summaryrefslogtreecommitdiffstats
path: root/third_party/aom/test/av1_convolve_scale_test.cc
blob: 9d8be888de9e53ea4ce8df2e13f9857552b4f757 (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
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
/*
 * Copyright (c) 2017, 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.
 */

#include <vector>

#include "third_party/googletest/src/googletest/include/gtest/gtest.h"

#include "./av1_rtcd.h"
#include "aom_ports/aom_timer.h"
#include "test/acm_random.h"
#include "test/clear_system_state.h"
#include "test/register_state_check.h"
#include "test/util.h"

namespace {
const int kTestIters = 10;
const int kPerfIters = 1000;

const int kVPad = 32;
const int kHPad = 32;
const int kXStepQn = 16;
const int kYStepQn = 20;

using std::tr1::tuple;
using std::tr1::make_tuple;
using libaom_test::ACMRandom;

enum NTaps { EIGHT_TAP, TEN_TAP, TWELVE_TAP };
int NTapsToInt(NTaps ntaps) { return 8 + static_cast<int>(ntaps) * 2; }

// A 16-bit filter with a configurable number of taps.
class TestFilter {
 public:
  void set(NTaps ntaps, bool backwards);

  InterpFilterParams params_;

 private:
  std::vector<int16_t> coeffs_;
};

void TestFilter::set(NTaps ntaps, bool backwards) {
  const int n = NTapsToInt(ntaps);
  assert(n >= 8 && n <= 12);

  // The filter has n * SUBPEL_SHIFTS proper elements and an extra 8 bogus
  // elements at the end so that convolutions can read off the end safely.
  coeffs_.resize(n * SUBPEL_SHIFTS + 8);

  // The coefficients are pretty much arbitrary, but convolutions shouldn't
  // over or underflow. For the first filter (subpels = 0), we use an
  // increasing or decreasing ramp (depending on the backwards parameter). We
  // don't want any zero coefficients, so we make it have an x-intercept at -1
  // or n. To ensure absence of under/overflow, we normalise the area under the
  // ramp to be I = 1 << FILTER_BITS (so that convolving a constant function
  // gives the identity).
  //
  // When increasing, the function has the form:
  //
  //   f(x) = A * (x + 1)
  //
  // Summing and rearranging for A gives A = 2 * I / (n * (n + 1)). If the
  // filter is reversed, we have the same A but with formula
  //
  //   g(x) = A * (n - x)
  const int I = 1 << FILTER_BITS;
  const float A = 2.f * I / (n * (n + 1.f));
  for (int i = 0; i < n; ++i) {
    coeffs_[i] = static_cast<int16_t>(A * (backwards ? (n - i) : (i + 1)));
  }

  // For the other filters, make them slightly different by swapping two
  // columns. Filter k will have the columns (k % n) and (7 * k) % n swapped.
  const size_t filter_size = sizeof(coeffs_[0] * n);
  int16_t *const filter0 = &coeffs_[0];
  for (int k = 1; k < SUBPEL_SHIFTS; ++k) {
    int16_t *filterk = &coeffs_[k * n];
    memcpy(filterk, filter0, filter_size);

    const int idx0 = k % n;
    const int idx1 = (7 * k) % n;

    const int16_t tmp = filterk[idx0];
    filterk[idx0] = filterk[idx1];
    filterk[idx1] = tmp;
  }

  // Finally, write some rubbish at the end to make sure we don't use it.
  for (int i = 0; i < 8; ++i) coeffs_[n * SUBPEL_SHIFTS + i] = 123 + i;

  // Fill in params
  params_.filter_ptr = &coeffs_[0];
  params_.taps = n;
  // These are ignored by the functions being tested. Set them to whatever.
  params_.subpel_shifts = SUBPEL_SHIFTS;
  params_.interp_filter = EIGHTTAP_REGULAR;
}

template <typename SrcPixel>
class TestImage {
 public:
  TestImage(int w, int h, int bd) : w_(w), h_(h), bd_(bd) {
    assert(bd < 16);
    assert(bd <= 8 * static_cast<int>(sizeof(SrcPixel)));

    // Pad width by 2*kHPad and then round up to the next multiple of 16
    // to get src_stride_. Add another 16 for dst_stride_ (to make sure
    // something goes wrong if we use the wrong one)
    src_stride_ = (w_ + 2 * kHPad + 15) & ~15;
    dst_stride_ = src_stride_ + 16;

