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_codec.h"
24#include "vpx/vpx_integer.h"
25
26using libvpx_test::ACMRandom;
27
28namespace {
29
30#ifdef _MSC_VER
31static int round(double x) {
32  if (x < 0)
33    return static_cast<int>(ceil(x - 0.5));
34  else
35    return static_cast<int>(floor(x + 0.5));
36}
37#endif
38
39const int kNumCoeffs = 256;
40const double PI = 3.1415926535898;
41void reference2_16x16_idct_2d(double *input, double *output) {
42  double x;
43  for (int l = 0; l < 16; ++l) {
44    for (int k = 0; k < 16; ++k) {
45      double s = 0;
46      for (int i = 0; i < 16; ++i) {
47        for (int j = 0; j < 16; ++j) {
48          x = cos(PI * j * (l + 0.5) / 16.0) *
49              cos(PI * i * (k + 0.5) / 16.0) *
50              input[i * 16 + j] / 256;
51          if (i != 0)
52            x *= sqrt(2.0);
53          if (j != 0)
54            x *= sqrt(2.0);
55          s += x;
56        }
57      }
58      output[k*16+l] = s;
59    }
60  }
61}
62
63
64const double C1 = 0.995184726672197;
65const double C2 = 0.98078528040323;
66const double C3 = 0.956940335732209;
67const double C4 = 0.923879532511287;
68const double C5 = 0.881921264348355;
69const double C6 = 0.831469612302545;
70const double C7 = 0.773010453362737;
71const double C8 = 0.707106781186548;
72const double C9 = 0.634393284163646;
73const double C10 = 0.555570233019602;
74const double C11 = 0.471396736825998;
75const double C12 = 0.38268343236509;
76const double C13 = 0.290284677254462;
77const double C14 = 0.195090322016128;
78const double C15 = 0.098017140329561;
79
80void butterfly_16x16_dct_1d(double input[16], double output[16]) {
81  double step[16];
82  double intermediate[16];
83  double temp1, temp2;
84
85  // step 1
86  step[ 0] = input[0] + input[15];
87  step[ 1] = input[1] + input[14];
88  step[ 2] = input[2] + input[13];
89  step[ 3] = input[3] + input[12];
90  step[ 4] = input[4] + input[11];
91  step[ 5] = input[5] + input[10];
92  step[ 6] = input[6] + input[ 9];
93  step[ 7] = input[7] + input[ 8];
94  step[ 8] = input[7] - input[ 8];
95  step[ 9] = input[6] - input[ 9];
96  step[10] = input[5] - input[10];
97  step[11] = input[4] - input[11];
98  step[12] = input[3] - input[12];
99  step[13] = input[2] - input[13];
100  step[14] = input[1] - input[14];
101  step[15] = input[0] - input[15];
102
103  // step 2
104  output[0] = step[0] + step[7];
105  output[1] = step[1] + step[6];
106  output[2] = step[2] + step[5];
107  output[3] = step[3] + step[4];
108  output[4] = step[3] - step[4];
109  output[5] = step[2] - step[5];
110  output[6] = step[1] - step[6];
111  output[7] = step[0] - step[7];
112
113  temp1 = step[ 8] * C7;
114  temp2 = step[15] * C9;
115  output[ 8] = temp1 + temp2;
116
117  temp1 = step[ 9] * C11;
118  temp2 = step[14] * C5;
119  output[ 9] = temp1 - temp2;
120
121  temp1 = step[10] * C3;
122  temp2 = step[13] * C13;
123  output[10] = temp1 + temp2;
124
125  temp1 = step[11] * C15;
126  temp2 = step[12] * C1;
127  output[11] = temp1 - temp2;
128
129  temp1 = step[11] * C1;
130  temp2 = step[12] * C15;
131  output[12] = temp2 + temp1;
132
133  temp1 = step[10] * C13;
134  temp2 = step[13] * C3;
135  output[13] = temp2 - temp1;
136
137  temp1 = step[ 9] * C5;
138  temp2 = step[14] * C11;
139  output[14] = temp2 + temp1;
