variance_test.cc 90 KB
Newer Older
1 2 3 4 5 6 7 8 9 10
/*
 *  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.
 */

11 12
#include <cstdlib>
#include <new>
13

14
#include "test/acm_random.h"
15
#include "test/clear_system_state.h"
16
#include "test/register_state_check.h"
17
#include "third_party/googletest/src/include/gtest/gtest.h"
18

Yaowu Xu's avatar
Yaowu Xu committed
19
#include "./vpx_config.h"
20 21
#include "vpx/vpx_codec.h"
#include "vpx/vpx_integer.h"
22
#include "vpx_mem/vpx_mem.h"
23
#include "vpx_ports/mem.h"
James Zern's avatar
James Zern committed
24
#if CONFIG_VP9_ENCODER
Yaowu Xu's avatar
Yaowu Xu committed
25
# include "./vp9_rtcd.h"
26
# include "vp9/encoder/vp9_variance.h"
Johann's avatar
Johann committed
27 28
#endif  // CONFIG_VP9_ENCODER
#include "./vpx_dsp_rtcd.h"
29 30 31

namespace {

Johann's avatar
Johann committed
32 33 34
typedef unsigned int (*VarianceMxNFunc)(const uint8_t *a, int a_stride,
                                        const uint8_t *b, int b_stride,
                                        unsigned int *sse);
Johann's avatar
Johann committed
35
typedef unsigned int (*Get4x4SseFunc)(const uint8_t *a, int a_stride,
Johann's avatar
Johann committed
36 37 38
                                      const uint8_t *b, int b_stride);


39 40 41
using ::std::tr1::get;
using ::std::tr1::make_tuple;
using ::std::tr1::tuple;
42 43
using libvpx_test::ACMRandom;

44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62
// Truncate high bit depth results by downshifting (with rounding) by:
// 2 * (bit_depth - 8) for sse
// (bit_depth - 8) for se
static void RoundHighBitDepth(int bit_depth, int64_t *se, uint64_t *sse) {
  switch (bit_depth) {
    case VPX_BITS_12:
      *sse = (*sse + 128) >> 8;
      *se = (*se + 8) >> 4;
      break;
    case VPX_BITS_10:
      *sse = (*sse + 8) >> 4;
      *se = (*se + 2) >> 2;
      break;
    case VPX_BITS_8:
    default:
      break;
  }
}

63 64 65 66 67 68 69 70
static unsigned int mb_ss_ref(const int16_t *src) {
  unsigned int res = 0;
  for (int i = 0; i < 256; ++i) {
    res += src[i] * src[i];
  }
  return res;
}

71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87
static unsigned int variance_ref(const uint8_t *src, const uint8_t *ref,
                                 int l2w, int l2h, int src_stride_coeff,
                                 int ref_stride_coeff, uint32_t *sse_ptr,
                                 bool use_high_bit_depth_,
                                 vpx_bit_depth_t bit_depth) {
  int64_t se = 0;
  uint64_t sse = 0;
  const int w = 1 << l2w;
  const int h = 1 << l2h;
  for (int y = 0; y < h; y++) {
    for (int x = 0; x < w; x++) {
      int diff;
      if (!use_high_bit_depth_) {
        diff = ref[w * y * ref_stride_coeff + x] -
               src[w * y * src_stride_coeff + x];
        se += diff;
        sse += diff * diff;
Johann's avatar
Johann committed
88
#if CONFIG_VP9_HIGHBITDEPTH
89 90 91 92 93
      } else {
        diff = CONVERT_TO_SHORTPTR(ref)[w * y * ref_stride_coeff + x] -
               CONVERT_TO_SHORTPTR(src)[w * y * src_stride_coeff + x];
        se += diff;
        sse += diff * diff;
Johann's avatar
Johann committed
94
#endif  // CONFIG_VP9_HIGHBITDEPTH
95 96 97
      }
    }
  }
98
  RoundHighBitDepth(bit_depth, &se, &sse);
99 100 101 102 103 104
  *sse_ptr = sse;
  return sse - (((int64_t) se * se) >> (l2w + l2h));
}

static unsigned int subpel_variance_ref(const uint8_t *ref, const uint8_t *src,
                                        int l2w, int l2h, int xoff, int yoff,
105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125
                                        unsigned int *sse_ptr,
                                        bool use_high_bit_depth_,
                                        vpx_bit_depth_t bit_depth) {
  int64_t se = 0;
  uint64_t sse = 0;
  const int w = 1 << l2w;
  const int h = 1 << l2h;
  for (int y = 0; y < h; y++) {
    for (int x = 0; x < w; x++) {
      // Bilinear interpolation at a 16th pel step.
      if (!use_high_bit_depth_) {
        const int a1 = ref[(w + 1) * (y + 0) + x + 0];
        const int a2 = ref[(w + 1) * (y + 0) + x + 1];
        const int b1 = ref[(w + 1) * (y + 1) + x + 0];
        const int b2 = ref[(w + 1) * (y + 1) + x + 1];
        const int a = a1 + (((a2 - a1) * xoff + 8) >> 4);
        const int b = b1 + (((b2 - b1) * xoff + 8) >> 4);
        const int r = a + (((b - a) * yoff + 8) >> 4);
        const int diff = r - src[w * y + x];
        se += diff;
        sse += diff * diff;
Johann's avatar
Johann committed
126
#if CONFIG_VP9_HIGHBITDEPTH
127 128 129 130 131 132 133 134 135 136 137 138 139
      } else {
        uint16_t *ref16 = CONVERT_TO_SHORTPTR(ref);
        uint16_t *src16 = CONVERT_TO_SHORTPTR(src);
        const int a1 = ref16[(w + 1) * (y + 0) + x + 0];
        const int a2 = ref16[(w + 1) * (y + 0) + x + 1];
        const int b1 = ref16[(w + 1) * (y + 1) + x + 0];
        const int b2 = ref16[(w + 1) * (y + 1) + x + 1];
        const int a = a1 + (((a2 - a1) * xoff + 8) >> 4);
        const int b = b1 + (((b2 - b1) * xoff + 8) >> 4);
        const int r = a + (((b - a) * yoff + 8) >> 4);
        const int diff = r - src16[w * y + x];
        se += diff;
        sse += diff * diff;
Johann's avatar
Johann committed
140
#endif  // CONFIG_VP9_HIGHBITDEPTH
141 142 143
      }
    }
  }
144
  RoundHighBitDepth(bit_depth, &se, &sse);
145 146 147
  *sse_ptr = sse;
  return sse - (((int64_t) se * se) >> (l2w + l2h));
}
148

149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192
typedef unsigned int (*SumOfSquaresFunction)(const int16_t *src);

class SumOfSquaresTest : public ::testing::TestWithParam<SumOfSquaresFunction> {
 public:
  SumOfSquaresTest() : func_(GetParam()) {}

  virtual ~SumOfSquaresTest() {
    libvpx_test::ClearSystemState();
  }

 protected:
  void ConstTest();
  void RefTest();

  SumOfSquaresFunction func_;
  ACMRandom rnd_;
};

void SumOfSquaresTest::ConstTest() {
  int16_t mem[256];
  unsigned int res;
  for (int v = 0; v < 256; ++v) {
    for (int i = 0; i < 256; ++i) {
      mem[i] = v;
    }
    ASM_REGISTER_STATE_CHECK(res = func_(mem));
    EXPECT_EQ(256u * (v * v), res);
  }
}

void SumOfSquaresTest::RefTest() {
  int16_t mem[256];
  for (int i = 0; i < 100; ++i) {
    for (int j = 0; j < 256; ++j) {
      mem[j] = rnd_.Rand8() - rnd_.Rand8();
    }

    const unsigned int expected = mb_ss_ref(mem);
    unsigned int res;
    ASM_REGISTER_STATE_CHECK(res = func_(mem));
    EXPECT_EQ(expected, res);
  }
}

James Zern's avatar
James Zern committed
193
template<typename VarianceFunctionType>
Yaowu Xu's avatar
Yaowu Xu committed
194
class VarianceTest
195 196
    : public ::testing::TestWithParam<tuple<int, int,
                                            VarianceFunctionType, int> > {
197 198
 public:
  virtual void SetUp() {
199
    const tuple<int, int, VarianceFunctionType, int>& params = this->GetParam();
200 201 202 203
    log2width_  = get<0>(params);
    width_ = 1 << log2width_;
    log2height_ = get<1>(params);
    height_ = 1 << log2height_;
204
    variance_ = get<2>(params);
205 206 207 208 209 210 211 212
    if (get<3>(params)) {
      bit_depth_ = static_cast<vpx_bit_depth_t>(get<3>(params));
      use_high_bit_depth_ = true;
    } else {
      bit_depth_ = VPX_BITS_8;
      use_high_bit_depth_ = false;
    }
    mask_ = (1 << bit_depth_) - 1;
213