    // Allocate image data
    src_data_.resize(2 * src_block_size());
    dst_data_.resize(2 * dst_block_size());
  }

  void Initialize(ACMRandom *rnd);
  void Check() const;

  int src_stride() const { return src_stride_; }
  int dst_stride() const { return dst_stride_; }

  int src_block_size() const { return (h_ + 2 * kVPad) * src_stride(); }
  int dst_block_size() const { return (h_ + 2 * kVPad) * dst_stride(); }

  const SrcPixel *GetSrcData(bool ref, bool borders) const {
    const SrcPixel *block = &src_data_[ref ? 0 : src_block_size()];
    return borders ? block : block + kHPad + src_stride_ * kVPad;
  }

  int32_t *GetDstData(bool ref, bool borders) {
    int32_t *block = &dst_data_[ref ? 0 : dst_block_size()];
    return borders ? block : block + kHPad + dst_stride_ * kVPad;
  }

 private:
  int w_, h_, bd_;
  int src_stride_, dst_stride_;

  std::vector<SrcPixel> src_data_;
  std::vector<int32_t> dst_data_;
};

template <typename Pixel>
void FillEdge(ACMRandom *rnd, int num_pixels, int bd, bool trash, Pixel *data) {
  if (!trash) {
    memset(data, 0, sizeof(*data) * num_pixels);
    return;
  }
  const Pixel mask = (1 << bd) - 1;
  for (int i = 0; i < num_pixels; ++i) data[i] = rnd->Rand16() & mask;
}

template <typename Pixel>
void PrepBuffers(ACMRandom *rnd, int w, int h, int stride, int bd,
                 bool trash_edges, Pixel *data) {
  assert(rnd);
  const Pixel mask = (1 << bd) - 1;

  // Fill in the first buffer with random data
  // Top border
  FillEdge(rnd, stride * kVPad, bd, trash_edges, data);
  for (int r = 0; r < h; ++r) {
    Pixel *row_data = data + (kVPad + r) * stride;
    // Left border, contents, right border
    FillEdge(rnd, kHPad, bd, trash_edges, row_data);
    for (int c = 0; c < w; ++c) row_data[kHPad + c] = rnd->Rand16() & mask;
    FillEdge(rnd, kHPad, bd, trash_edges, row_data + kHPad + w);
  }
  // Bottom border
  FillEdge(rnd, stride * kVPad, bd, trash_edges, data + stride * (kVPad + h));

  const int bpp = sizeof(*data);
  const int block_elts = stride * (h + 2 * kVPad);
  const int block_size = bpp * block_elts;

  // Now copy that to the second buffer
  memcpy(data + block_elts, data, block_size);
}

template <typename SrcPixel>
void TestImage<SrcPixel>::Initialize(ACMRandom *rnd) {
  PrepBuffers(rnd, w_, h_, src_stride_, bd_, false, &src_data_[0]);
  PrepBuffers(rnd, w_, h_, dst_stride_, bd_, true, &dst_data_[0]);
}

template <typename SrcPixel>
void TestImage<SrcPixel>::Check() const {
  // If memcmp returns 0, there's nothing to do.
  const int num_pixels = dst_block_size();
  const int32_t *ref_dst = &dst_data_[0];
  const int32_t *tst_dst = &dst_data_[num_pixels];

  if (0 == memcmp(ref_dst, tst_dst, sizeof(*ref_dst) * num_pixels)) return;

  // Otherwise, iterate through the buffer looking for differences (including
  // the edges)
  const int stride = dst_stride_;
  for (int r = 0; r < h_ + 2 * kVPad; ++r) {
    for (int c = 0; c < w_ + 2 * kHPad; ++c) {
      const int32_t ref_value = ref_dst[r * stride + c];
      const int32_t tst_value = tst_dst[r * stride + c];

      EXPECT_EQ(tst_value, ref_value)
          << "Error at row: " << (r - kVPad) << ", col: " << (c - kHPad);
    }
  }
}

typedef tuple<int, int> BlockDimension;

struct BaseParams {
  BaseParams(BlockDimension dims, NTaps ntaps_x, NTaps ntaps_y, bool avg)
      : dims(dims), ntaps_x(ntaps_x), ntaps_y(ntaps_y), avg(avg) {}

  BlockDimension dims;
  NTaps ntaps_x, ntaps_y;
  bool avg;
};

template <typename SrcPixel>
class ConvolveScaleTestBase : public ::testing::Test {
 public:
  ConvolveScaleTestBase() : image_(NULL) {}
  virtual ~ConvolveScaleTestBase() { delete image_; }
  virtual void TearDown() { libaom_test::ClearSystemState(); }

