test/dct16x16_test.cc

   1 /*
   2  *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
   3  *
   4  *  Use of this source code is governed by a BSD-style license
   5  *  that can be found in the LICENSE file in the root of the source
   6  *  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.
   9  */
  10
  11 #include <math.h>
  12 #include <stdlib.h>
  13 #include <string.h>
  14
  15 #include "third_party/googletest/src/include/gtest/gtest.h"
  16 #include "test/acm_random.h"
  17 #include "test/clear_system_state.h"
  18 #include "test/register_state_check.h"
  19 #include "test/util.h"
  20
  21 #include "./vp9_rtcd.h"
  22 #include "vp9/common/vp9_entropy.h"
  23 #include "vp9/common/vp9_scan.h"
  24 #include "vpx/vpx_codec.h"
  25 #include "vpx/vpx_integer.h"
  26 #include "vpx_ports/mem.h"
  27
  28 using libvpx_test::ACMRandom;
  29
  30 namespace {
  31
  32 #ifdef _MSC_VER
  33 static int round(double x) {
  34   if (x < 0)
  35     return static_cast<int>(ceil(x - 0.5));
  36   else
  37     return static_cast<int>(floor(x + 0.5));
  38 }
  39 #endif
  40
  41 const int kNumCoeffs = 256;
  42 const double PI = 3.1415926535898;
  43 void reference2_16x16_idct_2d(double *input, double *output) {
  44   double x;
  45   for (int l = 0; l < 16; ++l) {
  46     for (int k = 0; k < 16; ++k) {
  47       double s = 0;
  48       for (int i = 0; i < 16; ++i) {
  49         for (int j = 0; j < 16; ++j) {
  50           x = cos(PI * j * (l + 0.5) / 16.0) *
  51               cos(PI * i * (k + 0.5) / 16.0) *
  52               input[i * 16 + j] / 256;
  53           if (i != 0)
  54             x *= sqrt(2.0);
  55           if (j != 0)
  56             x *= sqrt(2.0);
  57           s += x;
  58         }
  59       }
  60       output[k*16+l] = s;
  61     }
  62   }
  63 }
  64
  65
  66 const double C1 = 0.995184726672197;
  67 const double C2 = 0.98078528040323;
  68 const double C3 = 0.956940335732209;
  69 const double C4 = 0.923879532511287;
  70 const double C5 = 0.881921264348355;
  71 const double C6 = 0.831469612302545;
  72 const double C7 = 0.773010453362737;
  73 const double C8 = 0.707106781186548;
  74 const double C9 = 0.634393284163646;
  75 const double C10 = 0.555570233019602;
  76 const double C11 = 0.471396736825998;
  77 const double C12 = 0.38268343236509;
  78 const double C13 = 0.290284677254462;
  79 const double C14 = 0.195090322016128;
  80 const double C15 = 0.098017140329561;
  81
  82 void butterfly_16x16_dct_1d(double input[16], double output[16]) {
  83   double step[16];
  84   double intermediate[16];
  85   double temp1, temp2;
  86
  87   // step 1
  88   step[ 0] = input[0] + input[15];
  89   step[ 1] = input[1] + input[14];
  90   step[ 2] = input[2] + input[13];
  91   step[ 3] = input[3] + input[12];
  92   step[ 4] = input[4] + input[11];
  93   step[ 5] = input[5] + input[10];
  94   step[ 6] = input[6] + input[ 9];
  95   step[ 7] = input[7] + input[ 8];
  96   step[ 8] = input[7] - input[ 8];
  97   step[ 9] = input[6] - input[ 9];
  98   step[10] = input[5] - input[10];
  99   step[11] = input[4] - input[11];
 100   step[12] = input[3] - input[12];
 101   step[13] = input[2] - input[13];
 102   step[14] = input[1] - input[14];
 103   step[15] = input[0] - input[15];
 104
 105   // step 2
 106   output[0] = step[0] + step[7];
 107   output[1] = step[1] + step[6];
 108   output[2] = step[2] + step[5];
 109   output[3] = step[3] + step[4];
 110   output[4] = step[3] - step[4];
 111   output[5] = step[2] - step[5];
 112   output[6] = step[1] - step[6];
 113   output[7] = step[0] - step[7];
 114
 115   temp1 = step[ 8] * C7;
 116   temp2 = step[15] * C9;
 117   output[ 8] = temp1 + temp2;
 118
 119   temp1 = step[ 9] * C11;
 120   temp2 = step[14] * C5;
 121   output[ 9] = temp1 - temp2;
 122
 123   temp1 = step[10] * C3;
 124   temp2 = step[13] * C13;
 125   output[10] = temp1 + temp2;
 126
 127   temp1 = step[11] * C15;
 128   temp2 = step[12] * C1;
 129   output[11] = temp1 - temp2;
 130
 131   temp1 = step[11] * C1;
 132   temp2 = step[12] * C15;
 133   output[12] = temp2 + temp1;
 134
