convolve.c 25.5 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
#define MAX_STEP (32)

Yaowu Xu's avatar
Yaowu Xu committed
26 27 28
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,
29 30
                          const int subpel_x_q4, int x_step_q4,
                          ConvolveParams *conv_params) {
31
  int x, y;
32
  int filter_size = filter_params.taps;
Angie Chiang's avatar
Angie Chiang committed
33
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
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

Angie Chiang's avatar
Angie Chiang committed
44 45 46 47 48
      sum = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
      if (conv_params->ref)
        dst[x] = ROUND_POWER_OF_TWO(dst[x] + sum, 1);
      else
        dst[x] = sum;
49

50 51 52 53 54 55 56
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
57 58 59
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,
60 61
                         const int subpel_y_q4, int y_step_q4,
                         ConvolveParams *conv_params) {
62
  int x, y;
63
  int filter_size = filter_params.taps;
Angie Chiang's avatar
Angie Chiang committed
64
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
65 66 67 68 69
  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
70
      const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
71
          filter_params, y_q4 & SUBPEL_MASK);
72 73 74
      int k, sum = 0;
      for (k = 0; k < filter_size; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
75

Angie Chiang's avatar
Angie Chiang committed
76 77 78 79 80
      sum = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
      if (conv_params->ref)
        dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] + sum, 1);
      else
        dst[y * dst_stride] = sum;
81

82 83 84 85 86 87 88
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

Angie Chiang's avatar
Angie Chiang committed
89
static void convolve_copy(const uint8_t *src, int src_stride, uint8_t *dst,
90 91
                          int dst_stride, int w, int h,
                          ConvolveParams *conv_params) {
Angie Chiang's avatar
Angie Chiang committed
92
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
93
  if (conv_params->ref == 0) {
Angie Chiang's avatar
Angie Chiang committed
94
    int r;
Angie Chiang's avatar
Angie Chiang committed
95
    for (r = 0; r < h; ++r) {
Angie Chiang's avatar
Angie Chiang committed
96
      memcpy(dst, src, w);
Angie Chiang's avatar
Angie Chiang committed
97 98 99 100 101 102
      src += src_stride;
      dst += dst_stride;
    }
  } else {
    int r, c;
    for (r = 0; r < h; ++r) {
103
      for (c = 0; c < w; ++c) {
Angie Chiang's avatar
Angie Chiang committed
104
        dst[c] = clip_pixel(ROUND_POWER_OF_TWO(dst[c] + src[c], 1));
105
      }
Angie Chiang's avatar
Angie Chiang committed
106 107 108 109 110 111
      src += src_stride;
      dst += dst_stride;
    }
  }
}

112 113 114
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,
115 116
                               const int subpel_x_q4, int x_step_q4,
                               ConvolveParams *conv_params) {
Angie Chiang's avatar
Angie Chiang committed
117 118
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
  if (filter_params.taps == SUBPEL_TAPS) {
119 120
    const int16_t *filter_x =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_x_q4);
121
    if (conv_params->ref == 0)
122 123 124 125 126 127
      aom_convolve8_horiz(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
                          NULL, -1, w, h);
    else
      aom_convolve8_avg_horiz(src, src_stride, dst, dst_stride, filter_x,
                              x_step_q4, NULL, -1, w, h);
  } else {
Angie Chiang's avatar
Angie Chiang committed
128 129
    av1_convolve_horiz(src, src_stride, dst, dst_stride, w, h, filter_params,
                       subpel_x_q4, x_step_q4, conv_params);
130 131 132 133 134 135
  }
}

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,
136 137
                              const int subpel_y_q4, int y_step_q4,
                              ConvolveParams *conv_params) {
Angie Chiang's avatar
Angie Chiang committed
138 139
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
  if (filter_params.taps == SUBPEL_TAPS) {
140 141
    const int16_t *filter_y =
        av1_get_interp_filter_subpel_kernel(filter_params, subpel_y_q4);
142
    if (conv_params->ref == 0) {
143 144 145 146 147 148 149
      aom_convolve8_vert(src, src_stride, dst, dst_stride, NULL, -1, filter_y,
                         y_step_q4, w, h);
    } else {
      aom_convolve8_avg_vert(src, src_stride, dst, dst_stride, NULL, -1,
                             filter_y, y_step_q4, w, h);
    }
  } else {
Angie Chiang's avatar
Angie Chiang committed
150 151
    av1_convolve_vert(src, src_stride, dst, dst_stride, w, h, filter_params,
                      subpel_y_q4, y_step_q4, conv_params);
152 153 154
  }
}

