1/*
2 *  Copyright (c) 2013 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
17#include "./vp10_rtcd.h"
18#include "./vpx_dsp_rtcd.h"
19#include "test/acm_random.h"
20#include "test/clear_system_state.h"
21#include "test/register_state_check.h"
22#include "test/util.h"
23#include "vp10/common/blockd.h"
24#include "vp10/common/scan.h"
25#include "vpx/vpx_integer.h"
26#include "vp10/common/vp10_inv_txfm.h"
27
28using libvpx_test::ACMRandom;
29
30namespace {
31const double PI = 3.141592653589793238462643383279502884;
32const double kInvSqrt2 = 0.707106781186547524400844362104;
33
34void reference_idct_1d(const double *in, double *out, int size) {
35  for (int n = 0; n < size; ++n) {
36    out[n] = 0;
37    for (int k = 0; k < size; ++k) {
38      if (k == 0)
39        out[n] += kInvSqrt2 * in[k] * cos(PI * (2 * n + 1) * k / (2 * size));
40      else
41        out[n] += in[k] * cos(PI * (2 * n + 1) * k / (2 * size));
42    }
43  }
44}
45
46typedef void (*IdctFuncRef)(const double *in, double *out, int size);
47typedef void (*IdctFunc)(const tran_low_t *in, tran_low_t *out);
48
49class TransTestBase {
50 public:
51  virtual ~TransTestBase() {}
52
53 protected:
54  void RunInvAccuracyCheck() {
55    tran_low_t *input  = new tran_low_t[txfm_size_];
56    tran_low_t *output = new tran_low_t[txfm_size_];
57    double *ref_input  = new double[txfm_size_];
58    double *ref_output = new double[txfm_size_];
59
60    ACMRandom rnd(ACMRandom::DeterministicSeed());
61    const int count_test_block = 5000;
62    for (int ti =  0; ti < count_test_block; ++ti) {
63      for (int ni = 0; ni < txfm_size_; ++ni) {
64        input[ni] = rnd.Rand8() - rnd.Rand8();
65        ref_input[ni] = static_cast<double>(input[ni]);
66      }
67
68      fwd_txfm_(input, output);
69      fwd_txfm_ref_(ref_input, ref_output, txfm_size_);
70
71      for (int ni = 0; ni < txfm_size_; ++ni) {
72        EXPECT_LE(
73            abs(output[ni] - static_cast<tran_low_t>(round(ref_output[ni]))),
74            max_error_);
75      }
76    }
77
78    delete[] input;
79    delete[] output;
80    delete[] ref_input;
81    delete[] ref_output;
82  }
83
84  double max_error_;
85  int txfm_size_;
86  IdctFunc fwd_txfm_;
87  IdctFuncRef fwd_txfm_ref_;
88};
89
90typedef std::tr1::tuple<IdctFunc, IdctFuncRef, int, int> IdctParam;
91class Vp10InvTxfm
92    : public TransTestBase,
93      public ::testing::TestWithParam<IdctParam> {
94 public:
95  virtual void SetUp() {
96    fwd_txfm_ = GET_PARAM(0);
97    fwd_txfm_ref_ = GET_PARAM(1);
98    txfm_size_ = GET_PARAM(2);
99    max_error_ = GET_PARAM(3);
100  }
101  virtual void TearDown() {}
102};
103
104TEST_P(Vp10InvTxfm, RunInvAccuracyCheck) {
105  RunInvAccuracyCheck();
106}
107
108INSTANTIATE_TEST_CASE_P(
109    C, Vp10InvTxfm,
110    ::testing::Values(
111        IdctParam(&vp10_idct4_c, &reference_idct_1d, 4, 1),
112        IdctParam(&vp10_idct8_c, &reference_idct_1d, 8, 2),
113        IdctParam(&vp10_idct16_c, &reference_idct_1d, 16, 4),
114        IdctParam(&vp10_idct32_c, &reference_idct_1d, 32, 6))
115);
116
117typedef void (*FwdTxfmFunc)(const int16_t *in, tran_low_t *out, int stride);
118typedef void (*InvTxfmFunc)(const tran_low_t *in, uint8_t *out, int stride);
119typedef std::tr1::tuple<FwdTxfmFunc,
120                        InvTxfmFunc,
121                        InvTxfmFunc,
122                        TX_SIZE, int> PartialInvTxfmParam;
123const int kMaxNumCoeffs = 1024;
124class Vp10PartialIDctTest
