dct16x16_test.cc 23.7 KB
Newer Older
Daniel Kang's avatar
Daniel Kang committed
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
/*
 *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

#include <math.h>
#include <stdlib.h>
#include <string.h>

#include "third_party/googletest/src/include/gtest/gtest.h"
16 17 18 19
#include "test/acm_random.h"
#include "test/clear_system_state.h"
#include "test/register_state_check.h"
#include "test/util.h"
Daniel Kang's avatar
Daniel Kang committed
20

Yaowu Xu's avatar
Yaowu Xu committed
21
#include "./vp9_rtcd.h"
22
#include "vp9/common/vp9_entropy.h"
23
#include "vpx/vpx_codec.h"
24 25
#include "vpx/vpx_integer.h"

Daniel Kang's avatar
Daniel Kang committed
26 27 28
using libvpx_test::ACMRandom;

namespace {
29 30 31 32

#ifdef _MSC_VER
static int round(double x) {
  if (x < 0)
Yaowu Xu's avatar
Yaowu Xu committed
33
    return static_cast<int>(ceil(x - 0.5));
34
  else
Yaowu Xu's avatar
Yaowu Xu committed
35
    return static_cast<int>(floor(x + 0.5));
36 37
}
#endif
Daniel Kang's avatar
Daniel Kang committed
38

39
const int kNumCoeffs = 256;
Daniel Kang's avatar
Daniel Kang committed
40 41 42 43 44 45 46 47
const double PI = 3.1415926535898;
void reference2_16x16_idct_2d(double *input, double *output) {
  double x;
  for (int l = 0; l < 16; ++l) {
    for (int k = 0; k < 16; ++k) {
      double s = 0;
      for (int i = 0; i < 16; ++i) {
        for (int j = 0; j < 16; ++j) {
Yaowu Xu's avatar
Yaowu Xu committed
48 49 50
          x = cos(PI * j * (l + 0.5) / 16.0) *
              cos(PI * i * (k + 0.5) / 16.0) *
              input[i * 16 + j] / 256;
Daniel Kang's avatar
Daniel Kang committed
51 52 53 54 55 56 57 58 59 60 61 62
          if (i != 0)
            x *= sqrt(2.0);
          if (j != 0)
            x *= sqrt(2.0);
          s += x;
        }
      }
      output[k*16+l] = s;
    }
  }
}

63

64 65 66 67 68 69 70 71 72 73 74 75 76 77 78
const double C1 = 0.995184726672197;
const double C2 = 0.98078528040323;
const double C3 = 0.956940335732209;
const double C4 = 0.923879532511287;
const double C5 = 0.881921264348355;
const double C6 = 0.831469612302545;
const double C7 = 0.773010453362737;
const double C8 = 0.707106781186548;
const double C9 = 0.634393284163646;
const double C10 = 0.555570233019602;
const double C11 = 0.471396736825998;
const double C12 = 0.38268343236509;
const double C13 = 0.290284677254462;
const double C14 = 0.195090322016128;
const double C15 = 0.098017140329561;
79

Jingning Han's avatar
Jingning Han committed
80
void butterfly_16x16_dct_1d(double input[16], double output[16]) {
Daniel Kang's avatar
Daniel Kang committed
81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112
  double step[16];
  double intermediate[16];
  double temp1, temp2;

  // step 1
  step[ 0] = input[0] + input[15];
  step[ 1] = input[1] + input[14];
  step[ 2] = input[2] + input[13];
  step[ 3] = input[3] + input[12];
  step[ 4] = input[4] + input[11];
  step[ 5] = input[5] + input[10];
  step[ 6] = input[6] + input[ 9];
  step[ 7] = input[7] + input[ 8];
  step[ 8] = input[7] - input[ 8];
  step[ 9] = input[6] - input[ 9];
  step[10] = input[5] - input[10];
  step[11] = input[4] - input[11];
  step[12] = input[3] - input[12];
  step[13] = input[2] - input[13];
  step[14] = input[1] - input[14];
  step[15] = input[0] - input[15];