  // Implemented by subclasses (SetUp depends on the parameters passed
  // in and RunOne depends on the function to be tested. These can't
  // be templated for low/high bit depths because they have different
  // numbers of parameters)
  virtual void SetUp() = 0;
  virtual void RunOne(bool ref) = 0;

 protected:
  void SetParams(const BaseParams &params, int bd) {
    width_ = std::tr1::get<0>(params.dims);
    height_ = std::tr1::get<1>(params.dims);
    ntaps_x_ = params.ntaps_x;
    ntaps_y_ = params.ntaps_y;
    bd_ = bd;
    avg_ = params.avg;

    filter_x_.set(ntaps_x_, false);
    filter_y_.set(ntaps_y_, true);
    convolve_params_ = get_conv_params_no_round(0, avg_ != false, 0, NULL, 0);

    delete image_;
    image_ = new TestImage<SrcPixel>(width_, height_, bd_);
  }

  void Run() {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
    for (int i = 0; i < kTestIters; ++i) {
      Prep(&rnd);
      RunOne(true);
      RunOne(false);
      image_->Check();
    }
  }

  void SpeedTest() {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
    Prep(&rnd);

    aom_usec_timer ref_timer;
    aom_usec_timer_start(&ref_timer);
    for (int i = 0; i < kPerfIters; ++i) RunOne(true);
    aom_usec_timer_mark(&ref_timer);
    const int64_t ref_time = aom_usec_timer_elapsed(&ref_timer);

    aom_usec_timer tst_timer;
    aom_usec_timer_start(&tst_timer);
    for (int i = 0; i < kPerfIters; ++i) RunOne(false);
    aom_usec_timer_mark(&tst_timer);
    const int64_t tst_time = aom_usec_timer_elapsed(&tst_timer);

    std::cout << "[          ] C time = " << ref_time / 1000
              << " ms, SIMD time = " << tst_time / 1000 << " ms\n";

    EXPECT_GT(ref_time, tst_time)
        << "Error: CDEFSpeedTest, SIMD slower than C.\n"
        << "C time: " << ref_time << " us\n"
        << "SIMD time: " << tst_time << " us\n";
  }

  static int RandomSubpel(ACMRandom *rnd) {
    const uint8_t subpel_mode = rnd->Rand8();
    if ((subpel_mode & 7) == 0) {
      return 0;
    } else if ((subpel_mode & 7) == 1) {
      return SCALE_SUBPEL_SHIFTS - 1;
    } else {
      return 1 + rnd->PseudoUniform(SCALE_SUBPEL_SHIFTS - 2);
    }
  }

  void Prep(ACMRandom *rnd) {
    assert(rnd);

    // Choose subpel_x_ and subpel_y_. They should be less than
    // SCALE_SUBPEL_SHIFTS; we also want to add extra weight to "interesting"
    // values: 0 and SCALE_SUBPEL_SHIFTS - 1
    subpel_x_ = RandomSubpel(rnd);
    subpel_y_ = RandomSubpel(rnd);

    image_->Initialize(rnd);
  }

  int width_, height_, bd_;
  NTaps ntaps_x_, ntaps_y_;
  bool avg_;
  int subpel_x_, subpel_y_;
  TestFilter filter_x_, filter_y_;
  TestImage<SrcPixel> *image_;
  ConvolveParams convolve_params_;
};

typedef tuple<int, int> BlockDimension;

typedef void (*LowbdConvolveFunc)(const uint8_t *src, int src_stride,
                                  int32_t *dst, int dst_stride, int w, int h,
                                  InterpFilterParams *filter_params_x,
                                  InterpFilterParams *filter_params_y,
                                  const int subpel_x_qn, const int x_step_qn,
                                  const int subpel_y_qn, const int y_step_qn,
                                  ConvolveParams *conv_params);

// Test parameter list:
//  <tst_fun, dims, ntaps_x, ntaps_y, avg>
typedef tuple<LowbdConvolveFunc, BlockDimension, NTaps, NTaps, bool>
    LowBDParams;

class LowBDConvolveScaleTest
    : public ConvolveScaleTestBase<uint8_t>,
      public ::testing::WithParamInterface<LowBDParams> {
 public:
  virtual ~LowBDConvolveScaleTest() {}

  void SetUp() {
    tst_fun_ = GET_PARAM(0);

    const BlockDimension &block = GET_PARAM(1);
    const NTaps ntaps_x = GET_PARAM(2);
    const NTaps ntaps_y = GET_PARAM(3);
    const int bd = 8;
    const bool avg = GET_PARAM(4);