140
141  temp1 = step[ 8] * C9;
142  temp2 = step[15] * C7;
143  output[15] = temp2 - temp1;
144
145  // step 3
146  step[ 0] = output[0] + output[3];
147  step[ 1] = output[1] + output[2];
148  step[ 2] = output[1] - output[2];
149  step[ 3] = output[0] - output[3];
150
151  temp1 = output[4] * C14;
152  temp2 = output[7] * C2;
153  step[ 4] = temp1 + temp2;
154
155  temp1 = output[5] * C10;
156  temp2 = output[6] * C6;
157  step[ 5] = temp1 + temp2;
158
159  temp1 = output[5] * C6;
160  temp2 = output[6] * C10;
161  step[ 6] = temp2 - temp1;
162
163  temp1 = output[4] * C2;
164  temp2 = output[7] * C14;
165  step[ 7] = temp2 - temp1;
166
167  step[ 8] = output[ 8] + output[11];
168  step[ 9] = output[ 9] + output[10];
169  step[10] = output[ 9] - output[10];
170  step[11] = output[ 8] - output[11];
171
172  step[12] = output[12] + output[15];
173  step[13] = output[13] + output[14];
174  step[14] = output[13] - output[14];
175  step[15] = output[12] - output[15];
176
177  // step 4
178  output[ 0] = (step[ 0] + step[ 1]);
179  output[ 8] = (step[ 0] - step[ 1]);
180
181  temp1 = step[2] * C12;
182  temp2 = step[3] * C4;
183  temp1 = temp1 + temp2;
184  output[ 4] = 2*(temp1 * C8);
185
186  temp1 = step[2] * C4;
187  temp2 = step[3] * C12;
188  temp1 = temp2 - temp1;
189  output[12] = 2 * (temp1 * C8);
190
191  output[ 2] = 2 * ((step[4] + step[ 5]) * C8);
192  output[14] = 2 * ((step[7] - step[ 6]) * C8);
193
194  temp1 = step[4] - step[5];
195  temp2 = step[6] + step[7];
196  output[ 6] = (temp1 + temp2);
197  output[10] = (temp1 - temp2);
198
199  intermediate[8] = step[8] + step[14];
200  intermediate[9] = step[9] + step[15];
201
202  temp1 = intermediate[8] * C12;
203  temp2 = intermediate[9] * C4;
204  temp1 = temp1 - temp2;
205  output[3] = 2 * (temp1 * C8);
206
207  temp1 = intermediate[8] * C4;
208  temp2 = intermediate[9] * C12;
209  temp1 = temp2 + temp1;
210  output[13] = 2 * (temp1 * C8);
211
212  output[ 9] = 2 * ((step[10] + step[11]) * C8);
213
214  intermediate[11] = step[10] - step[11];
215  intermediate[12] = step[12] + step[13];
216  intermediate[13] = step[12] - step[13];
217  intermediate[14] = step[ 8] - step[14];
218  intermediate[15] = step[ 9] - step[15];
219
220  output[15] = (intermediate[11] + intermediate[12]);
221  output[ 1] = -(intermediate[11] - intermediate[12]);
222
223  output[ 7] = 2 * (intermediate[13] * C8);
224
225  temp1 = intermediate[14] * C12;
226  temp2 = intermediate[15] * C4;
227  temp1 = temp1 - temp2;
228  output[11] = -2 * (temp1 * C8);
229
230  temp1 = intermediate[14] * C4;
231  temp2 = intermediate[15] * C12;
232  temp1 = temp2 + temp1;
233  output[ 5] = 2 * (temp1 * C8);
234}
235
236void reference_16x16_dct_2d(int16_t input[256], double output[256]) {
237  // First transform columns
238  for (int i = 0; i < 16; ++i) {
239    double temp_in[16], temp_out[16];
240    for (int j = 0; j < 16; ++j)
241      temp_in[j] = input[j * 16 + i];
242    butterfly_16x16_dct_1d(temp_in, temp_out);
243    for (int j = 0; j < 16; ++j)
244      output[j * 16 + i] = temp_out[j];
245  }
246  // Then transform rows
247  for (int i = 0; i < 16; ++i) {
248    double temp_in[16], temp_out[16];
249    for (int j = 0; j < 16; ++j)
250      temp_in[j] = output[j + i * 16];
251    butterfly_16x16_dct_1d(temp_in, temp_out);
