dct16x16_test.cc 29.7 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
typedef std::tr1::tuple<IdctFunc, IdctFunc, int, vpx_bit_depth_t>
    Idct16x16Param;
269

270
void fdct16x16_ref(const int16_t *in, tran_low_t *out, int stride,
271
                   int /*tx_type*/) {
272
  vp9_fdct16x16_c(in, out, stride);
273 274
}

275
void idct16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
276
                   int /*tx_type*/) {
277 278 279
  vp9_idct16x16_256_add_c(in, dest, stride);
}

280 281
void fht16x16_ref(const int16_t *in, tran_low_t *out, int stride,
                  int tx_type) {
282
  vp9_fht16x16_c(in, out, stride, tx_type);
283 284
}

285 286
void iht16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
                  int tx_type) {
287 288 289
  vp9_iht16x16_256_add_c(in, dest, stride, tx_type);
}

290 291
#if CONFIG_VP9_HIGHBITDEPTH
void idct16x16_10(const tran_low_t *in, uint8_t *out, int stride) {
292
  vp9_highbd_idct16x16_256_add_c(in, out, stride, 10);
293 294 295
}

void idct16x16_12(const tran_low_t *in, uint8_t *out, int stride) {
296
  vp9_highbd_idct16x16_256_add_c(in, out, stride, 12);
297 298 299 300 301 302 303 304 305 306 307 308 309
}

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) {
310
  vp9_highbd_iht16x16_256_add_c(in, out, stride, tx_type, 10);
311 312 313
}

void iht16x16_12(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
314
  vp9_highbd_iht16x16_256_add_c(in, out, stride, tx_type, 12);
315
}
316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342

void idct16x16_10_add_10_c(const tran_low_t *in, uint8_t *out, int stride) {
  vp9_highbd_idct16x16_10_add_c(in, out, stride, 10);
}

void idct16x16_10_add_12_c(const tran_low_t *in, uint8_t *out, int stride) {
  vp9_highbd_idct16x16_10_add_c(in, out, stride, 12);
}

#if HAVE_SSE2
void idct16x16_256_add_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
  vp9_highbd_idct16x16_256_add_sse2(in, out, stride, 10);
}

void idct16x16_256_add_12_sse2(const tran_low_t *in, uint8_t *out, int stride) {
  vp9_highbd_idct16x16_256_add_sse2(in, out, stride, 12);
}

void idct16x16_10_add_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
  vp9_highbd_idct16x16_10_add_sse2(in, out, stride, 10);
}

void idct16x16_10_add_12_sse2(const tran_low_t *in, uint8_t *out, int stride) {
  vp9_highbd_idct16x16_10_add_sse2(in, out, stride, 12);
}
#endif  // HAVE_SSE2
#endif  // CONFIG_VP9_HIGHBITDEPTH
343

344
class Trans16x16TestBase {
345
 public:
346
  virtual ~Trans16x16TestBase() {}
347

348
 protected:
349
  virtual void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) = 0;
350

351
  virtual void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) = 0;
352 353 354

  void RunAccuracyCheck() {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
355 356
    uint32_t max_error = 0;
    int64_t total_error = 0;
357 358 359
    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);
360
      DECLARE_ALIGNED_ARRAY(16, tran_low_t, test_temp_block, kNumCoeffs);
361 362
      DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
      DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
363 364 365 366
#if CONFIG_VP9_HIGHBITDEPTH
      DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
      DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
#endif
367

368
      // Initialize a test block with input range [-mask_, mask_].
369
      for (int j = 0; j < kNumCoeffs; ++j) {
370 371 372 373 374 375 376 377 378 379 380
        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
        }
381 382
      }

383 384
      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block,
                                          test_temp_block, pitch_));
385 386 387 388 389 390 391 392 393
      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
      }
394 395

      for (int j = 0; j < kNumCoeffs; ++j) {
396 397 398 399
#if CONFIG_VP9_HIGHBITDEPTH
        const uint32_t diff =
            bit_depth_ == VPX_BITS_8 ?  dst[j] - src[j] : dst16[j] - src16[j];
#else
400
        const uint32_t diff = dst[j] - src[j];
401
#endif
402
        const uint32_t error = diff * diff;
403 404 405 406
        if (max_error < error)
          max_error = error;
        total_error += error;
      }
Scott LaVarnway's avatar
Scott LaVarnway committed
407
    }
Daniel Kang's avatar
Daniel Kang committed
408