Angie Chiang's avatar
Angie Chiang committed
155
#if CONFIG_CONVOLVE_ROUND
Angie Chiang's avatar
Angie Chiang committed
156
void av1_convolve_rounding(const int32_t *src, int src_stride, uint8_t *dst,
Angie Chiang's avatar
Angie Chiang committed
157
                           int dst_stride, int w, int h, int bits) {
Angie Chiang's avatar
Angie Chiang committed
158 159 160 161
  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
162
          clip_pixel(ROUND_POWER_OF_TWO_SIGNED(src[r * src_stride + c], bits));
Angie Chiang's avatar
Angie Chiang committed
163 164 165
    }
  }
}
Angie Chiang's avatar
Angie Chiang committed
166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209

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);
    }
  }
}

210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227
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
228 229 230 231 232 233 234 235 236 237 238 239 240 241
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]);
242 243

#if CONFIG_DUAL_FILTER
244 245 246 247 248 249
  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);
  }
250
#endif
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

  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
278 279
}

Angie Chiang's avatar
Angie Chiang committed
280
#endif  // CONFIG_CONVOLVE_ROUND
Angie Chiang's avatar
Angie Chiang committed
281

Yaowu Xu's avatar
Yaowu Xu committed
282 283
void av1_convolve(const uint8_t *src, int src_stride, uint8_t *dst,
                  int dst_stride, int w, int h,
284
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
285
                  const InterpFilter *interp_filter,
286
#else
James Zern's avatar
James Zern committed
287
                  const InterpFilter interp_filter,
288
#endif
Yaowu Xu's avatar
Yaowu Xu committed
289
                  const int subpel_x_q4, int x_step_q4, const int subpel_y_q4,
290
                  int y_step_q4, ConvolveParams *conv_params) {
Angie Chiang's avatar
Angie Chiang committed
291 292
  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
293 294 295 296 297 298 299 300 301 302
#if CONFIG_DUAL_FILTER
  InterpFilterParams filter_params_x =
      av1_get_interp_filter_params(interp_filter[1 + 2 * conv_params->ref]);
  InterpFilterParams filter_params_y =
      av1_get_interp_filter_params(interp_filter[0 + 2 * conv_params->ref]);
  InterpFilterParams filter_params;
#else
  InterpFilterParams filter_params =
      av1_get_interp_filter_params(interp_filter);
#endif
Angie Chiang's avatar
Angie Chiang committed
303
  assert(conv_params->round == CONVOLVE_OPT_ROUND);
304 305 306 307 308 309

  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
310
  if (ignore_horiz && ignore_vert) {
311
    convolve_copy(src, src_stride, dst, dst_stride, w, h, conv_params);
Angie Chiang's avatar
Angie Chiang committed
312
  } else if (ignore_vert) {
313
#if CONFIG_DUAL_FILTER
Angie Chiang's avatar
Angie Chiang committed
314
    filter_params = filter_params_x;
315 316
#endif
    assert(filter_params.taps <= MAX_FILTER_TAP);
317
    av1_convolve_horiz_facade(src, src_stride, dst, dst_stride, w, h,
318 319
                              filter_params, subpel_x_q4, x_step_q4,
                              conv_params);
Angie Chiang's avatar
Angie Chiang committed
320
  } else if (ignore_horiz) {
321
#if CONFIG_DUAL_FILTER
Angie Chiang's avatar
Angie Chiang committed
322
    filter_params = filter_params_y;
323 324
#endif
    assert(filter_params.taps <= MAX_FILTER_TAP);
325
    av1_convolve_vert_facade(src, src_stride, dst, dst_stride, w, h,
326 327
                             filter_params, subpel_y_q4, y_step_q4,
                             conv_params);
Angie Chiang's avatar
Angie Chiang committed
328
  } else {
329 330 331 332 333 334
    // 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);
335
    int filter_size;
336
#if CONFIG_DUAL_FILTER
337 338
    if (interp_filter[0 + 2 * conv_params->ref] == MULTITAP_SHARP &&
        interp_filter[1 + 2 * conv_params->ref] == MULTITAP_SHARP) {
339 340 341 342
      // Avoid two directions both using 12-tap filter.
      // This will reduce hardware implementation cost.
      filter_params_y = av1_get_interp_filter_params(EIGHTTAP_SHARP);
    }
343

344 345 346 347
    // we do filter with fewer taps first to reduce hardware implementation
    // complexity
    if (filter_params_y.taps < filter_params_x.taps) {
      int intermediate_width;
348
      int temp_stride = max_intermediate_size;
349 350
      ConvolveParams temp_conv_params;
      temp_conv_params.ref = 0;
351
      temp_conv_params.round = CONVOLVE_OPT_ROUND;
352 353 354 355
      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;
356
      assert(intermediate_width <= max_intermediate_size);
Angie Chiang's avatar
Angie Chiang committed
357

358
      assert(filter_params.taps <= MAX_FILTER_TAP);
359

360 361
      av1_convolve_vert_facade(src - (filter_size / 2 - 1), src_stride, temp,
                               temp_stride, intermediate_width, h,
362 363
                               filter_params, subpel_y_q4, y_step_q4,
                               &temp_conv_params);
364