252    // Scale by some magic number
253    for (int j = 0; j < 16; ++j)
254      output[j + i * 16] = temp_out[j]/2;
255  }
256}
257
258typedef void (*FdctFunc)(const int16_t *in, tran_low_t *out, int stride);
259typedef void (*IdctFunc)(const tran_low_t *in, uint8_t *out, int stride);
260typedef void (*FhtFunc)(const int16_t *in, tran_low_t *out, int stride,
261                        int tx_type);
262typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
263                        int tx_type);
264
265typedef std::tr1::tuple<FdctFunc, IdctFunc, int, vpx_bit_depth_t> Dct16x16Param;
266typedef std::tr1::tuple<FhtFunc, IhtFunc, int, vpx_bit_depth_t> Ht16x16Param;
267
268void fdct16x16_ref(const int16_t *in, tran_low_t *out, int stride,
269                   int /*tx_type*/) {
270  vp9_fdct16x16_c(in, out, stride);
271}
272
273void idct16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
274                   int /*tx_type*/) {
275  vp9_idct16x16_256_add_c(in, dest, stride);
276}
277
278void fht16x16_ref(const int16_t *in, tran_low_t *out, int stride,
279                  int tx_type) {
280  vp9_fht16x16_c(in, out, stride, tx_type);
281}
282
283void iht16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
284                  int tx_type) {
285  vp9_iht16x16_256_add_c(in, dest, stride, tx_type);
286}
287
288#if CONFIG_VP9_HIGHBITDEPTH
289void idct16x16_10(const tran_low_t *in, uint8_t *out, int stride) {
290  vp9_high_idct16x16_256_add_c(in, out, stride, 10);
291}
292
293void idct16x16_12(const tran_low_t *in, uint8_t *out, int stride) {
294  vp9_high_idct16x16_256_add_c(in, out, stride, 12);
295}
296
297void idct16x16_10_ref(const tran_low_t *in, uint8_t *out, int stride,
298                      int tx_type) {
299  idct16x16_10(in, out, stride);
300}
301
302void idct16x16_12_ref(const tran_low_t *in, uint8_t *out, int stride,
303                      int tx_type) {
304  idct16x16_12(in, out, stride);
305}
306
307void iht16x16_10(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
308  vp9_high_iht16x16_256_add_c(in, out, stride, tx_type, 10);
309}
310
311void iht16x16_12(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
312  vp9_high_iht16x16_256_add_c(in, out, stride, tx_type, 12);
313}
314#endif
315
316class Trans16x16TestBase {
317 public:
318  virtual ~Trans16x16TestBase() {}
319
320 protected:
321  virtual void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) = 0;
322
323  virtual void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) = 0;
324
325  void RunAccuracyCheck() {
326    ACMRandom rnd(ACMRandom::DeterministicSeed());
327    uint32_t max_error = 0;
328    int64_t total_error = 0;
329    const int count_test_block = 10000;
330    for (int i = 0; i < count_test_block; ++i) {
331      DECLARE_ALIGNED_ARRAY(16, int16_t, test_input_block, kNumCoeffs);
332      DECLARE_ALIGNED_ARRAY(16, tran_low_t, test_temp_block, kNumCoeffs);
333      DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
334      DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
335#if CONFIG_VP9_HIGHBITDEPTH
336      DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
337      DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
338#endif
339
340      // Initialize a test block with input range [-mask_, mask_].
