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 "vpx/vpx_integer.h"
24
25 extern "C" {
26 void vp9_idct4x4_16_add_c(const int16_t *input, uint8_t *output, int pitch);
27 }
28
29 using libvpx_test::ACMRandom;
30
31 namespace {
32 const int kNumCoeffs = 16;
33 typedef void (*fdct_t)(const int16_t *in, int16_t *out, int stride);
34 typedef void (*idct_t)(const int16_t *in, uint8_t *out, int stride);
35 typedef void (*fht_t) (const int16_t *in, int16_t *out, int stride,
36 int tx_type);
37 typedef void (*iht_t) (const int16_t *in, uint8_t *out, int stride,
38 int tx_type);
39
40 typedef std::tr1::tuple<fdct_t, idct_t, int> dct_4x4_param_t;
41 typedef std::tr1::tuple<fht_t, iht_t, int> ht_4x4_param_t;
42
fdct4x4_ref(const int16_t * in,int16_t * out,int stride,int tx_type)43 void fdct4x4_ref(const int16_t *in, int16_t *out, int stride, int tx_type) {
44 vp9_fdct4x4_c(in, out, stride);
45 }
46
fht4x4_ref(const int16_t * in,int16_t * out,int stride,int tx_type)47 void fht4x4_ref(const int16_t *in, int16_t *out, int stride, int tx_type) {
48 vp9_fht4x4_c(in, out, stride, tx_type);
49 }
50
51 class Trans4x4TestBase {
52 public:
~Trans4x4TestBase()53 virtual ~Trans4x4TestBase() {}
54
55 protected:
56 virtual void RunFwdTxfm(const int16_t *in, int16_t *out, int stride) = 0;
57
58 virtual void RunInvTxfm(const int16_t *out, uint8_t *dst, int stride) = 0;
59
RunAccuracyCheck()60 void RunAccuracyCheck() {
61 ACMRandom rnd(ACMRandom::DeterministicSeed());
62 uint32_t max_error = 0;
63 int64_t total_error = 0;
64 const int count_test_block = 10000;
65 for (int i = 0; i < count_test_block; ++i) {
66 DECLARE_ALIGNED_ARRAY(16, int16_t, test_input_block, kNumCoeffs);
67 DECLARE_ALIGNED_ARRAY(16, int16_t, test_temp_block, kNumCoeffs);
68 DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
69 DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
70
71 // Initialize a test block with input range [-255, 255].
72 for (int j = 0; j < kNumCoeffs; ++j) {
73 src[j] = rnd.Rand8();
74 dst[j] = rnd.Rand8();
75 test_input_block[j] = src[j] - dst[j];
76 }
77
78 REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block,
79 test_temp_block, pitch_));
80 REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block, dst, pitch_));
81
82 for (int j = 0; j < kNumCoeffs; ++j) {
83 const uint32_t diff = dst[j] - src[j];
84 const uint32_t error = diff * diff;
85 if (max_error < error)
86 max_error = error;
87 total_error += error;
88 }
89 }
90
91 EXPECT_GE(1u, max_error)
92 << "Error: 4x4 FHT/IHT has an individual round trip error > 1";
93
94 EXPECT_GE(count_test_block , total_error)
95 << "Error: 4x4 FHT/IHT has average round trip error > 1 per block";
96 }
97
RunCoeffCheck()98 void RunCoeffCheck() {
99 ACMRandom rnd(ACMRandom::DeterministicSeed());
100 const int count_test_block = 5000;
101 DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
102 DECLARE_ALIGNED_ARRAY(16, int16_t, output_ref_block, kNumCoeffs);
103 DECLARE_ALIGNED_ARRAY(16, int16_t, output_block, kNumCoeffs);
104
105 for (int i = 0; i < count_test_block; ++i) {
106 // Initialize a test block with input range [-255, 255].
107 for (int j = 0; j < kNumCoeffs; ++j)
108 input_block[j] = rnd.Rand8() - rnd.Rand8();
109
110 fwd_txfm_ref(input_block, output_ref_block, pitch_, tx_type_);
111 REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
112
113 // The minimum quant value is 4.