409
    EXPECT_GE(1u  << 2 * (bit_depth_ - 8), max_error)
410 411
        << "Error: 16x16 FHT/IHT has an individual round trip error > 1";

412
    EXPECT_GE(count_test_block << 2 * (bit_depth_ - 8), total_error)
413
        << "Error: 16x16 FHT/IHT has average round trip error > 1 per block";
414 415
  }

416
  void RunCoeffCheck() {
417 418
    ACMRandom rnd(ACMRandom::DeterministicSeed());
    const int count_test_block = 1000;
419
    DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
420 421
    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_block, kNumCoeffs);
422

423
    for (int i = 0; i < count_test_block; ++i) {
424
      // Initialize a test block with input range [-mask_, mask_].
425
      for (int j = 0; j < kNumCoeffs; ++j)
426
        input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
427 428

      fwd_txfm_ref(input_block, output_ref_block, pitch_, tx_type_);
429
      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
430 431 432 433 434 435 436 437 438 439 440 441

      // 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);
442 443
    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
444

445
    for (int i = 0; i < count_test_block; ++i) {
446
      // Initialize a test block with input range [-mask_, mask_].
447
      for (int j = 0; j < kNumCoeffs; ++j) {
448 449
        input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
        input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
450
      }
451
      if (i == 0) {
452
        for (int j = 0; j < kNumCoeffs; ++j)
453
          input_extreme_block[j] = mask_;
454
      } else if (i == 1) {
455
        for (int j = 0; j < kNumCoeffs; ++j)
456
          input_extreme_block[j] = -mask_;
457
      }
458

459
      fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
460 461
      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_extreme_block,
                                          output_block, pitch_));
462 463 464

      // The minimum quant value is 4.
      for (int j = 0; j < kNumCoeffs; ++j) {
465
        EXPECT_EQ(output_block[j], output_ref_block[j]);
466
        EXPECT_GE(4 * DCT_MAX_VALUE << (bit_depth_ - 8), abs(output_block[j]))
467 468
            << "Error: 16x16 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
      }
469
    }
470 471
  }

472 473
  void RunQuantCheck(int dc_thred, int ac_thred) {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
474
    const int count_test_block = 100000;
475 476
    DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kNumCoeffs);
477
    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
478 479 480

    DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, uint8_t, ref, kNumCoeffs);
481 482 483 484
#if CONFIG_VP9_HIGHBITDEPTH
    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, uint16_t, ref16, kNumCoeffs);
#endif
485 486

    for (int i = 0; i < count_test_block; ++i) {
487
      // Initialize a test block with input range [-mask_, mask_].
488
      for (int j = 0; j < kNumCoeffs; ++j) {
489 490 491 492 493
        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_;
494 495 496
      }
      if (i == 0)
        for (int j = 0; j < kNumCoeffs; ++j)
497
          input_extreme_block[j] = mask_;
498 499
      if (i == 1)
        for (int j = 0; j < kNumCoeffs; ++j)
500
          input_extreme_block[j] = -mask_;
501 502 503 504

      fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);

      // clear reconstructed pixel buffers
James Zern's avatar
James Zern committed
505 506
      memset(dst, 0, kNumCoeffs * sizeof(uint8_t));
      memset(ref, 0, kNumCoeffs * sizeof(uint8_t));
507
#if CONFIG_VP9_HIGHBITDEPTH
James Zern's avatar
James Zern committed
508 509
      memset(dst16, 0, kNumCoeffs * sizeof(uint16_t));
      memset(ref16, 0, kNumCoeffs * sizeof(uint16_t));
510
#endif
511 512 513 514 515

      // 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;
516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535
      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
      }
536 537 538
    }
  }

539 540 541
  void RunInvAccuracyCheck() {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
    const int count_test_block = 1000;
542
    DECLARE_ALIGNED_ARRAY(16, int16_t, in, kNumCoeffs);
543
    DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff, kNumCoeffs);
544 545
    DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
546 547 548
#if CONFIG_VP9_HIGHBITDEPTH
    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
549
#endif  // CONFIG_VP9_HIGHBITDEPTH
Daniel Kang's avatar
Daniel Kang committed
550

551 552
    for (int i = 0; i < count_test_block; ++i) {
      double out_r[kNumCoeffs];
Daniel Kang's avatar
Daniel Kang committed
553