  // step 2
  output[0] = step[0] + step[7];
  output[1] = step[1] + step[6];
  output[2] = step[2] + step[5];
  output[3] = step[3] + step[4];
  output[4] = step[3] - step[4];
  output[5] = step[2] - step[5];
  output[6] = step[1] - step[6];
  output[7] = step[0] - step[7];

113 114
  temp1 = step[ 8] * C7;
  temp2 = step[15] * C9;
Daniel Kang's avatar
Daniel Kang committed
115 116
  output[ 8] = temp1 + temp2;

117 118
  temp1 = step[ 9] * C11;
  temp2 = step[14] * C5;
Daniel Kang's avatar
Daniel Kang committed
119 120
  output[ 9] = temp1 - temp2;

121 122
  temp1 = step[10] * C3;
  temp2 = step[13] * C13;
Daniel Kang's avatar
Daniel Kang committed
123 124
  output[10] = temp1 + temp2;

125 126
  temp1 = step[11] * C15;
  temp2 = step[12] * C1;
Daniel Kang's avatar
Daniel Kang committed
127 128
  output[11] = temp1 - temp2;

129 130
  temp1 = step[11] * C1;
  temp2 = step[12] * C15;
Daniel Kang's avatar
Daniel Kang committed
131 132
  output[12] = temp2 + temp1;

133 134
  temp1 = step[10] * C13;
  temp2 = step[13] * C3;
Daniel Kang's avatar
Daniel Kang committed
135 136
  output[13] = temp2 - temp1;

137 138
  temp1 = step[ 9] * C5;
  temp2 = step[14] * C11;
Daniel Kang's avatar
Daniel Kang committed
139 140
  output[14] = temp2 + temp1;

141 142
  temp1 = step[ 8] * C9;
  temp2 = step[15] * C7;
Daniel Kang's avatar
Daniel Kang committed
143 144 145 146 147 148 149 150
  output[15] = temp2 - temp1;

  // step 3
  step[ 0] = output[0] + output[3];
  step[ 1] = output[1] + output[2];
  step[ 2] = output[1] - output[2];
  step[ 3] = output[0] - output[3];

151 152
  temp1 = output[4] * C14;
  temp2 = output[7] * C2;
Daniel Kang's avatar
Daniel Kang committed
153 154
  step[ 4] = temp1 + temp2;

155 156
  temp1 = output[5] * C10;
  temp2 = output[6] * C6;
Daniel Kang's avatar
Daniel Kang committed
157 158
  step[ 5] = temp1 + temp2;

159 160
  temp1 = output[5] * C6;
  temp2 = output[6] * C10;
Daniel Kang's avatar
Daniel Kang committed
161 162
  step[ 6] = temp2 - temp1;

163 164
  temp1 = output[4] * C2;
  temp2 = output[7] * C14;
Daniel Kang's avatar
Daniel Kang committed
165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180
  step[ 7] = temp2 - temp1;

  step[ 8] = output[ 8] + output[11];
  step[ 9] = output[ 9] + output[10];
  step[10] = output[ 9] - output[10];
  step[11] = output[ 8] - output[11];

  step[12] = output[12] + output[15];
  step[13] = output[13] + output[14];
  step[14] = output[13] - output[14];
  step[15] = output[12] - output[15];

  // step 4
  output[ 0] = (step[ 0] + step[ 1]);
  output[ 8] = (step[ 0] - step[ 1]);

181 182
  temp1 = step[2] * C12;
  temp2 = step[3] * C4;
Daniel Kang's avatar
Daniel Kang committed
183
  temp1 = temp1 + temp2;
184
  output[ 4] = 2*(temp1 * C8);
Daniel Kang's avatar
Daniel Kang committed
185

186 187
  temp1 = step[2] * C4;
  temp2 = step[3] * C12;
Daniel Kang's avatar
Daniel Kang committed
188
  temp1 = temp2 - temp1;
189
  output[12] = 2 * (temp1 * C8);
Daniel Kang's avatar
Daniel Kang committed
190

