vp9_dct.c 44.3 KB
Newer Older
John Koleszar's avatar
John Koleszar committed
1
/*
2
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
John Koleszar's avatar
John Koleszar committed
3
 *
4
 *  Use of this source code is governed by a BSD-style license
5 6
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
7
 *  in the file PATENTS.  All contributing project authors may
8
 *  be found in the AUTHORS file in the root of the source tree.
John Koleszar's avatar
John Koleszar committed
9 10 11
 */


12
#include <assert.h>
John Koleszar's avatar
John Koleszar committed
13
#include <math.h>
14
#include "./vpx_config.h"
15
#include "vp9/common/vp9_systemdependent.h"
16

17
#include "vp9/common/vp9_blockd.h"
18
#include "vp9/common/vp9_idct.h"
Jingning Han's avatar
Jingning Han committed
19

20
static void fdct4(const int16_t *input, int16_t *output) {
Yaowu Xu's avatar
Yaowu Xu committed
21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
  int16_t step[4];
  int temp1, temp2;

  step[0] = input[0] + input[3];
  step[1] = input[1] + input[2];
  step[2] = input[1] - input[2];
  step[3] = input[0] - input[3];

  temp1 = (step[0] + step[1]) * cospi_16_64;
  temp2 = (step[0] - step[1]) * cospi_16_64;
  output[0] = dct_const_round_shift(temp1);
  output[2] = dct_const_round_shift(temp2);
  temp1 = step[2] * cospi_24_64 + step[3] * cospi_8_64;
  temp2 = -step[2] * cospi_8_64 + step[3] * cospi_24_64;
  output[1] = dct_const_round_shift(temp1);
  output[3] = dct_const_round_shift(temp2);
}
Jingning Han's avatar
Jingning Han committed
38

39
void vp9_fdct4x4_c(int16_t *input, int16_t *output, int stride) {
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59
  // The 2D transform is done with two passes which are actually pretty
  // similar. In the first one, we transform the columns and transpose
  // the results. In the second one, we transform the rows. To achieve that,
  // as the first pass results are transposed, we tranpose the columns (that
  // is the transposed rows) and transpose the results (so that it goes back
  // in normal/row positions).
  int pass;
  // We need an intermediate buffer between passes.
  int16_t intermediate[4 * 4];
  int16_t *in = input;
  int16_t *out = intermediate;
  // Do the two transform/transpose passes
  for (pass = 0; pass < 2; ++pass) {
    /*canbe16*/ int input[4];
    /*canbe16*/ int step[4];
    /*needs32*/ int temp1, temp2;
    int i;
    for (i = 0; i < 4; ++i) {
      // Load inputs.
      if (0 == pass) {
Yaowu Xu's avatar
Yaowu Xu committed
60 61 62 63
        input[0] = in[0 * stride] * 16;
        input[1] = in[1 * stride] * 16;
        input[2] = in[2 * stride] * 16;
        input[3] = in[3 * stride] * 16;
64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92
        if (i == 0 && input[0]) {
          input[0] += 1;
        }
      } else {
        input[0] = in[0 * 4];
        input[1] = in[1 * 4];
        input[2] = in[2 * 4];
        input[3] = in[3 * 4];
      }
      // Transform.
      step[0] = input[0] + input[3];
      step[1] = input[1] + input[2];
      step[2] = input[1] - input[2];
      step[3] = input[0] - input[3];
      temp1 = (step[0] + step[1]) * cospi_16_64;
      temp2 = (step[0] - step[1]) * cospi_16_64;
      out[0] = dct_const_round_shift(temp1);
      out[2] = dct_const_round_shift(temp2);
      temp1 = step[2] * cospi_24_64 + step[3] * cospi_8_64;
      temp2 = -step[2] * cospi_8_64 + step[3] * cospi_24_64;
      out[1] = dct_const_round_shift(temp1);
      out[3] = dct_const_round_shift(temp2);
      // Do next column (which is a transposed row in second/horizontal pass)
      in++;
      out += 4;
    }
    // Setup in/out for next pass.
    in = intermediate;
    out = output;
John Koleszar's avatar
John Koleszar committed
93
  }
94

95 96 97 98 99 100
  {
    int i, j;
    for (i = 0; i < 4; ++i) {
      for (j = 0; j < 4; ++j)
        output[j + i * 4] = (output[j + i * 4] + 1) >> 2;
    }
John Koleszar's avatar
John Koleszar committed
101
  }
John Koleszar's avatar
John Koleszar committed
102
}
103

104
static void fadst4(const int16_t *input, int16_t *output) {
105 106
  int x0, x1, x2, x3;
  int s0, s1, s2, s3, s4, s5, s6, s7;
107

108 109 110 111
  x0 = input[0];
  x1 = input[1];
  x2 = input[2];
  x3 = input[3];
112

113 114 115 116
  if (!(x0 | x1 | x2 | x3)) {
    output[0] = output[1] = output[2] = output[3] = 0;
    return;
  }
117

118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142
  s0 = sinpi_1_9 * x0;
  s1 = sinpi_4_9 * x0;
  s2 = sinpi_2_9 * x1;
  s3 = sinpi_1_9 * x1;
  s4 = sinpi_3_9 * x2;
  s5 = sinpi_4_9 * x3;
  s6 = sinpi_2_9 * x3;
  s7 = x0 + x1 - x3;

  x0 = s0 + s2 + s5;
  x1 = sinpi_3_9 * s7;
  x2 = s1 - s3 + s6;
  x3 = s4;

  s0 = x0 + x3;
  s1 = x1;
  s2 = x2 - x3;
  s3 = x2 - x0 + x3;

  // 1-D transform scaling factor is sqrt(2).
  output[0] = dct_const_round_shift(s0);
  output[1] = dct_const_round_shift(s1);
  output[2] = dct_const_round_shift(s2);
  output[3] = dct_const_round_shift(s3);
}
143

Dmitry Kovalev's avatar
Dmitry Kovalev committed
144
static const transform_2d FHT_4[] = {
145 146 147 148
  { fdct4,  fdct4  },  // DCT_DCT  = 0
  { fadst4, fdct4  },  // ADST_DCT = 1
  { fdct4,  fadst4 },  // DCT_ADST = 2
  { fadst4, fadst4 }   // ADST_ADST = 3
149 150
};

151 152 153 154 155 156
void vp9_short_fht4x4_c(int16_t *input, int16_t *output,
                        int pitch, TX_TYPE tx_type) {
  int16_t out[4 * 4];
  int16_t *outptr = &out[0];
  int i, j;
  int16_t temp_in[4], temp_out[4];
Dmitry Kovalev's avatar
Dmitry Kovalev committed
157
  const transform_2d ht = FHT_4[tx_type];
158

Dmitry Kovalev's avatar
Dmitry Kovalev committed
159
  // Columns
160 161
  for (i = 0; i < 4; ++i) {
    for (j = 0; j < 4; ++j)
Yaowu Xu's avatar
Yaowu Xu committed
162
      temp_in[j] = input[j * pitch + i] * 16;
163 164
    if (i == 0 && temp_in[0])
      temp_in[0] += 1;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
165
    ht.cols(temp_in, temp_out);
166 167
    for (j = 0; j < 4; ++j)
      outptr[j * 4 + i] = temp_out[j];
John Koleszar's avatar
John Koleszar committed
168
  }
Yaowu Xu's avatar
Yaowu Xu committed
169