554 555
      // Initialize a test block with input range [-255, 255].
      for (int j = 0; j < kNumCoeffs; ++j) {
556 557 558 559 560 561 562 563 564
        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];
565
#endif  // CONFIG_VP9_HIGHBITDEPTH
566
        }
567 568 569 570
      }

      reference_16x16_dct_2d(in, out_r);
      for (int j = 0; j < kNumCoeffs; ++j)
571
        coeff[j] = static_cast<tran_low_t>(round(out_r[j]));
572

573 574 575 576 577 578
      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));
579
#endif  // CONFIG_VP9_HIGHBITDEPTH
580
      }
581 582

      for (int j = 0; j < kNumCoeffs; ++j) {
583 584 585 586
#if CONFIG_VP9_HIGHBITDEPTH
        const uint32_t diff =
            bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
#else
587
        const uint32_t diff = dst[j] - src[j];
588
#endif  // CONFIG_VP9_HIGHBITDEPTH
589 590
        const uint32_t error = diff * diff;
        EXPECT_GE(1u, error)
591 592 593
            << "Error: 16x16 IDCT has error " << error
            << " at index " << j;
      }
Daniel Kang's avatar
Daniel Kang committed
594 595
    }
  }
596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653

  void CompareInvReference(IdctFunc ref_txfm, int thresh) {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
    const int count_test_block = 10000;
    const int eob = 10;
    const int16_t *scan = vp9_default_scan_orders[TX_16X16].scan;
    DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, uint8_t, ref, kNumCoeffs);
#if CONFIG_VP9_HIGHBITDEPTH
    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
    DECLARE_ALIGNED_ARRAY(16, uint16_t, ref16, kNumCoeffs);
#endif  // CONFIG_VP9_HIGHBITDEPTH

    for (int i = 0; i < count_test_block; ++i) {
      for (int j = 0; j < kNumCoeffs; ++j) {
        if (j < eob) {
          // Random values less than the threshold, either positive or negative
          coeff[scan[j]] = rnd(thresh) * (1 - 2 * (i % 2));
        } else {
          coeff[scan[j]] = 0;
        }
        if (bit_depth_ == VPX_BITS_8) {
          dst[j] = 0;
          ref[j] = 0;
#if CONFIG_VP9_HIGHBITDEPTH
        } else {
          dst16[j] = 0;
          ref16[j] = 0;
#endif  // CONFIG_VP9_HIGHBITDEPTH
        }
      }
      if (bit_depth_ == VPX_BITS_8) {
        ref_txfm(coeff, ref, pitch_);
        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
      } else {
#if CONFIG_VP9_HIGHBITDEPTH
        ref_txfm(coeff, CONVERT_TO_BYTEPTR(ref16), pitch_);
        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16),
                                 pitch_));
#endif  // CONFIG_VP9_HIGHBITDEPTH
      }

      for (int j = 0; j < kNumCoeffs; ++j) {
#if CONFIG_VP9_HIGHBITDEPTH
        const uint32_t diff =
            bit_depth_ == VPX_BITS_8 ? dst[j] - ref[j] : dst16[j] - ref16[j];
#else
        const uint32_t diff = dst[j] - ref[j];
#endif  // CONFIG_VP9_HIGHBITDEPTH
        const uint32_t error = diff * diff;
        EXPECT_EQ(0u, error)
            << "Error: 16x16 IDCT Comparison has error " << error
            << " at index " << j;
      }
    }
  }

654 655
  int pitch_;
  int tx_type_;
656 657
  vpx_bit_depth_t bit_depth_;
  int mask_;
658 659
  FhtFunc fwd_txfm_ref;
  IhtFunc inv_txfm_ref;
660
};
Daniel Kang's avatar
Daniel Kang committed
661

662 663
class Trans16x16DCT
    : public Trans16x16TestBase,
664
      public ::testing::TestWithParam<Dct16x16Param> {
665 666
 public:
  virtual ~Trans16x16DCT() {}
Daniel Kang's avatar
Daniel Kang committed
667

668 669 670 671
  virtual void SetUp() {
    fwd_txfm_ = GET_PARAM(0);
    inv_txfm_ = GET_PARAM(1);
    tx_type_  = GET_PARAM(2);
672
    bit_depth_ = GET_PARAM(3);
673
    pitch_    = 16;
674
    fwd_txfm_ref = fdct16x16_ref;
675
    inv_txfm_ref = idct16x16_ref;
676 677 678
    mask_ = (1 << bit_depth_) - 1;
#if CONFIG_VP9_HIGHBITDEPTH
    switch (bit_depth_) {
679
      case VPX_BITS_10:
680 681
        inv_txfm_ref = idct16x16_10_ref;
        break;
682
      case VPX_BITS_12:
683 684 685 686 687 688 689 690 691
        inv_txfm_ref = idct16x16_12_ref;
        break;
      default:
        inv_txfm_ref = idct16x16_ref;
        break;
    }
#else
    inv_txfm_ref = idct16x16_ref;
#endif
692 693
  }
  virtual void TearDown() { libvpx_test::ClearSystemState(); }
Daniel Kang's avatar
Daniel Kang committed
694