191 192
  output[ 2] = 2 * ((step[4] + step[ 5]) * C8);
  output[14] = 2 * ((step[7] - step[ 6]) * C8);
Daniel Kang's avatar
Daniel Kang committed
193 194 195 196 197 198 199 200 201

  temp1 = step[4] - step[5];
  temp2 = step[6] + step[7];
  output[ 6] = (temp1 + temp2);
  output[10] = (temp1 - temp2);

  intermediate[8] = step[8] + step[14];
  intermediate[9] = step[9] + step[15];

202 203
  temp1 = intermediate[8] * C12;
  temp2 = intermediate[9] * C4;
Daniel Kang's avatar
Daniel Kang committed
204
  temp1 = temp1 - temp2;
205
  output[3] = 2 * (temp1 * C8);
Daniel Kang's avatar
Daniel Kang committed
206

207 208
  temp1 = intermediate[8] * C4;
  temp2 = intermediate[9] * C12;
Daniel Kang's avatar
Daniel Kang committed
209
  temp1 = temp2 + temp1;
210
  output[13] = 2 * (temp1 * C8);
Daniel Kang's avatar
Daniel Kang committed
211

212
  output[ 9] = 2 * ((step[10] + step[11]) * C8);
Daniel Kang's avatar
Daniel Kang committed
213 214 215 216 217 218 219 220 221 222

  intermediate[11] = step[10] - step[11];
  intermediate[12] = step[12] + step[13];
  intermediate[13] = step[12] - step[13];
  intermediate[14] = step[ 8] - step[14];
  intermediate[15] = step[ 9] - step[15];

  output[15] = (intermediate[11] + intermediate[12]);
  output[ 1] = -(intermediate[11] - intermediate[12]);

223
  output[ 7] = 2 * (intermediate[13] * C8);
Daniel Kang's avatar
Daniel Kang committed
224

225 226
  temp1 = intermediate[14] * C12;
  temp2 = intermediate[15] * C4;
Daniel Kang's avatar
Daniel Kang committed
227
  temp1 = temp1 - temp2;
228
  output[11] = -2 * (temp1 * C8);
Daniel Kang's avatar
Daniel Kang committed
229

230 231
  temp1 = intermediate[14] * C4;
  temp2 = intermediate[15] * C12;
Daniel Kang's avatar
Daniel Kang committed
232
  temp1 = temp2 + temp1;
233
  output[ 5] = 2 * (temp1 * C8);
Daniel Kang's avatar
Daniel Kang committed
234 235
}

236
void reference_16x16_dct_2d(int16_t input[256], double output[256]) {
Daniel Kang's avatar
Daniel Kang committed
237 238 239 240
  // First transform columns
  for (int i = 0; i < 16; ++i) {
    double temp_in[16], temp_out[16];
    for (int j = 0; j < 16; ++j)
241
      temp_in[j] = input[j * 16 + i];
Daniel Kang's avatar
Daniel Kang committed
242 243
    butterfly_16x16_dct_1d(temp_in, temp_out);
    for (int j = 0; j < 16; ++j)
244
      output[j * 16 + i] = temp_out[j];
Daniel Kang's avatar
Daniel Kang committed
245 246 247 248 249
  }
  // Then transform rows
  for (int i = 0; i < 16; ++i) {
    double temp_in[16], temp_out[16];
    for (int j = 0; j < 16; ++j)
250
      temp_in[j] = output[j + i * 16];
Daniel Kang's avatar
Daniel Kang committed
251 252 253
    butterfly_16x16_dct_1d(temp_in, temp_out);
    // Scale by some magic number
    for (int j = 0; j < 16; ++j)
254
      output[j + i * 16] = temp_out[j]/2;
Daniel Kang's avatar
Daniel Kang committed
255 256 257
  }
}

258 259 260
typedef void (*FdctFunc)(const int16_t *in, tran_low_t *out, int stride);
typedef void (*IdctFunc)(const tran_low_t *in, uint8_t *out, int stride);
typedef void (*FhtFunc)(const int16_t *in, tran_low_t *out, int stride,
261
                        int tx_type);
262
typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
263
                        int tx_type);
Daniel Kang's avatar
Daniel Kang committed
264