Dmitry Kovalev's avatar
Dmitry Kovalev committed
170
  // Rows
171 172 173
  for (i = 0; i < 4; ++i) {
    for (j = 0; j < 4; ++j)
      temp_in[j] = out[j + i * 4];
Dmitry Kovalev's avatar
Dmitry Kovalev committed
174
    ht.rows(temp_in, temp_out);
175 176
    for (j = 0; j < 4; ++j)
      output[j + i * 4] = (temp_out[j] + 1) >> 2;
John Koleszar's avatar
John Koleszar committed
177
  }
Yaowu Xu's avatar
Yaowu Xu committed
178
}
179

180
static void fdct8(const int16_t *input, int16_t *output) {
Christian Duvivier's avatar
Christian Duvivier committed
181 182 183
  /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
  /*needs32*/ int t0, t1, t2, t3;
  /*canbe16*/ int x0, x1, x2, x3;
Yunqing Wang's avatar
Yunqing Wang committed
184 185

  // stage 1
Christian Duvivier's avatar
Christian Duvivier committed
186 187 188 189 190 191 192 193 194
  s0 = input[0] + input[7];
  s1 = input[1] + input[6];
  s2 = input[2] + input[5];
  s3 = input[3] + input[4];
  s4 = input[3] - input[4];
  s5 = input[2] - input[5];
  s6 = input[1] - input[6];
  s7 = input[0] - input[7];

195
  // fdct4(step, step);
Christian Duvivier's avatar
Christian Duvivier committed
196 197 198 199 200 201 202 203 204 205 206 207
  x0 = s0 + s3;
  x1 = s1 + s2;
  x2 = s1 - s2;
  x3 = s0 - s3;
  t0 = (x0 + x1) * cospi_16_64;
  t1 = (x0 - x1) * cospi_16_64;
  t2 =  x2 * cospi_24_64 + x3 *  cospi_8_64;
  t3 = -x2 * cospi_8_64  + x3 * cospi_24_64;
  output[0] = dct_const_round_shift(t0);
  output[2] = dct_const_round_shift(t2);
  output[4] = dct_const_round_shift(t1);
  output[6] = dct_const_round_shift(t3);
208

Yunqing Wang's avatar
Yunqing Wang committed
209
  // Stage 2
Christian Duvivier's avatar
Christian Duvivier committed
210 211 212 213
  t0 = (s6 - s5) * cospi_16_64;
  t1 = (s6 + s5) * cospi_16_64;
  t2 = dct_const_round_shift(t0);
  t3 = dct_const_round_shift(t1);
214

Yunqing Wang's avatar
Yunqing Wang committed
215
  // Stage 3
Christian Duvivier's avatar
Christian Duvivier committed
216 217 218 219
  x0 = s4 + t2;
  x1 = s4 - t2;
  x2 = s7 - t3;
  x3 = s7 + t3;
220

Yunqing Wang's avatar
Yunqing Wang committed
221
  // Stage 4
Christian Duvivier's avatar
Christian Duvivier committed
222 223 224 225 226 227 228 229
  t0 = x0 * cospi_28_64 + x3 *   cospi_4_64;
  t1 = x1 * cospi_12_64 + x2 *  cospi_20_64;
  t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
  t3 = x3 * cospi_28_64 + x0 *  -cospi_4_64;
  output[1] = dct_const_round_shift(t0);
  output[3] = dct_const_round_shift(t2);
  output[5] = dct_const_round_shift(t1);
  output[7] = dct_const_round_shift(t3);
Yunqing Wang's avatar
Yunqing Wang committed
230
}
231

232
void vp9_fdct8x8_c(int16_t *input, int16_t *final_output, int stride) {
Yunqing Wang's avatar
Yunqing Wang committed
233
  int i, j;
Christian Duvivier's avatar
Christian Duvivier committed
234
  int16_t intermediate[64];
235

Christian Duvivier's avatar
Christian Duvivier committed
236 237 238 239 240 241 242 243 244 245
  // Transform columns
  {
    int16_t *output = intermediate;
    /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
    /*needs32*/ int t0, t1, t2, t3;
    /*canbe16*/ int x0, x1, x2, x3;

    int i;
    for (i = 0; i < 8; i++) {
      // stage 1
Yaowu Xu's avatar
Yaowu Xu committed
246 247 248 249 250 251 252 253
      s0 = (input[0 * stride] + input[7 * stride]) * 4;
      s1 = (input[1 * stride] + input[6 * stride]) * 4;
      s2 = (input[2 * stride] + input[5 * stride]) * 4;
      s3 = (input[3 * stride] + input[4 * stride]) * 4;
      s4 = (input[3 * stride] - input[4 * stride]) * 4;
      s5 = (input[2 * stride] - input[5 * stride]) * 4;
      s6 = (input[1 * stride] - input[6 * stride]) * 4;
      s7 = (input[0 * stride] - input[7 * stride]) * 4;
Christian Duvivier's avatar
Christian Duvivier committed
254

255
      // fdct4(step, step);
Christian Duvivier's avatar
Christian Duvivier committed
256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291
      x0 = s0 + s3;
      x1 = s1 + s2;
      x2 = s1 - s2;
      x3 = s0 - s3;
      t0 = (x0 + x1) * cospi_16_64;
      t1 = (x0 - x1) * cospi_16_64;
      t2 =  x2 * cospi_24_64 + x3 *  cospi_8_64;
      t3 = -x2 * cospi_8_64  + x3 * cospi_24_64;
      output[0 * 8] = dct_const_round_shift(t0);
      output[2 * 8] = dct_const_round_shift(t2);
      output[4 * 8] = dct_const_round_shift(t1);
      output[6 * 8] = dct_const_round_shift(t3);

      // Stage 2
      t0 = (s6 - s5) * cospi_16_64;
      t1 = (s6 + s5) * cospi_16_64;
      t2 = dct_const_round_shift(t0);
      t3 = dct_const_round_shift(t1);

      // Stage 3
      x0 = s4 + t2;
      x1 = s4 - t2;
      x2 = s7 - t3;
      x3 = s7 + t3;

      // Stage 4
      t0 = x0 * cospi_28_64 + x3 *   cospi_4_64;
      t1 = x1 * cospi_12_64 + x2 *  cospi_20_64;
      t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
      t3 = x3 * cospi_28_64 + x0 *  -cospi_4_64;
      output[1 * 8] = dct_const_round_shift(t0);
      output[3 * 8] = dct_const_round_shift(t2);
      output[5 * 8] = dct_const_round_shift(t1);
      output[7 * 8] = dct_const_round_shift(t3);
      input++;
      output++;
292 293
    }
  }
294

Dmitry Kovalev's avatar
Dmitry Kovalev committed
295
  // Rows
Yunqing Wang's avatar
Yunqing Wang committed
296
  for (i = 0; i < 8; ++i) {
297
    fdct8(&intermediate[i * 8], &final_output[i * 8]);
Yunqing Wang's avatar
Yunqing Wang committed
298
    for (j = 0; j < 8; ++j)
Christian Duvivier's avatar
Christian Duvivier committed
299
      final_output[j + i * 8] /= 2;
300
  }
Yunqing Wang's avatar
Yunqing Wang committed
301
}
302

