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
25extern "C" {
26void vp9_idct16x16_256_add_c(const int16_t *input, uint8_t *output, int pitch);
27}
28
29using libvpx_test::ACMRandom;
30
31namespace {
32
33#ifdef _MSC_VER
34static int round(double x) {
35  if (x < 0)
36    return static_cast<int>(ceil(x - 0.5));
37  else
38    return static_cast<int>(floor(x + 0.5));
39}
40#endif
41
42const int kNumCoeffs = 256;
43const double PI = 3.1415926535898;
44void reference2_16x16_idct_2d(double *input, double *output) {
45  double x;
46  for (int l = 0; l < 16; ++l) {
47    for (int k = 0; k < 16; ++k) {
48      double s = 0;
49      for (int i = 0; i < 16; ++i) {
50        for (int j = 0; j < 16; ++j) {
51          x = cos(PI * j * (l + 0.5) / 16.0) *
52              cos(PI * i * (k + 0.5) / 16.0) *
53              input[i * 16 + j] / 256;
54          if (i != 0)
55            x *= sqrt(2.0);
56          if (j != 0)
57            x *= sqrt(2.0);
58          s += x;
59        }
60      }
61      output[k*16+l] = s;
62    }
63  }
64}
65
66
67const double C1 = 0.995184726672197;
68const double C2 = 0.98078528040323;
69const double C3 = 0.956940335732209;
70const double C4 = 0.923879532511287;
71const double C5 = 0.881921264348355;
72const double C6 = 0.831469612302545;
73const double C7 = 0.773010453362737;
74const double C8 = 0.707106781186548;
75const double C9 = 0.634393284163646;
76const double C10 = 0.555570233019602;
77const double C11 = 0.471396736825998;
78const double C12 = 0.38268343236509;
79const double C13 = 0.290284677254462;
80const double C14 = 0.195090322016128;
81const double C15 = 0.098017140329561;
82
83void butterfly_16x16_dct_1d(double input[16], double output[16]) {
84  double step[16];
85  double intermediate[16];
86  double temp1, temp2;
87
88  // step 1
89  step[ 0] = input[0] + input[15];
90  step[ 1] = input[1] + input[14];
91  step[ 2] = input[2] + input[13];
92  step[ 3] = input[3] + input[12];
93  step[ 4] = input[4] + input[11];
94  step[ 5] = input[5] + input[10];
95  step[ 6] = input[6] + input[ 9];
96  step[ 7] = input[7] + input[ 8];
97  step[ 8] = input[7] - input[ 8];
98  step[ 9] = input[6] - input[ 9];
99  step[10] = input[5] - input[10];
100  step[11] = input[4] - input[11];
101  step[12] = input[3] - input[12];
102  step[13] = input[2] - input[13];
103  step[14] = input[1] - input[14];
104  step[15] = input[0] - input[15];
105
106  // step 2
107  output[0] = step[0] + step[7];
108  output[1] = step[1] + step[6];
109  output[2] = step[2] + step[5];
110  output[3] = step[3] + step[4];
111  output[4] = step[3] - step[4];
112  output[5] = step[2] - step[5];
113  output[6] = step[1] - step[6];
114  output[7] = step[0] - step[7];
115
116  temp1 = step[ 8] * C7;
117  temp2 = step[15] * C9;
118  output[ 8] = temp1 + temp2;
119
120  temp1 = step[ 9] * C11;
121  temp2 = step[14] * C5;
122  output[ 9] = temp1 - temp2;
123
124  temp1 = step[10] * C3;
125  temp2 = step[13] * C13;
126  output[10] = temp1 + temp2;
127
128  temp1 = step[11] * C15;
129  temp2 = step[12] * C1;
130  output[11] = temp1 - temp2;
131
132  temp1 = step[11] * C1;
133  temp2 = step[12] * C15;
134  output[12] = temp2 + temp1;
135
136  temp1 = step[10] * C13;
137  temp2 = step[13] * C3;
138  output[13] = temp2 - temp1;
139
140  temp1 = step[ 9] * C5;
141  temp2 = step[14] * C11;
142  output[14] = temp2 + temp1;
143
144  temp1 = step[ 8] * C9;
145  temp2 = step[15] * C7;
146  output[15] = temp2 - temp1;
147
148  // step 3
149  step[ 0] = output[0] + output[3];
150  step[ 1] = output[1] + output[2];
151  step[ 2] = output[1] - output[2];
152  step[ 3] = output[0] - output[3];
153
154  temp1 = output[4] * C14;
155  temp2 = output[7] * C2;
156  step[ 4] = temp1 + temp2;
157
158  temp1 = output[5] * C10;
159  temp2 = output[6] * C6;
160  step[ 5] = temp1 + temp2;
161
162  temp1 = output[5] * C6;
163  temp2 = output[6] * C10;
164  step[ 6] = temp2 - temp1;
165
166  temp1 = output[4] * C2;
167  temp2 = output[7] * C14;
168  step[ 7] = temp2 - temp1;
169
170  step[ 8] = output[ 8] + output[11];
171  step[ 9] = output[ 9] + output[10];
172  step[10] = output[ 9] - output[10];
173  step[11] = output[ 8] - output[11];
174
175  step[12] = output[12] + output[15];