265 266
typedef std::tr1::tuple<FdctFunc, IdctFunc, int, vpx_bit_depth_t> Dct16x16Param;
typedef std::tr1::tuple<FhtFunc, IhtFunc, int, vpx_bit_depth_t> Ht16x16Param;
267

268
void fdct16x16_ref(const int16_t *in, tran_low_t *out, int stride,
269
                   int /*tx_type*/) {
270
  vp9_fdct16x16_c(in, out, stride);
271 272
}

273
void idct16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
274
                   int /*tx_type*/) {
275 276 277
  vp9_idct16x16_256_add_c(in, dest, stride);
}

278 279
void fht16x16_ref(const int16_t *in, tran_low_t *out, int stride,
                  int tx_type) {
280
  vp9_fht16x16_c(in, out, stride, tx_type);
281 282
}

283 284
void iht16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
                  int tx_type) {
285 286 287
  vp9_iht16x16_256_add_c(in, dest, stride, tx_type);
}

288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315
#if CONFIG_VP9_HIGHBITDEPTH
void idct16x16_10(const tran_low_t *in, uint8_t *out, int stride) {
  vp9_high_idct16x16_256_add_c(in, out, stride, 10);
}

void idct16x16_12(const tran_low_t *in, uint8_t *out, int stride) {
  vp9_high_idct16x16_256_add_c(in, out, stride, 12);
}

void idct16x16_10_ref(const tran_low_t *in, uint8_t *out, int stride,
                      int tx_type) {
  idct16x16_10(in, out, stride);
}

void idct16x16_12_ref(const tran_low_t *in, uint8_t *out, int stride,
                      int tx_type) {
  idct16x16_12(in, out, stride);
}

void iht16x16_10(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
  vp9_high_iht16x16_256_add_c(in, out, stride, tx_type, 10);
}

void iht16x16_12(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
  vp9_high_iht16x16_256_add_c(in, out, stride, tx_type, 12);
}
#endif

316
class Trans16x16TestBase {
317
 public:
318
  virtual ~Trans16x16TestBase() {}
319

320
 protected:
321
  virtual void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) = 0;
322

323
  virtual void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) = 0;
324 325 326

  void RunAccuracyCheck() {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
327 328
    uint32_t max_error = 0;
    int64_t total_error = 0;
329 330 331
    const int count_test_block = 10000;
    for (int i = 0; i < count_test_block; ++i) {
      DECLARE_ALIGNED_ARRAY(16, int16_t, test_input_block, kNumCoeffs);
332
      DECLARE_ALIGNED_ARRAY(16, tran_low_t, test_temp_block, kNumCoeffs);
333 334
      DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
      DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
335 336 337 338
#if CONFIG_VP9_HIGHBITDEPTH
      DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
      DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
#endif
339

340
      // Initialize a test block with input range [-mask_, mask_].
341
      for (int j = 0; j < kNumCoeffs; ++j) {
342 343 344 345 346 347 348 349 350 351 352
        if (bit_depth_ == VPX_BITS_8) {
          src[j] = rnd.Rand8();
          dst[j] = rnd.Rand8();
          test_input_block[j] = src[j] - dst[j];
#if CONFIG_VP9_HIGHBITDEPTH
        } else {
          src16[j] = rnd.Rand16() & mask_;
          dst16[j] = rnd.Rand16() & mask_;
          test_input_block[j] = src16[j] - dst16[j];
#endif
        }
353 354
      }

355 356
      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block,
                                          test_temp_block, pitch_));
357 358 359 360 361 362 363 364 365
      if (bit_depth_ == VPX_BITS_8) {
        ASM_REGISTER_STATE_CHECK(
            RunInvTxfm(test_temp_block, dst, pitch_));
#if CONFIG_VP9_HIGHBITDEPTH
      } else {
        ASM_REGISTER_STATE_CHECK(
            RunInvTxfm(test_temp_block, CONVERT_TO_BYTEPTR(dst16), pitch_));
#endif
      }
366 367

      for (int j = 0; j < kNumCoeffs; ++j) {
368 369 370 371
#if CONFIG_VP9_HIGHBITDEPTH
        const uint32_t diff =
            bit_depth_ == VPX_BITS_8 ?  dst[j] - src[j] : dst16[j] - src16[j];
#else
372
        const uint32_t diff = dst[j] - src[j];
373
#endif
374
        const uint32_t error = diff * diff;
375 376 377 378
        if (max_error < error)
          max_error = error;
        total_error += error;
      }
Scott LaVarnway's avatar
Scott LaVarnway committed
379
    }
Daniel Kang's avatar
Daniel Kang committed
380