303
void vp9_fdct16x16_c(int16_t *input, int16_t *output, int stride) {
304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325
  // The 2D transform is done with two passes which are actually pretty
  // similar. In the first one, we transform the columns and transpose
  // the results. In the second one, we transform the rows. To achieve that,
  // as the first pass results are transposed, we tranpose the columns (that
  // is the transposed rows) and transpose the results (so that it goes back
  // in normal/row positions).
  int pass;
  // We need an intermediate buffer between passes.
  int16_t intermediate[256];
  int16_t *in = input;
  int16_t *out = intermediate;
  // Do the two transform/transpose passes
  for (pass = 0; pass < 2; ++pass) {
    /*canbe16*/ int step1[8];
    /*canbe16*/ int step2[8];
    /*canbe16*/ int step3[8];
    /*canbe16*/ int input[8];
    /*needs32*/ int temp1, temp2;
    int i;
    for (i = 0; i < 16; i++) {
      if (0 == pass) {
        // Calculate input for the first 8 results.
Yaowu Xu's avatar
Yaowu Xu committed
326 327 328 329 330 331 332 333
        input[0] = (in[0 * stride] + in[15 * stride]) * 4;
        input[1] = (in[1 * stride] + in[14 * stride]) * 4;
        input[2] = (in[2 * stride] + in[13 * stride]) * 4;
        input[3] = (in[3 * stride] + in[12 * stride]) * 4;
        input[4] = (in[4 * stride] + in[11 * stride]) * 4;
        input[5] = (in[5 * stride] + in[10 * stride]) * 4;
        input[6] = (in[6 * stride] + in[ 9 * stride]) * 4;
        input[7] = (in[7 * stride] + in[ 8 * stride]) * 4;
334
        // Calculate input for the next 8 results.
Yaowu Xu's avatar
Yaowu Xu committed
335 336 337 338 339 340 341 342
        step1[0] = (in[7 * stride] - in[ 8 * stride]) * 4;
        step1[1] = (in[6 * stride] - in[ 9 * stride]) * 4;
        step1[2] = (in[5 * stride] - in[10 * stride]) * 4;
        step1[3] = (in[4 * stride] - in[11 * stride]) * 4;
        step1[4] = (in[3 * stride] - in[12 * stride]) * 4;
        step1[5] = (in[2 * stride] - in[13 * stride]) * 4;
        step1[6] = (in[1 * stride] - in[14 * stride]) * 4;
        step1[7] = (in[0 * stride] - in[15 * stride]) * 4;
343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361
      } else {
        // Calculate input for the first 8 results.
        input[0] = ((in[0 * 16] + 1) >> 2) + ((in[15 * 16] + 1) >> 2);
        input[1] = ((in[1 * 16] + 1) >> 2) + ((in[14 * 16] + 1) >> 2);
        input[2] = ((in[2 * 16] + 1) >> 2) + ((in[13 * 16] + 1) >> 2);
        input[3] = ((in[3 * 16] + 1) >> 2) + ((in[12 * 16] + 1) >> 2);
        input[4] = ((in[4 * 16] + 1) >> 2) + ((in[11 * 16] + 1) >> 2);
        input[5] = ((in[5 * 16] + 1) >> 2) + ((in[10 * 16] + 1) >> 2);
        input[6] = ((in[6 * 16] + 1) >> 2) + ((in[ 9 * 16] + 1) >> 2);
        input[7] = ((in[7 * 16] + 1) >> 2) + ((in[ 8 * 16] + 1) >> 2);
        // Calculate input for the next 8 results.
        step1[0] = ((in[7 * 16] + 1) >> 2) - ((in[ 8 * 16] + 1) >> 2);
        step1[1] = ((in[6 * 16] + 1) >> 2) - ((in[ 9 * 16] + 1) >> 2);
        step1[2] = ((in[5 * 16] + 1) >> 2) - ((in[10 * 16] + 1) >> 2);
        step1[3] = ((in[4 * 16] + 1) >> 2) - ((in[11 * 16] + 1) >> 2);
        step1[4] = ((in[3 * 16] + 1) >> 2) - ((in[12 * 16] + 1) >> 2);
        step1[5] = ((in[2 * 16] + 1) >> 2) - ((in[13 * 16] + 1) >> 2);
        step1[6] = ((in[1 * 16] + 1) >> 2) - ((in[14 * 16] + 1) >> 2);
        step1[7] = ((in[0 * 16] + 1) >> 2) - ((in[15 * 16] + 1) >> 2);
362
      }
363
      // Work on the first eight values; fdct8(input, even_results);
364 365 366 367 368 369 370 371 372 373 374 375 376 377 378
      {
        /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
        /*needs32*/ int t0, t1, t2, t3;
        /*canbe16*/ int x0, x1, x2, x3;

        // stage 1
        s0 = input[0] + input[7];
        s1 = input[1] + input[6];
        s2 = input[2] + input[5];
        s3 = input[3] + input[4];
        s4 = input[3] - input[4];
        s5 = input[2] - input[5];
        s6 = input[1] - input[6];
        s7 = input[0] - input[7];

379
        // fdct4(step, step);
380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413
        x0 = s0 + s3;
        x1 = s1 + s2;
        x2 = s1 - s2;
        x3 = s0 - s3;
        t0 = (x0 + x1) * cospi_16_64;
        t1 = (x0 - x1) * cospi_16_64;
        t2 = x3 * cospi_8_64  + x2 * cospi_24_64;
        t3 = x3 * cospi_24_64 - x2 * cospi_8_64;
        out[0] = dct_const_round_shift(t0);
        out[4] = dct_const_round_shift(t2);
        out[8] = dct_const_round_shift(t1);
        out[12] = dct_const_round_shift(t3);

        // Stage 2
        t0 = (s6 - s5) * cospi_16_64;
        t1 = (s6 + s5) * cospi_16_64;
        t2 = dct_const_round_shift(t0);
        t3 = dct_const_round_shift(t1);

        // Stage 3
        x0 = s4 + t2;
        x1 = s4 - t2;
        x2 = s7 - t3;
        x3 = s7 + t3;

