summaryrefslogtreecommitdiffstats
path: root/media/libaom/src/test/transform_test_base.h
blob: 8ebcf5ff7caf232fe43a54cbf75574e20d13118a (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
/*
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
 *
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 */

#ifndef AOM_TEST_TRANSFORM_TEST_BASE_H_
#define AOM_TEST_TRANSFORM_TEST_BASE_H_

#include "config/aom_config.h"

#include "aom_mem/aom_mem.h"
#include "aom/aom_codec.h"
#include "aom_dsp/txfm_common.h"

namespace libaom_test {

//  Note:
//   Same constant are defined in av1/common/av1_entropy.h and
//   av1/common/entropy.h.  Goal is to make this base class
//   to use for future codec transform testing.  But including
//   either of them would lead to compiling error when we do
//   unit test for another codec. Suggest to move the definition
//   to a aom header file.
const int kDctMaxValue = 16384;

typedef void (*FhtFunc)(const int16_t *in, tran_low_t *out, int stride,
                        TxfmParam *txfm_param);

typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
                        const TxfmParam *txfm_param);

class TransformTestBase {
 public:
  virtual ~TransformTestBase() {}

 protected:
  virtual void RunFwdTxfm(const int16_t *in, tran_low_t *out, int stride) = 0;

  virtual void RunInvTxfm(const tran_low_t *out, uint8_t *dst, int stride) = 0;

  void RunAccuracyCheck(uint32_t ref_max_error, double ref_avg_error) {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
    uint32_t max_error = 0;
    int64_t total_error = 0;
    const int count_test_block = 10000;

    int16_t *test_input_block = reinterpret_cast<int16_t *>(
        aom_memalign(16, sizeof(int16_t) * num_coeffs_));
    tran_low_t *test_temp_block = reinterpret_cast<tran_low_t *>(
        aom_memalign(16, sizeof(tran_low_t) * num_coeffs_));
    uint8_t *dst = reinterpret_cast<uint8_t *>(
        aom_memalign(16, sizeof(uint8_t) * num_coeffs_));
    uint8_t *src = reinterpret_cast<uint8_t *>(
        aom_memalign(16, sizeof(uint8_t) * num_coeffs_));
    uint16_t *dst16 = reinterpret_cast<uint16_t *>(
        aom_memalign(16, sizeof(uint16_t) * num_coeffs_));
    uint16_t *src16 = reinterpret_cast<uint16_t *>(
        aom_memalign(16, sizeof(uint16_t) * num_coeffs_));

    for (int i = 0; i < count_test_block; ++i) {
      // Initialize a test block with input range [-255, 255].
      for (int j = 0; j < num_coeffs_; ++j) {
        if (bit_depth_ == AOM_BITS_8) {
          src[j] = rnd.Rand8();
          dst[j] = rnd.Rand8();
          test_input_block[j] = src[j] - dst[j];
        } else {
          src16[j] = rnd.Rand16() & mask_;
          dst16[j] = rnd.Rand16() & mask_;
          test_input_block[j] = src16[j] - dst16[j];
        }
      }

      ASM_REGISTER_STATE_CHECK(
          RunFwdTxfm(test_input_block, test_temp_block, pitch_));
      if (bit_depth_ == AOM_BITS_8) {
        ASM_REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block, dst, pitch_));
      } else {
        ASM_REGISTER_STATE_CHECK(
            RunInvTxfm(test_temp_block, CONVERT_TO_BYTEPTR(dst16), pitch_));
      }

      for (int j = 0; j < num_coeffs_; ++j) {
        const int diff =
            bit_depth_ == AOM_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
        const uint32_t error = diff * diff;
        if (max_error < error) max_error = error;
        total_error += error;
      }
    }

    double avg_error = total_error * 1. / count_test_block / num_coeffs_;

    EXPECT_GE(ref_max_error, max_error)
        << "Error: FHT/IHT has an individual round trip error > "
        << ref_max_error;

    EXPECT_GE(ref_avg_error, avg_error)
        << "Error: FHT/IHT has average round trip error > " << ref_avg_error
        << " per block";

    aom_free(test_input_block);
    aom_free(test_temp_block);
    aom_free(dst);
    aom_free(src);
    aom_free(dst16);
    aom_free(src16);
  }

  void RunCoeffCheck() {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
    const int count_test_block = 5000;