 135   temp1 = step[10] * C13;
 136   temp2 = step[13] * C3;
 137   output[13] = temp2 - temp1;
 138
 139   temp1 = step[ 9] * C5;
 140   temp2 = step[14] * C11;
 141   output[14] = temp2 + temp1;
 142
 143   temp1 = step[ 8] * C9;
 144   temp2 = step[15] * C7;
 145   output[15] = temp2 - temp1;
 146
 147   // step 3
 148   step[ 0] = output[0] + output[3];
 149   step[ 1] = output[1] + output[2];
 150   step[ 2] = output[1] - output[2];
 151   step[ 3] = output[0] - output[3];
 152
 153   temp1 = output[4] * C14;
 154   temp2 = output[7] * C2;
 155   step[ 4] = temp1 + temp2;
 156
 157   temp1 = output[5] * C10;
 158   temp2 = output[6] * C6;
 159   step[ 5] = temp1 + temp2;
 160
 161   temp1 = output[5] * C6;
 162   temp2 = output[6] * C10;
 163   step[ 6] = temp2 - temp1;
 164
 165   temp1 = output[4] * C2;
 166   temp2 = output[7] * C14;
 167   step[ 7] = temp2 - temp1;
 168
 169   step[ 8] = output[ 8] + output[11];
 170   step[ 9] = output[ 9] + output[10];
 171   step[10] = output[ 9] - output[10];
 172   step[11] = output[ 8] - output[11];
 173
 174   step[12] = output[12] + output[15];
 175   step[13] = output[13] + output[14];
 176   step[14] = output[13] - output[14];
 177   step[15] = output[12] - output[15];
 178
 179   // step 4
 180   output[ 0] = (step[ 0] + step[ 1]);
 181   output[ 8] = (step[ 0] - step[ 1]);
 182
 183   temp1 = step[2] * C12;
 184   temp2 = step[3] * C4;
 185   temp1 = temp1 + temp2;
 186   output[ 4] = 2*(temp1 * C8);
 187
 188   temp1 = step[2] * C4;
 189   temp2 = step[3] * C12;
 190   temp1 = temp2 - temp1;
 191   output[12] = 2 * (temp1 * C8);
 192
 193   output[ 2] = 2 * ((step[4] + step[ 5]) * C8);
 194   output[14] = 2 * ((step[7] - step[ 6]) * C8);
 195
 196   temp1 = step[4] - step[5];
 197   temp2 = step[6] + step[7];
 198   output[ 6] = (temp1 + temp2);
 199   output[10] = (temp1 - temp2);
 200
 201   intermediate[8] = step[8] + step[14];
 202   intermediate[9] = step[9] + step[15];
 203
 204   temp1 = intermediate[8] * C12;
 205   temp2 = intermediate[9] * C4;
 206   temp1 = temp1 - temp2;
 207   output[3] = 2 * (temp1 * C8);
 208
 209   temp1 = intermediate[8] * C4;
 210   temp2 = intermediate[9] * C12;
 211   temp1 = temp2 + temp1;
 212   output[13] = 2 * (temp1 * C8);
 213
 214   output[ 9] = 2 * ((step[10] + step[11]) * C8);
 215
 216   intermediate[11] = step[10] - step[11];
 217   intermediate[12] = step[12] + step[13];
 218   intermediate[13] = step[12] - step[13];
 219   intermediate[14] = step[ 8] - step[14];
 220   intermediate[15] = step[ 9] - step[15];
 221
 222   output[15] = (intermediate[11] + intermediate[12]);
 223   output[ 1] = -(intermediate[11] - intermediate[12]);
 224
 225   output[ 7] = 2 * (intermediate[13] * C8);
 226
 227   temp1 = intermediate[14] * C12;
 228   temp2 = intermediate[15] * C4;
 229   temp1 = temp1 - temp2;
 230   output[11] = -2 * (temp1 * C8);
 231
 232   temp1 = intermediate[14] * C4;
 233   temp2 = intermediate[15] * C12;
 234   temp1 = temp2 + temp1;
 235   output[ 5] = 2 * (temp1 * C8);
 236 }
 237
 238 void reference_16x16_dct_2d(int16_t input[256], double output[256]) {
 239   // First transform columns
 240   for (int i = 0; i < 16; ++i) {
 241     double temp_in[16], temp_out[16];
 242     for (int j = 0; j < 16; ++j)
 243       temp_in[j] = input[j * 16 + i];
 244     butterfly_16x16_dct_1d(temp_in, temp_out);
 245     for (int j = 0; j < 16; ++j)
 246       output[j * 16 + i] = temp_out[j];
 247   }
 248   // Then transform rows
 249   for (int i = 0; i < 16; ++i) {
 250     double temp_in[16], temp_out[16];
 251     for (int j = 0; j < 16; ++j)
 252       temp_in[j] = output[j + i * 16];
 253     butterfly_16x16_dct_1d(temp_in, temp_out);
 254     // Scale by some magic number
 255     for (int j = 0; j < 16; ++j)
 256       output[j + i * 16] = temp_out[j]/2;
 257   }
 258 }
 259
 260 typedef void (*FdctFunc)(const int16_t *in, tran_low_t *out, int stride);
 261 typedef void (*IdctFunc)(const tran_low_t *in, uint8_t *out, int stride);
 262 typedef void (*FhtFunc)(const int16_t *in, tran_low_t *out, int stride,
 263                         int tx_type);