214
    rnd_.Reset(ACMRandom::DeterministicSeed());
215
    block_size_ = width_ * height_;
216 217 218
    if (!use_high_bit_depth_) {
      src_ = reinterpret_cast<uint8_t *>(vpx_memalign(16, block_size_ * 2));
      ref_ = new uint8_t[block_size_ * 2];
Johann's avatar
Johann committed
219
#if CONFIG_VP9_HIGHBITDEPTH
220 221 222 223
    } else {
      src_ = CONVERT_TO_BYTEPTR(reinterpret_cast<uint16_t *>(
          vpx_memalign(16, block_size_ * 2 * sizeof(uint16_t))));
      ref_ = CONVERT_TO_BYTEPTR(new uint16_t[block_size_ * 2]);
Johann's avatar
Johann committed
224
#endif  // CONFIG_VP9_HIGHBITDEPTH
225
    }
226 227 228 229 230
    ASSERT_TRUE(src_ != NULL);
    ASSERT_TRUE(ref_ != NULL);
  }

  virtual void TearDown() {
231 232 233
    if (!use_high_bit_depth_) {
      vpx_free(src_);
      delete[] ref_;
Johann's avatar
Johann committed
234
#if CONFIG_VP9_HIGHBITDEPTH
235 236 237
    } else {
      vpx_free(CONVERT_TO_SHORTPTR(src_));
      delete[] CONVERT_TO_SHORTPTR(ref_);
Johann's avatar
Johann committed
238
#endif  // CONFIG_VP9_HIGHBITDEPTH
239
    }
240
    libvpx_test::ClearSystemState();
241 242 243
  }

 protected:
James Zern's avatar
James Zern committed
244
  void ZeroTest();
245
  void RefTest();
246
  void RefStrideTest();
James Zern's avatar
James Zern committed
247 248
  void OneQuarterTest();

249
  ACMRandom rnd_;
250 251
  uint8_t *src_;
  uint8_t *ref_;
252 253
  int width_, log2width_;
  int height_, log2height_;
254 255 256
  vpx_bit_depth_t bit_depth_;
  int mask_;
  bool use_high_bit_depth_;
257
  int block_size_;
James Zern's avatar
James Zern committed
258
  VarianceFunctionType variance_;
259 260
};

James Zern's avatar
James Zern committed
261 262
template<typename VarianceFunctionType>
void VarianceTest<VarianceFunctionType>::ZeroTest() {
263
  for (int i = 0; i <= 255; ++i) {
264 265
    if (!use_high_bit_depth_) {
      memset(src_, i, block_size_);
Johann's avatar
Johann committed
266
#if CONFIG_VP9_HIGHBITDEPTH
267 268 269
    } else {
      vpx_memset16(CONVERT_TO_SHORTPTR(src_), i << (bit_depth_ - 8),
                   block_size_);
Johann's avatar
Johann committed
270
#endif  // CONFIG_VP9_HIGHBITDEPTH
271
    }
272
    for (int j = 0; j <= 255; ++j) {
273 274
      if (!use_high_bit_depth_) {
        memset(ref_, j, block_size_);
Johann's avatar
Johann committed
275
#if CONFIG_VP9_HIGHBITDEPTH
276 277 278
      } else {
        vpx_memset16(CONVERT_TO_SHORTPTR(ref_), j  << (bit_depth_ - 8),
                     block_size_);
Johann's avatar
Johann committed
279
#endif  // CONFIG_VP9_HIGHBITDEPTH
280
      }
281
      unsigned int sse;
282
      unsigned int var;
283 284
      ASM_REGISTER_STATE_CHECK(
          var = variance_(src_, width_, ref_, width_, &sse));
285
      EXPECT_EQ(0u, var) << "src values: " << i << " ref values: " << j;
286 287 288 289
    }
  }
}

290 291 292 293
template<typename VarianceFunctionType>
void VarianceTest<VarianceFunctionType>::RefTest() {
  for (int i = 0; i < 10; ++i) {
    for (int j = 0; j < block_size_; j++) {
294
    if (!use_high_bit_depth_) {
295 296
      src_[j] = rnd_.Rand8();
      ref_[j] = rnd_.Rand8();
Johann's avatar
Johann committed
297
#if CONFIG_VP9_HIGHBITDEPTH
298 299 300
    } else {
      CONVERT_TO_SHORTPTR(src_)[j] = rnd_.Rand16() && mask_;
      CONVERT_TO_SHORTPTR(ref_)[j] = rnd_.Rand16() && mask_;
Johann's avatar
Johann committed
301
#endif  // CONFIG_VP9_HIGHBITDEPTH
302
    }
303 304
    }
    unsigned int sse1, sse2;
305
    unsigned int var1;
306
    const int stride_coeff = 1;
307 308
    ASM_REGISTER_STATE_CHECK(
        var1 = variance_(src_, width_, ref_, width_, &sse1));
309
    const unsigned int var2 = variance_ref(src_, ref_, log2width_,
310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328
                                           log2height_, stride_coeff,
                                           stride_coeff, &sse2,
                                           use_high_bit_depth_, bit_depth_);
    EXPECT_EQ(sse1, sse2);
    EXPECT_EQ(var1, var2);
  }
}

template<typename VarianceFunctionType>
void VarianceTest<VarianceFunctionType>::RefStrideTest() {
  for (int i = 0; i < 10; ++i) {
    int ref_stride_coeff = i % 2;
    int src_stride_coeff = (i >> 1) % 2;
    for (int j = 0; j < block_size_; j++) {
      int ref_ind = (j / width_) * ref_stride_coeff * width_ + j % width_;
      int src_ind = (j / width_) * src_stride_coeff * width_ + j % width_;
      if (!use_high_bit_depth_) {
        src_[src_ind] = rnd_.Rand8();
        ref_[ref_ind] = rnd_.Rand8();
Johann's avatar
Johann committed
329
#if CONFIG_VP9_HIGHBITDEPTH
330 331 332
      } else {
        CONVERT_TO_SHORTPTR(src_)[src_ind] = rnd_.Rand16() && mask_;
        CONVERT_TO_SHORTPTR(ref_)[ref_ind] = rnd_.Rand16() && mask_;
Johann's avatar
Johann committed
333
#endif  // CONFIG_VP9_HIGHBITDEPTH
334 335 336 337 338 339 340 341 342 343 344 345
      }
    }
    unsigned int sse1, sse2;
    unsigned int var1;

    ASM_REGISTER_STATE_CHECK(
        var1 = variance_(src_, width_ * src_stride_coeff,
                         ref_, width_ * ref_stride_coeff, &sse1));
    const unsigned int var2 = variance_ref(src_, ref_, log2width_,
                                           log2height_, src_stride_coeff,
                                           ref_stride_coeff, &sse2,
                                           use_high_bit_depth_, bit_depth_);
346 347 348 349 350
    EXPECT_EQ(sse1, sse2);
    EXPECT_EQ(var1, var2);
  }
}

James Zern's avatar
James Zern committed
351 352
template<typename VarianceFunctionType>
void VarianceTest<VarianceFunctionType>::OneQuarterTest() {
353
  const int half = block_size_ / 2;
354 355 356 357
  if (!use_high_bit_depth_) {
    memset(src_, 255, block_size_);
    memset(ref_, 255, half);
    memset(ref_ + half, 0, half);
Johann's avatar
Johann committed
358
#if CONFIG_VP9_HIGHBITDEPTH
359 360 361 362 363
  } else {
    vpx_memset16(CONVERT_TO_SHORTPTR(src_), 255 << (bit_depth_ - 8),
                 block_size_);
    vpx_memset16(CONVERT_TO_SHORTPTR(ref_), 255 << (bit_depth_ - 8), half);
    vpx_memset16(CONVERT_TO_SHORTPTR(ref_) + half, 0, half);
Johann's avatar
Johann committed
364
#endif  // CONFIG_VP9_HIGHBITDEPTH
365
  }
366
  unsigned int sse;
367
  unsigned int var;
368
  ASM_REGISTER_STATE_CHECK(var = variance_(src_, width_, ref_, width_, &sse));
369 370 371 372
  const unsigned int expected = block_size_ * 255 * 255 / 4;
  EXPECT_EQ(expected, var);
}