    // Use a stride value which is not the width of any transform, to catch
    // cases where the transforms use the stride incorrectly.
    int stride = 96;

    int16_t *input_block = reinterpret_cast<int16_t *>(
        aom_memalign(16, sizeof(int16_t) * stride * height_));
    tran_low_t *output_ref_block = reinterpret_cast<tran_low_t *>(
        aom_memalign(16, sizeof(tran_low_t) * num_coeffs_));
    tran_low_t *output_block = reinterpret_cast<tran_low_t *>(
        aom_memalign(16, sizeof(tran_low_t) * num_coeffs_));

    for (int i = 0; i < count_test_block; ++i) {
      int j, k;
      for (j = 0; j < height_; ++j) {
        for (k = 0; k < pitch_; ++k) {
          int in_idx = j * stride + k;
          int out_idx = j * pitch_ + k;
          input_block[in_idx] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
          if (bit_depth_ == AOM_BITS_8) {
            output_block[out_idx] = output_ref_block[out_idx] = rnd.Rand8();
          } else {
            output_block[out_idx] = output_ref_block[out_idx] =
                rnd.Rand16() & mask_;
          }
        }
      }

      fwd_txfm_ref(input_block, output_ref_block, stride, &txfm_param_);
      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, stride));

      // The minimum quant value is 4.
      for (j = 0; j < height_; ++j) {
        for (k = 0; k < pitch_; ++k) {
          int out_idx = j * pitch_ + k;
          ASSERT_EQ(output_block[out_idx], output_ref_block[out_idx])
              << "Error: not bit-exact result at index: " << out_idx
              << " at test block: " << i;
        }
      }
    }
    aom_free(input_block);
    aom_free(output_ref_block);
    aom_free(output_block);
  }

  void RunInvCoeffCheck() {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
    const int count_test_block = 5000;

    // Use a stride value which is not the width of any transform, to catch
    // cases where the transforms use the stride incorrectly.
    int stride = 96;

    int16_t *input_block = reinterpret_cast<int16_t *>(
        aom_memalign(16, sizeof(int16_t) * num_coeffs_));
    tran_low_t *trans_block = reinterpret_cast<tran_low_t *>(
        aom_memalign(16, sizeof(tran_low_t) * num_coeffs_));
    uint8_t *output_block = reinterpret_cast<uint8_t *>(
        aom_memalign(16, sizeof(uint8_t) * stride * height_));
    uint8_t *output_ref_block = reinterpret_cast<uint8_t *>(
        aom_memalign(16, sizeof(uint8_t) * stride * height_));

    for (int i = 0; i < count_test_block; ++i) {
      // Initialize a test block with input range [-mask_, mask_].
      int j, k;
      for (j = 0; j < height_; ++j) {
        for (k = 0; k < pitch_; ++k) {
          int in_idx = j * pitch_ + k;
          int out_idx = j * stride + k;
          input_block[in_idx] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
          output_ref_block[out_idx] = rnd.Rand16() & mask_;
          output_block[out_idx] = output_ref_block[out_idx];
        }
      }

      fwd_txfm_ref(input_block, trans_block, pitch_, &txfm_param_);

      inv_txfm_ref(trans_block, output_ref_block, stride, &txfm_param_);
      ASM_REGISTER_STATE_CHECK(RunInvTxfm(trans_block, output_block, stride));

      for (j = 0; j < height_; ++j) {
        for (k = 0; k < pitch_; ++k) {
          int out_idx = j * stride + k;
          ASSERT_EQ(output_block[out_idx], output_ref_block[out_idx])
              << "Error: not bit-exact result at index: " << out_idx
              << " j = " << j << " k = " << k << " at test block: " << i;
        }
      }
    }
    aom_free(input_block);
    aom_free(trans_block);
    aom_free(output_ref_block);
    aom_free(output_block);
  }

  void RunMemCheck() {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
    const int count_test_block = 5000;

    int16_t *input_extreme_block = reinterpret_cast<int16_t *>(
        aom_memalign(16, sizeof(int16_t) * num_coeffs_));
    tran_low_t *output_ref_block = reinterpret_cast<tran_low_t *>(
        aom_memalign(16, sizeof(tran_low_t) * num_coeffs_));
    tran_low_t *output_block = reinterpret_cast<tran_low_t *>(
        aom_memalign(16, sizeof(tran_low_t) * num_coeffs_));