114 for (int j = 0; j < kNumCoeffs; ++j)
115 EXPECT_EQ(output_block[j], output_ref_block[j]);
116 }
117 }
118
RunMemCheck()119 void RunMemCheck() {
120 ACMRandom rnd(ACMRandom::DeterministicSeed());
121 const int count_test_block = 5000;
122 DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
123 DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kNumCoeffs);
124 DECLARE_ALIGNED_ARRAY(16, int16_t, output_ref_block, kNumCoeffs);
125 DECLARE_ALIGNED_ARRAY(16, int16_t, output_block, kNumCoeffs);
126
127 for (int i = 0; i < count_test_block; ++i) {
128 // Initialize a test block with input range [-255, 255].
129 for (int j = 0; j < kNumCoeffs; ++j) {
130 input_block[j] = rnd.Rand8() - rnd.Rand8();
131 input_extreme_block[j] = rnd.Rand8() % 2 ? 255 : -255;
132 }
133 if (i == 0)
134 for (int j = 0; j < kNumCoeffs; ++j)
135 input_extreme_block[j] = 255;
136 if (i == 1)
137 for (int j = 0; j < kNumCoeffs; ++j)
138 input_extreme_block[j] = -255;
139
140 fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
141 REGISTER_STATE_CHECK(RunFwdTxfm(input_extreme_block,
142 output_block, pitch_));
143
144 // The minimum quant value is 4.
145 for (int j = 0; j < kNumCoeffs; ++j) {
146 EXPECT_EQ(output_block[j], output_ref_block[j]);
147 EXPECT_GE(4 * DCT_MAX_VALUE, abs(output_block[j]))
148 << "Error: 16x16 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
149 }
150 }
151 }
152
RunInvAccuracyCheck()153 void RunInvAccuracyCheck() {
154 ACMRandom rnd(ACMRandom::DeterministicSeed());
155 const int count_test_block = 1000;
156 DECLARE_ALIGNED_ARRAY(16, int16_t, in, kNumCoeffs);
157 DECLARE_ALIGNED_ARRAY(16, int16_t, coeff, kNumCoeffs);
158 DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
159 DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
160
161 for (int i = 0; i < count_test_block; ++i) {
162 // Initialize a test block with input range [-255, 255].
163 for (int j = 0; j < kNumCoeffs; ++j) {
164 src[j] = rnd.Rand8();
165 dst[j] = rnd.Rand8();
166 in[j] = src[j] - dst[j];
167 }
168
169 fwd_txfm_ref(in, coeff, pitch_, tx_type_);
170
171 REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
172
173 for (int j = 0; j < kNumCoeffs; ++j) {
174 const uint32_t diff = dst[j] - src[j];
175 const uint32_t error = diff * diff;
176 EXPECT_GE(1u, error)
177 << "Error: 16x16 IDCT has error " << error
178 << " at index " << j;
179 }
180 }
181 }
182
183 int pitch_;
184 int tx_type_;
185 fht_t fwd_txfm_ref;
186 };
187
188 class Trans4x4DCT
189 : public Trans4x4TestBase,
190 public ::testing::TestWithParam<dct_4x4_param_t> {
191 public:
~Trans4x4DCT()192 virtual ~Trans4x4DCT() {}
193
SetUp()194 virtual void SetUp() {
195 fwd_txfm_ = GET_PARAM(0);
196 inv_txfm_ = GET_PARAM(1);
197 tx_type_ = GET_PARAM(2);
198 pitch_ = 4;
199 fwd_txfm_ref = fdct4x4_ref;
200 }
TearDown()201 virtual void TearDown() { libvpx_test::ClearSystemState(); }
202
203 protected:
RunFwdTxfm(const int16_t * in,int16_t * out,int stride)204 void RunFwdTxfm(const int16_t *in, int16_t *out, int stride) {
205 fwd_txfm_(in, out, stride);
206 }
RunInvTxfm(const int16_t * out,uint8_t * dst,int stride)207 void RunInvTxfm(const int16_t *out, uint8_t *dst, int stride) {
208 inv_txfm_(out, dst, stride);
209 }
210
211 fdct_t fwd_txfm_;
212 idct_t inv_txfm_;