125    : public ::testing::TestWithParam<PartialInvTxfmParam> {
126 public:
127  virtual ~Vp10PartialIDctTest() {}
128  virtual void SetUp() {
129    ftxfm_ = GET_PARAM(0);
130    full_itxfm_ = GET_PARAM(1);
131    partial_itxfm_ = GET_PARAM(2);
132    tx_size_  = GET_PARAM(3);
133    last_nonzero_ = GET_PARAM(4);
134  }
135
136  virtual void TearDown() { libvpx_test::ClearSystemState(); }
137
138 protected:
139  int last_nonzero_;
140  TX_SIZE tx_size_;
141  FwdTxfmFunc ftxfm_;
142  InvTxfmFunc full_itxfm_;
143  InvTxfmFunc partial_itxfm_;
144};
145
146TEST_P(Vp10PartialIDctTest, RunQuantCheck) {
147  ACMRandom rnd(ACMRandom::DeterministicSeed());
148  int size;
149  switch (tx_size_) {
150    case TX_4X4:
151      size = 4;
152      break;
153    case TX_8X8:
154      size = 8;
155      break;
156    case TX_16X16:
157      size = 16;
158      break;
159    case TX_32X32:
160      size = 32;
161      break;
162    default:
163      FAIL() << "Wrong Size!";
164      break;
165  }
166  DECLARE_ALIGNED(16, tran_low_t, test_coef_block1[kMaxNumCoeffs]);
167  DECLARE_ALIGNED(16, tran_low_t, test_coef_block2[kMaxNumCoeffs]);
168  DECLARE_ALIGNED(16, uint8_t, dst1[kMaxNumCoeffs]);
169  DECLARE_ALIGNED(16, uint8_t, dst2[kMaxNumCoeffs]);
170
171  const int count_test_block = 1000;
172  const int block_size = size * size;
173
174  DECLARE_ALIGNED(16, int16_t, input_extreme_block[kMaxNumCoeffs]);
175  DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kMaxNumCoeffs]);
176
177  int max_error = 0;
178  for (int i = 0; i < count_test_block; ++i) {
179    // clear out destination buffer
180    memset(dst1, 0, sizeof(*dst1) * block_size);
181    memset(dst2, 0, sizeof(*dst2) * block_size);
182    memset(test_coef_block1, 0, sizeof(*test_coef_block1) * block_size);
183    memset(test_coef_block2, 0, sizeof(*test_coef_block2) * block_size);
184
185    ACMRandom rnd(ACMRandom::DeterministicSeed());
186
187    for (int i = 0; i < count_test_block; ++i) {
188      // Initialize a test block with input range [-255, 255].
189      if (i == 0) {
190        for (int j = 0; j < block_size; ++j)
191          input_extreme_block[j] = 255;
192      } else if (i == 1) {
193        for (int j = 0; j < block_size; ++j)
194          input_extreme_block[j] = -255;
195      } else {
196        for (int j = 0; j < block_size; ++j) {
197          input_extreme_block[j] = rnd.Rand8() % 2 ? 255 : -255;
198        }
199      }
200
201      ftxfm_(input_extreme_block, output_ref_block, size);
202
203      // quantization with maximum allowed step sizes
204      test_coef_block1[0] = (output_ref_block[0] / 1336) * 1336;
205      for (int j = 1; j < last_nonzero_; ++j)
206        test_coef_block1[vp10_default_scan_orders[tx_size_].scan[j]]
207                         = (output_ref_block[j] / 1828) * 1828;
208    }
209
210    ASM_REGISTER_STATE_CHECK(full_itxfm_(test_coef_block1, dst1, size));
211    ASM_REGISTER_STATE_CHECK(partial_itxfm_(test_coef_block1, dst2, size));
212
213    for (int j = 0; j < block_size; ++j) {
214      const int diff = dst1[j] - dst2[j];
215      const int error = diff * diff;
216      if (max_error < error)
217        max_error = error;
218    }
219  }
220
221  EXPECT_EQ(0, max_error)
222      << "Error: partial inverse transform produces different results";
223}
224
225TEST_P(Vp10PartialIDctTest, ResultsMatch) {
226  ACMRandom rnd(ACMRandom::DeterministicSeed());
227  int size;
228  switch (tx_size_) {