    SetParams(BaseParams(block, ntaps_x, ntaps_y, avg), bd);
  }

  void RunOne(bool ref) {
    const uint8_t *src = image_->GetSrcData(ref, false);
    CONV_BUF_TYPE *dst = image_->GetDstData(ref, false);
    const int src_stride = image_->src_stride();
    const int dst_stride = image_->dst_stride();

    if (ref) {
      av1_convolve_2d_scale_c(src, src_stride, dst, dst_stride, width_, height_,
                              &filter_x_.params_, &filter_y_.params_, subpel_x_,
                              kXStepQn, subpel_y_, kYStepQn, &convolve_params_);
    } else {
      tst_fun_(src, src_stride, dst, dst_stride, width_, height_,
               &filter_x_.params_, &filter_y_.params_, subpel_x_, kXStepQn,
               subpel_y_, kYStepQn, &convolve_params_);
    }
  }

 private:
  LowbdConvolveFunc tst_fun_;
};

const BlockDimension kBlockDim[] = {
  make_tuple(2, 2),    make_tuple(2, 4),    make_tuple(4, 4),
  make_tuple(4, 8),    make_tuple(8, 4),    make_tuple(8, 8),
  make_tuple(8, 16),   make_tuple(16, 8),   make_tuple(16, 16),
  make_tuple(16, 32),  make_tuple(32, 16),  make_tuple(32, 32),
  make_tuple(32, 64),  make_tuple(64, 32),  make_tuple(64, 64),
  make_tuple(64, 128), make_tuple(128, 64), make_tuple(128, 128),
};

const NTaps kNTaps[] = { EIGHT_TAP, TEN_TAP, TWELVE_TAP };

TEST_P(LowBDConvolveScaleTest, Check) { Run(); }
TEST_P(LowBDConvolveScaleTest, DISABLED_Speed) { SpeedTest(); }

INSTANTIATE_TEST_CASE_P(
    SSE4_1, LowBDConvolveScaleTest,
    ::testing::Combine(::testing::Values(av1_convolve_2d_scale_sse4_1),
                       ::testing::ValuesIn(kBlockDim),
                       ::testing::ValuesIn(kNTaps), ::testing::ValuesIn(kNTaps),
                       ::testing::Bool()));

#if CONFIG_HIGHBITDEPTH
typedef void (*HighbdConvolveFunc)(const uint16_t *src, int src_stride,
                                   int32_t *dst, int dst_stride, int w, int h,
                                   InterpFilterParams *filter_params_x,
                                   InterpFilterParams *filter_params_y,
                                   const int subpel_x_qn, const int x_step_qn,
                                   const int subpel_y_qn, const int y_step_qn,
                                   ConvolveParams *conv_params, int bd);

// Test parameter list:
//  <tst_fun, dims, ntaps_x, ntaps_y, avg, bd>
typedef tuple<HighbdConvolveFunc, BlockDimension, NTaps, NTaps, bool, int>
    HighBDParams;

class HighBDConvolveScaleTest
    : public ConvolveScaleTestBase<uint16_t>,
      public ::testing::WithParamInterface<HighBDParams> {
 public:
  virtual ~HighBDConvolveScaleTest() {}

  void SetUp() {
    tst_fun_ = GET_PARAM(0);

    const BlockDimension &block = GET_PARAM(1);
    const NTaps ntaps_x = GET_PARAM(2);
    const NTaps ntaps_y = GET_PARAM(3);
    const bool avg = GET_PARAM(4);
    const int bd = GET_PARAM(5);

    SetParams(BaseParams(block, ntaps_x, ntaps_y, avg), bd);
  }

  void RunOne(bool ref) {
    const uint16_t *src = image_->GetSrcData(ref, false);
    CONV_BUF_TYPE *dst = image_->GetDstData(ref, false);
    const int src_stride = image_->src_stride();
    const int dst_stride = image_->dst_stride();

    if (ref) {
      av1_highbd_convolve_2d_scale_c(
          src, src_stride, dst, dst_stride, width_, height_, &filter_x_.params_,
          &filter_y_.params_, subpel_x_, kXStepQn, subpel_y_, kYStepQn,
          &convolve_params_, bd_);
    } else {
      tst_fun_(src, src_stride, dst, dst_stride, width_, height_,
               &filter_x_.params_, &filter_y_.params_, subpel_x_, kXStepQn,
               subpel_y_, kYStepQn, &convolve_params_, bd_);
    }
  }

 private:
  HighbdConvolveFunc tst_fun_;
};

const int kBDs[] = { 8, 10, 12 };

TEST_P(HighBDConvolveScaleTest, Check) { Run(); }
TEST_P(HighBDConvolveScaleTest, DISABLED_Speed) { SpeedTest(); }

INSTANTIATE_TEST_CASE_P(
    SSE4_1, HighBDConvolveScaleTest,
    ::testing::Combine(::testing::Values(av1_highbd_convolve_2d_scale_sse4_1),
                       ::testing::ValuesIn(kBlockDim),
                       ::testing::ValuesIn(kNTaps), ::testing::ValuesIn(kNTaps),
                       ::testing::Bool(), ::testing::ValuesIn(kBDs)));

#endif  // CONFIG_HIGHBITDEPTH
}  // namespace