381
    EXPECT_GE(1u  << 2 * (bit_depth_ - 8), max_error)
382 383
        << "Error: 16x16 FHT/IHT has an individual round trip error > 1";

384
    EXPECT_GE(count_test_block << 2 * (bit_depth_ - 8), total_error)
385
        << "Error: 16x16 FHT/IHT has average round trip error > 1 per block";
386 387
  }

388
  void RunCoeffCheck() {
389 390
    ACMRandom rnd(ACMRandom::DeterministicSeed());
    const int count_test_block = 1000;
391
    DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
392 393
    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_block, kNumCoeffs);
394

395
    for (int i = 0; i < count_test_block; ++i) {
396
      // Initialize a test block with input range [-mask_, mask_].
397
      for (int j = 0; j < kNumCoeffs; ++j)
398
        input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
399 400

      fwd_txfm_ref(input_block, output_ref_block, pitch_, tx_type_);
401
      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
402 403 404 405 406 407 408 409 410 411 412 413

      // The minimum quant value is 4.
      for (int j = 0; j < kNumCoeffs; ++j)
        EXPECT_EQ(output_block[j], output_ref_block[j]);
    }
  }

  void RunMemCheck() {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
    const int count_test_block = 1000;
    DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kNumCoeffs);
414 415
    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_block, kNumCoeffs);
Scott LaVarnway's avatar
Scott LaVarnway committed
416

417
    for (int i = 0; i < count_test_block; ++i) {
418
      // Initialize a test block with input range [-mask_, mask_].
419
      for (int j = 0; j < kNumCoeffs; ++j) {
420 421
        input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
        input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
422
      }
423
      if (i == 0) {
424
        for (int j = 0; j < kNumCoeffs; ++j)
425
          input_extreme_block[j] = mask_;
426
      } else if (i == 1) {
427
        for (int j = 0; j < kNumCoeffs; ++j)
428
          input_extreme_block[j] = -mask_;
429
      }
430

431
      fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
432 433
      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_extreme_block,
                                          output_block, pitch_));
434 435 436

      // The minimum quant value is 4.
      for (int j = 0; j < kNumCoeffs; ++j) {
437
        EXPECT_EQ(output_block[j], output_ref_block[j]);
438
        EXPECT_GE(4 * DCT_MAX_VALUE << (bit_depth_ - 8), abs(output_block[j]))
439 440
            << "Error: 16x16 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
      }
441
    }
442 443
  }

444 445
  void RunQuantCheck(int dc_thred, int ac_thred) {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
446
    const int count_test_block = 100000;
447 448
    DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kNumCoeffs);
449
    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
450 451 452

    DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, uint8_t, ref, kNumCoeffs);
453 454 455 456
#if CONFIG_VP9_HIGHBITDEPTH
    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, uint16_t, ref16, kNumCoeffs);
#endif
457 458

    for (int i = 0; i < count_test_block; ++i) {
459
      // Initialize a test block with input range [-mask_, mask_].
460
      for (int j = 0; j < kNumCoeffs; ++j) {
461 462 463 464 465
        if (bit_depth_ == VPX_BITS_8)
          input_block[j] = rnd.Rand8() - rnd.Rand8();
        else
          input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
        input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
466 467 468
      }
      if (i == 0)
        for (int j = 0; j < kNumCoeffs; ++j)
469
          input_extreme_block[j] = mask_;
470 471
      if (i == 1)
        for (int j = 0; j < kNumCoeffs; ++j)
472
          input_extreme_block[j] = -mask_;
473 474 475 476 477 478

      fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);