373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421
template<typename MseFunctionType>
class MseTest
    : public ::testing::TestWithParam<tuple<int, int, MseFunctionType> > {
 public:
  virtual void SetUp() {
    const tuple<int, int, MseFunctionType>& params = this->GetParam();
    log2width_  = get<0>(params);
    width_ = 1 << log2width_;
    log2height_ = get<1>(params);
    height_ = 1 << log2height_;
    mse_ = get<2>(params);

    rnd(ACMRandom::DeterministicSeed());
    block_size_ = width_ * height_;
    src_ = reinterpret_cast<uint8_t *>(vpx_memalign(16, block_size_));
    ref_ = new uint8_t[block_size_];
    ASSERT_TRUE(src_ != NULL);
    ASSERT_TRUE(ref_ != NULL);
  }

  virtual void TearDown() {
    vpx_free(src_);
    delete[] ref_;
    libvpx_test::ClearSystemState();
  }

 protected:
  void RefTest_mse();
  void RefTest_sse();
  void MaxTest_mse();
  void MaxTest_sse();

  ACMRandom rnd;
  uint8_t* src_;
  uint8_t* ref_;
  int width_, log2width_;
  int height_, log2height_;
  int block_size_;
  MseFunctionType mse_;
};

template<typename MseFunctionType>
void MseTest<MseFunctionType>::RefTest_mse() {
  for (int i = 0; i < 10; ++i) {
    for (int j = 0; j < block_size_; j++) {
      src_[j] = rnd.Rand8();
      ref_[j] = rnd.Rand8();
    }
    unsigned int sse1, sse2;
422
    const int stride_coeff = 1;
423
    ASM_REGISTER_STATE_CHECK(mse_(src_, width_, ref_, width_, &sse1));
424 425
    variance_ref(src_, ref_, log2width_, log2height_, stride_coeff,
                 stride_coeff, &sse2, false, VPX_BITS_8);
426 427 428 429 430 431 432 433 434 435 436 437 438
    EXPECT_EQ(sse1, sse2);
  }
}

template<typename MseFunctionType>
void MseTest<MseFunctionType>::RefTest_sse() {
  for (int i = 0; i < 10; ++i) {
    for (int j = 0; j < block_size_; j++) {
      src_[j] = rnd.Rand8();
      ref_[j] = rnd.Rand8();
    }
    unsigned int sse2;
    unsigned int var1;
439 440 441 442
    const int stride_coeff = 1;
    ASM_REGISTER_STATE_CHECK(var1 = mse_(src_, width_, ref_, width_));
    variance_ref(src_, ref_, log2width_, log2height_, stride_coeff,
                 stride_coeff, &sse2, false, VPX_BITS_8);
443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466
    EXPECT_EQ(var1, sse2);
  }
}

template<typename MseFunctionType>
void MseTest<MseFunctionType>::MaxTest_mse() {
  memset(src_, 255, block_size_);
  memset(ref_, 0, block_size_);
  unsigned int sse;
  ASM_REGISTER_STATE_CHECK(mse_(src_, width_, ref_, width_, &sse));
  const unsigned int expected = block_size_ * 255 * 255;
  EXPECT_EQ(expected, sse);
}

template<typename MseFunctionType>
void MseTest<MseFunctionType>::MaxTest_sse() {
  memset(src_, 255, block_size_);
  memset(ref_, 0, block_size_);
  unsigned int var;
  ASM_REGISTER_STATE_CHECK(var = mse_(src_, width_, ref_, width_));
  const unsigned int expected = block_size_ * 255 * 255;
  EXPECT_EQ(expected, var);
}

James Zern's avatar
James Zern committed
467 468 469 470 471
unsigned int subpel_avg_variance_ref(const uint8_t *ref,
                                     const uint8_t *src,
                                     const uint8_t *second_pred,
                                     int l2w, int l2h,
                                     int xoff, int yoff,
472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492
                                     unsigned int *sse_ptr,
                                     bool use_high_bit_depth,
                                     vpx_bit_depth_t bit_depth) {
  int64_t se = 0;
  uint64_t sse = 0;
  const int w = 1 << l2w;
  const int h = 1 << l2h;
  for (int y = 0; y < h; y++) {
    for (int x = 0; x < w; x++) {
      // bilinear interpolation at a 16th pel step
      if (!use_high_bit_depth) {
        const int a1 = ref[(w + 1) * (y + 0) + x + 0];
        const int a2 = ref[(w + 1) * (y + 0) + x + 1];
        const int b1 = ref[(w + 1) * (y + 1) + x + 0];
        const int b2 = ref[(w + 1) * (y + 1) + x + 1];
        const int a = a1 + (((a2 - a1) * xoff + 8) >> 4);
        const int b = b1 + (((b2 - b1) * xoff + 8) >> 4);
        const int r = a + (((b - a) * yoff + 8) >> 4);
        const int diff = ((r + second_pred[w * y + x] + 1) >> 1) - src[w * y + x];
        se += diff;
        sse += diff * diff;
Johann's avatar
Johann committed
493
#if CONFIG_VP9_HIGHBITDEPTH
494 495 496 497 498 499 500 501 502 503 504 505 506 507
      } else {
        uint16_t *ref16 = CONVERT_TO_SHORTPTR(ref);
        uint16_t *src16 = CONVERT_TO_SHORTPTR(src);
        uint16_t *sec16   = CONVERT_TO_SHORTPTR(second_pred);
        const int a1 = ref16[(w + 1) * (y + 0) + x + 0];
        const int a2 = ref16[(w + 1) * (y + 0) + x + 1];
        const int b1 = ref16[(w + 1) * (y + 1) + x + 0];
        const int b2 = ref16[(w + 1) * (y + 1) + x + 1];
        const int a = a1 + (((a2 - a1) * xoff + 8) >> 4);
        const int b = b1 + (((b2 - b1) * xoff + 8) >> 4);
        const int r = a + (((b - a) * yoff + 8) >> 4);
        const int diff = ((r + sec16[w * y + x] + 1) >> 1) - src16[w * y + x];
        se += diff;
        sse += diff * diff;
Johann's avatar
Johann committed
508
#endif  // CONFIG_VP9_HIGHBITDEPTH
509 510 511
      }
    }
  }
512
  RoundHighBitDepth(bit_depth, &se, &sse);
James Zern's avatar
James Zern committed
513 514 515 516
  *sse_ptr = sse;
  return sse - (((int64_t) se * se) >> (l2w + l2h));
}

517
template<typename SubpelVarianceFunctionType>
Yaowu Xu's avatar
Yaowu Xu committed
518 519
class SubpelVarianceTest
    : public ::testing::TestWithParam<tuple<int, int,
520
                                            SubpelVarianceFunctionType, int> > {
521 522
 public:
  virtual void SetUp() {
523
    const tuple<int, int, SubpelVarianceFunctionType, int>& params =
524 525 526 527 528 529
        this->GetParam();
    log2width_  = get<0>(params);
    width_ = 1 << log2width_;
    log2height_ = get<1>(params);
    height_ = 1 << log2height_;
    subpel_variance_ = get<2>(params);
530 531 532 533 534 535 536 537
    if (get<3>(params)) {
      bit_depth_ = (vpx_bit_depth_t) get<3>(params);
      use_high_bit_depth_ = true;
    } else {
      bit_depth_ = VPX_BITS_8;
      use_high_bit_depth_ = false;
    }
    mask_ = (1 << bit_depth_)-1;
538

