = filter_params_x;
      assert(filter_params.taps <= MAX_FILTER_TAP);
408 409
      av1_convolve_horiz_facade(temp + (filter_size / 2 - 1), temp_stride, dst,
                                dst_stride, w, h, filter_params, subpel_x_q4,
410
                                x_step_q4, conv_params);
411
    } else
412
#endif  // CONFIG_DUAL_FILTER
413 414
    {
      int intermediate_height;
415
      int temp_stride = MAX_SB_SIZE;
416 417
      ConvolveParams temp_conv_params;
      temp_conv_params.ref = 0;
418
      temp_conv_params.round = CONVOLVE_OPT_ROUND;
419 420 421 422 423 424 425 426
#if CONFIG_DUAL_FILTER
      filter_params = filter_params_x;
      filter_size = filter_params_y.taps;
#else
      filter_size = filter_params.taps;
#endif
      intermediate_height =
          (((h - 1) * y_step_q4 + subpel_y_q4) >> SUBPEL_BITS) + filter_size;
427 428
      assert(intermediate_height <= max_intermediate_size);
      (void)max_intermediate_size;
429 430 431

      assert(filter_params.taps <= MAX_FILTER_TAP);

432 433 434
      av1_convolve_horiz_facade(src - src_stride * (filter_size / 2 - 1),
                                src_stride, temp, temp_stride, w,
                                intermediate_height, filter_params, subpel_x_q4,
435
                                x_step_q4, &temp_conv_params);
436 437 438 439 440 441

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

442 443 444
      av1_convolve_vert_facade(
          temp + temp_stride * (filter_size / 2 - 1), temp_stride, dst,
          dst_stride, w, h, filter_params, subpel_y_q4, y_step_q4, conv_params);
445
    }
Angie Chiang's avatar
Angie Chiang committed
446
  }
447 448
}

Angie Chiang's avatar
Angie Chiang committed
449 450 451 452 453
void av1_convolve_init_c(void) {
  // A placeholder for SIMD initialization
  return;
}

Yaowu Xu's avatar
Yaowu Xu committed
454 455 456 457 458 459
#if CONFIG_AOM_HIGHBITDEPTH
void av1_highbd_convolve_horiz_c(const uint16_t *src, int src_stride,
                                 uint16_t *dst, int dst_stride, int w, int h,
                                 const InterpFilterParams filter_params,
                                 const int subpel_x_q4, int x_step_q4, int avg,
                                 int bd) {
460
  int x, y;
461
  int filter_size = filter_params.taps;
462 463 464 465 466
  src -= filter_size / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = subpel_x_q4;
    for (x = 0; x < w; ++x) {
      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
Yaowu Xu's avatar
Yaowu Xu committed
467
      const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
468
          filter_params, x_q4 & SUBPEL_MASK);
469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k) sum += src_x[k] * x_filter[k];
      if (avg)
        dst[x] = ROUND_POWER_OF_TWO(
            dst[x] +
                clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd),
            1);
      else
        dst[x] = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
485 486 487 488 489
void av1_highbd_convolve_vert_c(const uint16_t *src, int src_stride,
                                uint16_t *dst, int dst_stride, int w, int h,
                                const InterpFilterParams filter_params,
                                const int subpel_y_q4, int y_step_q4, int avg,
                                int bd) {
490
  int x, y;
491
  int filter_size = filter_params.taps;
492 493 494 495 496 497
  src -= src_stride * (filter_size / 2 - 1);

  for (x = 0; x < w; ++x) {
    int y_q4 = subpel_y_q4;
    for (y = 0; y < h; ++y) {
      const uint16_t *const src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
Yaowu Xu's avatar
Yaowu Xu committed
498
      const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
499
          filter_params, y_q4 & SUBPEL_MASK);
500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      if (avg) {
        dst[y * dst_stride] = ROUND_POWER_OF_TWO(
            dst[y * dst_stride] +
                clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd),
            1);
      } else {
        dst[y * dst_stride] =
            clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
      }
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