      // clear reconstructed pixel buffers
      vpx_memset(dst, 0, kNumCoeffs * sizeof(uint8_t));
      vpx_memset(ref, 0, kNumCoeffs * sizeof(uint8_t));
479 480 481 482
#if CONFIG_VP9_HIGHBITDEPTH
      vpx_memset(dst16, 0, kNumCoeffs * sizeof(uint16_t));
      vpx_memset(ref16, 0, kNumCoeffs * sizeof(uint16_t));
#endif
483 484 485 486 487

      // quantization with maximum allowed step sizes
      output_ref_block[0] = (output_ref_block[0] / dc_thred) * dc_thred;
      for (int j = 1; j < kNumCoeffs; ++j)
        output_ref_block[j] = (output_ref_block[j] / ac_thred) * ac_thred;
488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507
      if (bit_depth_ == VPX_BITS_8) {
        inv_txfm_ref(output_ref_block, ref, pitch_, tx_type_);
        ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block, dst, pitch_));
#if CONFIG_VP9_HIGHBITDEPTH
      } else {
        inv_txfm_ref(output_ref_block, CONVERT_TO_BYTEPTR(ref16), pitch_,
                     tx_type_);
        ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block,
                                            CONVERT_TO_BYTEPTR(dst16), pitch_));
#endif
      }
      if (bit_depth_ == VPX_BITS_8) {
        for (int j = 0; j < kNumCoeffs; ++j)
          EXPECT_EQ(ref[j], dst[j]);
#if CONFIG_VP9_HIGHBITDEPTH
      } else {
        for (int j = 0; j < kNumCoeffs; ++j)
          EXPECT_EQ(ref16[j], dst16[j]);
#endif
      }
508 509 510
    }
  }

511 512 513
  void RunInvAccuracyCheck() {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
    const int count_test_block = 1000;
514
    DECLARE_ALIGNED_ARRAY(16, int16_t, in, kNumCoeffs);
515
    DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff, kNumCoeffs);
516 517
    DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
518 519 520 521
#if CONFIG_VP9_HIGHBITDEPTH
    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
#endif
Daniel Kang's avatar
Daniel Kang committed
522

523 524
    for (int i = 0; i < count_test_block; ++i) {
      double out_r[kNumCoeffs];
Daniel Kang's avatar
Daniel Kang committed
525

526 527
      // Initialize a test block with input range [-255, 255].
      for (int j = 0; j < kNumCoeffs; ++j) {
528 529 530 531 532 533 534 535 536 537 538
        if (bit_depth_ == VPX_BITS_8) {
          src[j] = rnd.Rand8();
          dst[j] = rnd.Rand8();
          in[j] = src[j] - dst[j];
#if CONFIG_VP9_HIGHBITDEPTH
        } else {
          src16[j] = rnd.Rand16() & mask_;
          dst16[j] = rnd.Rand16() & mask_;
          in[j] = src16[j] - dst16[j];
#endif
        }
539 540 541 542 543 544
      }

      reference_16x16_dct_2d(in, out_r);
      for (int j = 0; j < kNumCoeffs; ++j)
        coeff[j] = round(out_r[j]);

545 546 547 548 549 550 551 552
      if (bit_depth_ == VPX_BITS_8) {
        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, 16));
#if CONFIG_VP9_HIGHBITDEPTH
      } else {
        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16),
                                            16));
#endif
      }
553 554

      for (int j = 0; j < kNumCoeffs; ++j) {
555 556 557 558
#if CONFIG_VP9_HIGHBITDEPTH
        const uint32_t diff =
            bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
#else
559
        const uint32_t diff = dst[j] - src[j];
560
#endif
561 562
        const uint32_t error = diff * diff;
        EXPECT_GE(1u, error)
563 564 565
            << "Error: 16x16 IDCT has error " << error
            << " at index " << j;
      }
Daniel Kang's avatar
Daniel Kang committed
566 567
    }
  }
568 569
  int pitch_;
  int tx_type_;
570 571
  vpx_bit_depth_t bit_depth_;
  int mask_;
572 573
  FhtFunc fwd_txfm_ref;
  IhtFunc inv_txfm_ref;
574
};
Daniel Kang's avatar
Daniel Kang committed
575