 264 typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
 265                         int tx_type);
 266
 267 typedef std::tr1::tuple<FdctFunc, IdctFunc, int, vpx_bit_depth_t> Dct16x16Param;
 268 typedef std::tr1::tuple<FhtFunc, IhtFunc, int, vpx_bit_depth_t> Ht16x16Param;
 269 typedef std::tr1::tuple<IdctFunc, IdctFunc, int, vpx_bit_depth_t>
 270     Idct16x16Param;
 271
 272 void fdct16x16_ref(const int16_t *in, tran_low_t *out, int stride,
 273                    int /*tx_type*/) {
 274   vp9_fdct16x16_c(in, out, stride);
 275 }
 276
 277 void idct16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
 278                    int /*tx_type*/) {
 279   vp9_idct16x16_256_add_c(in, dest, stride);
 280 }
 281
 282 void fht16x16_ref(const int16_t *in, tran_low_t *out, int stride,
 283                   int tx_type) {
 284   vp9_fht16x16_c(in, out, stride, tx_type);
 285 }
 286
 287 void iht16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
 288                   int tx_type) {
 289   vp9_iht16x16_256_add_c(in, dest, stride, tx_type);
 290 }
 291
 292 #if CONFIG_VP9_HIGHBITDEPTH
 293 void idct16x16_10(const tran_low_t *in, uint8_t *out, int stride) {
 294   vp9_highbd_idct16x16_256_add_c(in, out, stride, 10);
 295 }
 296
 297 void idct16x16_12(const tran_low_t *in, uint8_t *out, int stride) {
 298   vp9_highbd_idct16x16_256_add_c(in, out, stride, 12);
 299 }
 300
 301 void idct16x16_10_ref(const tran_low_t *in, uint8_t *out, int stride,
 302                       int tx_type) {
 303   idct16x16_10(in, out, stride);
 304 }
 305
 306 void idct16x16_12_ref(const tran_low_t *in, uint8_t *out, int stride,
 307                       int tx_type) {
 308   idct16x16_12(in, out, stride);
 309 }
 310
 311 void iht16x16_10(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
 312   vp9_highbd_iht16x16_256_add_c(in, out, stride, tx_type, 10);
 313 }
 314
 315 void iht16x16_12(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
 316   vp9_highbd_iht16x16_256_add_c(in, out, stride, tx_type, 12);
 317 }
 318
 319 void idct16x16_10_add_10_c(const tran_low_t *in, uint8_t *out, int stride) {
 320   vp9_highbd_idct16x16_10_add_c(in, out, stride, 10);
 321 }
 322
 323 void idct16x16_10_add_12_c(const tran_low_t *in, uint8_t *out, int stride) {
 324   vp9_highbd_idct16x16_10_add_c(in, out, stride, 12);
 325 }
 326
 327 #if HAVE_SSE2
 328 void idct16x16_256_add_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
 329   vp9_highbd_idct16x16_256_add_sse2(in, out, stride, 10);
 330 }
 331
 332 void idct16x16_256_add_12_sse2(const tran_low_t *in, uint8_t *out, int stride) {
 333   vp9_highbd_idct16x16_256_add_sse2(in, out, stride, 12);
 334 }
 335
 336 void idct16x16_10_add_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
 337   vp9_highbd_idct16x16_10_add_sse2(in, out, stride, 10);
 338 }
 339
 340 void idct16x16_10_add_12_sse2(const tran_low_t *in, uint8_t *out, int stride) {
 341   vp9_highbd_idct16x16_10_add_sse2(in, out, stride, 12);
 342 }
 343 #endif  // HAVE_SSE2
 344 #endif  // CONFIG_VP9_HIGHBITDEPTH
 345
 346 class Trans16x16TestBase {
 347  public:
 348   virtual ~Trans16x16TestBase() {}
 349
 350  protected:
 351   virtual void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) = 0;
 352
 353   virtual void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) = 0;
 354
 355   void RunAccuracyCheck() {
 356     ACMRandom rnd(ACMRandom::DeterministicSeed());
 357     uint32_t max_error = 0;
 358     int64_t total_error = 0;
 359     const int count_test_block = 10000;
 360     for (int i = 0; i < count_test_block; ++i) {
 361       DECLARE_ALIGNED(16, int16_t, test_input_block[kNumCoeffs]);
 362       DECLARE_ALIGNED(16, tran_low_t, test_temp_block[kNumCoeffs]);
 363       DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
 364       DECLARE_ALIGNED(16, uint8_t, src[kNumCoeffs]);
 365 #if CONFIG_VP9_HIGHBITDEPTH
 366       DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
 367       DECLARE_ALIGNED(16, uint16_t, src16[kNumCoeffs]);
 368 #endif
 369
 370       // Initialize a test block with input range [-mask_, mask_].