539
    rnd_.Reset(ACMRandom::DeterministicSeed());
540
    block_size_ = width_ * height_;
541 542 543 544
    if (!use_high_bit_depth_) {
      src_ = reinterpret_cast<uint8_t *>(vpx_memalign(16, block_size_));
      sec_ = reinterpret_cast<uint8_t *>(vpx_memalign(16, block_size_));
      ref_ = new uint8_t[block_size_ + width_ + height_ + 1];
Johann's avatar
Johann committed
545
#if CONFIG_VP9_HIGHBITDEPTH
546 547 548 549 550 551 552 553 554 555
    } else {
      src_ = CONVERT_TO_BYTEPTR(
          reinterpret_cast<uint16_t *>(
              vpx_memalign(16, block_size_*sizeof(uint16_t))));
      sec_ = CONVERT_TO_BYTEPTR(
          reinterpret_cast<uint16_t *>(
              vpx_memalign(16, block_size_*sizeof(uint16_t))));
      ref_ = CONVERT_TO_BYTEPTR(
          new uint16_t[block_size_ + width_ + height_ + 1]);
#endif  // CONFIG_VP9_HIGHBITDEPTH
Johann's avatar
Johann committed
556
    }
557
    ASSERT_TRUE(src_ != NULL);
558
    ASSERT_TRUE(sec_ != NULL);
559 560 561 562
    ASSERT_TRUE(ref_ != NULL);
  }

  virtual void TearDown() {
563 564 565 566
    if (!use_high_bit_depth_) {
      vpx_free(src_);
      delete[] ref_;
      vpx_free(sec_);
Johann's avatar
Johann committed
567
#if CONFIG_VP9_HIGHBITDEPTH
568 569 570 571
    } else {
      vpx_free(CONVERT_TO_SHORTPTR(src_));
      delete[] CONVERT_TO_SHORTPTR(ref_);
      vpx_free(CONVERT_TO_SHORTPTR(sec_));
Johann's avatar
Johann committed
572
#endif  // CONFIG_VP9_HIGHBITDEPTH
573
    }
574
    libvpx_test::ClearSystemState();
575 576 577 578
  }

 protected:
  void RefTest();
579
  void ExtremeRefTest();
580

581
  ACMRandom rnd_;
582 583 584
  uint8_t *src_;
  uint8_t *ref_;
  uint8_t *sec_;
585 586
  bool use_high_bit_depth_;
  vpx_bit_depth_t bit_depth_;
587 588
  int width_, log2width_;
  int height_, log2height_;
589
  int block_size_,  mask_;
590 591 592 593 594 595 596
  SubpelVarianceFunctionType subpel_variance_;
};

template<typename SubpelVarianceFunctionType>
void SubpelVarianceTest<SubpelVarianceFunctionType>::RefTest() {
  for (int x = 0; x < 16; ++x) {
    for (int y = 0; y < 16; ++y) {
597 598 599 600 601 602 603
      if (!use_high_bit_depth_) {
        for (int j = 0; j < block_size_; j++) {
          src_[j] = rnd_.Rand8();
        }
        for (int j = 0; j < block_size_ + width_ + height_ + 1; j++) {
          ref_[j] = rnd_.Rand8();
        }
Johann's avatar
Johann committed
604
#if CONFIG_VP9_HIGHBITDEPTH
605 606 607 608 609 610 611 612
      } else {
        for (int j = 0; j < block_size_; j++) {
          CONVERT_TO_SHORTPTR(src_)[j] = rnd_.Rand16() & mask_;
        }
        for (int j = 0; j < block_size_ + width_ + height_ + 1; j++) {
          CONVERT_TO_SHORTPTR(ref_)[j] = rnd_.Rand16() & mask_;
        }
#endif  // CONFIG_VP9_HIGHBITDEPTH
Johann's avatar
Johann committed
613
      }
614
      unsigned int sse1, sse2;
615
      unsigned int var1;
616 617
      ASM_REGISTER_STATE_CHECK(var1 = subpel_variance_(ref_, width_ + 1, x, y,
                                                       src_, width_, &sse1));
618
      const unsigned int var2 = subpel_variance_ref(ref_, src_, log2width_,
619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640
                                                    log2height_, x, y, &sse2,
                                                    use_high_bit_depth_,
                                                    bit_depth_);
      EXPECT_EQ(sse1, sse2) << "at position " << x << ", " << y;
      EXPECT_EQ(var1, var2) << "at position " << x << ", " << y;
    }
  }
}

template<typename SubpelVarianceFunctionType>
void SubpelVarianceTest<SubpelVarianceFunctionType>::ExtremeRefTest() {
  // Compare against reference.
  // Src: Set the first half of values to 0, the second half to the maximum.
  // Ref: Set the first half of values to the maximum, the second half to 0.
  for (int x = 0; x < 16; ++x) {
    for (int y = 0; y < 16; ++y) {
      const int half = block_size_ / 2;
      if (!use_high_bit_depth_) {
        memset(src_, 0, half);
        memset(src_ + half, 255, half);
        memset(ref_, 255, half);
        memset(ref_ + half, 0, half + width_ + height_ + 1);
Johann's avatar
Johann committed
641
#if CONFIG_VP9_HIGHBITDEPTH
642 643 644 645 646 647 648
      } else {
        vpx_memset16(CONVERT_TO_SHORTPTR(src_), mask_, half);
        vpx_memset16(CONVERT_TO_SHORTPTR(src_) + half, 0, half);
        vpx_memset16(CONVERT_TO_SHORTPTR(ref_), 0, half);
        vpx_memset16(CONVERT_TO_SHORTPTR(ref_) + half, mask_,
                     half + width_ + height_ + 1);
#endif  // CONFIG_VP9_HIGHBITDEPTH
Johann's avatar
Johann committed
649
      }
650 651 652 653 654 655 656
      unsigned int sse1, sse2;
      unsigned int var1;
      ASM_REGISTER_STATE_CHECK(
          var1 = subpel_variance_(ref_, width_ + 1, x, y, src_, width_, &sse1));
      const unsigned int var2 =
          subpel_variance_ref(ref_, src_, log2width_, log2height_, x, y, &sse2,
                              use_high_bit_depth_, bit_depth_);
657 658 659 660 661 662
      EXPECT_EQ(sse1, sse2) << "at position " << x << ", " << y;
      EXPECT_EQ(var1, var2) << "at position " << x << ", " << y;
    }
  }
}

Johann's avatar
Johann committed
663
#if CONFIG_VP9_ENCODER
664 665 666 667
template<>
void SubpelVarianceTest<vp9_subp_avg_variance_fn_t>::RefTest() {
  for (int x = 0; x < 16; ++x) {
    for (int y = 0; y < 16; ++y) {
668 669 670 671 672 673 674 675
      if (!use_high_bit_depth_) {
        for (int j = 0; j < block_size_; j++) {
          src_[j] = rnd_.Rand8();
          sec_[j] = rnd_.Rand8();
        }
        for (int j = 0; j < block_size_ + width_ + height_ + 1; j++) {
          ref_[j] = rnd_.Rand8();
        }
Johann's avatar
Johann committed
676
#if CONFIG_VP9_HIGHBITDEPTH
677 678 679 680 681 682 683 684
      } else {
        for (int j = 0; j < block_size_; j++) {
          CONVERT_TO_SHORTPTR(src_)[j] = rnd_.Rand16() & mask_;
          CONVERT_TO_SHORTPTR(sec_)[j] = rnd_.Rand16() & mask_;
        }
        for (int j = 0; j < block_size_ + width_ + height_ + 1; j++) {
          CONVERT_TO_SHORTPTR(ref_)[j] = rnd_.Rand16() & mask_;
        }
Johann's avatar
Johann committed
685
#endif  // CONFIG_VP9_HIGHBITDEPTH
686
      }
687
      unsigned int sse1, sse2;
688
      unsigned int var1;
689 690 691
      ASM_REGISTER_STATE_CHECK(
          var1 = subpel_variance_(ref_, width_ + 1, x, y,
                                  src_, width_, &sse1, sec_));
692 693
      const unsigned int var2 = subpel_avg_variance_ref(ref_, src_, sec_,
                                                        log2width_, log2height_,
694 695 696
                                                        x, y, &sse2,
                                                        use_high_bit_depth_,
                                                        bit_depth_);
697 698 699 700 701
      EXPECT_EQ(sse1, sse2) << "at position " << x << ", " << y;
      EXPECT_EQ(var1, var2) << "at position " << x << ", " << y;
    }
  }
}
James Zern's avatar
James Zern committed
702 703
#endif  // CONFIG_VP9_ENCODER

Johann's avatar
Johann committed
704
typedef MseTest<Get4x4SseFunc> VpxSseTest;
Johann's avatar
Johann committed
705 706 707 708 709 710 711 712 713 714 715 716 717
typedef MseTest<VarianceMxNFunc> VpxMseTest;
typedef VarianceTest<VarianceMxNFunc> VpxVarianceTest;

TEST_P(VpxSseTest, Ref_sse) { RefTest_sse(); }
TEST_P(VpxSseTest, Max_sse) { MaxTest_sse(); }
TEST_P(VpxMseTest, Ref_mse) { RefTest_mse(); }
TEST_P(VpxMseTest, Max_mse) { MaxTest_mse(); }
TEST_P(VpxVarianceTest, Zero) { ZeroTest(); }
TEST_P(VpxVarianceTest, Ref) { RefTest(); }
TEST_P(VpxVarianceTest, RefStride) { RefStrideTest(); }
TEST_P(VpxVarianceTest, OneQuarter) { OneQuarterTest(); }
TEST_P(SumOfSquaresTest, Const) { ConstTest(); }
TEST_P(SumOfSquaresTest, Ref) { RefTest(); }
James Zern's avatar
James Zern committed
718