576 577
class Trans16x16DCT
    : public Trans16x16TestBase,
578
      public ::testing::TestWithParam<Dct16x16Param> {
579 580
 public:
  virtual ~Trans16x16DCT() {}
Daniel Kang's avatar
Daniel Kang committed
581

582 583 584 585
  virtual void SetUp() {
    fwd_txfm_ = GET_PARAM(0);
    inv_txfm_ = GET_PARAM(1);
    tx_type_  = GET_PARAM(2);
586
    bit_depth_ = GET_PARAM(3);
587
    pitch_    = 16;
588
    fwd_txfm_ref = fdct16x16_ref;
589
    inv_txfm_ref = idct16x16_ref;
590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605
    mask_ = (1 << bit_depth_) - 1;
#if CONFIG_VP9_HIGHBITDEPTH
    switch (bit_depth_) {
      case 10:
        inv_txfm_ref = idct16x16_10_ref;
        break;
      case 12:
        inv_txfm_ref = idct16x16_12_ref;
        break;
      default:
        inv_txfm_ref = idct16x16_ref;
        break;
    }
#else
    inv_txfm_ref = idct16x16_ref;
#endif
606 607
  }
  virtual void TearDown() { libvpx_test::ClearSystemState(); }
Daniel Kang's avatar
Daniel Kang committed
608

609
 protected:
610
  void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
611 612
    fwd_txfm_(in, out, stride);
  }
613
  void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
614
    inv_txfm_(out, dst, stride);
Daniel Kang's avatar
Daniel Kang committed
615
  }
616

617 618
  FdctFunc fwd_txfm_;
  IdctFunc inv_txfm_;
619 620 621 622
};

TEST_P(Trans16x16DCT, AccuracyCheck) {
  RunAccuracyCheck();
Daniel Kang's avatar
Daniel Kang committed
623
}
624

625 626 627 628 629 630
TEST_P(Trans16x16DCT, CoeffCheck) {
  RunCoeffCheck();
}

TEST_P(Trans16x16DCT, MemCheck) {
  RunMemCheck();
631 632
}

633 634 635 636 637 638
TEST_P(Trans16x16DCT, QuantCheck) {
  // Use maximally allowed quantization step sizes for DC and AC
  // coefficients respectively.
  RunQuantCheck(1336, 1828);
}

639 640 641 642
TEST_P(Trans16x16DCT, InvAccuracyCheck) {
  RunInvAccuracyCheck();
}

643 644
class Trans16x16HT
    : public Trans16x16TestBase,
645
      public ::testing::TestWithParam<Ht16x16Param> {
646 647 648 649 650 651 652
 public:
  virtual ~Trans16x16HT() {}

  virtual void SetUp() {
    fwd_txfm_ = GET_PARAM(0);
    inv_txfm_ = GET_PARAM(1);
    tx_type_  = GET_PARAM(2);
653
    bit_depth_ = GET_PARAM(3);
654 655
    pitch_    = 16;
    fwd_txfm_ref = fht16x16_ref;
656
    inv_txfm_ref = iht16x16_ref;
657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672
    mask_ = (1 << bit_depth_) - 1;
#if CONFIG_VP9_HIGHBITDEPTH
    switch (bit_depth_) {
      case VPX_BITS_10:
        inv_txfm_ref = iht16x16_10;
        break;
      case VPX_BITS_12:
        inv_txfm_ref = iht16x16_12;
        break;
      default:
        inv_txfm_ref = iht16x16_ref;
        break;
    }
#else
    inv_txfm_ref = iht16x16_ref;
#endif
673
  }
674 675 676
  virtual void TearDown() { libvpx_test::ClearSystemState(); }

 protected:
677
  void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
678
    fwd_txfm_(in, out, stride, tx_type_);
679
  }
680
  void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
681
    inv_txfm_(out, dst, stride, tx_type_);
682 683
  }

684 685
  FhtFunc fwd_txfm_;
  IhtFunc inv_txfm_;
686 687 688 689 690 691
};