695
 protected:
696
  void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
697 698
    fwd_txfm_(in, out, stride);
  }
699
  void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
700
    inv_txfm_(out, dst, stride);
Daniel Kang's avatar
Daniel Kang committed
701
  }
702

703 704
  FdctFunc fwd_txfm_;
  IdctFunc inv_txfm_;
705 706 707 708
};

TEST_P(Trans16x16DCT, AccuracyCheck) {
  RunAccuracyCheck();
Daniel Kang's avatar
Daniel Kang committed
709
}
710

711 712 713 714 715 716
TEST_P(Trans16x16DCT, CoeffCheck) {
  RunCoeffCheck();
}

TEST_P(Trans16x16DCT, MemCheck) {
  RunMemCheck();
717 718
}

719 720 721 722 723 724
TEST_P(Trans16x16DCT, QuantCheck) {
  // Use maximally allowed quantization step sizes for DC and AC
  // coefficients respectively.
  RunQuantCheck(1336, 1828);
}

725 726 727 728
TEST_P(Trans16x16DCT, InvAccuracyCheck) {
  RunInvAccuracyCheck();
}

729 730
class Trans16x16HT
    : public Trans16x16TestBase,
731
      public ::testing::TestWithParam<Ht16x16Param> {
732 733 734 735 736 737 738
 public:
  virtual ~Trans16x16HT() {}

  virtual void SetUp() {
    fwd_txfm_ = GET_PARAM(0);
    inv_txfm_ = GET_PARAM(1);
    tx_type_  = GET_PARAM(2);
739
    bit_depth_ = GET_PARAM(3);
740 741
    pitch_    = 16;
    fwd_txfm_ref = fht16x16_ref;
742
    inv_txfm_ref = iht16x16_ref;
743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758
    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
759
  }
760 761 762
  virtual void TearDown() { libvpx_test::ClearSystemState(); }

 protected:
763
  void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
764
    fwd_txfm_(in, out, stride, tx_type_);
765
  }
766
  void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
767
    inv_txfm_(out, dst, stride, tx_type_);
768 769
  }

770 771
  FhtFunc fwd_txfm_;
  IhtFunc inv_txfm_;
772 773 774 775 776 777
};

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

778 779 780 781 782 783
TEST_P(Trans16x16HT, CoeffCheck) {
  RunCoeffCheck();
}

TEST_P(Trans16x16HT, MemCheck) {
  RunMemCheck();
784 785
}

786 787 788
TEST_P(Trans16x16HT, QuantCheck) {
  // The encoder skips any non-DC intra prediction modes,
  // when the quantization step size goes beyond 988.
789
  RunQuantCheck(429, 729);
790 791
}

792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822
class InvTrans16x16DCT
    : public Trans16x16TestBase,
      public ::testing::TestWithParam<Idct16x16Param> {
 public:
  virtual ~InvTrans16x16DCT() {}

  virtual void SetUp() {
    ref_txfm_ = GET_PARAM(0);
    inv_txfm_ = GET_PARAM(1);
    thresh_ = GET_PARAM(2);
    bit_depth_ = GET_PARAM(3);
    pitch_ = 16;
    mask_ = (1 << bit_depth_) - 1;
}
  virtual void TearDown() { libvpx_test::ClearSystemState(); }

 protected:
  void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {}
  void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
    inv_txfm_(out, dst, stride);
  }

  IdctFunc ref_txfm_;
  IdctFunc inv_txfm_;
  int thresh_;
};

TEST_P(InvTrans16x16DCT, CompareReference) {
  CompareInvReference(ref_txfm_, thresh_);
}

823 824
using std::tr1::make_tuple;