        // Stage 4
        t0 = x0 * cospi_28_64 + x3 *   cospi_4_64;
        t1 = x1 * cospi_12_64 + x2 *  cospi_20_64;
        t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
        t3 = x3 * cospi_28_64 + x0 *  -cospi_4_64;
        out[2] = dct_const_round_shift(t0);
        out[6] = dct_const_round_shift(t2);
        out[10] = dct_const_round_shift(t1);
        out[14] = dct_const_round_shift(t3);
414
      }
415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473
      // Work on the next eight values; step1 -> odd_results
      {
        // step 2
        temp1 = (step1[5] - step1[2]) * cospi_16_64;
        temp2 = (step1[4] - step1[3]) * cospi_16_64;
        step2[2] = dct_const_round_shift(temp1);
        step2[3] = dct_const_round_shift(temp2);
        temp1 = (step1[4] + step1[3]) * cospi_16_64;
        temp2 = (step1[5] + step1[2]) * cospi_16_64;
        step2[4] = dct_const_round_shift(temp1);
        step2[5] = dct_const_round_shift(temp2);
        // step 3
        step3[0] = step1[0] + step2[3];
        step3[1] = step1[1] + step2[2];
        step3[2] = step1[1] - step2[2];
        step3[3] = step1[0] - step2[3];
        step3[4] = step1[7] - step2[4];
        step3[5] = step1[6] - step2[5];
        step3[6] = step1[6] + step2[5];
        step3[7] = step1[7] + step2[4];
        // step 4
        temp1 = step3[1] *  -cospi_8_64 + step3[6] * cospi_24_64;
        temp2 = step3[2] * -cospi_24_64 - step3[5] *  cospi_8_64;
        step2[1] = dct_const_round_shift(temp1);
        step2[2] = dct_const_round_shift(temp2);
        temp1 = step3[2] * -cospi_8_64 + step3[5] * cospi_24_64;
        temp2 = step3[1] * cospi_24_64 + step3[6] *  cospi_8_64;
        step2[5] = dct_const_round_shift(temp1);
        step2[6] = dct_const_round_shift(temp2);
        // step 5
        step1[0] = step3[0] + step2[1];
        step1[1] = step3[0] - step2[1];
        step1[2] = step3[3] - step2[2];
        step1[3] = step3[3] + step2[2];
        step1[4] = step3[4] + step2[5];
        step1[5] = step3[4] - step2[5];
        step1[6] = step3[7] - step2[6];
        step1[7] = step3[7] + step2[6];
        // step 6
        temp1 = step1[0] * cospi_30_64 + step1[7] *  cospi_2_64;
        temp2 = step1[1] * cospi_14_64 + step1[6] * cospi_18_64;
        out[1] = dct_const_round_shift(temp1);
        out[9] = dct_const_round_shift(temp2);
        temp1 = step1[2] * cospi_22_64 + step1[5] * cospi_10_64;
        temp2 = step1[3] *  cospi_6_64 + step1[4] * cospi_26_64;
        out[5] = dct_const_round_shift(temp1);
        out[13] = dct_const_round_shift(temp2);
        temp1 = step1[3] * -cospi_26_64 + step1[4] *  cospi_6_64;
        temp2 = step1[2] * -cospi_10_64 + step1[5] * cospi_22_64;
        out[3] = dct_const_round_shift(temp1);
        out[11] = dct_const_round_shift(temp2);
        temp1 = step1[1] * -cospi_18_64 + step1[6] * cospi_14_64;
        temp2 = step1[0] *  -cospi_2_64 + step1[7] * cospi_30_64;
        out[7] = dct_const_round_shift(temp1);
        out[15] = dct_const_round_shift(temp2);
      }
      // Do next column (which is a transposed row in second/horizontal pass)
      in++;
      out += 16;
474
    }
475 476 477
    // Setup in/out for next pass.
    in = intermediate;
    out = output;
478
  }
479 480
}

481
static void fadst8(const int16_t *input, int16_t *output) {
482 483
  int s0, s1, s2, s3, s4, s5, s6, s7;

Dmitry Kovalev's avatar
Dmitry Kovalev committed
484 485 486 487 488 489 490 491
  int x0 = input[7];
  int x1 = input[0];
  int x2 = input[5];
  int x3 = input[2];
  int x4 = input[3];
  int x5 = input[4];
  int x6 = input[1];
  int x7 = input[6];
492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549

  // stage 1
  s0 = cospi_2_64  * x0 + cospi_30_64 * x1;
  s1 = cospi_30_64 * x0 - cospi_2_64  * x1;
  s2 = cospi_10_64 * x2 + cospi_22_64 * x3;
  s3 = cospi_22_64 * x2 - cospi_10_64 * x3;
  s4 = cospi_18_64 * x4 + cospi_14_64 * x5;
  s5 = cospi_14_64 * x4 - cospi_18_64 * x5;
  s6 = cospi_26_64 * x6 + cospi_6_64  * x7;
  s7 = cospi_6_64  * x6 - cospi_26_64 * x7;

  x0 = dct_const_round_shift(s0 + s4);
  x1 = dct_const_round_shift(s1 + s5);
  x2 = dct_const_round_shift(s2 + s6);
  x3 = dct_const_round_shift(s3 + s7);
  x4 = dct_const_round_shift(s0 - s4);
  x5 = dct_const_round_shift(s1 - s5);
  x6 = dct_const_round_shift(s2 - s6);
  x7 = dct_const_round_shift(s3 - s7);

  // stage 2
  s0 = x0;
  s1 = x1;
  s2 = x2;
  s3 = x3;
  s4 = cospi_8_64  * x4 + cospi_24_64 * x5;
  s5 = cospi_24_64 * x4 - cospi_8_64  * x5;
  s6 = - cospi_24_64 * x6 + cospi_8_64  * x7;
  s7 =   cospi_8_64  * x6 + cospi_24_64 * x7;

  x0 = s0 + s2;
  x1 = s1 + s3;
  x2 = s0 - s2;
  x3 = s1 - s3;
  x4 = dct_const_round_shift(s4 + s6);
  x5 = dct_const_round_shift(s5 + s7);
  x6 = dct_const_round_shift(s4 - s6);
  x7 = dct_const_round_shift(s5 - s7);

  // stage 3
  s2 = cospi_16_64 * (x2 + x3);
  s3 = cospi_16_64 * (x2 - x3);
  s6 = cospi_16_64 * (x6 + x7);
  s7 = cospi_16_64 * (x6 - x7);

  x2 = dct_const_round_shift(s2);
  x3 = dct_const_round_shift(s3);
  x6 = dct_const_round_shift(s6);
  x7 = dct_const_round_shift(s7);

  output[0] =   x0;
  output[1] = - x4;
  output[2] =   x6;
  output[3] = - x2;
  output[4] =   x3;
  output[5] = - x7;
  output[6] =   x5;
  output[7] = - x1;
John Koleszar's avatar
John Koleszar committed
550 551
}

Dmitry Kovalev's avatar
Dmitry Kovalev committed
552
static const transform_2d FHT_8[] = {
553 554 555 556
  { fdct8,  fdct8  },  // DCT_DCT  = 0
  { fadst8, fdct8  },  // ADST_DCT = 1
  { fdct8,  fadst8 },  // DCT_ADST = 2
  { fadst8, fadst8 }   // ADST_ADST = 3
Dmitry Kovalev's avatar
Dmitry Kovalev committed
557
};
John Koleszar's avatar
John Koleszar committed
558

559 560 561 562 563 564
void vp9_short_fht8x8_c(int16_t *input, int16_t *output,
                        int pitch, TX_TYPE tx_type) {
  int16_t out[64];
  int16_t *outptr = &out[0];
  int i, j;
  int16_t temp_in[8], temp_out[8];
Dmitry Kovalev's avatar
Dmitry Kovalev committed
565
  const transform_2d ht = FHT_8[tx_type];
566

Dmitry Kovalev's avatar
Dmitry Kovalev committed
567
  // Columns
568 569
  for (i = 0; i < 8; ++i) {
    for (j = 0; j < 8; ++j)
Yaowu Xu's avatar
Yaowu Xu committed
570
      temp_in[j] = input[j * pitch + i] * 4;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
571
    ht.cols(temp_in, temp_out);
572 573
    for (j = 0; j < 8; ++j)
      outptr[j * 8 + i] = temp_out[j];
John Koleszar's avatar
John Koleszar committed
574 575
  }

