av1_fht16x16_test.cc 9.83 KB
Newer Older
1
/*
2
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3
 *
4 5 6 7 8 9
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10 11
 */

12
#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
13

14 15
#include "./av1_rtcd.h"
#include "./aom_dsp_rtcd.h"
16 17 18 19 20 21

#include "test/acm_random.h"
#include "test/clear_system_state.h"
#include "test/register_state_check.h"
#include "test/transform_test_base.h"
#include "test/util.h"
22
#include "aom_ports/mem.h"
23

24
using libaom_test::ACMRandom;
25 26 27

namespace {
typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
28
                        const TxfmParam *txfm_param);
29
using libaom_test::FhtFunc;
Johann's avatar
Johann committed
30
using std::tr1::tuple;
31
typedef tuple<FhtFunc, IhtFunc, TX_TYPE, aom_bit_depth_t, int> Ht16x16Param;
32

33
void fht16x16_ref(const int16_t *in, tran_low_t *out, int stride,
34 35
                  TxfmParam *txfm_param) {
  av1_fht16x16_c(in, out, stride, txfm_param);
36 37
}

38
void iht16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
39 40
                  const TxfmParam *txfm_param) {
  av1_iht16x16_256_add_c(in, dest, stride, txfm_param);
41 42
}

43
typedef void (*IHbdHtFunc)(const tran_low_t *in, uint8_t *out, int stride,
44
                           TX_TYPE tx_type, int bd);
45
typedef void (*HbdHtFunc)(const int16_t *input, int32_t *output, int stride,
46
                          TX_TYPE tx_type, int bd);
47 48

// Target optimized function, tx_type, bit depth
49
typedef tuple<HbdHtFunc, TX_TYPE, int> HighbdHt16x16Param;
50 51

void highbd_fht16x16_ref(const int16_t *in, int32_t *out, int stride,
52
                         TX_TYPE tx_type, int bd) {
53
  av1_fwd_txfm2d_16x16_c(in, out, stride, tx_type, bd);
54 55
}

56 57
class AV1Trans16x16HT : public libaom_test::TransformTestBase,
                        public ::testing::TestWithParam<Ht16x16Param> {
58
 public:
59
  virtual ~AV1Trans16x16HT() {}
60 61 62 63

  virtual void SetUp() {
    fwd_txfm_ = GET_PARAM(0);
    inv_txfm_ = GET_PARAM(1);
clang-format's avatar
clang-format committed
64
    pitch_ = 16;
65
    height_ = 16;
66
    fwd_txfm_ref = fht16x16_ref;
67
    inv_txfm_ref = iht16x16_ref;
68 69 70
    bit_depth_ = GET_PARAM(3);
    mask_ = (1 << bit_depth_) - 1;
    num_coeffs_ = GET_PARAM(4);
71
    txfm_param_.tx_type = GET_PARAM(2);
72
  }
73
  virtual void TearDown() { libaom_test::ClearSystemState(); }
74 75 76

 protected:
  void RunFwdTxfm(const int16_t *in, tran_low_t *out, int stride) {
77
    fwd_txfm_(in, out, stride, &txfm_param_);
78 79 80
  }

  void RunInvTxfm(const tran_low_t *out, uint8_t *dst, int stride) {
81
    inv_txfm_(out, dst, stride, &txfm_param_);
82 83 84 85 86 87
  }

  FhtFunc fwd_txfm_;
  IhtFunc inv_txfm_;
};

88
TEST_P(AV1Trans16x16HT, MemCheck) { RunMemCheck(); }
Angie Chiang's avatar
Angie Chiang committed
89
TEST_P(AV1Trans16x16HT, AccuracyCheck) { RunAccuracyCheck(1, 0.001); }
90
TEST_P(AV1Trans16x16HT, InvAccuracyCheck) { RunInvAccuracyCheck(1); }
91
TEST_P(AV1Trans16x16HT, CoeffCheck) { RunCoeffCheck(); }
92
TEST_P(AV1Trans16x16HT, InvCoeffCheck) { RunInvCoeffCheck(); }
93

94
class AV1HighbdTrans16x16HT
95 96
    : public ::testing::TestWithParam<HighbdHt16x16Param> {
 public:
97
  virtual ~AV1HighbdTrans16x16HT() {}
98 99 100 101

  virtual void SetUp() {
    fwd_txfm_ = GET_PARAM(0);
    fwd_txfm_ref_ = highbd_fht16x16_ref;
clang-format's avatar
clang-format committed
102
    tx_type_ = GET_PARAM(1);
103 104 105 106 107
    bit_depth_ = GET_PARAM(2);
    mask_ = (1 << bit_depth_) - 1;
    num_coeffs_ = 256;

    input_ = reinterpret_cast<int16_t *>(
108
        aom_memalign(32, sizeof(int16_t) * num_coeffs_));
109
    output_ = reinterpret_cast<int32_t *>(
110
        aom_memalign(32, sizeof(int32_t) * num_coeffs_));
111
    output_ref_ = reinterpret_cast<int32_t *>(
112
        aom_memalign(32, sizeof(int32_t) * num_coeffs_));
113 114 115
  }

  virtual void TearDown() {
116 117 118
    aom_free(input_);
    aom_free(output_);
    aom_free(output_ref_);
119
    libaom_test::ClearSystemState();
120 121 122 123 124 125 126 127
  }

 protected:
  void RunBitexactCheck();

 private:
  HbdHtFunc fwd_txfm_;
  HbdHtFunc fwd_txfm_ref_;
128
  TX_TYPE tx_type_;
129 130 131 132 133 134 135 136
  int bit_depth_;
  int mask_;
  int num_coeffs_;
  int16_t *input_;
  int32_t *output_;
  int32_t *output_ref_;
};