176  step[13] = output[13] + output[14];
177  step[14] = output[13] - output[14];
178  step[15] = output[12] - output[15];
179
180  // step 4
181  output[ 0] = (step[ 0] + step[ 1]);
182  output[ 8] = (step[ 0] - step[ 1]);
183
184  temp1 = step[2] * C12;
185  temp2 = step[3] * C4;
186  temp1 = temp1 + temp2;
187  output[ 4] = 2*(temp1 * C8);
188
189  temp1 = step[2] * C4;
190  temp2 = step[3] * C12;
191  temp1 = temp2 - temp1;
192  output[12] = 2 * (temp1 * C8);
193
194  output[ 2] = 2 * ((step[4] + step[ 5]) * C8);
195  output[14] = 2 * ((step[7] - step[ 6]) * C8);
196
197  temp1 = step[4] - step[5];
198  temp2 = step[6] + step[7];
199  output[ 6] = (temp1 + temp2);
200  output[10] = (temp1 - temp2);
201
202  intermediate[8] = step[8] + step[14];
203  intermediate[9] = step[9] + step[15];
204
205  temp1 = intermediate[8] * C12;
206  temp2 = intermediate[9] * C4;
207  temp1 = temp1 - temp2;
208  output[3] = 2 * (temp1 * C8);
209
210  temp1 = intermediate[8] * C4;
211  temp2 = intermediate[9] * C12;
212  temp1 = temp2 + temp1;
213  output[13] = 2 * (temp1 * C8);
214
215  output[ 9] = 2 * ((step[10] + step[11]) * C8);
216
217  intermediate[11] = step[10] - step[11];
218  intermediate[12] = step[12] + step[13];
219  intermediate[13] = step[12] - step[13];
220  intermediate[14] = step[ 8] - step[14];
221  intermediate[15] = step[ 9] - step[15];
222
223  output[15] = (intermediate[11] + intermediate[12]);
224  output[ 1] = -(intermediate[11] - intermediate[12]);
225
226  output[ 7] = 2 * (intermediate[13] * C8);
227
228  temp1 = intermediate[14] * C12;
229  temp2 = intermediate[15] * C4;
230  temp1 = temp1 - temp2;
231  output[11] = -2 * (temp1 * C8);
232
233  temp1 = intermediate[14] * C4;
234  temp2 = intermediate[15] * C12;
235  temp1 = temp2 + temp1;
236  output[ 5] = 2 * (temp1 * C8);
237}
238
239void reference_16x16_dct_2d(int16_t input[256], double output[256]) {
240  // First transform columns
241  for (int i = 0; i < 16; ++i) {
242    double temp_in[16], temp_out[16];
243    for (int j = 0; j < 16; ++j)
244      temp_in[j] = input[j * 16 + i];
245    butterfly_16x16_dct_1d(temp_in, temp_out);
246    for (int j = 0; j < 16; ++j)
247      output[j * 16 + i] = temp_out[j];
248  }
249  // Then transform rows
250  for (int i = 0; i < 16; ++i) {
251    double temp_in[16], temp_out[16];
252    for (int j = 0; j < 16; ++j)
253      temp_in[j] = output[j + i * 16];
254    butterfly_16x16_dct_1d(temp_in, temp_out);
255    // Scale by some magic number
256    for (int j = 0; j < 16; ++j)
257      output[j + i * 16] = temp_out[j]/2;
258  }
259}
260
261typedef void (*fdct_t)(const int16_t *in, int16_t *out, int stride);
262typedef void (*idct_t)(const int16_t *in, uint8_t *out, int stride);
263typedef void (*fht_t) (const int16_t *in, int16_t *out, int stride,
264                       int tx_type);
265typedef void (*iht_t) (const int16_t *in, uint8_t *out, int stride,
266                       int tx_type);
267
268typedef std::tr1::tuple<fdct_t, idct_t, int> dct_16x16_param_t;
269typedef std::tr1::tuple<fht_t, iht_t, int> ht_16x16_param_t;
270
271void fdct16x16_ref(const int16_t *in, int16_t *out, int stride, int tx_type) {
272  vp9_fdct16x16_c(in, out, stride);
273}
274
275void fht16x16_ref(const int16_t *in, int16_t *out, int stride, int tx_type) {
276  vp9_fht16x16_c(in, out, stride, tx_type);
277}
278
279class Trans16x16TestBase {
280 public:
281  virtual ~Trans16x16TestBase() {}
282
283 protected:
284  virtual void RunFwdTxfm(int16_t *in, int16_t *out, int stride) = 0;
285
286  virtual void RunInvTxfm(int16_t *out, uint8_t *dst, int stride) = 0;
287
288  void RunAccuracyCheck() {
289    ACMRandom rnd(ACMRandom::DeterministicSeed());
290    uint32_t max_error = 0;
291    int64_t total_error = 0;
292    const int count_test_block = 10000;
293    for (int i = 0; i < count_test_block; ++i) {
294      DECLARE_ALIGNED_ARRAY(16, int16_t, test_input_block, kNumCoeffs);
295      DECLARE_ALIGNED_ARRAY(16, int16_t, test_temp_block, kNumCoeffs);
296      DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
297      DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
298
299      // Initialize a test block with input range [-255, 255].