341      for (int j = 0; j < kNumCoeffs; ++j) {
342        if (bit_depth_ == VPX_BITS_8) {
343          src[j] = rnd.Rand8();
344          dst[j] = rnd.Rand8();
345          test_input_block[j] = src[j] - dst[j];
346#if CONFIG_VP9_HIGHBITDEPTH
347        } else {
348          src16[j] = rnd.Rand16() & mask_;
349          dst16[j] = rnd.Rand16() & mask_;
350          test_input_block[j] = src16[j] - dst16[j];
351#endif
352        }
353      }
354
355      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block,
356                                          test_temp_block, pitch_));
357      if (bit_depth_ == VPX_BITS_8) {
358        ASM_REGISTER_STATE_CHECK(
359            RunInvTxfm(test_temp_block, dst, pitch_));
360#if CONFIG_VP9_HIGHBITDEPTH
361      } else {
362        ASM_REGISTER_STATE_CHECK(
363            RunInvTxfm(test_temp_block, CONVERT_TO_BYTEPTR(dst16), pitch_));
364#endif
365      }
366
367      for (int j = 0; j < kNumCoeffs; ++j) {
368#if CONFIG_VP9_HIGHBITDEPTH
369        const uint32_t diff =
370            bit_depth_ == VPX_BITS_8 ?  dst[j] - src[j] : dst16[j] - src16[j];
371#else
372        const uint32_t diff = dst[j] - src[j];
373#endif
374        const uint32_t error = diff * diff;
375        if (max_error < error)
376          max_error = error;
377        total_error += error;
378      }
379    }
380
381    EXPECT_GE(1u  << 2 * (bit_depth_ - 8), max_error)
382        << "Error: 16x16 FHT/IHT has an individual round trip error > 1";
383
384    EXPECT_GE(count_test_block << 2 * (bit_depth_ - 8), total_error)
385        << "Error: 16x16 FHT/IHT has average round trip error > 1 per block";
386  }
387
388  void RunCoeffCheck() {
389    ACMRandom rnd(ACMRandom::DeterministicSeed());
390    const int count_test_block = 1000;
391    DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
392    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
393    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_block, kNumCoeffs);
394
395    for (int i = 0; i < count_test_block; ++i) {
396      // Initialize a test block with input range [-mask_, mask_].
397      for (int j = 0; j < kNumCoeffs; ++j)
398        input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
399
400      fwd_txfm_ref(input_block, output_ref_block, pitch_, tx_type_);
401      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
402
403      // The minimum quant value is 4.
404      for (int j = 0; j < kNumCoeffs; ++j)
405        EXPECT_EQ(output_block[j], output_ref_block[j]);
406    }
407  }
408
409  void RunMemCheck() {
410    ACMRandom rnd(ACMRandom::DeterministicSeed());
411    const int count_test_block = 1000;
412    DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
413    DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kNumCoeffs);
414    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
415    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_block, kNumCoeffs);
416
417    for (int i = 0; i < count_test_block; ++i) {
418      // Initialize a test block with input range [-mask_, mask_].
419      for (int j = 0; j < kNumCoeffs; ++j) {
420        input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
421        input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
422      }
423      if (i == 0) {
424        for (int j = 0; j < kNumCoeffs; ++j)
425          input_extreme_block[j] = mask_;
426      } else if (i == 1) {
427        for (int j = 0; j < kNumCoeffs; ++j)
428          input_extreme_block[j] = -mask_;
429      }
430
431      fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
432      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_extreme_block,
433                                          output_block, pitch_));
434
435      // The minimum quant value is 4.
436      for (int j = 0; j < kNumCoeffs; ++j) {
437        EXPECT_EQ(output_block[j], output_ref_block[j]);
438        EXPECT_GE(4 * DCT_MAX_VALUE << (bit_depth_ - 8), abs(output_block[j]))
439            << "Error: 16x16 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
440      }
441    }
442  }
443
444  void RunQuantCheck(int dc_thred, int ac_thred) {
445    ACMRandom rnd(ACMRandom::DeterministicSeed());
446    const int count_test_block = 1000;
447    DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
448    DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kNumCoeffs);
449    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
450
451    DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
452    DECLARE_ALIGNED_ARRAY(16, uint8_t, ref, kNumCoeffs);
453#if CONFIG_VP9_HIGHBITDEPTH
454    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
455    DECLARE_ALIGNED_ARRAY(16, uint16_t, ref16, kNumCoeffs);
456#endif
457
458    for (int i = 0; i < count_test_block; ++i) {
459      // Initialize a test block with input range [-mask_, mask_].