Dmitry Kovalev's avatar
Dmitry Kovalev committed
576
  // Rows
577 578 579
  for (i = 0; i < 8; ++i) {
    for (j = 0; j < 8; ++j)
      temp_in[j] = out[j + i * 8];
Dmitry Kovalev's avatar
Dmitry Kovalev committed
580
    ht.rows(temp_in, temp_out);
581
    for (j = 0; j < 8; ++j)
Jingning Han's avatar
Jingning Han committed
582
      output[j + i * 8] = (temp_out[j] + (temp_out[j] < 0)) >> 1;
John Koleszar's avatar
John Koleszar committed
583
  }
Hui Su's avatar
Hui Su committed
584 585
}

Timothy B. Terriberry's avatar
Timothy B. Terriberry committed
586 587
/* 4-point reversible, orthonormal Walsh-Hadamard in 3.5 adds, 0.5 shifts per
   pixel. */
588
void vp9_fwht4x4_c(int16_t *input, int16_t *output, int stride) {
John Koleszar's avatar
John Koleszar committed
589
  int i;
Timothy B. Terriberry's avatar
Timothy B. Terriberry committed
590
  int a1, b1, c1, d1, e1;
591 592
  int16_t *ip = input;
  int16_t *op = output;
John Koleszar's avatar
John Koleszar committed
593 594

  for (i = 0; i < 4; i++) {
595 596 597 598
    a1 = ip[0 * stride];
    b1 = ip[1 * stride];
    c1 = ip[2 * stride];
    d1 = ip[3 * stride];
Timothy B. Terriberry's avatar
Timothy B. Terriberry committed
599

Yaowu Xu's avatar
Yaowu Xu committed
600 601 602 603 604 605 606
    a1 += b1;
    d1 = d1 - c1;
    e1 = (a1 - d1) >> 1;
    b1 = e1 - b1;
    c1 = e1 - c1;
    a1 -= c1;
    d1 += b1;
Timothy B. Terriberry's avatar
Timothy B. Terriberry committed
607 608 609 610
    op[0] = a1;
    op[4] = c1;
    op[8] = d1;
    op[12] = b1;
John Koleszar's avatar
John Koleszar committed
611 612 613 614 615 616 617 618

    ip++;
    op++;
  }
  ip = output;
  op = output;

  for (i = 0; i < 4; i++) {
Timothy B. Terriberry's avatar
Timothy B. Terriberry committed
619 620 621 622 623
    a1 = ip[0];
    b1 = ip[1];
    c1 = ip[2];
    d1 = ip[3];

Yaowu Xu's avatar
Yaowu Xu committed
624 625 626 627 628 629 630
    a1 += b1;
    d1 -= c1;
    e1 = (a1 - d1) >> 1;
    b1 = e1 - b1;
    c1 = e1 - c1;
    a1 -= c1;
    d1 += b1;
Yaowu Xu's avatar
Yaowu Xu committed
631 632 633 634
    op[0] = a1 * UNIT_QUANT_FACTOR;
    op[1] = c1 * UNIT_QUANT_FACTOR;
    op[2] = d1 * UNIT_QUANT_FACTOR;
    op[3] = b1 * UNIT_QUANT_FACTOR;
John Koleszar's avatar
John Koleszar committed
635 636 637 638

    ip += 4;
    op += 4;
  }
Hui Su's avatar
Hui Su committed
639 640
}

Yunqing Wang's avatar
Yunqing Wang committed
641
// Rewrote to use same algorithm as others.
642
static void fdct16(const int16_t in[16], int16_t out[16]) {
643 644 645 646 647
  /*canbe16*/ int step1[8];
  /*canbe16*/ int step2[8];
  /*canbe16*/ int step3[8];
  /*canbe16*/ int input[8];
  /*needs32*/ int temp1, temp2;
Yunqing Wang's avatar
Yunqing Wang committed
648 649

  // step 1
650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667
  input[0] = in[0] + in[15];
  input[1] = in[1] + in[14];
  input[2] = in[2] + in[13];
  input[3] = in[3] + in[12];
  input[4] = in[4] + in[11];
  input[5] = in[5] + in[10];
  input[6] = in[6] + in[ 9];
  input[7] = in[7] + in[ 8];

  step1[0] = in[7] - in[ 8];
  step1[1] = in[6] - in[ 9];
  step1[2] = in[5] - in[10];
  step1[3] = in[4] - in[11];
  step1[4] = in[3] - in[12];
  step1[5] = in[2] - in[13];
  step1[6] = in[1] - in[14];
  step1[7] = in[0] - in[15];

668
  // fdct8(step, step);
669
  {
670 671 672 673 674 675 676 677 678 679 680 681 682 683
    /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
    /*needs32*/ int t0, t1, t2, t3;
    /*canbe16*/ int x0, x1, x2, x3;

    // stage 1
    s0 = input[0] + input[7];
    s1 = input[1] + input[6];
    s2 = input[2] + input[5];
    s3 = input[3] + input[4];
    s4 = input[3] - input[4];
    s5 = input[2] - input[5];
    s6 = input[1] - input[6];
    s7 = input[0] - input[7];

684
    // fdct4(step, step);
685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718
    x0 = s0 + s3;
    x1 = s1 + s2;
    x2 = s1 - s2;
    x3 = s0 - s3;
    t0 = (x0 + x1) * cospi_16_64;
    t1 = (x0 - x1) * cospi_16_64;
    t2 = x3 * cospi_8_64  + x2 * cospi_24_64;
    t3 = x3 * cospi_24_64 - x2 * cospi_8_64;
    out[0] = dct_const_round_shift(t0);
    out[4] = dct_const_round_shift(t2);
    out[8] = dct_const_round_shift(t1);
    out[12] = dct_const_round_shift(t3);

    // Stage 2
    t0 = (s6 - s5) * cospi_16_64;
    t1 = (s6 + s5) * cospi_16_64;
    t2 = dct_const_round_shift(t0);
    t3 = dct_const_round_shift(t1);

    // Stage 3
    x0 = s4 + t2;
    x1 = s4 - t2;
    x2 = s7 - t3;
    x3 = s7 + t3;

    // Stage 4
    t0 = x0 * cospi_28_64 + x3 *   cospi_4_64;
    t1 = x1 * cospi_12_64 + x2 *  cospi_20_64;
    t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
    t3 = x3 * cospi_28_64 + x0 *  -cospi_4_64;
    out[2] = dct_const_round_shift(t0);
    out[6] = dct_const_round_shift(t2);
    out[10] = dct_const_round_shift(t1);
    out[14] = dct_const_round_shift(t3);
719
  }
Yunqing Wang's avatar
Yunqing Wang committed
720 721

  // step 2
722 723 724 725 726 727 728 729
  temp1 = (step1[5] - step1[2]) * cospi_16_64;
  temp2 = (step1[4] - step1[3]) * cospi_16_64;
  step2[2] = dct_const_round_shift(temp1);
  step2[3] = dct_const_round_shift(temp2);
  temp1 = (step1[4] + step1[3]) * cospi_16_64;
  temp2 = (step1[5] + step1[2]) * cospi_16_64;
  step2[4] = dct_const_round_shift(temp1);
  step2[5] = dct_const_round_shift(temp2);
Yunqing Wang's avatar
Yunqing Wang committed
730 731