 371       for (int j = 0; j < kNumCoeffs; ++j) {
 372         if (bit_depth_ == VPX_BITS_8) {
 373           src[j] = rnd.Rand8();
 374           dst[j] = rnd.Rand8();
 375           test_input_block[j] = src[j] - dst[j];
 376 #if CONFIG_VP9_HIGHBITDEPTH
 377         } else {
 378           src16[j] = rnd.Rand16() & mask_;
 379           dst16[j] = rnd.Rand16() & mask_;
 380           test_input_block[j] = src16[j] - dst16[j];
 381 #endif
 382         }
 383       }
 384
 385       ASM_REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block,
 386                                           test_temp_block, pitch_));
 387       if (bit_depth_ == VPX_BITS_8) {
 388         ASM_REGISTER_STATE_CHECK(
 389             RunInvTxfm(test_temp_block, dst, pitch_));
 390 #if CONFIG_VP9_HIGHBITDEPTH
 391       } else {
 392         ASM_REGISTER_STATE_CHECK(
 393             RunInvTxfm(test_temp_block, CONVERT_TO_BYTEPTR(dst16), pitch_));
 394 #endif
 395       }
 396
 397       for (int j = 0; j < kNumCoeffs; ++j) {
 398 #if CONFIG_VP9_HIGHBITDEPTH
 399         const uint32_t diff =
 400             bit_depth_ == VPX_BITS_8 ?  dst[j] - src[j] : dst16[j] - src16[j];
 401 #else
 402         const uint32_t diff = dst[j] - src[j];
 403 #endif
 404         const uint32_t error = diff * diff;
 405         if (max_error < error)
 406           max_error = error;
 407         total_error += error;
 408       }
 409     }
 410
 411     EXPECT_GE(1u  << 2 * (bit_depth_ - 8), max_error)
 412         << "Error: 16x16 FHT/IHT has an individual round trip error > 1";
 413
 414     EXPECT_GE(count_test_block << 2 * (bit_depth_ - 8), total_error)
 415         << "Error: 16x16 FHT/IHT has average round trip error > 1 per block";
 416   }
 417
 418   void RunCoeffCheck() {
 419     ACMRandom rnd(ACMRandom::DeterministicSeed());
 420     const int count_test_block = 1000;
 421     DECLARE_ALIGNED(16, int16_t, input_block[kNumCoeffs]);
 422     DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kNumCoeffs]);
 423     DECLARE_ALIGNED(16, tran_low_t, output_block[kNumCoeffs]);
 424
 425     for (int i = 0; i < count_test_block; ++i) {
 426       // Initialize a test block with input range [-mask_, mask_].
 427       for (int j = 0; j < kNumCoeffs; ++j)
 428         input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
 429
 430       fwd_txfm_ref(input_block, output_ref_block, pitch_, tx_type_);
 431       ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
 432
 433       // The minimum quant value is 4.
 434       for (int j = 0; j < kNumCoeffs; ++j)
 435         EXPECT_EQ(output_block[j], output_ref_block[j]);
 436     }
 437   }
 438
 439   void RunMemCheck() {
 440     ACMRandom rnd(ACMRandom::DeterministicSeed());
 441     const int count_test_block = 1000;
 442     DECLARE_ALIGNED(16, int16_t, input_extreme_block[kNumCoeffs]);
 443     DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kNumCoeffs]);
 444     DECLARE_ALIGNED(16, tran_low_t, output_block[kNumCoeffs]);
 445
 446     for (int i = 0; i < count_test_block; ++i) {
 447       // Initialize a test block with input range [-mask_, mask_].
 448       for (int j = 0; j < kNumCoeffs; ++j) {
 449         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       ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_extreme_block,
 461                                           output_block, pitch_));
 462
 463       // The minimum quant value is 4.
 464       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             << "Error: 16x16 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
 468       }
 469     }
 470   }
 471
 472   void RunQuantCheck(int dc_thred, int ac_thred) {
 473     ACMRandom rnd(ACMRandom::DeterministicSeed());
 474     const int count_test_block = 100000;
 475     DECLARE_ALIGNED(16, int16_t, input_extreme_block[kNumCoeffs]);
 476     DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kNumCoeffs]);
 477
 478     DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
 479     DECLARE_ALIGNED(16, uint8_t, ref[kNumCoeffs]);
 480 #if CONFIG_VP9_HIGHBITDEPTH
 481     DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
 482     DECLARE_ALIGNED(16, uint16_t, ref16[kNumCoeffs]);
 483 #endif
 484
 485     for (int i = 0; i < count_test_block; ++i) {
 486       // Initialize a test block with input range [-mask_, mask_].