Angie Chiang's avatar
Angie Chiang committed
519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540
static void highbd_convolve_copy(const uint16_t *src, int src_stride,
                                 uint16_t *dst, int dst_stride, int w, int h,
                                 int avg, int bd) {
  if (avg == 0) {
    int r;
    for (r = 0; r < h; ++r) {
      memcpy(dst, src, w * sizeof(*src));
      src += src_stride;
      dst += dst_stride;
    }
  } else {
    int r, c;
    for (r = 0; r < h; ++r) {
      for (c = 0; c < w; ++c) {
        dst[c] = clip_pixel_highbd(ROUND_POWER_OF_TWO(dst[c] + src[c], 1), bd);
      }
      src += src_stride;
      dst += dst_stride;
    }
  }
}

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

void av1_highbd_convolve_vert_facade(const uint8_t *src8, int src_stride,
                                     uint8_t *dst8, int dst_stride, int w,
                                     int h,
                                     const InterpFilterParams filter_params,
                                     const int subpel_y_q4, int y_step_q4,
                                     int avg, int bd) {
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
572

573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588
  if (filter_params.taps == SUBPEL_TAPS) {
    const int16_t *filter_y =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_y_q4);
    if (avg == 0) {
      aom_highbd_convolve8_vert(src8, src_stride, dst8, dst_stride, NULL, -1,
                                filter_y, y_step_q4, w, h, bd);
    } else {
      aom_highbd_convolve8_avg_vert(src8, src_stride, dst8, dst_stride, NULL,
                                    -1, filter_y, y_step_q4, w, h, bd);
    }
  } else {
    av1_highbd_convolve_vert(src, src_stride, dst, dst_stride, w, h,
                             filter_params, subpel_y_q4, y_step_q4, avg, bd);
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
589 590
void av1_highbd_convolve(const uint8_t *src8, int src_stride, uint8_t *dst8,
                         int dst_stride, int w, int h,
591
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
592
                         const InterpFilter *interp_filter,
593
#else
James Zern's avatar
James Zern committed
594
                         const InterpFilter interp_filter,
595
#endif
Yaowu Xu's avatar
Yaowu Xu committed
596 597 598
                         const int subpel_x_q4, int x_step_q4,
                         const int subpel_y_q4, int y_step_q4, int ref_idx,
                         int bd) {
Angie Chiang's avatar
Angie Chiang committed
599 600 601 602
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  int ignore_horiz = x_step_q4 == 16 && subpel_x_q4 == 0;
  int ignore_vert = y_step_q4 == 16 && subpel_y_q4 == 0;
603 604 605 606 607 608

  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
609
  if (ignore_horiz && ignore_vert) {
610
    highbd_convolve_copy(src, src_stride, dst, dst_stride, w, h, ref_idx, bd);
Angie Chiang's avatar
Angie Chiang committed
611
  } else if (ignore_vert) {
612 613
#if CONFIG_DUAL_FILTER
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
614
        av1_get_interp_filter_params(interp_filter[1 + 2 * ref_idx]);
615 616
#else
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
617
        av1_get_interp_filter_params(interp_filter);
618
#endif
619 620 621
    av1_highbd_convolve_horiz_facade(src8, src_stride, dst8, dst_stride, w, h,
                                     filter_params, subpel_x_q4, x_step_q4,
                                     ref_idx, bd);
Angie Chiang's avatar
Angie Chiang committed
622
  } else if (ignore_horiz) {
623 624
#if CONFIG_DUAL_FILTER
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
625
        av1_get_interp_filter_params(interp_filter[0 + 2 * ref_idx]);
626 627
#else
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
628
        av1_get_interp_filter_params(interp_filter);
629
#endif
630 631 632
    av1_highbd_convolve_vert_facade(src8, src_stride, dst8, dst_stride, w, h,
                                    filter_params, subpel_y_q4, y_step_q4,
                                    ref_idx, bd);
Angie Chiang's avatar
Angie Chiang committed
633
  } else {
634 635 636 637 638
    // temp's size is set to a 256 aligned value to facilitate SIMD
    // implementation. The value is greater than (maximum possible intermediate
    // height or width) * MAX_SB_SIZE
    DECLARE_ALIGNED(16, uint16_t,
                    temp[((MAX_SB_SIZE * 2 + 16) + 16) * MAX_SB_SIZE]);
Yaowu Xu's avatar
Yaowu Xu committed
639
    uint8_t *temp8 = CONVERT_TO_BYTEPTR(temp);
640 641 642
    int max_intermediate_size = ((MAX_SB_SIZE * 2 + 16) + 16);
    int filter_size;
    InterpFilterParams filter_params;
643
#if CONFIG_DUAL_FILTER
Yaowu Xu's avatar
Yaowu Xu committed
644 645 646 647
    InterpFilterParams filter_params_x =
        av1_get_interp_filter_params(interp_filter[1 + 2 * ref_idx]);
    InterpFilterParams filter_params_y =
        av1_get_interp_filter_params(interp_filter[0 + 2 * ref_idx]);
648 649 650 651 652 653
    if (interp_filter[0 + 2 * ref_idx] == MULTITAP_SHARP &&
        interp_filter[1 + 2 * ref_idx] == MULTITAP_SHARP) {
      // Avoid two directions both using 12-tap filter.
      // This will reduce hardware implementation cost.
      filter_params_y = av1_get_interp_filter_params(EIGHTTAP_SHARP);
    }
654
#endif
655