TEST_P(Trans16x16HT, AccuracyCheck) {
  RunAccuracyCheck();
}

692 693 694 695 696 697
TEST_P(Trans16x16HT, CoeffCheck) {
  RunCoeffCheck();
}

TEST_P(Trans16x16HT, MemCheck) {
  RunMemCheck();
698 699
}

700 701 702
TEST_P(Trans16x16HT, QuantCheck) {
  // The encoder skips any non-DC intra prediction modes,
  // when the quantization step size goes beyond 988.
703
  RunQuantCheck(429, 729);
704 705
}

706 707
using std::tr1::make_tuple;

708 709 710 711 712 713 714 715
#if CONFIG_VP9_HIGHBITDEPTH
INSTANTIATE_TEST_CASE_P(
    C, Trans16x16DCT,
    ::testing::Values(
        make_tuple(&vp9_high_fdct16x16_c, &idct16x16_10, 0, VPX_BITS_10),
        make_tuple(&vp9_high_fdct16x16_c, &idct16x16_12, 0, VPX_BITS_12),
        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_c, 0, VPX_BITS_8)));
#else
716 717 718
INSTANTIATE_TEST_CASE_P(
    C, Trans16x16DCT,
    ::testing::Values(
719 720 721 722
        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_c, 0, VPX_BITS_8)));
#endif

#if CONFIG_VP9_HIGHBITDEPTH
723 724 725
INSTANTIATE_TEST_CASE_P(
    C, Trans16x16HT,
    ::testing::Values(
726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746
        make_tuple(&vp9_high_fht16x16_c, &iht16x16_10, 0, VPX_BITS_10),
        make_tuple(&vp9_high_fht16x16_c, &iht16x16_10, 1, VPX_BITS_10),
        make_tuple(&vp9_high_fht16x16_c, &iht16x16_10, 2, VPX_BITS_10),
        make_tuple(&vp9_high_fht16x16_c, &iht16x16_10, 3, VPX_BITS_10),
        make_tuple(&vp9_high_fht16x16_c, &iht16x16_12, 0, VPX_BITS_12),
        make_tuple(&vp9_high_fht16x16_c, &iht16x16_12, 1, VPX_BITS_12),
        make_tuple(&vp9_high_fht16x16_c, &iht16x16_12, 2, VPX_BITS_12),
        make_tuple(&vp9_high_fht16x16_c, &iht16x16_12, 3, VPX_BITS_12),
        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
#else
INSTANTIATE_TEST_CASE_P(
    C, Trans16x16HT,
    ::testing::Values(
        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
#endif
747

748
#if HAVE_NEON_ASM && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
James Zern's avatar
James Zern committed
749 750 751 752
INSTANTIATE_TEST_CASE_P(
    NEON, Trans16x16DCT,
    ::testing::Values(
        make_tuple(&vp9_fdct16x16_c,
753
                   &vp9_idct16x16_256_add_neon, 0, VPX_BITS_8)));
James Zern's avatar
James Zern committed
754 755
#endif

756
#if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
757 758 759
INSTANTIATE_TEST_CASE_P(
    SSE2, Trans16x16DCT,
    ::testing::Values(
760
        make_tuple(&vp9_fdct16x16_sse2,
761
                   &vp9_idct16x16_256_add_sse2, 0, VPX_BITS_8)));
762 763 764
INSTANTIATE_TEST_CASE_P(
    SSE2, Trans16x16HT,
    ::testing::Values(
765 766 767 768 769 770 771 772
        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 0,
                   VPX_BITS_8),
        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 1,
                   VPX_BITS_8),
        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 2,
                   VPX_BITS_8),
        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 3,
                   VPX_BITS_8)));
773
#endif
774

775
#if HAVE_SSSE3 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
776 777 778
INSTANTIATE_TEST_CASE_P(
    SSSE3, Trans16x16DCT,
    ::testing::Values(
779 780
        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_ssse3, 0,
                   VPX_BITS_8)));
781
#endif
Daniel Kang's avatar
Daniel Kang committed
782
}  // namespace