 487       for (int j = 0; j < kNumCoeffs; ++j) {
 488         input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
 489       }
 490       if (i == 0)
 491         for (int j = 0; j < kNumCoeffs; ++j)
 492           input_extreme_block[j] = mask_;
 493       if (i == 1)
 494         for (int j = 0; j < kNumCoeffs; ++j)
 495           input_extreme_block[j] = -mask_;
 496
 497       fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
 498
 499       // clear reconstructed pixel buffers
 500       memset(dst, 0, kNumCoeffs * sizeof(uint8_t));
 501       memset(ref, 0, kNumCoeffs * sizeof(uint8_t));
 502 #if CONFIG_VP9_HIGHBITDEPTH
 503       memset(dst16, 0, kNumCoeffs * sizeof(uint16_t));
 504       memset(ref16, 0, kNumCoeffs * sizeof(uint16_t));
 505 #endif
 506
 507       // quantization with maximum allowed step sizes
 508       output_ref_block[0] = (output_ref_block[0] / dc_thred) * dc_thred;
 509       for (int j = 1; j < kNumCoeffs; ++j)
 510         output_ref_block[j] = (output_ref_block[j] / ac_thred) * ac_thred;
 511       if (bit_depth_ == VPX_BITS_8) {
 512         inv_txfm_ref(output_ref_block, ref, pitch_, tx_type_);
 513         ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block, dst, pitch_));
 514 #if CONFIG_VP9_HIGHBITDEPTH
 515       } else {
 516         inv_txfm_ref(output_ref_block, CONVERT_TO_BYTEPTR(ref16), pitch_,
 517                      tx_type_);
 518         ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block,
 519                                             CONVERT_TO_BYTEPTR(dst16), pitch_));
 520 #endif
 521       }
 522       if (bit_depth_ == VPX_BITS_8) {
 523         for (int j = 0; j < kNumCoeffs; ++j)
 524           EXPECT_EQ(ref[j], dst[j]);
 525 #if CONFIG_VP9_HIGHBITDEPTH
 526       } else {
 527         for (int j = 0; j < kNumCoeffs; ++j)
 528           EXPECT_EQ(ref16[j], dst16[j]);
 529 #endif
 530       }
 531     }
 532   }
 533
 534   void RunInvAccuracyCheck() {
 535     ACMRandom rnd(ACMRandom::DeterministicSeed());
 536     const int count_test_block = 1000;
 537     DECLARE_ALIGNED(16, int16_t, in[kNumCoeffs]);
 538     DECLARE_ALIGNED(16, tran_low_t, coeff[kNumCoeffs]);
 539     DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
 540     DECLARE_ALIGNED(16, uint8_t, src[kNumCoeffs]);
 541 #if CONFIG_VP9_HIGHBITDEPTH
 542     DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
 543     DECLARE_ALIGNED(16, uint16_t, src16[kNumCoeffs]);
 544 #endif  // CONFIG_VP9_HIGHBITDEPTH
 545
 546     for (int i = 0; i < count_test_block; ++i) {
 547       double out_r[kNumCoeffs];
 548
 549       // Initialize a test block with input range [-255, 255].
 550       for (int j = 0; j < kNumCoeffs; ++j) {
 551         if (bit_depth_ == VPX_BITS_8) {
 552           src[j] = rnd.Rand8();
 553           dst[j] = rnd.Rand8();
 554           in[j] = src[j] - dst[j];
 555 #if CONFIG_VP9_HIGHBITDEPTH
 556         } else {
 557           src16[j] = rnd.Rand16() & mask_;
 558           dst16[j] = rnd.Rand16() & mask_;
 559           in[j] = src16[j] - dst16[j];
 560 #endif  // CONFIG_VP9_HIGHBITDEPTH
 561         }
 562       }
 563
 564       reference_16x16_dct_2d(in, out_r);
 565       for (int j = 0; j < kNumCoeffs; ++j)
 566         coeff[j] = static_cast<tran_low_t>(round(out_r[j]));
 567
 568       if (bit_depth_ == VPX_BITS_8) {
 569         ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, 16));
 570 #if CONFIG_VP9_HIGHBITDEPTH
 571       } else {
 572         ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16),
 573                                             16));
 574 #endif  // CONFIG_VP9_HIGHBITDEPTH
 575       }
 576
 577       for (int j = 0; j < kNumCoeffs; ++j) {
 578 #if CONFIG_VP9_HIGHBITDEPTH
 579         const uint32_t diff =
 580             bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
 581 #else
 582         const uint32_t diff = dst[j] - src[j];
 583 #endif  // CONFIG_VP9_HIGHBITDEPTH
 584         const uint32_t error = diff * diff;
 585         EXPECT_GE(1u, error)
 586             << "Error: 16x16 IDCT has error " << error
 587             << " at index " << j;