Johann's avatar
Johann committed
719 720 721
INSTANTIATE_TEST_CASE_P(C, SumOfSquaresTest,
                        ::testing::Values(vpx_get_mb_ss_c));

Johann's avatar
Johann committed
722
const Get4x4SseFunc get4x4sse_cs_c = vpx_get4x4sse_cs_c;
Johann's avatar
Johann committed
723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748
INSTANTIATE_TEST_CASE_P(C, VpxSseTest,
                        ::testing::Values(make_tuple(2, 2, get4x4sse_cs_c)));

const VarianceMxNFunc mse16x16_c = vpx_mse16x16_c;
const VarianceMxNFunc mse16x8_c = vpx_mse16x8_c;
const VarianceMxNFunc mse8x16_c = vpx_mse8x16_c;
const VarianceMxNFunc mse8x8_c = vpx_mse8x8_c;
INSTANTIATE_TEST_CASE_P(C, VpxMseTest,
                        ::testing::Values(make_tuple(4, 4, mse16x16_c),
                                          make_tuple(4, 3, mse16x8_c),
                                          make_tuple(3, 4, mse8x16_c),
                                          make_tuple(3, 3, mse8x8_c)));

const VarianceMxNFunc variance64x64_c = vpx_variance64x64_c;
const VarianceMxNFunc variance64x32_c = vpx_variance64x32_c;
const VarianceMxNFunc variance32x64_c = vpx_variance32x64_c;
const VarianceMxNFunc variance32x32_c = vpx_variance32x32_c;
const VarianceMxNFunc variance32x16_c = vpx_variance32x16_c;
const VarianceMxNFunc variance16x32_c = vpx_variance16x32_c;
const VarianceMxNFunc variance16x16_c = vpx_variance16x16_c;
const VarianceMxNFunc variance16x8_c = vpx_variance16x8_c;
const VarianceMxNFunc variance8x16_c = vpx_variance8x16_c;
const VarianceMxNFunc variance8x8_c = vpx_variance8x8_c;
const VarianceMxNFunc variance8x4_c = vpx_variance8x4_c;
const VarianceMxNFunc variance4x8_c = vpx_variance4x8_c;
const VarianceMxNFunc variance4x4_c = vpx_variance4x4_c;
749 750

INSTANTIATE_TEST_CASE_P(
Johann's avatar
Johann committed
751 752 753 754 755 756 757 758
    C, VpxVarianceTest,
    ::testing::Values(make_tuple(6, 6, variance64x64_c, 0),
                      make_tuple(6, 5, variance64x32_c, 0),
                      make_tuple(5, 6, variance32x64_c, 0),
                      make_tuple(5, 5, variance32x32_c, 0),
                      make_tuple(5, 4, variance32x16_c, 0),
                      make_tuple(4, 5, variance16x32_c, 0),
                      make_tuple(4, 4, variance16x16_c, 0),
759
                      make_tuple(4, 3, variance16x8_c, 0),
Johann's avatar
Johann committed
760 761 762 763 764
                      make_tuple(3, 4, variance8x16_c, 0),
                      make_tuple(3, 3, variance8x8_c, 0),
                      make_tuple(3, 2, variance8x4_c, 0),
                      make_tuple(2, 3, variance4x8_c, 0),
                      make_tuple(2, 2, variance4x4_c, 0)));
James Zern's avatar
James Zern committed
765

Johann's avatar
Johann committed
766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791
#if CONFIG_VP9_HIGHBITDEPTH
typedef MseTest<VarianceMxNFunc> VpxHBDMseTest;
typedef VarianceTest<VarianceMxNFunc> VpxHBDVarianceTest;

TEST_P(VpxHBDMseTest, Ref_mse) { RefTest_mse(); }
TEST_P(VpxHBDMseTest, Max_mse) { MaxTest_mse(); }
TEST_P(VpxHBDVarianceTest, Zero) { ZeroTest(); }
TEST_P(VpxHBDVarianceTest, Ref) { RefTest(); }
TEST_P(VpxHBDVarianceTest, RefStride) { RefStrideTest(); }
TEST_P(VpxHBDVarianceTest, OneQuarter) { OneQuarterTest(); }

/* TODO(debargha): This test does not support the highbd version
const VarianceMxNFunc highbd_12_mse16x16_c = vpx_highbd_12_mse16x16_c;
const VarianceMxNFunc highbd_12_mse16x8_c = vpx_highbd_12_mse16x8_c;
const VarianceMxNFunc highbd_12_mse8x16_c = vpx_highbd_12_mse8x16_c;
const VarianceMxNFunc highbd_12_mse8x8_c = vpx_highbd_12_mse8x8_c;

const VarianceMxNFunc highbd_10_mse16x16_c = vpx_highbd_10_mse16x16_c;
const VarianceMxNFunc highbd_10_mse16x8_c = vpx_highbd_10_mse16x8_c;
const VarianceMxNFunc highbd_10_mse8x16_c = vpx_highbd_10_mse8x16_c;
const VarianceMxNFunc highbd_10_mse8x8_c = vpx_highbd_10_mse8x8_c;

const VarianceMxNFunc highbd_8_mse16x16_c = vpx_highbd_8_mse16x16_c;
const VarianceMxNFunc highbd_8_mse16x8_c = vpx_highbd_8_mse16x8_c;
const VarianceMxNFunc highbd_8_mse8x16_c = vpx_highbd_8_mse8x16_c;
const VarianceMxNFunc highbd_8_mse8x8_c = vpx_highbd_8_mse8x8_c;
792 793

INSTANTIATE_TEST_CASE_P(
Johann's avatar
Johann committed
794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849
    C, VpxHBDMseTest, ::testing::Values(make_tuple(4, 4, highbd_12_mse16x16_c),
                                        make_tuple(4, 4, highbd_12_mse16x8_c),
                                        make_tuple(4, 4, highbd_12_mse8x16_c),
                                        make_tuple(4, 4, highbd_12_mse8x8_c),
                                        make_tuple(4, 4, highbd_10_mse16x16_c),
                                        make_tuple(4, 4, highbd_10_mse16x8_c),
                                        make_tuple(4, 4, highbd_10_mse8x16_c),
                                        make_tuple(4, 4, highbd_10_mse8x8_c),
                                        make_tuple(4, 4, highbd_8_mse16x16_c),
                                        make_tuple(4, 4, highbd_8_mse16x8_c),
                                        make_tuple(4, 4, highbd_8_mse8x16_c),
                                        make_tuple(4, 4, highbd_8_mse8x8_c)));
*/


const VarianceMxNFunc highbd_12_variance64x64_c = vpx_highbd_12_variance64x64_c;
const VarianceMxNFunc highbd_12_variance64x32_c = vpx_highbd_12_variance64x32_c;
const VarianceMxNFunc highbd_12_variance32x64_c = vpx_highbd_12_variance32x64_c;
const VarianceMxNFunc highbd_12_variance32x32_c = vpx_highbd_12_variance32x32_c;
const VarianceMxNFunc highbd_12_variance32x16_c = vpx_highbd_12_variance32x16_c;
const VarianceMxNFunc highbd_12_variance16x32_c = vpx_highbd_12_variance16x32_c;
const VarianceMxNFunc highbd_12_variance16x16_c = vpx_highbd_12_variance16x16_c;
const VarianceMxNFunc highbd_12_variance16x8_c = vpx_highbd_12_variance16x8_c;
const VarianceMxNFunc highbd_12_variance8x16_c = vpx_highbd_12_variance8x16_c;
const VarianceMxNFunc highbd_12_variance8x8_c = vpx_highbd_12_variance8x8_c;
const VarianceMxNFunc highbd_12_variance8x4_c = vpx_highbd_12_variance8x4_c;
const VarianceMxNFunc highbd_12_variance4x8_c = vpx_highbd_12_variance4x8_c;
const VarianceMxNFunc highbd_12_variance4x4_c = vpx_highbd_12_variance4x4_c;

const VarianceMxNFunc highbd_10_variance64x64_c = vpx_highbd_10_variance64x64_c;
const VarianceMxNFunc highbd_10_variance64x32_c = vpx_highbd_10_variance64x32_c;
const VarianceMxNFunc highbd_10_variance32x64_c = vpx_highbd_10_variance32x64_c;
const VarianceMxNFunc highbd_10_variance32x32_c = vpx_highbd_10_variance32x32_c;
const VarianceMxNFunc highbd_10_variance32x16_c = vpx_highbd_10_variance32x16_c;
const VarianceMxNFunc highbd_10_variance16x32_c = vpx_highbd_10_variance16x32_c;
const VarianceMxNFunc highbd_10_variance16x16_c = vpx_highbd_10_variance16x16_c;
const VarianceMxNFunc highbd_10_variance16x8_c = vpx_highbd_10_variance16x8_c;
const VarianceMxNFunc highbd_10_variance8x16_c = vpx_highbd_10_variance8x16_c;
const VarianceMxNFunc highbd_10_variance8x8_c = vpx_highbd_10_variance8x8_c;
const VarianceMxNFunc highbd_10_variance8x4_c = vpx_highbd_10_variance8x4_c;
const VarianceMxNFunc highbd_10_variance4x8_c = vpx_highbd_10_variance4x8_c;
const VarianceMxNFunc highbd_10_variance4x4_c = vpx_highbd_10_variance4x4_c;