  // step 3
732 733 734 735 736 737 738 739
  step3[0] = step1[0] + step2[3];
  step3[1] = step1[1] + step2[2];
  step3[2] = step1[1] - step2[2];
  step3[3] = step1[0] - step2[3];
  step3[4] = step1[7] - step2[4];
  step3[5] = step1[6] - step2[5];
  step3[6] = step1[6] + step2[5];
  step3[7] = step1[7] + step2[4];
Yunqing Wang's avatar
Yunqing Wang committed
740 741

  // step 4
742 743 744 745 746 747 748 749
  temp1 = step3[1] *  -cospi_8_64 + step3[6] * cospi_24_64;
  temp2 = step3[2] * -cospi_24_64 - step3[5] *  cospi_8_64;
  step2[1] = dct_const_round_shift(temp1);
  step2[2] = dct_const_round_shift(temp2);
  temp1 = step3[2] * -cospi_8_64 + step3[5] * cospi_24_64;
  temp2 = step3[1] * cospi_24_64 + step3[6] *  cospi_8_64;
  step2[5] = dct_const_round_shift(temp1);
  step2[6] = dct_const_round_shift(temp2);
Yunqing Wang's avatar
Yunqing Wang committed
750 751

  // step 5
752 753 754 755 756 757 758 759
  step1[0] = step3[0] + step2[1];
  step1[1] = step3[0] - step2[1];
  step1[2] = step3[3] - step2[2];
  step1[3] = step3[3] + step2[2];
  step1[4] = step3[4] + step2[5];
  step1[5] = step3[4] - step2[5];
  step1[6] = step3[7] - step2[6];
  step1[7] = step3[7] + step2[6];
Yunqing Wang's avatar
Yunqing Wang committed
760 761

  // step 6
762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780
  temp1 = step1[0] * cospi_30_64 + step1[7] *  cospi_2_64;
  temp2 = step1[1] * cospi_14_64 + step1[6] * cospi_18_64;
  out[1] = dct_const_round_shift(temp1);
  out[9] = dct_const_round_shift(temp2);

  temp1 = step1[2] * cospi_22_64 + step1[5] * cospi_10_64;
  temp2 = step1[3] *  cospi_6_64 + step1[4] * cospi_26_64;
  out[5] = dct_const_round_shift(temp1);
  out[13] = dct_const_round_shift(temp2);

  temp1 = step1[3] * -cospi_26_64 + step1[4] *  cospi_6_64;
  temp2 = step1[2] * -cospi_10_64 + step1[5] * cospi_22_64;
  out[3] = dct_const_round_shift(temp1);
  out[11] = dct_const_round_shift(temp2);

  temp1 = step1[1] * -cospi_18_64 + step1[6] * cospi_14_64;
  temp2 = step1[0] *  -cospi_2_64 + step1[7] * cospi_30_64;
  out[7] = dct_const_round_shift(temp1);
  out[15] = dct_const_round_shift(temp2);
Daniel Kang's avatar
Daniel Kang committed
781
}
782

783
static void fadst16(const int16_t *input, int16_t *output) {
784 785
  int s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12, s13, s14, s15;

Dmitry Kovalev's avatar
Dmitry Kovalev committed
786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801
  int x0 = input[15];
  int x1 = input[0];
  int x2 = input[13];
  int x3 = input[2];
  int x4 = input[11];
  int x5 = input[4];
  int x6 = input[9];
  int x7 = input[6];
  int x8 = input[7];
  int x9 = input[8];
  int x10 = input[5];
  int x11 = input[10];
  int x12 = input[3];
  int x13 = input[12];
  int x14 = input[1];
  int x15 = input[14];
802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942

  // stage 1
  s0 = x0 * cospi_1_64  + x1 * cospi_31_64;
  s1 = x0 * cospi_31_64 - x1 * cospi_1_64;
  s2 = x2 * cospi_5_64  + x3 * cospi_27_64;
  s3 = x2 * cospi_27_64 - x3 * cospi_5_64;
  s4 = x4 * cospi_9_64  + x5 * cospi_23_64;
  s5 = x4 * cospi_23_64 - x5 * cospi_9_64;
  s6 = x6 * cospi_13_64 + x7 * cospi_19_64;
  s7 = x6 * cospi_19_64 - x7 * cospi_13_64;
  s8 = x8 * cospi_17_64 + x9 * cospi_15_64;
  s9 = x8 * cospi_15_64 - x9 * cospi_17_64;
  s10 = x10 * cospi_21_64 + x11 * cospi_11_64;
  s11 = x10 * cospi_11_64 - x11 * cospi_21_64;
  s12 = x12 * cospi_25_64 + x13 * cospi_7_64;
  s13 = x12 * cospi_7_64  - x13 * cospi_25_64;
  s14 = x14 * cospi_29_64 + x15 * cospi_3_64;
  s15 = x14 * cospi_3_64  - x15 * cospi_29_64;

  x0 = dct_const_round_shift(s0 + s8);
  x1 = dct_const_round_shift(s1 + s9);
  x2 = dct_const_round_shift(s2 + s10);
  x3 = dct_const_round_shift(s3 + s11);
  x4 = dct_const_round_shift(s4 + s12);
  x5 = dct_const_round_shift(s5 + s13);
  x6 = dct_const_round_shift(s6 + s14);
  x7 = dct_const_round_shift(s7 + s15);
  x8  = dct_const_round_shift(s0 - s8);
  x9  = dct_const_round_shift(s1 - s9);
  x10 = dct_const_round_shift(s2 - s10);
  x11 = dct_const_round_shift(s3 - s11);
  x12 = dct_const_round_shift(s4 - s12);
  x13 = dct_const_round_shift(s5 - s13);
  x14 = dct_const_round_shift(s6 - s14);
  x15 = dct_const_round_shift(s7 - s15);

  // stage 2
  s0 = x0;
  s1 = x1;
  s2 = x2;
  s3 = x3;
  s4 = x4;
  s5 = x5;
  s6 = x6;
  s7 = x7;
  s8 =    x8 * cospi_4_64   + x9 * cospi_28_64;
  s9 =    x8 * cospi_28_64  - x9 * cospi_4_64;
  s10 =   x10 * cospi_20_64 + x11 * cospi_12_64;
  s11 =   x10 * cospi_12_64 - x11 * cospi_20_64;
  s12 = - x12 * cospi_28_64 + x13 * cospi_4_64;
  s13 =   x12 * cospi_4_64  + x13 * cospi_28_64;
  s14 = - x14 * cospi_12_64 + x15 * cospi_20_64;
  s15 =   x14 * cospi_20_64 + x15 * cospi_12_64;

  x0 = s0 + s4;
  x1 = s1 + s5;
  x2 = s2 + s6;
  x3 = s3 + s7;
  x4 = s0 - s4;
  x5 = s1 - s5;
  x6 = s2 - s6;
  x7 = s3 - s7;
  x8 = dct_const_round_shift(s8 + s12);
  x9 = dct_const_round_shift(s9 + s13);
  x10 = dct_const_round_shift(s10 + s14);
  x11 = dct_const_round_shift(s11 + s15);
  x12 = dct_const_round_shift(s8 - s12);
  x13 = dct_const_round_shift(s9 - s13);
  x14 = dct_const_round_shift(s10 - s14);
  x15 = dct_const_round_shift(s11 - s15);