 588       }
 589     }
 590   }
 591
 592   void CompareInvReference(IdctFunc ref_txfm, int thresh) {
 593     ACMRandom rnd(ACMRandom::DeterministicSeed());
 594     const int count_test_block = 10000;
 595     const int eob = 10;
 596     const int16_t *scan = vp9_default_scan_orders[TX_16X16].scan;
 597     DECLARE_ALIGNED(16, tran_low_t, coeff[kNumCoeffs]);
 598     DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
 599     DECLARE_ALIGNED(16, uint8_t, ref[kNumCoeffs]);
 600 #if CONFIG_VP9_HIGHBITDEPTH
 601     DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
 602     DECLARE_ALIGNED(16, uint16_t, ref16[kNumCoeffs]);
 603 #endif  // CONFIG_VP9_HIGHBITDEPTH
 604
 605     for (int i = 0; i < count_test_block; ++i) {
 606       for (int j = 0; j < kNumCoeffs; ++j) {
 607         if (j < eob) {
 608           // Random values less than the threshold, either positive or negative
 609           coeff[scan[j]] = rnd(thresh) * (1 - 2 * (i % 2));
 610         } else {
 611           coeff[scan[j]] = 0;
 612         }
 613         if (bit_depth_ == VPX_BITS_8) {
 614           dst[j] = 0;
 615           ref[j] = 0;
 616 #if CONFIG_VP9_HIGHBITDEPTH
 617         } else {
 618           dst16[j] = 0;
 619           ref16[j] = 0;
 620 #endif  // CONFIG_VP9_HIGHBITDEPTH
 621         }
 622       }
 623       if (bit_depth_ == VPX_BITS_8) {
 624         ref_txfm(coeff, ref, pitch_);
 625         ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
 626       } else {
 627 #if CONFIG_VP9_HIGHBITDEPTH
 628         ref_txfm(coeff, CONVERT_TO_BYTEPTR(ref16), pitch_);
 629         ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16),
 630                                  pitch_));
 631 #endif  // CONFIG_VP9_HIGHBITDEPTH
 632       }
 633
 634       for (int j = 0; j < kNumCoeffs; ++j) {
 635 #if CONFIG_VP9_HIGHBITDEPTH
 636         const uint32_t diff =
 637             bit_depth_ == VPX_BITS_8 ? dst[j] - ref[j] : dst16[j] - ref16[j];
 638 #else
 639         const uint32_t diff = dst[j] - ref[j];
 640 #endif  // CONFIG_VP9_HIGHBITDEPTH
 641         const uint32_t error = diff * diff;
 642         EXPECT_EQ(0u, error)
 643             << "Error: 16x16 IDCT Comparison has error " << error
 644             << " at index " << j;
 645       }
 646     }
 647   }
 648
 649   int pitch_;
 650   int tx_type_;
 651   vpx_bit_depth_t bit_depth_;
 652   int mask_;
 653   FhtFunc fwd_txfm_ref;
 654   IhtFunc inv_txfm_ref;
 655 };
 656
 657 class Trans16x16DCT
 658     : public Trans16x16TestBase,
 659       public ::testing::TestWithParam<Dct16x16Param> {
 660  public:
 661   virtual ~Trans16x16DCT() {}
 662
 663   virtual void SetUp() {
 664     fwd_txfm_ = GET_PARAM(0);
 665     inv_txfm_ = GET_PARAM(1);
 666     tx_type_  = GET_PARAM(2);
 667     bit_depth_ = GET_PARAM(3);
 668     pitch_    = 16;
 669     fwd_txfm_ref = fdct16x16_ref;
 670     inv_txfm_ref = idct16x16_ref;
 671     mask_ = (1 << bit_depth_) - 1;
 672 #if CONFIG_VP9_HIGHBITDEPTH
 673     switch (bit_depth_) {
 674       case VPX_BITS_10:
 675         inv_txfm_ref = idct16x16_10_ref;
 676         break;
 677       case VPX_BITS_12:
 678         inv_txfm_ref = idct16x16_12_ref;
 679         break;
 680       default:
 681         inv_txfm_ref = idct16x16_ref;
 682         break;
 683     }
 684 #else
 685     inv_txfm_ref = idct16x16_ref;
 686 #endif
 687   }
 688   virtual void TearDown() { libvpx_test::ClearSystemState(); }
 689
 690  protected:
 691   void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
 692     fwd_txfm_(in, out, stride);
 693   }
 694   void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
 695     inv_txfm_(out, dst, stride);
 696   }
 697
 698   FdctFunc fwd_txfm_;
 699   IdctFunc inv_txfm_;
 700 };
 701
 702 TEST_P(Trans16x16DCT, AccuracyCheck) {
 703   RunAccuracyCheck();
 704 }
 705
 706 TEST_P(Trans16x16DCT, CoeffCheck) {
 707   RunCoeffCheck();
 708 }
 709
 710 TEST_P(Trans16x16DCT, MemCheck) {
 711   RunMemCheck();
 712 }
 713
 714 TEST_P(Trans16x16DCT, QuantCheck) {
 715   // Use maximally allowed quantization step sizes for DC and AC
 716   // coefficients respectively.