const VarianceMxNFunc highbd_8_variance64x64_c = vpx_highbd_8_variance64x64_c;
const VarianceMxNFunc highbd_8_variance64x32_c = vpx_highbd_8_variance64x32_c;
const VarianceMxNFunc highbd_8_variance32x64_c = vpx_highbd_8_variance32x64_c;
const VarianceMxNFunc highbd_8_variance32x32_c = vpx_highbd_8_variance32x32_c;
const VarianceMxNFunc highbd_8_variance32x16_c = vpx_highbd_8_variance32x16_c;
const VarianceMxNFunc highbd_8_variance16x32_c = vpx_highbd_8_variance16x32_c;
const VarianceMxNFunc highbd_8_variance16x16_c = vpx_highbd_8_variance16x16_c;
const VarianceMxNFunc highbd_8_variance16x8_c = vpx_highbd_8_variance16x8_c;
const VarianceMxNFunc highbd_8_variance8x16_c = vpx_highbd_8_variance8x16_c;
const VarianceMxNFunc highbd_8_variance8x8_c = vpx_highbd_8_variance8x8_c;
const VarianceMxNFunc highbd_8_variance8x4_c = vpx_highbd_8_variance8x4_c;
const VarianceMxNFunc highbd_8_variance4x8_c = vpx_highbd_8_variance4x8_c;
const VarianceMxNFunc highbd_8_variance4x4_c = vpx_highbd_8_variance4x4_c;
James Zern's avatar
James Zern committed
850
INSTANTIATE_TEST_CASE_P(
Johann's avatar
Johann committed
851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891
    C, VpxHBDVarianceTest,
    ::testing::Values(make_tuple(6, 6, highbd_12_variance64x64_c, 12),
                      make_tuple(6, 5, highbd_12_variance64x32_c, 12),
                      make_tuple(5, 6, highbd_12_variance32x64_c, 12),
                      make_tuple(5, 5, highbd_12_variance32x32_c, 12),
                      make_tuple(5, 4, highbd_12_variance32x16_c, 12),
                      make_tuple(4, 5, highbd_12_variance16x32_c, 12),
                      make_tuple(4, 4, highbd_12_variance16x16_c, 12),
                      make_tuple(4, 3, highbd_12_variance16x8_c, 12),
                      make_tuple(3, 4, highbd_12_variance8x16_c, 12),
                      make_tuple(3, 3, highbd_12_variance8x8_c, 12),
                      make_tuple(3, 2, highbd_12_variance8x4_c, 12),
                      make_tuple(2, 3, highbd_12_variance4x8_c, 12),
                      make_tuple(2, 2, highbd_12_variance4x4_c, 12),
                      make_tuple(6, 6, highbd_10_variance64x64_c, 10),
                      make_tuple(6, 5, highbd_10_variance64x32_c, 10),
                      make_tuple(5, 6, highbd_10_variance32x64_c, 10),
                      make_tuple(5, 5, highbd_10_variance32x32_c, 10),
                      make_tuple(5, 4, highbd_10_variance32x16_c, 10),
                      make_tuple(4, 5, highbd_10_variance16x32_c, 10),
                      make_tuple(4, 4, highbd_10_variance16x16_c, 10),
                      make_tuple(4, 3, highbd_10_variance16x8_c, 10),
                      make_tuple(3, 4, highbd_10_variance8x16_c, 10),
                      make_tuple(3, 3, highbd_10_variance8x8_c, 10),
                      make_tuple(3, 2, highbd_10_variance8x4_c, 10),
                      make_tuple(2, 3, highbd_10_variance4x8_c, 10),
                      make_tuple(2, 2, highbd_10_variance4x4_c, 10),
                      make_tuple(6, 6, highbd_8_variance64x64_c, 8),
                      make_tuple(6, 5, highbd_8_variance64x32_c, 8),
                      make_tuple(5, 6, highbd_8_variance32x64_c, 8),
                      make_tuple(5, 5, highbd_8_variance32x32_c, 8),
                      make_tuple(5, 4, highbd_8_variance32x16_c, 8),
                      make_tuple(4, 5, highbd_8_variance16x32_c, 8),
                      make_tuple(4, 4, highbd_8_variance16x16_c, 8),
                      make_tuple(4, 3, highbd_8_variance16x8_c, 8),
                      make_tuple(3, 4, highbd_8_variance8x16_c, 8),
                      make_tuple(3, 3, highbd_8_variance8x8_c, 8),
                      make_tuple(3, 2, highbd_8_variance8x4_c, 8),
                      make_tuple(2, 3, highbd_8_variance4x8_c, 8),
                      make_tuple(2, 2, highbd_8_variance4x4_c, 8)));
#endif  // CONFIG_VP9_HIGHBITDEPTH
James Zern's avatar
James Zern committed
892

James Zern's avatar
James Zern committed
893
#if HAVE_MMX
Johann's avatar
Johann committed
894 895 896 897 898 899 900 901 902 903 904 905
const VarianceMxNFunc mse16x16_mmx = vpx_mse16x16_mmx;
INSTANTIATE_TEST_CASE_P(MMX, VpxMseTest,
                        ::testing::Values(make_tuple(4, 4, mse16x16_mmx)));

INSTANTIATE_TEST_CASE_P(MMX, SumOfSquaresTest,
                        ::testing::Values(vpx_get_mb_ss_mmx));

const VarianceMxNFunc variance16x16_mmx = vpx_variance16x16_mmx;
const VarianceMxNFunc variance16x8_mmx = vpx_variance16x8_mmx;
const VarianceMxNFunc variance8x16_mmx = vpx_variance8x16_mmx;
const VarianceMxNFunc variance8x8_mmx = vpx_variance8x8_mmx;
const VarianceMxNFunc variance4x4_mmx = vpx_variance4x4_mmx;
James Zern's avatar
James Zern committed
906
INSTANTIATE_TEST_CASE_P(
Johann's avatar
Johann committed
907 908
    MMX, VpxVarianceTest,
    ::testing::Values(make_tuple(4, 4, variance16x16_mmx, 0),
909
                      make_tuple(4, 3, variance16x8_mmx, 0),
Johann's avatar
Johann committed
910 911 912 913
                      make_tuple(3, 4, variance8x16_mmx, 0),
                      make_tuple(3, 3, variance8x8_mmx, 0),
                      make_tuple(2, 2, variance4x4_mmx, 0)));
#endif  // HAVE_MMX
James Zern's avatar
James Zern committed
914 915

#if HAVE_SSE2
Johann's avatar
Johann committed
916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941
INSTANTIATE_TEST_CASE_P(SSE2, SumOfSquaresTest,
                        ::testing::Values(vpx_get_mb_ss_sse2));

const VarianceMxNFunc mse16x16_sse2 = vpx_mse16x16_sse2;
const VarianceMxNFunc mse16x8_sse2 = vpx_mse16x8_sse2;
const VarianceMxNFunc mse8x16_sse2 = vpx_mse8x16_sse2;
const VarianceMxNFunc mse8x8_sse2 = vpx_mse8x8_sse2;
INSTANTIATE_TEST_CASE_P(SSE2, VpxMseTest,
                        ::testing::Values(make_tuple(4, 4, mse16x16_sse2),
                                          make_tuple(4, 3, mse16x8_sse2),
                                          make_tuple(3, 4, mse8x16_sse2),
                                          make_tuple(3, 3, mse8x8_sse2)));

