test/av1_fwd_txfm2d_test.cc

/*
 * 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.
 */

#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <vector>

#include "test/acm_random.h"
#include "test/util.h"
#include "test/av1_txfm_test.h"
#include "av1/common/av1_txfm.h"
#include "./av1_rtcd.h"

using libaom_test::ACMRandom;
using libaom_test::TYPE_TXFM;
using libaom_test::bd;
using libaom_test::compute_avg_abs_error;
using libaom_test::input_base;

using std::vector;

namespace {
// tx_type_, tx_size_, max_error_, max_avg_error_
typedef ::testing::tuple<TX_TYPE, TX_SIZE, double, double> AV1FwdTxfm2dParam;

class AV1FwdTxfm2d : public ::testing::TestWithParam<AV1FwdTxfm2dParam> {
 public:
  virtual void SetUp() {
    tx_type_ = GET_PARAM(0);
    tx_size_ = GET_PARAM(1);
    max_error_ = GET_PARAM(2);
    max_avg_error_ = GET_PARAM(3);
    count_ = 500;
    TXFM_2D_FLIP_CFG fwd_txfm_flip_cfg;
    av1_get_fwd_txfm_cfg(tx_type_, tx_size_, &fwd_txfm_flip_cfg);
    amplify_factor_ = libaom_test::get_amplification_factor(tx_type_, tx_size_);
    tx_width_ = tx_size_wide[fwd_txfm_flip_cfg.tx_size];
    tx_height_ = tx_size_high[fwd_txfm_flip_cfg.tx_size];
    ud_flip_ = fwd_txfm_flip_cfg.ud_flip;
    lr_flip_ = fwd_txfm_flip_cfg.lr_flip;

    fwd_txfm_ = libaom_test::fwd_txfm_func_ls[tx_size_];
    txfm2d_size_ = tx_width_ * tx_height_;
    input_ = reinterpret_cast<int16_t *>(
        aom_memalign(16, sizeof(input_[0]) * txfm2d_size_));
    output_ = reinterpret_cast<int32_t *>(
        aom_memalign(16, sizeof(output_[0]) * txfm2d_size_));
    ref_input_ = reinterpret_cast<double *>(
        aom_memalign(16, sizeof(ref_input_[0]) * txfm2d_size_));
    ref_output_ = reinterpret_cast<double *>(
        aom_memalign(16, sizeof(ref_output_[0]) * txfm2d_size_));
  }

  void RunFwdAccuracyCheck() {
    ACMRandom rnd(ACMRandom::DeterministicSeed());
    double avg_abs_error = 0;
    for (int ci = 0; ci < count_; ci++) {
      for (int ni = 0; ni < txfm2d_size_; ++ni) {
        input_[ni] = rnd.Rand16() % input_base;
        ref_input_[ni] = static_cast<double>(input_[ni]);
        output_[ni] = 0;
        ref_output_[ni] = 0;
      }

      fwd_txfm_(input_, output_, tx_width_, tx_type_, bd);

      if (lr_flip_ && ud_flip_) {
        libaom_test::fliplrud(ref_input_, tx_width_, tx_height_, tx_width_);
      } else if (lr_flip_) {
        libaom_test::fliplr(ref_input_, tx_width_, tx_height_, tx_width_);
      } else if (ud_flip_) {
        libaom_test::flipud(ref_input_, tx_width_, tx_height_, tx_width_);
      }

      libaom_test::reference_hybrid_2d(ref_input_, ref_output_, tx_type_,
                                       tx_size_);

      double actual_max_error = 0;
      for (int ni = 0; ni < txfm2d_size_; ++ni) {
        ref_output_[ni] = round(ref_output_[ni]);
        const double this_error =
            fabs(output_[ni] - ref_output_[ni]) / amplify_factor_;
        actual_max_error = AOMMAX(actual_max_error, this_error);
      }
      EXPECT_GE(max_error_, actual_max_error)
          << "tx_size = " << tx_size_ << ", tx_type = " << tx_type_;
      if (actual_max_error > max_error_) {  // exit early.
        break;
      }

      avg_abs_error += compute_avg_abs_error<int32_t, double>(
          output_, ref_output_, txfm2d_size_);
    }