    for (int i = 0; i < count_test_block; ++i) {
      // Initialize a test block with input range [-mask_, mask_].
      for (int j = 0; j < num_coeffs_; ++j) {
        input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
      }
      if (i == 0) {
        for (int j = 0; j < num_coeffs_; ++j) input_extreme_block[j] = mask_;
      } else if (i == 1) {
        for (int j = 0; j < num_coeffs_; ++j) input_extreme_block[j] = -mask_;
      }

      fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, &txfm_param_);
      ASM_REGISTER_STATE_CHECK(
          RunFwdTxfm(input_extreme_block, output_block, pitch_));

      int row_length = FindRowLength();
      // The minimum quant value is 4.
      for (int j = 0; j < num_coeffs_; ++j) {
        ASSERT_EQ(output_block[j], output_ref_block[j])
            << "Not bit-exact at test index: " << i << ", "
            << "j = " << j << std::endl;
        EXPECT_GE(row_length * kDctMaxValue << (bit_depth_ - 8),
                  abs(output_block[j]))
            << "Error: NxN FDCT has coefficient larger than N*DCT_MAX_VALUE";
      }
    }
    aom_free(input_extreme_block);
    aom_free(output_ref_block);
    aom_free(output_block);
  }

  void RunInvAccuracyCheck(int limit) {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
    const int count_test_block = 1000;

    int16_t *in = reinterpret_cast<int16_t *>(
        aom_memalign(16, sizeof(int16_t) * num_coeffs_));
    tran_low_t *coeff = reinterpret_cast<tran_low_t *>(
        aom_memalign(16, sizeof(tran_low_t) * num_coeffs_));
    uint8_t *dst = reinterpret_cast<uint8_t *>(
        aom_memalign(16, sizeof(uint8_t) * num_coeffs_));
    uint8_t *src = reinterpret_cast<uint8_t *>(
        aom_memalign(16, sizeof(uint8_t) * num_coeffs_));

    uint16_t *dst16 = reinterpret_cast<uint16_t *>(
        aom_memalign(16, sizeof(uint16_t) * num_coeffs_));
    uint16_t *src16 = reinterpret_cast<uint16_t *>(
        aom_memalign(16, sizeof(uint16_t) * num_coeffs_));

    for (int i = 0; i < count_test_block; ++i) {
      // Initialize a test block with input range [-mask_, mask_].
      for (int j = 0; j < num_coeffs_; ++j) {
        if (bit_depth_ == AOM_BITS_8) {
          src[j] = rnd.Rand8();
          dst[j] = rnd.Rand8();
          in[j] = src[j] - dst[j];
        } else {
          src16[j] = rnd.Rand16() & mask_;
          dst16[j] = rnd.Rand16() & mask_;
          in[j] = src16[j] - dst16[j];
        }
      }

      fwd_txfm_ref(in, coeff, pitch_, &txfm_param_);

      if (bit_depth_ == AOM_BITS_8) {
        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
      } else {
        ASM_REGISTER_STATE_CHECK(
            RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16), pitch_));
      }

      for (int j = 0; j < num_coeffs_; ++j) {
        const int diff =
            bit_depth_ == AOM_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
        const uint32_t error = diff * diff;
        ASSERT_GE(static_cast<uint32_t>(limit), error)
            << "Error: 4x4 IDCT has error " << error << " at index " << j;
      }
    }
    aom_free(in);
    aom_free(coeff);
    aom_free(dst);
    aom_free(src);
    aom_free(src16);
    aom_free(dst16);
  }

  int pitch_;
  int height_;
  FhtFunc fwd_txfm_ref;
  IhtFunc inv_txfm_ref;
  aom_bit_depth_t bit_depth_;
  int mask_;
  int num_coeffs_;
  TxfmParam txfm_param_;

 private:
  //  Assume transform size is 4x4, 8x8, 16x16,...
  int FindRowLength() const {
    int row = 4;
    if (16 == num_coeffs_) {
      row = 4;
    } else if (64 == num_coeffs_) {
      row = 8;
    } else if (256 == num_coeffs_) {
      row = 16;
    } else if (1024 == num_coeffs_) {
      row = 32;
    }
    return row;
  }
};

}  // namespace libaom_test

#endif  // AOM_TEST_TRANSFORM_TEST_BASE_H_