213 };
214
TEST_P(Trans4x4DCT,AccuracyCheck)215 TEST_P(Trans4x4DCT, AccuracyCheck) {
216 RunAccuracyCheck();
217 }
218
TEST_P(Trans4x4DCT,CoeffCheck)219 TEST_P(Trans4x4DCT, CoeffCheck) {
220 RunCoeffCheck();
221 }
222
TEST_P(Trans4x4DCT,MemCheck)223 TEST_P(Trans4x4DCT, MemCheck) {
224 RunMemCheck();
225 }
226
TEST_P(Trans4x4DCT,InvAccuracyCheck)227 TEST_P(Trans4x4DCT, InvAccuracyCheck) {
228 RunInvAccuracyCheck();
229 }
230
231 class Trans4x4HT
232 : public Trans4x4TestBase,
233 public ::testing::TestWithParam<ht_4x4_param_t> {
234 public:
~Trans4x4HT()235 virtual ~Trans4x4HT() {}
236
SetUp()237 virtual void SetUp() {
238 fwd_txfm_ = GET_PARAM(0);
239 inv_txfm_ = GET_PARAM(1);
240 tx_type_ = GET_PARAM(2);
241 pitch_ = 4;
242 fwd_txfm_ref = fht4x4_ref;
243 }
TearDown()244 virtual void TearDown() { libvpx_test::ClearSystemState(); }
245
246 protected:
RunFwdTxfm(const int16_t * in,int16_t * out,int stride)247 void RunFwdTxfm(const int16_t *in, int16_t *out, int stride) {
248 fwd_txfm_(in, out, stride, tx_type_);
249 }
250
RunInvTxfm(const int16_t * out,uint8_t * dst,int stride)251 void RunInvTxfm(const int16_t *out, uint8_t *dst, int stride) {
252 inv_txfm_(out, dst, stride, tx_type_);
253 }
254
255 fht_t fwd_txfm_;
256 iht_t inv_txfm_;
257 };
258
TEST_P(Trans4x4HT,AccuracyCheck)259 TEST_P(Trans4x4HT, AccuracyCheck) {
260 RunAccuracyCheck();
261 }
262
TEST_P(Trans4x4HT,CoeffCheck)263 TEST_P(Trans4x4HT, CoeffCheck) {
264 RunCoeffCheck();
265 }
266
TEST_P(Trans4x4HT,MemCheck)267 TEST_P(Trans4x4HT, MemCheck) {
268 RunMemCheck();
269 }
270
TEST_P(Trans4x4HT,InvAccuracyCheck)271 TEST_P(Trans4x4HT, InvAccuracyCheck) {
272 RunInvAccuracyCheck();
273 }
274
275 using std::tr1::make_tuple;
276
277 INSTANTIATE_TEST_CASE_P(
278 C, Trans4x4DCT,
279 ::testing::Values(
280 make_tuple(&vp9_fdct4x4_c, &vp9_idct4x4_16_add_c, 0)));
281 INSTANTIATE_TEST_CASE_P(
282 C, Trans4x4HT,
283 ::testing::Values(
284 make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 0),
285 make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 1),
286 make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 2),
287 make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 3)));
288
289 #if HAVE_NEON
290 INSTANTIATE_TEST_CASE_P(
291 NEON, Trans4x4DCT,
292 ::testing::Values(
293 make_tuple(&vp9_fdct4x4_c,
294 &vp9_idct4x4_16_add_neon, 0)));
295 INSTANTIATE_TEST_CASE_P(
296 DISABLED_NEON, Trans4x4HT,
297 ::testing::Values(
298 make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 0),
299 make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 1),
300 make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 2),
301 make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 3)));
302 #endif
303
304 #if HAVE_SSE2
305 INSTANTIATE_TEST_CASE_P(
306 SSE2, Trans4x4DCT,
307 ::testing::Values(
308 make_tuple(&vp9_fdct4x4_sse2,
309 &vp9_idct4x4_16_add_sse2, 0)));
310 INSTANTIATE_TEST_CASE_P(
311 SSE2, Trans4x4HT,
312 ::testing::Values(
313 make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 0),
314 make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 1),
315 make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 2),
316 make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 3)));
317 #endif
318
319 } // namespace
320