const VarianceMxNFunc variance64x64_sse2 = vpx_variance64x64_sse2;
const VarianceMxNFunc variance64x32_sse2 = vpx_variance64x32_sse2;
const VarianceMxNFunc variance32x64_sse2 = vpx_variance32x64_sse2;
const VarianceMxNFunc variance32x32_sse2 = vpx_variance32x32_sse2;
const VarianceMxNFunc variance32x16_sse2 = vpx_variance32x16_sse2;
const VarianceMxNFunc variance16x32_sse2 = vpx_variance16x32_sse2;
const VarianceMxNFunc variance16x16_sse2 = vpx_variance16x16_sse2;
const VarianceMxNFunc variance16x8_sse2 = vpx_variance16x8_sse2;
const VarianceMxNFunc variance8x16_sse2 = vpx_variance8x16_sse2;
const VarianceMxNFunc variance8x8_sse2 = vpx_variance8x8_sse2;
const VarianceMxNFunc variance8x4_sse2 = vpx_variance8x4_sse2;
const VarianceMxNFunc variance4x8_sse2 = vpx_variance4x8_sse2;
const VarianceMxNFunc variance4x4_sse2 = vpx_variance4x4_sse2;
James Zern's avatar
James Zern committed
942
INSTANTIATE_TEST_CASE_P(
Johann's avatar
Johann committed
943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972
    SSE2, VpxVarianceTest,
    ::testing::Values(make_tuple(6, 6, variance64x64_sse2, 0),
                      make_tuple(6, 5, variance64x32_sse2, 0),
                      make_tuple(5, 6, variance32x64_sse2, 0),
                      make_tuple(5, 5, variance32x32_sse2, 0),
                      make_tuple(5, 4, variance32x16_sse2, 0),
                      make_tuple(4, 5, variance16x32_sse2, 0),
                      make_tuple(4, 4, variance16x16_sse2, 0),
                      make_tuple(4, 3, variance16x8_sse2, 0),
                      make_tuple(3, 4, variance8x16_sse2, 0),
                      make_tuple(3, 3, variance8x8_sse2, 0),
                      make_tuple(3, 2, variance8x4_sse2, 0),
                      make_tuple(2, 3, variance4x8_sse2, 0),
                      make_tuple(2, 2, variance4x4_sse2, 0)));
#if CONFIG_VP9_HIGHBITDEPTH
/* TODO(debargha): This test does not support the highbd version
const VarianceMxNFunc highbd_12_mse16x16_sse2 = vpx_highbd_12_mse16x16_sse2;
const VarianceMxNFunc highbd_12_mse16x8_sse2 = vpx_highbd_12_mse16x8_sse2;
const VarianceMxNFunc highbd_12_mse8x16_sse2 = vpx_highbd_12_mse8x16_sse2;
const VarianceMxNFunc highbd_12_mse8x8_sse2 = vpx_highbd_12_mse8x8_sse2;

const VarianceMxNFunc highbd_10_mse16x16_sse2 = vpx_highbd_10_mse16x16_sse2;
const VarianceMxNFunc highbd_10_mse16x8_sse2 = vpx_highbd_10_mse16x8_sse2;
const VarianceMxNFunc highbd_10_mse8x16_sse2 = vpx_highbd_10_mse8x16_sse2;
const VarianceMxNFunc highbd_10_mse8x8_sse2 = vpx_highbd_10_mse8x8_sse2;

const VarianceMxNFunc highbd_8_mse16x16_sse2 = vpx_highbd_8_mse16x16_sse2;
const VarianceMxNFunc highbd_8_mse16x8_sse2 = vpx_highbd_8_mse16x8_sse2;
const VarianceMxNFunc highbd_8_mse8x16_sse2 = vpx_highbd_8_mse8x16_sse2;
const VarianceMxNFunc highbd_8_mse8x8_sse2 = vpx_highbd_8_mse8x8_sse2;
James Zern's avatar
James Zern committed
973

Johann's avatar
Johann committed
974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048
INSTANTIATE_TEST_CASE_P(
    SSE2, VpxHBDMseTest, ::testing::Values(make_tuple(4, 4, highbd_12_mse16x16_sse2),
                                           make_tuple(4, 3, highbd_12_mse16x8_sse2),
                                           make_tuple(3, 4, highbd_12_mse8x16_sse2),
                                           make_tuple(3, 3, highbd_12_mse8x8_sse2),
                                           make_tuple(4, 4, highbd_10_mse16x16_sse2),
                                           make_tuple(4, 3, highbd_10_mse16x8_sse2),
                                           make_tuple(3, 4, highbd_10_mse8x16_sse2),
                                           make_tuple(3, 3, highbd_10_mse8x8_sse2),
                                           make_tuple(4, 4, highbd_8_mse16x16_sse2),
                                           make_tuple(4, 3, highbd_8_mse16x8_sse2),
                                           make_tuple(3, 4, highbd_8_mse8x16_sse2),
                                           make_tuple(3, 3, highbd_8_mse8x8_sse2)));
*/

const VarianceMxNFunc highbd_12_variance64x64_sse2 =
    vpx_highbd_12_variance64x64_sse2;
const VarianceMxNFunc highbd_12_variance64x32_sse2 =
    vpx_highbd_12_variance64x32_sse2;
const VarianceMxNFunc highbd_12_variance32x64_sse2 =
    vpx_highbd_12_variance32x64_sse2;
const VarianceMxNFunc highbd_12_variance32x32_sse2 =
    vpx_highbd_12_variance32x32_sse2;
const VarianceMxNFunc highbd_12_variance32x16_sse2 =
    vpx_highbd_12_variance32x16_sse2;
const VarianceMxNFunc highbd_12_variance16x32_sse2 =
    vpx_highbd_12_variance16x32_sse2;
const VarianceMxNFunc highbd_12_variance16x16_sse2 =
    vpx_highbd_12_variance16x16_sse2;
const VarianceMxNFunc highbd_12_variance16x8_sse2 =
    vpx_highbd_12_variance16x8_sse2;
const VarianceMxNFunc highbd_12_variance8x16_sse2 =
    vpx_highbd_12_variance8x16_sse2;
const VarianceMxNFunc highbd_12_variance8x8_sse2 =
    vpx_highbd_12_variance8x8_sse2;
const VarianceMxNFunc highbd_10_variance64x64_sse2 =
    vpx_highbd_10_variance64x64_sse2;
const VarianceMxNFunc highbd_10_variance64x32_sse2 =
    vpx_highbd_10_variance64x32_sse2;
const VarianceMxNFunc highbd_10_variance32x64_sse2 =
    vpx_highbd_10_variance32x64_sse2;
const VarianceMxNFunc highbd_10_variance32x32_sse2 =
    vpx_highbd_10_variance32x32_sse2;
const VarianceMxNFunc highbd_10_variance32x16_sse2 =
    vpx_highbd_10_variance32x16_sse2;
const VarianceMxNFunc highbd_10_variance16x32_sse2 =
    vpx_highbd_10_variance16x32_sse2;
const VarianceMxNFunc highbd_10_variance16x16_sse2 =
    vpx_highbd_10_variance16x16_sse2;
const VarianceMxNFunc highbd_10_variance16x8_sse2 =
    vpx_highbd_10_variance16x8_sse2;
const VarianceMxNFunc highbd_10_variance8x16_sse2 =
    vpx_highbd_10_variance8x16_sse2;
const VarianceMxNFunc highbd_10_variance8x8_sse2 =
    vpx_highbd_10_variance8x8_sse2;
const VarianceMxNFunc highbd_8_variance64x64_sse2 =
    vpx_highbd_8_variance64x64_sse2;
const VarianceMxNFunc highbd_8_variance64x32_sse2 =
    vpx_highbd_8_variance64x32_sse2;
const VarianceMxNFunc highbd_8_variance32x64_sse2 =
    vpx_highbd_8_variance32x64_sse2;
const VarianceMxNFunc highbd_8_variance32x32_sse2 =
    vpx_highbd_8_variance32x32_sse2;
const VarianceMxNFunc highbd_8_variance32x16_sse2 =
    vpx_highbd_8_variance32x16_sse2;
const VarianceMxNFunc highbd_8_variance16x32_sse2 =
    vpx_highbd_8_variance16x32_sse2;
const VarianceMxNFunc highbd_8_variance16x16_sse2 =
    vpx_highbd_8_variance16x16_sse2;
const VarianceMxNFunc highbd_8_variance16x8_sse2 =
    vpx_highbd_8_variance16x8_sse2;
const VarianceMxNFunc highbd_8_variance8x16_sse2 =
    vpx_highbd_8_variance8x16_sse2;
const VarianceMxNFunc highbd_8_variance8x8_sse2 =
    vpx_highbd_8_variance8x8_sse2;
James Zern's avatar
James Zern committed
1049