460      for (int j = 0; j < kNumCoeffs; ++j) {
461        if (bit_depth_ == VPX_BITS_8)
462          input_block[j] = rnd.Rand8() - rnd.Rand8();
463        else
464          input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
465        input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
466      }
467      if (i == 0)
468        for (int j = 0; j < kNumCoeffs; ++j)
469          input_extreme_block[j] = mask_;
470      if (i == 1)
471        for (int j = 0; j < kNumCoeffs; ++j)
472          input_extreme_block[j] = -mask_;
473
474      fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
475
476      // clear reconstructed pixel buffers
477      vpx_memset(dst, 0, kNumCoeffs * sizeof(uint8_t));
478      vpx_memset(ref, 0, kNumCoeffs * sizeof(uint8_t));
479#if CONFIG_VP9_HIGHBITDEPTH
480      vpx_memset(dst16, 0, kNumCoeffs * sizeof(uint16_t));
481      vpx_memset(ref16, 0, kNumCoeffs * sizeof(uint16_t));
482#endif
483
484      // quantization with maximum allowed step sizes
485      output_ref_block[0] = (output_ref_block[0] / dc_thred) * dc_thred;
486      for (int j = 1; j < kNumCoeffs; ++j)
487        output_ref_block[j] = (output_ref_block[j] / ac_thred) * ac_thred;
488      if (bit_depth_ == VPX_BITS_8) {
489        inv_txfm_ref(output_ref_block, ref, pitch_, tx_type_);
490        ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block, dst, pitch_));
491#if CONFIG_VP9_HIGHBITDEPTH
492      } else {
493        inv_txfm_ref(output_ref_block, CONVERT_TO_BYTEPTR(ref16), pitch_,
494                     tx_type_);
495        ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block,
496                                            CONVERT_TO_BYTEPTR(dst16), pitch_));
497#endif
498      }
499      if (bit_depth_ == VPX_BITS_8) {
500        for (int j = 0; j < kNumCoeffs; ++j)
501          EXPECT_EQ(ref[j], dst[j]);
502#if CONFIG_VP9_HIGHBITDEPTH
503      } else {
504        for (int j = 0; j < kNumCoeffs; ++j)
505          EXPECT_EQ(ref16[j], dst16[j]);
506#endif
507      }
508    }
509  }
510
511  void RunInvAccuracyCheck() {
512    ACMRandom rnd(ACMRandom::DeterministicSeed());
513    const int count_test_block = 1000;
514    DECLARE_ALIGNED_ARRAY(16, int16_t, in, kNumCoeffs);
515    DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff, kNumCoeffs);
516    DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
517    DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
518#if CONFIG_VP9_HIGHBITDEPTH
519    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
520    DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
521#endif
522
523    for (int i = 0; i < count_test_block; ++i) {
524      double out_r[kNumCoeffs];
525
526      // Initialize a test block with input range [-255, 255].