 717   RunQuantCheck(1336, 1828);
 718 }
 719
 720 TEST_P(Trans16x16DCT, InvAccuracyCheck) {
 721   RunInvAccuracyCheck();
 722 }
 723
 724 class Trans16x16HT
 725     : public Trans16x16TestBase,
 726       public ::testing::TestWithParam<Ht16x16Param> {
 727  public:
 728   virtual ~Trans16x16HT() {}
 729
 730   virtual void SetUp() {
 731     fwd_txfm_ = GET_PARAM(0);
 732     inv_txfm_ = GET_PARAM(1);
 733     tx_type_  = GET_PARAM(2);
 734     bit_depth_ = GET_PARAM(3);
 735     pitch_    = 16;
 736     fwd_txfm_ref = fht16x16_ref;
 737     inv_txfm_ref = iht16x16_ref;
 738     mask_ = (1 << bit_depth_) - 1;
 739 #if CONFIG_VP9_HIGHBITDEPTH
 740     switch (bit_depth_) {
 741       case VPX_BITS_10:
 742         inv_txfm_ref = iht16x16_10;
 743         break;
 744       case VPX_BITS_12:
 745         inv_txfm_ref = iht16x16_12;
 746         break;
 747       default:
 748         inv_txfm_ref = iht16x16_ref;
 749         break;
 750     }
 751 #else
 752     inv_txfm_ref = iht16x16_ref;
 753 #endif
 754   }
 755   virtual void TearDown() { libvpx_test::ClearSystemState(); }
 756
 757  protected:
 758   void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
 759     fwd_txfm_(in, out, stride, tx_type_);
 760   }
 761   void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
 762     inv_txfm_(out, dst, stride, tx_type_);
 763   }
 764
 765   FhtFunc fwd_txfm_;
 766   IhtFunc inv_txfm_;
 767 };
 768
 769 TEST_P(Trans16x16HT, AccuracyCheck) {
 770   RunAccuracyCheck();
 771 }
 772
 773 TEST_P(Trans16x16HT, CoeffCheck) {
 774   RunCoeffCheck();
 775 }
 776
 777 TEST_P(Trans16x16HT, MemCheck) {
 778   RunMemCheck();
 779 }
 780
 781 TEST_P(Trans16x16HT, QuantCheck) {
 782   // The encoder skips any non-DC intra prediction modes,
 783   // when the quantization step size goes beyond 988.
 784   RunQuantCheck(429, 729);
 785 }
 786
 787 class InvTrans16x16DCT
 788     : public Trans16x16TestBase,
 789       public ::testing::TestWithParam<Idct16x16Param> {
 790  public:
 791   virtual ~InvTrans16x16DCT() {}
 792
 793   virtual void SetUp() {
 794     ref_txfm_ = GET_PARAM(0);
 795     inv_txfm_ = GET_PARAM(1);
 796     thresh_ = GET_PARAM(2);
 797     bit_depth_ = GET_PARAM(3);
 798     pitch_ = 16;
 799     mask_ = (1 << bit_depth_) - 1;
 800 }
 801   virtual void TearDown() { libvpx_test::ClearSystemState(); }
 802
 803  protected:
 804   void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {}
 805   void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
 806     inv_txfm_(out, dst, stride);
 807   }
 808
 809   IdctFunc ref_txfm_;
 810   IdctFunc inv_txfm_;
 811   int thresh_;
 812 };
 813
 814 TEST_P(InvTrans16x16DCT, CompareReference) {
 815   CompareInvReference(ref_txfm_, thresh_);
 816 }
 817
 818 using std::tr1::make_tuple;
 819
 820 #if CONFIG_VP9_HIGHBITDEPTH
 821 INSTANTIATE_TEST_CASE_P(
 822     C, Trans16x16DCT,
 823     ::testing::Values(
 824         make_tuple(&vp9_highbd_fdct16x16_c, &idct16x16_10, 0, VPX_BITS_10),
 825         make_tuple(&vp9_highbd_fdct16x16_c, &idct16x16_12, 0, VPX_BITS_12),
 826         make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_c, 0, VPX_BITS_8)));
 827 #else
 828 INSTANTIATE_TEST_CASE_P(
 829     C, Trans16x16DCT,
 830     ::testing::Values(
 831         make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_c, 0, VPX_BITS_8)));
 832 #endif  // CONFIG_VP9_HIGHBITDEPTH
 833
 834 #if CONFIG_VP9_HIGHBITDEPTH
 835 INSTANTIATE_TEST_CASE_P(
 836     C, Trans16x16HT,
 837     ::testing::Values(
 838         make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 0, VPX_BITS_10),
 839         make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 1, VPX_BITS_10),
 840         make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 2, VPX_BITS_10),
 841         make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 3, VPX_BITS_10),
 842         make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 0, VPX_BITS_12),
 843         make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 1, VPX_BITS_12),
 844         make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 2, VPX_BITS_12),
 845         make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 3, VPX_BITS_12),
 846         make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
 847         make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
 848         make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
 849         make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
 850 #else
 851 INSTANTIATE_TEST_CASE_P(
 852     C, Trans16x16HT,
 853     ::testing::Values(
 854         make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
 855         make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