    avg_abs_error /= amplify_factor_;
    avg_abs_error /= count_;
    EXPECT_GE(max_avg_error_, avg_abs_error)
        << "tx_size = " << tx_size_ << ", tx_type = " << tx_type_;
  }

  virtual void TearDown() {
    aom_free(input_);
    aom_free(output_);
    aom_free(ref_input_);
    aom_free(ref_output_);
  }

 private:
  double max_error_;
  double max_avg_error_;
  int count_;
  double amplify_factor_;
  TX_TYPE tx_type_;
  TX_SIZE tx_size_;
  int tx_width_;
  int tx_height_;
  int txfm2d_size_;
  FwdTxfm2dFunc fwd_txfm_;
  int16_t *input_;
  int32_t *output_;
  double *ref_input_;
  double *ref_output_;
  int ud_flip_;  // flip upside down
  int lr_flip_;  // flip left to right
};

vector<AV1FwdTxfm2dParam> GetTxfm2dParamList() {
  vector<AV1FwdTxfm2dParam> param_list;
  for (int t = 0; t < TX_TYPES; ++t) {
    const TX_TYPE tx_type = static_cast<TX_TYPE>(t);
    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_4X4, 3, 0.5));
    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_8X8, 5, 0.5));
    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_16X16, 11, 1.2));
    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_32X32, 70, 6.1));
    if (tx_type == DCT_DCT) {  // Other types not supported by these tx sizes.
      param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_64X64, 64, 3.4));
    }

    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_4X8, 3.9, 0.57));
    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_8X4, 4.3, 0.68));
    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_8X16, 12, 0.92));
    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_16X8, 12, 1.1));
    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_16X32, 32, 4.1));
    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_32X16, 46, 6));

    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_4X16, 5, 0.6));
    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_16X4, 6, 0.9));
    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_8X32, 21, 1.2));
    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_32X8, 13, 1.7));

    if (tx_type == DCT_DCT) {  // Other types not supported by these tx sizes.
      param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_32X64, 136, 3.5));
      param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_64X32, 136, 5.7));
      param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_16X64, 30, 2.0));
      param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_64X16, 36, 4.7));
    }
  }
  return param_list;
}

INSTANTIATE_TEST_CASE_P(C, AV1FwdTxfm2d,
                        ::testing::ValuesIn(GetTxfm2dParamList()));

TEST_P(AV1FwdTxfm2d, RunFwdAccuracyCheck) { RunFwdAccuracyCheck(); }

TEST(AV1FwdTxfm2d, CfgTest) {
  for (int bd_idx = 0; bd_idx < BD_NUM; ++bd_idx) {
    int bd = libaom_test::bd_arr[bd_idx];
    int8_t low_range = libaom_test::low_range_arr[bd_idx];
    int8_t high_range = libaom_test::high_range_arr[bd_idx];
    for (int tx_size = 0; tx_size < TX_SIZES_ALL; ++tx_size) {
      for (int tx_type = 0; tx_type < TX_TYPES; ++tx_type) {
        if ((tx_size_wide[tx_size] == 64 || tx_size_high[tx_size] == 64) &&
            (tx_type != DCT_DCT && tx_type != IDTX && tx_type != V_DCT &&
             tx_type != H_DCT)) {
          continue;
        }
        TXFM_2D_FLIP_CFG cfg;
        av1_get_fwd_txfm_cfg(static_cast<TX_TYPE>(tx_type),
                             static_cast<TX_SIZE>(tx_size), &cfg);
        int8_t stage_range_col[MAX_TXFM_STAGE_NUM];
        int8_t stage_range_row[MAX_TXFM_STAGE_NUM];
        av1_gen_fwd_stage_range(stage_range_col, stage_range_row, &cfg, bd);
        libaom_test::txfm_stage_range_check(stage_range_col, cfg.stage_num_col,
                                            cfg.cos_bit_col, low_range,
                                            high_range);
        libaom_test::txfm_stage_range_check(stage_range_row, cfg.stage_num_row,
                                            cfg.cos_bit_row, low_range,
                                            high_range);
      }
    }
  }
}

typedef void (*lowbd_fwd_txfm_func)(const int16_t *src_diff, tran_low_t *coeff,
                                    int diff_stride, TxfmParam *txfm_param);