527      for (int j = 0; j < kNumCoeffs; ++j) {
528        if (bit_depth_ == VPX_BITS_8) {
529          src[j] = rnd.Rand8();
530          dst[j] = rnd.Rand8();
531          in[j] = src[j] - dst[j];
532#if CONFIG_VP9_HIGHBITDEPTH
533        } else {
534          src16[j] = rnd.Rand16() & mask_;
535          dst16[j] = rnd.Rand16() & mask_;
536          in[j] = src16[j] - dst16[j];
537#endif
538        }
539      }
540
541      reference_16x16_dct_2d(in, out_r);
542      for (int j = 0; j < kNumCoeffs; ++j)
543        coeff[j] = round(out_r[j]);
544
545      if (bit_depth_ == VPX_BITS_8) {
546        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, 16));
547#if CONFIG_VP9_HIGHBITDEPTH
548      } else {
549        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16),
550                                            16));
551#endif
552      }
553
554      for (int j = 0; j < kNumCoeffs; ++j) {
555#if CONFIG_VP9_HIGHBITDEPTH
556        const uint32_t diff =
557            bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
558#else
559        const uint32_t diff = dst[j] - src[j];
560#endif
561        const uint32_t error = diff * diff;
562        EXPECT_GE(1u, error)
563            << "Error: 16x16 IDCT has error " << error
564            << " at index " << j;
565      }
566    }
567  }
568  int pitch_;
569  int tx_type_;
570  vpx_bit_depth_t bit_depth_;
571  int mask_;
572  FhtFunc fwd_txfm_ref;
573  IhtFunc inv_txfm_ref;
574};
575
576class Trans16x16DCT
577    : public Trans16x16TestBase,
578      public ::testing::TestWithParam<Dct16x16Param> {
579 public:
580  virtual ~Trans16x16DCT() {}
581
582  virtual void SetUp() {
583    fwd_txfm_ = GET_PARAM(0);
584    inv_txfm_ = GET_PARAM(1);
585    tx_type_  = GET_PARAM(2);
586    bit_depth_ = GET_PARAM(3);
587    pitch_    = 16;
588    fwd_txfm_ref = fdct16x16_ref;
589    inv_txfm_ref = idct16x16_ref;
590    mask_ = (1 << bit_depth_) - 1;
591#if CONFIG_VP9_HIGHBITDEPTH
592    switch (bit_depth_) {
593      case 10:
594        inv_txfm_ref = idct16x16_10_ref;
595        break;
596      case 12:
597        inv_txfm_ref = idct16x16_12_ref;
598        break;
599      default:
600        inv_txfm_ref = idct16x16_ref;
601        break;
602    }
603#else
604    inv_txfm_ref = idct16x16_ref;
605#endif
606  }
607  virtual void TearDown() { libvpx_test::ClearSystemState(); }
608
609 protected:
610  void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
611    fwd_txfm_(in, out, stride);
612  }
613  void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
614    inv_txfm_(out, dst, stride);
615  }
616
617  FdctFunc fwd_txfm_;
618  IdctFunc inv_txfm_;
619};
620
621TEST_P(Trans16x16DCT, AccuracyCheck) {
622  RunAccuracyCheck();
623}
624
625TEST_P(Trans16x16DCT, CoeffCheck) {
626  RunCoeffCheck();
627}
628
629TEST_P(Trans16x16DCT, MemCheck) {
630  RunMemCheck();
631}
632
633TEST_P(Trans16x16DCT, QuantCheck) {
634  // Use maximally allowed quantization step sizes for DC and AC
635  // coefficients respectively.
636  RunQuantCheck(1336, 1828);
637}
638
639TEST_P(Trans16x16DCT, InvAccuracyCheck) {
640  RunInvAccuracyCheck();
641}
642
643class Trans16x16HT
644    : public Trans16x16TestBase,
645      public ::testing::TestWithParam<Ht16x16Param> {
646 public:
647  virtual ~Trans16x16HT() {}
648
649  virtual void SetUp() {
650    fwd_txfm_ = GET_PARAM(0);
651    inv_txfm_ = GET_PARAM(1);
652    tx_type_  = GET_PARAM(2);
653    bit_depth_ = GET_PARAM(3);
654    pitch_    = 16;
655    fwd_txfm_ref = fht16x16_ref;
656    inv_txfm_ref = iht16x16_ref;
657    mask_ = (1 << bit_depth_) - 1;
658#if CONFIG_VP9_HIGHBITDEPTH
659    switch (bit_depth_) {
660      case VPX_BITS_10:
661        inv_txfm_ref = iht16x16_10;
662        break;
663      case VPX_BITS_12:
664        inv_txfm_ref = iht16x16_12;
665        break;
666      default:
667        inv_txfm_ref = iht16x16_ref;
668        break;
669    }
670#else
671    inv_txfm_ref = iht16x16_ref;
672#endif
673  }
674  virtual void TearDown() { libvpx_test::ClearSystemState(); }
675
676 protected:
677  void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
678    fwd_txfm_(in, out, stride, tx_type_);
679  }
680  void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
681    inv_txfm_(out, dst, stride, tx_type_);
682  }
683
684  FhtFunc fwd_txfm_;
685  IhtFunc inv_txfm_;
686};
687
688TEST_P(Trans16x16HT, AccuracyCheck) {
689  RunAccuracyCheck();
690}
691
692TEST_P(Trans16x16HT, CoeffCheck) {
693  RunCoeffCheck();
694}
695
696TEST_P(Trans16x16HT, MemCheck) {
697  RunMemCheck();
698}
699
700TEST_P(Trans16x16HT, QuantCheck) {
701  // The encoder skips any non-DC intra prediction modes,
702  // when the quantization step size goes beyond 988.