 856         make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
 857         make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
 858 #endif  // CONFIG_VP9_HIGHBITDEPTH
 859
 860 #if HAVE_NEON_ASM && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 861 INSTANTIATE_TEST_CASE_P(
 862     NEON, Trans16x16DCT,
 863     ::testing::Values(
 864         make_tuple(&vp9_fdct16x16_c,
 865                    &vp9_idct16x16_256_add_neon, 0, VPX_BITS_8)));
 866 #endif
 867
 868 #if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 869 INSTANTIATE_TEST_CASE_P(
 870     SSE2, Trans16x16DCT,
 871     ::testing::Values(
 872         make_tuple(&vp9_fdct16x16_sse2,
 873                    &vp9_idct16x16_256_add_sse2, 0, VPX_BITS_8)));
 874 INSTANTIATE_TEST_CASE_P(
 875     SSE2, Trans16x16HT,
 876     ::testing::Values(
 877         make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 0,
 878                    VPX_BITS_8),
 879         make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 1,
 880                    VPX_BITS_8),
 881         make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 2,
 882                    VPX_BITS_8),
 883         make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 3,
 884                    VPX_BITS_8)));
 885 #endif  // HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 886
 887 #if HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 888 INSTANTIATE_TEST_CASE_P(
 889     SSE2, Trans16x16DCT,
 890     ::testing::Values(
 891         make_tuple(&vp9_highbd_fdct16x16_sse2,
 892                    &idct16x16_10, 0, VPX_BITS_10),
 893         make_tuple(&vp9_highbd_fdct16x16_c,
 894                    &idct16x16_256_add_10_sse2, 0, VPX_BITS_10),
 895         make_tuple(&vp9_highbd_fdct16x16_sse2,
 896                    &idct16x16_12, 0, VPX_BITS_12),
 897         make_tuple(&vp9_highbd_fdct16x16_c,
 898                    &idct16x16_256_add_12_sse2, 0, VPX_BITS_12),
 899         make_tuple(&vp9_fdct16x16_sse2,
 900                    &vp9_idct16x16_256_add_c, 0, VPX_BITS_8)));
 901 INSTANTIATE_TEST_CASE_P(
 902     SSE2, Trans16x16HT,
 903     ::testing::Values(
 904         make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_10, 0, VPX_BITS_10),
 905         make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_10, 1, VPX_BITS_10),
 906         make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_10, 2, VPX_BITS_10),
 907         make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_10, 3, VPX_BITS_10),
 908         make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_12, 0, VPX_BITS_12),
 909         make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_12, 1, VPX_BITS_12),
 910         make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_12, 2, VPX_BITS_12),
 911         make_tuple(&vp9_highbd_fht16x16_sse2, &iht16x16_12, 3, VPX_BITS_12),
 912         make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
 913         make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
 914         make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
 915         make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 3,
 916                    VPX_BITS_8)));
 917 // Optimizations take effect at a threshold of 3155, so we use a value close to
 918 // that to test both branches.
 919 INSTANTIATE_TEST_CASE_P(
 920     SSE2, InvTrans16x16DCT,
 921     ::testing::Values(
 922         make_tuple(&idct16x16_10_add_10_c,
 923                    &idct16x16_10_add_10_sse2, 3167, VPX_BITS_10),
 924         make_tuple(&idct16x16_10,
 925                    &idct16x16_256_add_10_sse2, 3167, VPX_BITS_10),
 926         make_tuple(&idct16x16_10_add_12_c,
 927                    &idct16x16_10_add_12_sse2, 3167, VPX_BITS_12),
 928         make_tuple(&idct16x16_12,
 929                    &idct16x16_256_add_12_sse2, 3167, VPX_BITS_12)));
 930 #endif  // HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 931
 932 #if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 933 INSTANTIATE_TEST_CASE_P(
 934     MSA, Trans16x16DCT,
 935     ::testing::Values(
 936         make_tuple(&vp9_fdct16x16_c,
 937                    &vp9_idct16x16_256_add_msa, 0, VPX_BITS_8)));
 938 INSTANTIATE_TEST_CASE_P(
 939     MSA, Trans16x16HT,
 940     ::testing::Values(
 941         make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_msa, 0, VPX_BITS_8),
 942         make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_msa, 1, VPX_BITS_8),
 943         make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_msa, 2, VPX_BITS_8),
 944         make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_msa, 3, VPX_BITS_8)));
 945 #endif  // HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 946 }  // namespace