300      for (int j = 0; j < kNumCoeffs; ++j) {
301        src[j] = rnd.Rand8();
302        dst[j] = rnd.Rand8();
303        test_input_block[j] = src[j] - dst[j];
304      }
305
306      REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block,
307                                      test_temp_block, pitch_));
308      REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block, dst, pitch_));
309
310      for (int j = 0; j < kNumCoeffs; ++j) {
311        const uint32_t diff = dst[j] - src[j];
312        const uint32_t error = diff * diff;
313        if (max_error < error)
314          max_error = error;
315        total_error += error;
316      }
317    }
318
319    EXPECT_GE(1u, max_error)
320        << "Error: 16x16 FHT/IHT has an individual round trip error > 1";
321
322    EXPECT_GE(count_test_block , total_error)
323        << "Error: 16x16 FHT/IHT has average round trip error > 1 per block";
324  }
325
326  void RunCoeffCheck() {
327    ACMRandom rnd(ACMRandom::DeterministicSeed());
328    const int count_test_block = 1000;
329    DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
330    DECLARE_ALIGNED_ARRAY(16, int16_t, output_ref_block, kNumCoeffs);
331    DECLARE_ALIGNED_ARRAY(16, int16_t, output_block, kNumCoeffs);
332
333    for (int i = 0; i < count_test_block; ++i) {
334      // Initialize a test block with input range [-255, 255].
335      for (int j = 0; j < kNumCoeffs; ++j)
336        input_block[j] = rnd.Rand8() - rnd.Rand8();
337
338      fwd_txfm_ref(input_block, output_ref_block, pitch_, tx_type_);
339      REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
340
341      // The minimum quant value is 4.
342      for (int j = 0; j < kNumCoeffs; ++j)
343        EXPECT_EQ(output_block[j], output_ref_block[j]);
344    }
345  }
346
347  void RunMemCheck() {
348    ACMRandom rnd(ACMRandom::DeterministicSeed());
349    const int count_test_block = 1000;
350    DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
351    DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kNumCoeffs);
352    DECLARE_ALIGNED_ARRAY(16, int16_t, output_ref_block, kNumCoeffs);
353    DECLARE_ALIGNED_ARRAY(16, int16_t, output_block, kNumCoeffs);
354
355    for (int i = 0; i < count_test_block; ++i) {
356      // Initialize a test block with input range [-255, 255].
357      for (int j = 0; j < kNumCoeffs; ++j) {
358        input_block[j] = rnd.Rand8() - rnd.Rand8();
359        input_extreme_block[j] = rnd.Rand8() % 2 ? 255 : -255;
360      }
361      if (i == 0)
362        for (int j = 0; j < kNumCoeffs; ++j)
363          input_extreme_block[j] = 255;
364      if (i == 1)
365        for (int j = 0; j < kNumCoeffs; ++j)
366          input_extreme_block[j] = -255;
367
368      fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
369      REGISTER_STATE_CHECK(RunFwdTxfm(input_extreme_block,
370                                      output_block, pitch_));
371
372      // The minimum quant value is 4.
373      for (int j = 0; j < kNumCoeffs; ++j) {
374        EXPECT_EQ(output_block[j], output_ref_block[j]);
375        EXPECT_GE(4 * DCT_MAX_VALUE, abs(output_block[j]))
376            << "Error: 16x16 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
377      }
378    }
379  }
380
381  void RunInvAccuracyCheck() {
382    ACMRandom rnd(ACMRandom::DeterministicSeed());
383    const int count_test_block = 1000;
384    DECLARE_ALIGNED_ARRAY(16, int16_t, in, kNumCoeffs);
385    DECLARE_ALIGNED_ARRAY(16, int16_t, coeff, kNumCoeffs);
386    DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
387    DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
388
389    for (int i = 0; i < count_test_block; ++i) {
390      double out_r[kNumCoeffs];
391
392      // Initialize a test block with input range [-255, 255].