703  RunQuantCheck(549, 988);
704}
705
706using std::tr1::make_tuple;
707
708#if CONFIG_VP9_HIGHBITDEPTH
709INSTANTIATE_TEST_CASE_P(
710    C, Trans16x16DCT,
711    ::testing::Values(
712        make_tuple(&vp9_high_fdct16x16_c, &idct16x16_10, 0, VPX_BITS_10),
713        make_tuple(&vp9_high_fdct16x16_c, &idct16x16_12, 0, VPX_BITS_12),
714        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_c, 0, VPX_BITS_8)));
715#else
716INSTANTIATE_TEST_CASE_P(
717    C, Trans16x16DCT,
718    ::testing::Values(
719        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_c, 0, VPX_BITS_8)));
720#endif
721
722#if CONFIG_VP9_HIGHBITDEPTH
723INSTANTIATE_TEST_CASE_P(
724    C, Trans16x16HT,
725    ::testing::Values(
726        make_tuple(&vp9_high_fht16x16_c, &iht16x16_10, 0, VPX_BITS_10),
727        make_tuple(&vp9_high_fht16x16_c, &iht16x16_10, 1, VPX_BITS_10),
728        make_tuple(&vp9_high_fht16x16_c, &iht16x16_10, 2, VPX_BITS_10),
729        make_tuple(&vp9_high_fht16x16_c, &iht16x16_10, 3, VPX_BITS_10),
730        make_tuple(&vp9_high_fht16x16_c, &iht16x16_12, 0, VPX_BITS_12),
731        make_tuple(&vp9_high_fht16x16_c, &iht16x16_12, 1, VPX_BITS_12),
732        make_tuple(&vp9_high_fht16x16_c, &iht16x16_12, 2, VPX_BITS_12),
733        make_tuple(&vp9_high_fht16x16_c, &iht16x16_12, 3, VPX_BITS_12),
734        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
735        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
736        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
737        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
738#else
739INSTANTIATE_TEST_CASE_P(
740    C, Trans16x16HT,
741    ::testing::Values(
742        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
743        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
744        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
745        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
746#endif
747
748#if HAVE_NEON_ASM && !CONFIG_VP9_HIGHBITDEPTH
749INSTANTIATE_TEST_CASE_P(
750    NEON, Trans16x16DCT,
751    ::testing::Values(
752        make_tuple(&vp9_fdct16x16_c,
753                   &vp9_idct16x16_256_add_neon, 0, VPX_BITS_8)));
754#endif
755
756#if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH
757INSTANTIATE_TEST_CASE_P(
758    SSE2, Trans16x16DCT,
759    ::testing::Values(
760        make_tuple(&vp9_fdct16x16_sse2,
761                   &vp9_idct16x16_256_add_sse2, 0, VPX_BITS_8)));
762INSTANTIATE_TEST_CASE_P(
763    SSE2, Trans16x16HT,
764    ::testing::Values(
765        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 0,
766                   VPX_BITS_8),
767        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 1,
768                   VPX_BITS_8),
769        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 2,
770                   VPX_BITS_8),
771        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 3,
772                   VPX_BITS_8)));
773#endif
774
775#if HAVE_SSSE3 && !CONFIG_VP9_HIGHBITDEPTH
776INSTANTIATE_TEST_CASE_P(
777    SSSE3, Trans16x16DCT,
778    ::testing::Values(
779        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_ssse3, 0,
780                   VPX_BITS_8)));
781#endif
782}  // namespace
783