Johann's avatar
Johann committed
1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083
INSTANTIATE_TEST_CASE_P(
    SSE2, VpxHBDVarianceTest,
    ::testing::Values(make_tuple(6, 6, highbd_12_variance64x64_sse2, 12),
                      make_tuple(6, 5, highbd_12_variance64x32_sse2, 12),
                      make_tuple(5, 6, highbd_12_variance32x64_sse2, 12),
                      make_tuple(5, 5, highbd_12_variance32x32_sse2, 12),
                      make_tuple(5, 4, highbd_12_variance32x16_sse2, 12),
                      make_tuple(4, 5, highbd_12_variance16x32_sse2, 12),
                      make_tuple(4, 4, highbd_12_variance16x16_sse2, 12),
                      make_tuple(4, 3, highbd_12_variance16x8_sse2, 12),
                      make_tuple(3, 4, highbd_12_variance8x16_sse2, 12),
                      make_tuple(3, 3, highbd_12_variance8x8_sse2, 12),
                      make_tuple(6, 6, highbd_10_variance64x64_sse2, 10),
                      make_tuple(6, 5, highbd_10_variance64x32_sse2, 10),
                      make_tuple(5, 6, highbd_10_variance32x64_sse2, 10),
                      make_tuple(5, 5, highbd_10_variance32x32_sse2, 10),
                      make_tuple(5, 4, highbd_10_variance32x16_sse2, 10),
                      make_tuple(4, 5, highbd_10_variance16x32_sse2, 10),
                      make_tuple(4, 4, highbd_10_variance16x16_sse2, 10),
                      make_tuple(4, 3, highbd_10_variance16x8_sse2, 10),
                      make_tuple(3, 4, highbd_10_variance8x16_sse2, 10),
                      make_tuple(3, 3, highbd_10_variance8x8_sse2, 10),
                      make_tuple(6, 6, highbd_8_variance64x64_sse2, 8),
                      make_tuple(6, 5, highbd_8_variance64x32_sse2, 8),
                      make_tuple(5, 6, highbd_8_variance32x64_sse2, 8),
                      make_tuple(5, 5, highbd_8_variance32x32_sse2, 8),
                      make_tuple(5, 4, highbd_8_variance32x16_sse2, 8),
                      make_tuple(4, 5, highbd_8_variance16x32_sse2, 8),
                      make_tuple(4, 4, highbd_8_variance16x16_sse2, 8),
                      make_tuple(4, 3, highbd_8_variance16x8_sse2, 8),
                      make_tuple(3, 4, highbd_8_variance8x16_sse2, 8),
                      make_tuple(3, 3, highbd_8_variance8x8_sse2, 8)));
#endif  // CONFIG_VP9_HIGHBITDEPTH
#endif  // HAVE_SSE2
James Zern's avatar
James Zern committed
1084 1085

#if CONFIG_VP9_ENCODER
1086
typedef SubpelVarianceTest<vp9_subpixvariance_fn_t> VP9SubpelVarianceTest;
1087
typedef SubpelVarianceTest<vp9_subp_avg_variance_fn_t> VP9SubpelAvgVarianceTest;
James Zern's avatar
James Zern committed
1088

1089
TEST_P(VP9SubpelVarianceTest, Ref) { RefTest(); }
1090
TEST_P(VP9SubpelVarianceTest, ExtremeRef) { ExtremeRefTest(); }
1091
TEST_P(VP9SubpelAvgVarianceTest, Ref) { RefTest(); }
James Zern's avatar
James Zern committed
1092

1093 1094 1095 1096 1097 1098 1099 1100 1101 1102
#if CONFIG_VP9_HIGHBITDEPTH
typedef SubpelVarianceTest<vp9_subpixvariance_fn_t> VP9SubpelVarianceHighTest;
typedef SubpelVarianceTest<vp9_subp_avg_variance_fn_t>
    VP9SubpelAvgVarianceHighTest;

TEST_P(VP9SubpelVarianceHighTest, Ref) { RefTest(); }
TEST_P(VP9SubpelVarianceHighTest, ExtremeRef) { ExtremeRefTest(); }
TEST_P(VP9SubpelAvgVarianceHighTest, Ref) { RefTest(); }
#endif  // CONFIG_VP9_HIGHBITDEPTH

1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128
const vp9_subpixvariance_fn_t subpel_variance4x4_c =
    vp9_sub_pixel_variance4x4_c;
const vp9_subpixvariance_fn_t subpel_variance4x8_c =
    vp9_sub_pixel_variance4x8_c;
const vp9_subpixvariance_fn_t subpel_variance8x4_c =
    vp9_sub_pixel_variance8x4_c;
const vp9_subpixvariance_fn_t subpel_variance8x8_c =
    vp9_sub_pixel_variance8x8_c;
const vp9_subpixvariance_fn_t subpel_variance8x16_c =
    vp9_sub_pixel_variance8x16_c;
const vp9_subpixvariance_fn_t subpel_variance16x8_c =
    vp9_sub_pixel_variance16x8_c;
const vp9_subpixvariance_fn_t subpel_variance16x16_c =
    vp9_sub_pixel_variance16x16_c;
const vp9_subpixvariance_fn_t subpel_variance16x32_c =
    vp9_sub_pixel_variance16x32_c;
const vp9_subpixvariance_fn_t subpel_variance32x16_c =
    vp9_sub_pixel_variance32x16_c;
const vp9_subpixvariance_fn_t subpel_variance32x32_c =
    vp9_sub_pixel_variance32x32_c;
const vp9_subpixvariance_fn_t subpel_variance32x64_c =
    vp9_sub_pixel_variance32x64_c;
const vp9_subpixvariance_fn_t subpel_variance64x32_c =
    vp9_sub_pixel_variance64x32_c;
const vp9_subpixvariance_fn_t subpel_variance64x64_c =
    vp9_sub_pixel_variance64x64_c;
1129 1130
INSTANTIATE_TEST_CASE_P(
    C, VP9SubpelVarianceTest,
1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143
    ::testing::Values(make_tuple(2, 2, subpel_variance4x4_c, 0),
                      make_tuple(2, 3, subpel_variance4x8_c, 0),
                      make_tuple(3, 2, subpel_variance8x4_c, 0),
                      make_tuple(3, 3, subpel_variance8x8_c, 0),
                      make_tuple(3, 4, subpel_variance8x16_c, 0),
                      make_tuple(4, 3, subpel_variance16x8_c, 0),
                      make_tuple(4, 4, subpel_variance16x16_c, 0),
                      make_tuple(4, 5, subpel_variance16x32_c, 0),
                      make_tuple(5, 4, subpel_variance32x16_c, 0),
                      make_tuple(5, 5, subpel_variance32x32_c, 0),
                      make_tuple(5, 6, subpel_variance32x64_c, 0),
                      make_tuple(6, 5, subpel_variance64x32_c, 0),
                      make_tuple(6, 6, subpel_variance64x64_c, 0)));
1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169
const vp9_subp_avg_variance_fn_t subpel_avg_variance4x4_c =
    vp9_sub_pixel_avg_variance4x4_c;
const vp9_subp_avg_variance_fn_t subpel_avg_variance4x8_c =
    vp9_sub_pixel_avg_variance4x8_c;
const vp9_subp_avg_variance_fn_t subpel_avg_variance8x4_c =
    vp9_sub_pixel_avg_variance8x4_c;
const vp9_subp_avg_variance_fn_t subpel_avg_variance8x8_c =
    vp9_sub_pixel_avg_variance8x8_c;
const vp9_subp_avg_variance_fn_t subpel_avg_variance8x16_c =
    vp9_sub_pixel_avg_variance8x16_c;
const vp9_subp_avg_variance_fn_t subpel_avg_variance16x8_c =
    vp9_sub_pixel_avg_variance16x8_c;
const vp9_subp_avg_variance_fn_t subpel_avg_variance16x16_c =
    vp9_sub_pixel_avg_variance16x16_c;
const vp9_subp_avg_variance_fn_t subpel_avg_variance16x32_c =
    vp9_sub_pixel_avg_variance16x32_c;
const vp9_subp_avg_variance_fn_t subpel_avg_variance32x16_c =
    vp9_sub_pixel_avg_variance32x16_c;
const vp9_subp_avg_variance_fn_t subpel_avg_variance32x32_c =
    vp9_sub_pixel_avg_variance32x32_c;
const vp9_subp_avg_variance_fn_t subpel_avg_variance32x64_c =
    vp9_sub_pixel_avg_variance32x64_c;
const vp9_subp_avg_variance_fn_t subpel_avg_variance64x32_c =
    vp9_sub_pixel_avg_variance64x32_c;
const vp9_subp_avg_variance_fn_t subpel_avg_variance64x64_c =
    vp9_sub_pixel_avg_variance64x64_c;
1170 1171
INSTANTIATE_TEST_CASE_P(
    C, VP9SubpelAvgVarianceTest,
1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201