393      for (int j = 0; j < kNumCoeffs; ++j) {
394        src[j] = rnd.Rand8();
395        dst[j] = rnd.Rand8();
396        in[j] = src[j] - dst[j];
397      }
398
399      reference_16x16_dct_2d(in, out_r);
400      for (int j = 0; j < kNumCoeffs; ++j)
401        coeff[j] = round(out_r[j]);
402
403      REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, 16));
404
405      for (int j = 0; j < kNumCoeffs; ++j) {
406        const uint32_t diff = dst[j] - src[j];
407        const uint32_t error = diff * diff;
408        EXPECT_GE(1u, error)
409            << "Error: 16x16 IDCT has error " << error
410            << " at index " << j;
411      }
412    }
413  }
414  int pitch_;
415  int tx_type_;
416  fht_t fwd_txfm_ref;
417};
418
419class Trans16x16DCT
420    : public Trans16x16TestBase,
421      public ::testing::TestWithParam<dct_16x16_param_t> {
422 public:
423  virtual ~Trans16x16DCT() {}
424
425  virtual void SetUp() {
426    fwd_txfm_ = GET_PARAM(0);
427    inv_txfm_ = GET_PARAM(1);
428    tx_type_  = GET_PARAM(2);
429    pitch_    = 16;
430    fwd_txfm_ref = fdct16x16_ref;
431  }
432  virtual void TearDown() { libvpx_test::ClearSystemState(); }
433
434 protected:
435  void RunFwdTxfm(int16_t *in, int16_t *out, int stride) {
436    fwd_txfm_(in, out, stride);
437  }
438  void RunInvTxfm(int16_t *out, uint8_t *dst, int stride) {
439    inv_txfm_(out, dst, stride);
440  }
441
442  fdct_t fwd_txfm_;
443  idct_t inv_txfm_;
444};
445
446TEST_P(Trans16x16DCT, AccuracyCheck) {
447  RunAccuracyCheck();
448}
449
450TEST_P(Trans16x16DCT, CoeffCheck) {
451  RunCoeffCheck();
452}
453
454TEST_P(Trans16x16DCT, MemCheck) {
455  RunMemCheck();
456}
457
458TEST_P(Trans16x16DCT, InvAccuracyCheck) {
459  RunInvAccuracyCheck();
460}
461
462class Trans16x16HT
463    : public Trans16x16TestBase,
464      public ::testing::TestWithParam<ht_16x16_param_t> {
465 public:
466  virtual ~Trans16x16HT() {}
467
468  virtual void SetUp() {
469    fwd_txfm_ = GET_PARAM(0);
470    inv_txfm_ = GET_PARAM(1);
471    tx_type_  = GET_PARAM(2);
472    pitch_    = 16;
473    fwd_txfm_ref = fht16x16_ref;
474  }
475  virtual void TearDown() { libvpx_test::ClearSystemState(); }
476
477 protected:
478  void RunFwdTxfm(int16_t *in, int16_t *out, int stride) {
479    fwd_txfm_(in, out, stride, tx_type_);
480  }
481  void RunInvTxfm(int16_t *out, uint8_t *dst, int stride) {
482    inv_txfm_(out, dst, stride, tx_type_);
483  }
484
485  fht_t fwd_txfm_;
486  iht_t inv_txfm_;
487};
488
489TEST_P(Trans16x16HT, AccuracyCheck) {
490  RunAccuracyCheck();
491}
492
493TEST_P(Trans16x16HT, CoeffCheck) {
494  RunCoeffCheck();
495}
496
497TEST_P(Trans16x16HT, MemCheck) {
498  RunMemCheck();
499}
500
501using std::tr1::make_tuple;
502
503INSTANTIATE_TEST_CASE_P(
504    C, Trans16x16DCT,
505    ::testing::Values(
506        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_c, 0)));
507INSTANTIATE_TEST_CASE_P(
508    C, Trans16x16HT,
509    ::testing::Values(
510        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 0),
511        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1),
512        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2),
513        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3)));
514
515#if HAVE_NEON
516INSTANTIATE_TEST_CASE_P(
517    NEON, Trans16x16DCT,
518    ::testing::Values(
519        make_tuple(&vp9_fdct16x16_c,
520                   &vp9_idct16x16_256_add_neon, 0)));
521#endif
522
523#if HAVE_SSE2
524INSTANTIATE_TEST_CASE_P(
525    SSE2, Trans16x16DCT,
526    ::testing::Values(
527        make_tuple(&vp9_fdct16x16_sse2,
528                   &vp9_idct16x16_256_add_sse2, 0)));
529INSTANTIATE_TEST_CASE_P(
530    SSE2, Trans16x16HT,
531    ::testing::Values(
532        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 0),
533        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 1),
534        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 2),
535        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 3)));
536#endif
537}  // namespace
538