av1_convolve_ x,y _avx2() -- use 256 bit load/store
[aom.git] / test / av1_fht8x8_test.cc
bloba73053f01174c4288941aa4d989652fa83dc7002
1 /*
2 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
4 * This source code is subject to the terms of the BSD 2 Clause License and
5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6 * was not distributed with this source code in the LICENSE file, you can
7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8 * Media Patent License 1.0 was not distributed with this source code in the
9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
12 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
14 #include "./av1_rtcd.h"
15 #include "./aom_dsp_rtcd.h"
17 #include "test/acm_random.h"
18 #include "test/clear_system_state.h"
19 #include "test/register_state_check.h"
20 #include "test/transform_test_base.h"
21 #include "test/util.h"
22 #include "aom_ports/mem.h"
24 using libaom_test::ACMRandom;
26 #if !CONFIG_DAALA_TX
27 namespace {
28 typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
29 const TxfmParam *txfm_param);
31 using libaom_test::FhtFunc;
32 using std::tr1::tuple;
33 typedef tuple<FhtFunc, IhtFunc, TX_TYPE, aom_bit_depth_t, int> Ht8x8Param;
35 void fht8x8_ref(const int16_t *in, tran_low_t *out, int stride,
36 TxfmParam *txfm_param) {
37 av1_fht8x8_c(in, out, stride, txfm_param);
40 void iht8x8_ref(const tran_low_t *in, uint8_t *out, int stride,
41 const TxfmParam *txfm_param) {
42 av1_iht8x8_64_add_c(in, out, stride, txfm_param);
45 typedef void (*IHbdHtFunc)(const tran_low_t *in, uint8_t *out, int stride,
46 TX_TYPE tx_type, int bd);
47 typedef void (*HbdHtFunc)(const int16_t *input, int32_t *output, int stride,
48 TX_TYPE tx_type, int bd);
49 // Target optimized function, tx_type, bit depth
50 typedef tuple<HbdHtFunc, TX_TYPE, int> HighbdHt8x8Param;
52 void highbd_fht8x8_ref(const int16_t *in, int32_t *out, int stride,
53 TX_TYPE tx_type, int bd) {
54 av1_fwd_txfm2d_8x8_c(in, out, stride, tx_type, bd);
57 class AV1Trans8x8HT : public libaom_test::TransformTestBase,
58 public ::testing::TestWithParam<Ht8x8Param> {
59 public:
60 virtual ~AV1Trans8x8HT() {}
62 virtual void SetUp() {
63 fwd_txfm_ = GET_PARAM(0);
64 inv_txfm_ = GET_PARAM(1);
65 pitch_ = 8;
66 height_ = 8;
67 fwd_txfm_ref = fht8x8_ref;
68 inv_txfm_ref = iht8x8_ref;
69 bit_depth_ = GET_PARAM(3);
70 mask_ = (1 << bit_depth_) - 1;
71 num_coeffs_ = GET_PARAM(4);
72 txfm_param_.tx_type = GET_PARAM(2);
74 virtual void TearDown() { libaom_test::ClearSystemState(); }
76 protected:
77 void RunFwdTxfm(const int16_t *in, tran_low_t *out, int stride) {
78 fwd_txfm_(in, out, stride, &txfm_param_);
81 void RunInvTxfm(const tran_low_t *out, uint8_t *dst, int stride) {
82 inv_txfm_(out, dst, stride, &txfm_param_);
85 FhtFunc fwd_txfm_;
86 IhtFunc inv_txfm_;
89 TEST_P(AV1Trans8x8HT, MemCheck) { RunMemCheck(); }
90 TEST_P(AV1Trans8x8HT, CoeffCheck) { RunCoeffCheck(); }
91 // Note:
92 // TODO(luoyi): Add tx_type, 9-15 for inverse transform.