  // stage 3
  s0 = x0;
  s1 = x1;
  s2 = x2;
  s3 = x3;
  s4 = x4 * cospi_8_64  + x5 * cospi_24_64;
  s5 = x4 * cospi_24_64 - x5 * cospi_8_64;
  s6 = - x6 * cospi_24_64 + x7 * cospi_8_64;
  s7 =   x6 * cospi_8_64  + x7 * cospi_24_64;
  s8 = x8;
  s9 = x9;
  s10 = x10;
  s11 = x11;
  s12 = x12 * cospi_8_64  + x13 * cospi_24_64;
  s13 = x12 * cospi_24_64 - x13 * cospi_8_64;
  s14 = - x14 * cospi_24_64 + x15 * cospi_8_64;
  s15 =   x14 * cospi_8_64  + x15 * cospi_24_64;

  x0 = s0 + s2;
  x1 = s1 + s3;
  x2 = s0 - s2;
  x3 = s1 - s3;
  x4 = dct_const_round_shift(s4 + s6);
  x5 = dct_const_round_shift(s5 + s7);
  x6 = dct_const_round_shift(s4 - s6);
  x7 = dct_const_round_shift(s5 - s7);
  x8 = s8 + s10;
  x9 = s9 + s11;
  x10 = s8 - s10;
  x11 = s9 - s11;
  x12 = dct_const_round_shift(s12 + s14);
  x13 = dct_const_round_shift(s13 + s15);
  x14 = dct_const_round_shift(s12 - s14);
  x15 = dct_const_round_shift(s13 - s15);

  // stage 4
  s2 = (- cospi_16_64) * (x2 + x3);
  s3 = cospi_16_64 * (x2 - x3);
  s6 = cospi_16_64 * (x6 + x7);
  s7 = cospi_16_64 * (- x6 + x7);
  s10 = cospi_16_64 * (x10 + x11);
  s11 = cospi_16_64 * (- x10 + x11);
  s14 = (- cospi_16_64) * (x14 + x15);
  s15 = cospi_16_64 * (x14 - x15);

  x2 = dct_const_round_shift(s2);
  x3 = dct_const_round_shift(s3);
  x6 = dct_const_round_shift(s6);
  x7 = dct_const_round_shift(s7);
  x10 = dct_const_round_shift(s10);
  x11 = dct_const_round_shift(s11);
  x14 = dct_const_round_shift(s14);
  x15 = dct_const_round_shift(s15);

  output[0] = x0;
  output[1] = - x8;
  output[2] = x12;
  output[3] = - x4;
  output[4] = x6;
  output[5] = x14;
  output[6] = x10;
  output[7] = x2;
  output[8] = x3;
  output[9] =  x11;
  output[10] = x15;
  output[11] = x7;
  output[12] = x5;
  output[13] = - x13;
  output[14] = x9;
  output[15] = - x1;
943 944
}

Dmitry Kovalev's avatar
Dmitry Kovalev committed
945
static const transform_2d FHT_16[] = {
946 947 948 949
  { fdct16,  fdct16  },  // DCT_DCT  = 0
  { fadst16, fdct16  },  // ADST_DCT = 1
  { fdct16,  fadst16 },  // DCT_ADST = 2
  { fadst16, fadst16 }   // ADST_ADST = 3
Dmitry Kovalev's avatar
Dmitry Kovalev committed
950
};
951

952 953 954 955 956 957
void vp9_short_fht16x16_c(int16_t *input, int16_t *output,
                          int pitch, TX_TYPE tx_type) {
  int16_t out[256];
  int16_t *outptr = &out[0];
  int i, j;
  int16_t temp_in[16], temp_out[16];
Dmitry Kovalev's avatar
Dmitry Kovalev committed
958
  const transform_2d ht = FHT_16[tx_type];
959

Dmitry Kovalev's avatar
Dmitry Kovalev committed
960
  // Columns
961 962
  for (i = 0; i < 16; ++i) {
    for (j = 0; j < 16; ++j)
Yaowu Xu's avatar
Yaowu Xu committed
963
      temp_in[j] = input[j * pitch + i] * 4;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
964
    ht.cols(temp_in, temp_out);
965
    for (j = 0; j < 16; ++j)
966 967
      outptr[j * 16 + i] = (temp_out[j] + 1 + (temp_out[j] < 0)) >> 2;
//      outptr[j * 16 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
968 969
  }

Dmitry Kovalev's avatar
Dmitry Kovalev committed
970
  // Rows
971 972 973
  for (i = 0; i < 16; ++i) {
    for (j = 0; j < 16; ++j)
      temp_in[j] = out[j + i * 16];
Dmitry Kovalev's avatar
Dmitry Kovalev committed
974
    ht.rows(temp_in, temp_out);
975 976 977
    for (j = 0; j < 16; ++j)
      output[j + i * 16] = temp_out[j];
  }
978
}
979

980 981 982 983 984 985 986 987 988 989
static INLINE int dct_32_round(int input) {
  int rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
  assert(-131072 <= rv && rv <= 131071);
  return rv;
}

static INLINE int half_round_shift(int input) {
  int rv = (input + 1 + (input < 0)) >> 2;
  return rv;
}
Yunqing Wang's avatar
Yunqing Wang committed
990

991
static void dct32_1d(const int *input, int *output, int round) {
Yunqing Wang's avatar
Yunqing Wang committed
992
  int step[32];
993
  // Stage 1
Yunqing Wang's avatar
Yunqing Wang committed
994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025
  step[0] = input[0] + input[(32 - 1)];
  step[1] = input[1] + input[(32 - 2)];
  step[2] = input[2] + input[(32 - 3)];
  step[3] = input[3] + input[(32 - 4)];
  step[4] = input[4] + input[(32 - 5)];
  step[5] = input[5] + input[(32 - 6)];
  step[6] = input[6] + input[(32 - 7)];
  step[7] = input[7] + input[(32 - 8)];
  step[8] = input[8] + input[(32 - 9)];
  step[9] = input[9] + input[(32 - 10)];
  step[10] = input[10] + input[(32 - 11)];
  step[11] = input[11] + input[(32 - 12)];
  step[12] = input[12] + input[(32 - 13)];
  step[13] = input[13] + input[(32 - 14)];
  step[14] = input[14] + input[(32 - 15)];
  step[15] = input[15] + input[(32 - 16)];
  step[16] = -input[16] + input[(32 - 17)];
  step[17] = -input[17] + input[(32 - 18)];
  step[18] = -input[18] + input[(32 - 19)];
  step[19] = -input[19] + input[(32 - 20)];
  step[20] = -input[20] + input[(32 - 21)];
  step[21] = -input[21] + input[(32 - 22)];
  step[22] = -input[22] + input[(32 - 23)];
  step[23] = -input[23] + input[(32 - 24)];
  step[24] = -input[24] + input[(32 - 25)];
  step[25] = -input[25] + input[(32 - 26)];
  step[26] = -input[26] + input[(32 - 27)];
  step[27] = -input[27] + input[(32 - 28)];
  step[28] = -input[28] + input[(32 - 29)];
  step[29] = -input[29] + input[(32 - 30)];
  step[30] = -input[30] + input[(32 - 31)];
  step[31] = -input[31] + input[(32 - 32)];
1026 1027