656 657 658 659 660 661 662 663 664 665 666
#if CONFIG_DUAL_FILTER
    if (filter_params_y.taps < filter_params_x.taps) {
      int intermediate_width;
      int temp_stride = max_intermediate_size;
      filter_params = filter_params_y;
      filter_size = filter_params_x.taps;
      intermediate_width =
          (((w - 1) * x_step_q4 + subpel_x_q4) >> SUBPEL_BITS) + filter_size;
      assert(intermediate_width <= max_intermediate_size);

      assert(filter_params.taps <= MAX_FILTER_TAP);
Angie Chiang's avatar
Angie Chiang committed
667

668 669 670
      av1_highbd_convolve_vert_facade(
          src8 - (filter_size / 2 - 1), src_stride, temp8, temp_stride,
          intermediate_width, h, filter_params, subpel_y_q4, y_step_q4, 0, bd);
671

672 673 674 675 676 677 678 679 680 681 682
      filter_params = filter_params_x;
      assert(filter_params.taps <= MAX_FILTER_TAP);

      av1_highbd_convolve_horiz_facade(
          temp8 + (filter_size / 2 - 1), temp_stride, dst8, dst_stride, w, h,
          filter_params, subpel_x_q4, x_step_q4, ref_idx, bd);
    } else
#endif  // CONFIG_DUAL_FILTER
    {
      int intermediate_height;
      int temp_stride = MAX_SB_SIZE;
683
#if CONFIG_DUAL_FILTER
684 685 686 687 688
      filter_params = filter_params_x;
      filter_size = filter_params_y.taps;
#else
      filter_params = av1_get_interp_filter_params(interp_filter);
      filter_size = filter_params.taps;
689
#endif
690 691 692 693 694 695 696 697 698 699 700 701 702 703 704
      intermediate_height =
          (((h - 1) * y_step_q4 + subpel_y_q4) >> SUBPEL_BITS) + filter_size;
      assert(intermediate_height <= max_intermediate_size);
      (void)max_intermediate_size;

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

#if CONFIG_DUAL_FILTER
      filter_params = filter_params_y;
#endif
      filter_size = filter_params.taps;
      assert(filter_params.taps <= MAX_FILTER_TAP);
705

706 707 708 709
      av1_highbd_convolve_vert_facade(
          temp8 + temp_stride * (filter_size / 2 - 1), temp_stride, dst8,
          dst_stride, w, h, filter_params, subpel_y_q4, y_step_q4, ref_idx, bd);
    }
Angie Chiang's avatar
Angie Chiang committed
710
  }
711
}
Yaowu Xu's avatar
Yaowu Xu committed
712
#endif  // CONFIG_AOM_HIGHBITDEPTH