void AV1FwdTxfm2dMatchTest(TX_SIZE tx_size, lowbd_fwd_txfm_func target_func) {
  const int bd = 8;
  TxfmParam param;
  memset(&param, 0, sizeof(param));
  const int rows = tx_size_high[tx_size];
  const int cols = tx_size_wide[tx_size];
  // printf("%d x %d\n", cols, rows);
  for (int tx_type = 0; tx_type < TX_TYPES; ++tx_type) {
    if ((rows >= 32 || cols >= 32) && tx_type != DCT_DCT && tx_type != IDTX &&
        tx_type != V_DCT && tx_type != H_DCT) {
      // No ADST for large size transforms.
      continue;
    }
    FwdTxfm2dFunc ref_func = libaom_test::fwd_txfm_func_ls[tx_size];
    if (ref_func != NULL) {
      DECLARE_ALIGNED(16, int16_t, input[64 * 64]) = { 0 };
      DECLARE_ALIGNED(16, int32_t, output[64 * 64]);
      DECLARE_ALIGNED(16, int32_t, ref_output[64 * 64]);
      int input_stride = 64;
      ACMRandom rnd(ACMRandom::DeterministicSeed());
      for (int cnt = 0; cnt < 500; ++cnt) {
        if (cnt == 0) {
          for (int r = 0; r < rows; ++r) {
            for (int c = 0; c < cols; ++c) {
              input[r * input_stride + c] = (1 << bd) - 1;
            }
          }
        } else {
          for (int r = 0; r < rows; ++r) {
            for (int c = 0; c < cols; ++c) {
              input[r * input_stride + c] = rnd.Rand16() % (1 << bd);
            }
          }
        }
        param.tx_type = (TX_TYPE)tx_type;
        param.tx_size = (TX_SIZE)tx_size;
        param.tx_set_type = EXT_TX_SET_ALL16;
        param.bd = bd;
        ref_func(input, ref_output, input_stride, (TX_TYPE)tx_type, bd);
        target_func(input, output, input_stride, &param);
        const int check_rows = AOMMIN(32, rows);
        const int check_cols = AOMMIN(32, rows * cols / check_rows);
        for (int r = 0; r < check_rows; ++r) {
          for (int c = 0; c < check_cols; ++c) {
            ASSERT_EQ(ref_output[r * check_cols + c],
                      output[r * check_cols + c])
                << "[" << r << "," << c << "] cnt:" << cnt
                << " tx_size: " << tx_size << " tx_type: " << tx_type;
          }
        }
      }
    }
  }
}

#if HAVE_SSE2
static TX_SIZE fwd_txfm_for_sse2[] = {
  TX_4X4,
  TX_8X8,
  TX_16X16,
  TX_32X32,
  // TX_64X64,
  TX_4X8,
  TX_8X4,
  TX_8X16,
  TX_16X8,
  TX_16X32,
  TX_32X16,
  // TX_32X64,
  // TX_64X32,
  TX_4X16,
  TX_16X4,
  TX_8X32,
  TX_32X8,
  TX_16X64,
  TX_64X16,
};
TEST(av1_fwd_txfm2d_sse2, match) {
  const int cnt = sizeof(fwd_txfm_for_sse2) / sizeof(TX_SIZE);
  for (int i = 0; i < cnt; ++i) {
    AV1FwdTxfm2dMatchTest(fwd_txfm_for_sse2[i], av1_lowbd_fwd_txfm_sse2);
  }
}
#endif  // HAVE_SSE2

#if HAVE_SSE4_1
static TX_SIZE fwd_txfm_for_sse41[] = {
  TX_4X4,
  TX_64X64,
  TX_32X64,
  TX_64X32,
};
TEST(av1_fwd_txfm2d_sse4_1, match) {
  const int cnt = sizeof(fwd_txfm_for_sse41) / sizeof(TX_SIZE);
  for (int i = 0; i < cnt; ++i) {
    AV1FwdTxfm2dMatchTest(fwd_txfm_for_sse41[i], av1_lowbd_fwd_txfm_sse4_1);
  }
}
#endif  // HAVE_SSE4_1
}  // namespace