365 366
      filter_params = filter_params_x;
      assert(filter_params.taps <= MAX_FILTER_TAP);
367 368
      av1_convolve_horiz_facade(temp + (filter_size / 2 - 1), temp_stride, dst,
                                dst_stride, w, h, filter_params, subpel_x_q4,
369
                                x_step_q4, conv_params);
370
    } else
371
#endif  // CONFIG_DUAL_FILTER
372 373
    {
      int intermediate_height;
374
      int temp_stride = MAX_SB_SIZE;
375 376
      ConvolveParams temp_conv_params;
      temp_conv_params.ref = 0;
377
      temp_conv_params.round = CONVOLVE_OPT_ROUND;
378 379 380 381 382 383 384 385
#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;
386 387
      assert(intermediate_height <= max_intermediate_size);
      (void)max_intermediate_size;
388 389 390

      assert(filter_params.taps <= MAX_FILTER_TAP);

391 392 393
      av1_convolve_horiz_facade(src - src_stride * (filter_size / 2 - 1),
                                src_stride, temp, temp_stride, w,
                                intermediate_height, filter_params, subpel_x_q4,
394
                                x_step_q4, &temp_conv_params);
395 396 397 398 399 400

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

401 402 403
      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);
404
    }
Angie Chiang's avatar
Angie Chiang committed
405
  }
406 407
}

Angie Chiang's avatar
Angie Chiang committed
408 409 410 411 412
void av1_convolve_init_c(void) {
  // A placeholder for SIMD initialization
  return;
}

Yaowu Xu's avatar
Yaowu Xu committed
413 414 415 416 417 418
#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) {
419
  int x, y;
420
  int filter_size = filter_params.taps;
421 422 423 424 425
  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
426
      const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
427
          filter_params, x_q4 & SUBPEL_MASK);
428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443
      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
444 445 446 447 448
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) {
449
  int x, y;
450
  int filter_size = filter_params.taps;
451 452 453 454 455 456
  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
457
      const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
clang-format's avatar
clang-format committed
458
          filter_params, y_q4 & SUBPEL_MASK);
459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477
      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
478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499
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;
    }
  }
}

500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530
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);
531

532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547
  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
548 549
void av1_highbd_convolve(const uint8_t *src8, int src_stride, uint8_t *dst8,
                         int dst_stride, int w, int h,
550
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
551
                         const InterpFilter *interp_filter,
552
#else
James Zern's avatar
James Zern committed
553
                         const InterpFilter interp_filter,
554
#endif
Yaowu Xu's avatar
Yaowu Xu committed
555 556 557
                         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
558 559 560 561
  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;
562 563 564 565 566 567

  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
568
  if (ignore_horiz && ignore_vert) {
569
    highbd_convolve_copy(src, src_stride, dst, dst_stride, w, h, ref_idx, bd);
Angie Chiang's avatar
Angie Chiang committed
570
  } else if (ignore_vert) {
571 572
#if CONFIG_DUAL_FILTER
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
573
        av1_get_interp_filter_params(interp_filter[1 + 2 * ref_idx]);
574 575
#else
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
576
        av1_get_interp_filter_params(interp_filter);
577
#endif
578 579 580
    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
581
  } else if (ignore_horiz) {
582 583
#if CONFIG_DUAL_FILTER
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
584
        av1_get_interp_filter_params(interp_filter[0 + 2 * ref_idx]);
585 586
#else
    InterpFilterParams filter_params =
Yaowu Xu's avatar
Yaowu Xu committed
587
        av1_get_interp_filter_params(interp_filter);
588
#endif
589 590 591
    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
592
  } else {
593 594 595 596 597
    // 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
598
    uint8_t *temp8 = CONVERT_TO_BYTEPTR(temp);
599 600 601
    int max_intermediate_size = ((MAX_SB_SIZE * 2 + 16) + 16);
    int filter_size;
    InterpFilterParams filter_params;
602
#if CONFIG_DUAL_FILTER
Yaowu Xu's avatar
Yaowu Xu committed
603 604 605 606
    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]);
607 608 609 610 611 612
    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);
    }
613
#endif
614

615 616 617 618 619 620 621 622 623 624 625
#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
626

627 628 629
      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);
630

631 632 633 634 635 636 637 638 639 640 641
      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;
642
#if CONFIG_DUAL_FILTER
643 644 645 646 647
      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;
648
#endif
649 650 651 652 653 654 655 656 657 658 659 660 661 662 663
      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);
664

665 666 667 668
      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
669
  }
670
}
Yaowu Xu's avatar
Yaowu Xu committed
671
#endif  // CONFIG_AOM_HIGHBITDEPTH