  // Stage 2
Yunqing Wang's avatar
Yunqing Wang committed
1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049
  output[0] = step[0] + step[16 - 1];
  output[1] = step[1] + step[16 - 2];
  output[2] = step[2] + step[16 - 3];
  output[3] = step[3] + step[16 - 4];
  output[4] = step[4] + step[16 - 5];
  output[5] = step[5] + step[16 - 6];
  output[6] = step[6] + step[16 - 7];
  output[7] = step[7] + step[16 - 8];
  output[8] = -step[8] + step[16 - 9];
  output[9] = -step[9] + step[16 - 10];
  output[10] = -step[10] + step[16 - 11];
  output[11] = -step[11] + step[16 - 12];
  output[12] = -step[12] + step[16 - 13];
  output[13] = -step[13] + step[16 - 14];
  output[14] = -step[14] + step[16 - 15];
  output[15] = -step[15] + step[16 - 16];

  output[16] = step[16];
  output[17] = step[17];
  output[18] = step[18];
  output[19] = step[19];

Yaowu Xu's avatar
Yaowu Xu committed
1050 1051 1052 1053
  output[20] = dct_32_round((-step[20] + step[27]) * cospi_16_64);
  output[21] = dct_32_round((-step[21] + step[26]) * cospi_16_64);
  output[22] = dct_32_round((-step[22] + step[25]) * cospi_16_64);
  output[23] = dct_32_round((-step[23] + step[24]) * cospi_16_64);
Yunqing Wang's avatar
Yunqing Wang committed
1054

Yaowu Xu's avatar
Yaowu Xu committed
1055 1056 1057 1058
  output[24] = dct_32_round((step[24] + step[23]) * cospi_16_64);
  output[25] = dct_32_round((step[25] + step[22]) * cospi_16_64);
  output[26] = dct_32_round((step[26] + step[21]) * cospi_16_64);
  output[27] = dct_32_round((step[27] + step[20]) * cospi_16_64);
Yunqing Wang's avatar
Yunqing Wang committed
1059 1060 1061 1062 1063

  output[28] = step[28];
  output[29] = step[29];
  output[30] = step[30];
  output[31] = step[31];
1064

1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102
  // dump the magnitude by 4, hence the intermediate values are within
  // the range of 16 bits.
  if (round) {
    output[0] = half_round_shift(output[0]);
    output[1] = half_round_shift(output[1]);
    output[2] = half_round_shift(output[2]);
    output[3] = half_round_shift(output[3]);
    output[4] = half_round_shift(output[4]);
    output[5] = half_round_shift(output[5]);
    output[6] = half_round_shift(output[6]);
    output[7] = half_round_shift(output[7]);
    output[8] = half_round_shift(output[8]);
    output[9] = half_round_shift(output[9]);
    output[10] = half_round_shift(output[10]);
    output[11] = half_round_shift(output[11]);
    output[12] = half_round_shift(output[12]);
    output[13] = half_round_shift(output[13]);
    output[14] = half_round_shift(output[14]);
    output[15] = half_round_shift(output[15]);

    output[16] = half_round_shift(output[16]);
    output[17] = half_round_shift(output[17]);
    output[18] = half_round_shift(output[18]);
    output[19] = half_round_shift(output[19]);
    output[20] = half_round_shift(output[20]);
    output[21] = half_round_shift(output[21]);
    output[22] = half_round_shift(output[22]);
    output[23] = half_round_shift(output[23]);
    output[24] = half_round_shift(output[24]);
    output[25] = half_round_shift(output[25]);
    output[26] = half_round_shift(output[26]);
    output[27] = half_round_shift(output[27]);
    output[28] = half_round_shift(output[28]);
    output[29] = half_round_shift(output[29]);
    output[30] = half_round_shift(output[30]);
    output[31] = half_round_shift(output[31]);
  }

1103
  // Stage 3
Yunqing Wang's avatar
Yunqing Wang committed
1104 1105 1106 1107 1108 1109 1110 1111 1112 1113
  step[0] = output[0] + output[(8 - 1)];
  step[1] = output[1] + output[(8 - 2)];
  step[2] = output[2] + output[(8 - 3)];
  step[3] = output[3] + output[(8 - 4)];
  step[4] = -output[4] + output[(8 - 5)];
  step[5] = -output[5] + output[(8 - 6)];
  step[6] = -output[6] + output[(8 - 7)];
  step[7] = -output[7] + output[(8 - 8)];
  step[8] = output[8];
  step[9] = output[9];
Yaowu Xu's avatar
Yaowu Xu committed
1114 1115 1116 1117
  step[10] = dct_32_round((-output[10] + output[13]) * cospi_16_64);
  step[11] = dct_32_round((-output[11] + output[12]) * cospi_16_64);
  step[12] = dct_32_round((output[12] + output[11]) * cospi_16_64);
  step[13] = dct_32_round((output[13] + output[10]) * cospi_16_64);
Yunqing Wang's avatar
Yunqing Wang committed
1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136
  step[14] = output[14];
  step[15] = output[15];

  step[16] = output[16] + output[23];
  step[17] = output[17] + output[22];
  step[18] = output[18] + output[21];
  step[19] = output[19] + output[20];
  step[20] = -output[20] + output[19];
  step[21] = -output[21] + output[18];
  step[22] = -output[22] + output[17];
  step[23] = -output[23] + output[16];
  step[24] = -output[24] + output[31];
  step[25] = -output[25] + output[30];
  step[26] = -output[26] + output[29];
  step[27] = -output[27] + output[28];
  step[28] = output[28] + output[27];
  step[29] = output[29] + output[26];
  step[30] = output[30] + output[25];
  step[31] = output[31] + output[24];
1137 1138

  // Stage 4
Yunqing Wang's avatar
Yunqing Wang committed
1139 1140 1141 1142 1143
  output[0] = step[0] + step[3];
  output[1] = step[1] + step[2];
  output[2] = -step[2] + step[1];
  output[3] = -step[3] + step[0];
  output[4] = step[4];
Yaowu Xu's avatar
Yaowu Xu committed
1144 1145
  output[5] = dct_32_round((-step[5] + step[6]) * cospi_16_64);
  output[6] = dct_32_round((step[6] + step[5]) * cospi_16_64);
Yunqing Wang's avatar
Yunqing Wang committed
1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157
  output[7] = step[7];
  output[8] = step[8] + step[11];
  output[9] = step[9] + step[10];
  output[10] = -step[10] + step[9];
  output[11] = -step[11] + step[8];
  output[12] = -step[12] + step[15];
  output[13] = -step[13] + step[14];
  output[14] = step[14] + step[13];
  output[15] = step[15] + step[12];

  output[16] = step[16];
  output[17] = step[17];
Yaowu Xu's avatar
Yaowu Xu committed
1158 1159 1160 1161
  output[18] =