229    case TX_4X4:
230      size = 4;
231      break;
232    case TX_8X8:
233      size = 8;
234      break;
235    case TX_16X16:
236      size = 16;
237      break;
238    case TX_32X32:
239      size = 32;
240      break;
241    default:
242      FAIL() << "Wrong Size!";
243      break;
244  }
245  DECLARE_ALIGNED(16, tran_low_t, test_coef_block1[kMaxNumCoeffs]);
246  DECLARE_ALIGNED(16, tran_low_t, test_coef_block2[kMaxNumCoeffs]);
247  DECLARE_ALIGNED(16, uint8_t, dst1[kMaxNumCoeffs]);
248  DECLARE_ALIGNED(16, uint8_t, dst2[kMaxNumCoeffs]);
249  const int count_test_block = 1000;
250  const int max_coeff = 32766 / 4;
251  const int block_size = size * size;
252  int max_error = 0;
253  for (int i = 0; i < count_test_block; ++i) {
254    // clear out destination buffer
255    memset(dst1, 0, sizeof(*dst1) * block_size);
256    memset(dst2, 0, sizeof(*dst2) * block_size);
257    memset(test_coef_block1, 0, sizeof(*test_coef_block1) * block_size);
258    memset(test_coef_block2, 0, sizeof(*test_coef_block2) * block_size);
259    int max_energy_leftover = max_coeff * max_coeff;
260    for (int j = 0; j < last_nonzero_; ++j) {
261      int16_t coef = static_cast<int16_t>(sqrt(1.0 * max_energy_leftover) *
262                                          (rnd.Rand16() - 32768) / 65536);
263      max_energy_leftover -= coef * coef;
264      if (max_energy_leftover < 0) {
265        max_energy_leftover = 0;
266        coef = 0;
267      }
268      test_coef_block1[vp10_default_scan_orders[tx_size_].scan[j]] = coef;
269    }
270
271    memcpy(test_coef_block2, test_coef_block1,
272           sizeof(*test_coef_block2) * block_size);
273
274    ASM_REGISTER_STATE_CHECK(full_itxfm_(test_coef_block1, dst1, size));
275    ASM_REGISTER_STATE_CHECK(partial_itxfm_(test_coef_block2, dst2, size));
276
277    for (int j = 0; j < block_size; ++j) {
278      const int diff = dst1[j] - dst2[j];
279      const int error = diff * diff;
280      if (max_error < error)
281        max_error = error;
282    }
283  }
284
285  EXPECT_EQ(0, max_error)
286      << "Error: partial inverse transform produces different results";
287}
288using std::tr1::make_tuple;
289
290INSTANTIATE_TEST_CASE_P(
291    C, Vp10PartialIDctTest,
292    ::testing::Values(
293        make_tuple(&vpx_fdct32x32_c,
294                   &vp10_idct32x32_1024_add_c,
295                   &vp10_idct32x32_34_add_c,
296                   TX_32X32, 34),
297        make_tuple(&vpx_fdct32x32_c,
298                   &vp10_idct32x32_1024_add_c,
299                   &vp10_idct32x32_1_add_c,
300                   TX_32X32, 1),
301        make_tuple(&vpx_fdct16x16_c,
302                   &vp10_idct16x16_256_add_c,
303                   &vp10_idct16x16_10_add_c,
304                   TX_16X16, 10),
305        make_tuple(&vpx_fdct16x16_c,
306                   &vp10_idct16x16_256_add_c,
307                   &vp10_idct16x16_1_add_c,
308                   TX_16X16, 1),
309        make_tuple(&vpx_fdct8x8_c,
310                   &vp10_idct8x8_64_add_c,
311                   &vp10_idct8x8_12_add_c,
312                   TX_8X8, 12),
313        make_tuple(&vpx_fdct8x8_c,
314                   &vp10_idct8x8_64_add_c,
315                   &vp10_idct8x8_1_add_c,
316                   TX_8X8, 1),
317        make_tuple(&vpx_fdct4x4_c,
318                   &vp10_idct4x4_16_add_c,
319                   &vp10_idct4x4_1_add_c,
320                   TX_4X4, 1)));
321}  // namespace
322