93 // Need cleanup since same tests may be done in fdct8x8_test.cc
94 // TEST_P(AV1Trans8x8HT, AccuracyCheck) { RunAccuracyCheck(0); }
95 // TEST_P(AV1Trans8x8HT, InvAccuracyCheck) { RunInvAccuracyCheck(0); }
96 // TEST_P(AV1Trans8x8HT, InvCoeffCheck) { RunInvCoeffCheck(); }
98 class AV1HighbdTrans8x8HT : public ::testing::TestWithParam<HighbdHt8x8Param> {
99 public:
100 virtual ~AV1HighbdTrans8x8HT() {}
102 virtual void SetUp() {
103 fwd_txfm_ = GET_PARAM(0);
104 fwd_txfm_ref_ = highbd_fht8x8_ref;
105 tx_type_ = GET_PARAM(1);
106 bit_depth_ = GET_PARAM(2);
107 mask_ = (1 << bit_depth_) - 1;
108 num_coeffs_ = 64;
110 input_ = reinterpret_cast<int16_t *>(
111 aom_memalign(16, sizeof(int16_t) * num_coeffs_));
112 output_ = reinterpret_cast<int32_t *>(
113 aom_memalign(16, sizeof(int32_t) * num_coeffs_));
114 output_ref_ = reinterpret_cast<int32_t *>(
115 aom_memalign(16, sizeof(int32_t) * num_coeffs_));
118 virtual void TearDown() {
119 aom_free(input_);
120 aom_free(output_);
121 aom_free(output_ref_);
122 libaom_test::ClearSystemState();
125 protected:
126 void RunBitexactCheck();
128 private:
129 HbdHtFunc fwd_txfm_;
130 HbdHtFunc fwd_txfm_ref_;
131 TX_TYPE tx_type_;
132 int bit_depth_;
133 int mask_;
134 int num_coeffs_;
135 int16_t *input_;
136 int32_t *output_;
137 int32_t *output_ref_;
140 void AV1HighbdTrans8x8HT::RunBitexactCheck() {
141 ACMRandom rnd(ACMRandom::DeterministicSeed());
142 int i, j;
143 const int stride = 8;
144 const int num_tests = 1000;
145 const int num_coeffs = 64;
147 for (i = 0; i < num_tests; ++i) {
148 for (j = 0; j < num_coeffs; ++j) {
149 input_[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
152 fwd_txfm_ref_(input_, output_ref_, stride, tx_type_, bit_depth_);
153 ASM_REGISTER_STATE_CHECK(
154 fwd_txfm_(input_, output_, stride, tx_type_, bit_depth_));
156 for (j = 0; j < num_coeffs; ++j) {
157 EXPECT_EQ(output_ref_[j], output_[j])
158 << "Not bit-exact result at index: " << j << " at test block: " << i;
163 TEST_P(AV1HighbdTrans8x8HT, HighbdCoeffCheck) { RunBitexactCheck(); }
165 using std::tr1::make_tuple;
167 #if HAVE_SSE2 && !CONFIG_DAALA_TX8
168 const Ht8x8Param kArrayHt8x8Param_sse2[] = {
169 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, DCT_DCT, AOM_BITS_8,
170 64),
171 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, ADST_DCT, AOM_BITS_8,
172 64),
173 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, DCT_ADST, AOM_BITS_8,
174 64),
175 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, ADST_ADST, AOM_BITS_8,
176 64),
177 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, FLIPADST_DCT,
178 AOM_BITS_8, 64),
179 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, DCT_FLIPADST,
180 AOM_BITS_8, 64),
181 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, FLIPADST_FLIPADST,
182 AOM_BITS_8, 64),
183 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, ADST_FLIPADST,
184 AOM_BITS_8, 64),
185 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, FLIPADST_ADST,
186 AOM_BITS_8, 64),
187 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, IDTX, AOM_BITS_8, 64),
188 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, V_DCT, AOM_BITS_8, 64),
189 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, H_DCT, AOM_BITS_8, 64),
190 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, V_ADST, AOM_BITS_8, 64),
191 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, H_ADST, AOM_BITS_8, 64),
192 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, V_FLIPADST, AOM_BITS_8,
193 64),
194 make_tuple(&av1_fht8x8_sse2, &av1_iht8x8_64_add_sse2, H_FLIPADST, AOM_BITS_8,
197 INSTANTIATE_TEST_CASE_P(SSE2, AV1Trans8x8HT,
198 ::testing::ValuesIn(kArrayHt8x8Param_sse2));
199 #endif // HAVE_SSE2
201 #if HAVE_SSE4_1 && !CONFIG_DAALA_TX8
202 const HighbdHt8x8Param kArrayHBDHt8x8Param_sse4_1[] = {
203 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, DCT_DCT, 10),
204 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, DCT_DCT, 12),
205 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, ADST_DCT, 10),
206 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, ADST_DCT, 12),
207 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, DCT_ADST, 10),
208 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, DCT_ADST, 12),
209 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, ADST_ADST, 10),
210 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, ADST_ADST, 12),
211 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, FLIPADST_DCT, 10),
212 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, FLIPADST_DCT, 12),
213 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, DCT_FLIPADST, 10),
214 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, DCT_FLIPADST, 12),
215 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, FLIPADST_FLIPADST, 10),
216 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, FLIPADST_FLIPADST, 12),
217 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, ADST_FLIPADST, 10),
218 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, ADST_FLIPADST, 12),
219 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, FLIPADST_ADST, 10),
220 make_tuple(&av1_fwd_txfm2d_8x8_sse4_1, FLIPADST_ADST, 12),
222 INSTANTIATE_TEST_CASE_P(SSE4_1, AV1HighbdTrans8x8HT,
223 ::testing::ValuesIn(kArrayHBDHt8x8Param_sse4_1));
224 #endif // HAVE_SSE4_1 && !CONFIG_DAALA_TX8
226 } // namespace
227 #endif // !CONFIG_DAALA_TX