825 826 827 828
#if CONFIG_VP9_HIGHBITDEPTH
INSTANTIATE_TEST_CASE_P(
    C, Trans16x16DCT,
    ::testing::Values(
829 830
        make_tuple(&vp9_highbd_fdct16x16_c, &idct16x16_10, 0, VPX_BITS_10),
        make_tuple(&vp9_highbd_fdct16x16_c, &idct16x16_12, 0, VPX_BITS_12),
831 832
        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_c, 0, VPX_BITS_8)));
#else
833 834 835
INSTANTIATE_TEST_CASE_P(
    C, Trans16x16DCT,
    ::testing::Values(
836
        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_c, 0, VPX_BITS_8)));
837
#endif  // CONFIG_VP9_HIGHBITDEPTH
838 839

#if CONFIG_VP9_HIGHBITDEPTH
840 841 842
INSTANTIATE_TEST_CASE_P(
    C, Trans16x16HT,
    ::testing::Values(
843 844 845 846 847 848 849 850
        make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 0, VPX_BITS_10),
        make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 1, VPX_BITS_10),
        make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 2, VPX_BITS_10),
        make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 3, VPX_BITS_10),
        make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 0, VPX_BITS_12),
        make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 1, VPX_BITS_12),
        make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 2, VPX_BITS_12),
        make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 3, VPX_BITS_12),
851 852 853 854 855 856 857 858 859 860 861 862
        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)));
863
#endif  // CONFIG_VP9_HIGHBITDEPTH
864

865
#if HAVE_NEON_ASM && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
James Zern's avatar
James Zern committed
866 867 868 869
INSTANTIATE_TEST_CASE_P(
    NEON, Trans16x16DCT,
    ::testing::Values(
        make_tuple(&vp9_fdct16x16_c,
870
                   &vp9_idct16x16_256_add_neon, 0, VPX_BITS_8)));
James Zern's avatar
James Zern committed
871 872
#endif

873
#if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
874 875 876
INSTANTIATE_TEST_CASE_P(
    SSE2, Trans16x16DCT,
    ::testing::Values(
877
        make_tuple(&vp9_fdct16x16_sse2,
878
                   &vp9_idct16x16_256_add_sse2, 0, VPX_BITS_8)));
879 880 881
INSTANTIATE_TEST_CASE_P(
    SSE2, Trans16x16HT,
    ::testing::Values(
882 883 884 885 886 887 888 889
        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)));
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
#endif  // HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE

#if HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
INSTANTIATE_TEST_CASE_P(
    SSE2, Trans16x16DCT,
    ::testing::Values(
        make_tuple(&vp9_highbd_fdct16x16_sse2,
                   &idct16x16_10, 0, VPX_BITS_10),
        make_tuple(&vp9_highbd_fdct16x16_c,
                   &idct16x16_256_add_10_sse2, 0, VPX_BITS_10),
        make_tuple(&vp9_highbd_fdct16x16_sse2,
                   &idct16x16_12, 0, VPX_BITS_12),
        make_tuple(&vp9_highbd_fdct16x16_c,
                   &idct16x16_256_add_12_sse2, 0, VPX_BITS_12),
        make_tuple(&vp9_fdct16x16_sse2,
                   &vp9_idct16x16_256_add_c, 0, VPX_BITS_8)));
INSTANTIATE_TEST_CASE_P(
    SSE2, Trans16x16HT,
    ::testing::Values(
        make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_10, 0, VPX_BITS_10),
        make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_10, 1, VPX_BITS_10),
        make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_10, 2, VPX_BITS_10),
        make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_10, 3, VPX_BITS_10),
        make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_12, 0, VPX_BITS_12),
        make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_12, 1, VPX_BITS_12),
        make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_12, 2, VPX_BITS_12),
        make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_12, 3, VPX_BITS_12),
        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 3,
                   VPX_BITS_8)));
// Optimizations take effect at a threshold of 3155, so we use a value close to
// that to test both branches.
INSTANTIATE_TEST_CASE_P(
    SSE2, InvTrans16x16DCT,
    ::testing::Values(
        make_tuple(&idct16x16_10_add_10_c,
                   &idct16x16_10_add_10_sse2, 3167, VPX_BITS_10),
        make_tuple(&idct16x16_10,
                   &idct16x16_256_add_10_sse2, 3167, VPX_BITS_10),
        make_tuple(&idct16x16_10_add_12_c,
                   &idct16x16_10_add_12_sse2, 3167, VPX_BITS_12),
        make_tuple(&idct16x16_12,
                   &idct16x16_256_add_12_sse2, 3167, VPX_BITS_12)));
#endif  // HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
Daniel Kang's avatar
Daniel Kang committed
936
}  // namespace