137
void AV1HighbdTrans16x16HT::RunBitexactCheck() {
138 139 140
  ACMRandom rnd(ACMRandom::DeterministicSeed());
  int i, j;
  const int stride = 16;
141
  const int num_tests = 1000;
142 143 144 145

  for (i = 0; i < num_tests; ++i) {
    for (j = 0; j < num_coeffs_; ++j) {
      input_[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
146 147
    }

148
    fwd_txfm_ref_(input_, output_ref_, stride, tx_type_, bit_depth_);
clang-format's avatar
clang-format committed
149 150
    ASM_REGISTER_STATE_CHECK(
        fwd_txfm_(input_, output_, stride, tx_type_, bit_depth_));
151 152 153

    for (j = 0; j < num_coeffs_; ++j) {
      EXPECT_EQ(output_ref_[j], output_[j])
clang-format's avatar
clang-format committed
154
          << "Not bit-exact result at index: " << j << " at test block: " << i;
155
    }
156 157
  }
}
158

159
TEST_P(AV1HighbdTrans16x16HT, HighbdCoeffCheck) { RunBitexactCheck(); }
160

161 162
using std::tr1::make_tuple;

163
#if HAVE_SSE2
164
const Ht16x16Param kArrayHt16x16Param_sse2[] = {
165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, DCT_DCT,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, ADST_DCT,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, DCT_ADST,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, ADST_ADST,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, FLIPADST_DCT,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, DCT_FLIPADST,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, FLIPADST_FLIPADST,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, ADST_FLIPADST,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, FLIPADST_ADST,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, IDTX, AOM_BITS_8,
clang-format's avatar
clang-format committed
184
             256),
185
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, V_DCT, AOM_BITS_8,
clang-format's avatar
clang-format committed
186
             256),
187
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, H_DCT, AOM_BITS_8,
clang-format's avatar
clang-format committed
188
             256),
189
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, V_ADST, AOM_BITS_8,
clang-format's avatar
clang-format committed
190
             256),
191
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, H_ADST, AOM_BITS_8,
clang-format's avatar
clang-format committed
192
             256),
193 194 195 196
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, V_FLIPADST,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_sse2, &av1_iht16x16_256_add_sse2, H_FLIPADST,
             AOM_BITS_8, 256)
197
};
198
INSTANTIATE_TEST_CASE_P(SSE2, AV1Trans16x16HT,
clang-format's avatar
clang-format committed
199
                        ::testing::ValuesIn(kArrayHt16x16Param_sse2));
200
#endif  // HAVE_SSE2
201

202
#if HAVE_AVX2
203
const Ht16x16Param kArrayHt16x16Param_avx2[] = {
204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, DCT_DCT,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, ADST_DCT,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, DCT_ADST,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, ADST_ADST,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, FLIPADST_DCT,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, DCT_FLIPADST,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, FLIPADST_FLIPADST,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, ADST_FLIPADST,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, FLIPADST_ADST,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, IDTX, AOM_BITS_8,
223
             256),
224
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, V_DCT, AOM_BITS_8,
Tom Finegan's avatar
Tom Finegan committed
225
             256),
226
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, H_DCT, AOM_BITS_8,
Tom Finegan's avatar
Tom Finegan committed
227
             256),
228
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, V_ADST, AOM_BITS_8,
Tom Finegan's avatar
Tom Finegan committed
229
             256),
230
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, H_ADST, AOM_BITS_8,
Tom Finegan's avatar
Tom Finegan committed
231
             256),
232 233 234 235
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, V_FLIPADST,
             AOM_BITS_8, 256),
  make_tuple(&av1_fht16x16_avx2, &av1_iht16x16_256_add_avx2, H_FLIPADST,
             AOM_BITS_8, 256)
236 237 238
};
INSTANTIATE_TEST_CASE_P(AVX2, AV1Trans16x16HT,
                        ::testing::ValuesIn(kArrayHt16x16Param_avx2));
239
#endif  // HAVE_AVX2
240

241
#if HAVE_SSE4_1
242
const HighbdHt16x16Param kArrayHBDHt16x16Param_sse4_1[] = {
243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, DCT_DCT, 10),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, DCT_DCT, 12),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, ADST_DCT, 10),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, ADST_DCT, 12),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, DCT_ADST, 10),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, DCT_ADST, 12),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, ADST_ADST, 10),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, ADST_ADST, 12),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, FLIPADST_DCT, 10),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, FLIPADST_DCT, 12),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, DCT_FLIPADST, 10),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, DCT_FLIPADST, 12),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, FLIPADST_FLIPADST, 10),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, FLIPADST_FLIPADST, 12),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, ADST_FLIPADST, 10),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, ADST_FLIPADST, 12),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, FLIPADST_ADST, 10),
  make_tuple(&av1_fwd_txfm2d_16x16_sse4_1, FLIPADST_ADST, 12),
261
};
262
INSTANTIATE_TEST_CASE_P(SSE4_1, AV1HighbdTrans16x16HT,
clang-format's avatar
clang-format committed
263
                        ::testing::ValuesIn(kArrayHBDHt16x16Param_sse4_1));
264
#endif  // HAVE_SSE4_1
265

266
}  // namespace