avg.c 6.78 KB
Newer Older
Jingning Han's avatar
Jingning Han committed
1 2 3 4 5 6 7 8 9
/*
 *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */
James Zern's avatar
James Zern committed
10 11 12
#include <stdlib.h>

#include "./vpx_dsp_rtcd.h"
Jingning Han's avatar
Jingning Han committed
13 14
#include "vpx_ports/mem.h"

James Zern's avatar
James Zern committed
15
unsigned int vpx_avg_8x8_c(const uint8_t *s, int p) {
Jingning Han's avatar
Jingning Han committed
16 17
  int i, j;
  int sum = 0;
clang-format's avatar
clang-format committed
18 19 20
  for (i = 0; i < 8; ++i, s += p)
    for (j = 0; j < 8; sum += s[j], ++j) {
    }
Jingning Han's avatar
Jingning Han committed
21 22 23 24

  return (sum + 32) >> 6;
}

James Zern's avatar
James Zern committed
25
unsigned int vpx_avg_4x4_c(const uint8_t *s, int p) {
Jingning Han's avatar
Jingning Han committed
26 27
  int i, j;
  int sum = 0;
clang-format's avatar
clang-format committed
28 29 30
  for (i = 0; i < 4; ++i, s += p)
    for (j = 0; j < 4; sum += s[j], ++j) {
    }
Jingning Han's avatar
Jingning Han committed
31 32 33 34 35 36

  return (sum + 8) >> 4;
}

// src_diff: first pass, 9 bit, dynamic range [-255, 255]
//           second pass, 12 bit, dynamic range [-2040, 2040]
37
static void hadamard_col8(const int16_t *src_diff, ptrdiff_t src_stride,
Jingning Han's avatar
Jingning Han committed
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
                          int16_t *coeff) {
  int16_t b0 = src_diff[0 * src_stride] + src_diff[1 * src_stride];
  int16_t b1 = src_diff[0 * src_stride] - src_diff[1 * src_stride];
  int16_t b2 = src_diff[2 * src_stride] + src_diff[3 * src_stride];
  int16_t b3 = src_diff[2 * src_stride] - src_diff[3 * src_stride];
  int16_t b4 = src_diff[4 * src_stride] + src_diff[5 * src_stride];
  int16_t b5 = src_diff[4 * src_stride] - src_diff[5 * src_stride];
  int16_t b6 = src_diff[6 * src_stride] + src_diff[7 * src_stride];
  int16_t b7 = src_diff[6 * src_stride] - src_diff[7 * src_stride];

  int16_t c0 = b0 + b2;
  int16_t c1 = b1 + b3;
  int16_t c2 = b0 - b2;
  int16_t c3 = b1 - b3;
  int16_t c4 = b4 + b6;
  int16_t c5 = b5 + b7;
  int16_t c6 = b4 - b6;
  int16_t c7 = b5 - b7;

  coeff[0] = c0 + c4;
  coeff[7] = c1 + c5;
  coeff[3] = c2 + c6;
  coeff[4] = c3 + c7;
  coeff[2] = c0 - c4;
  coeff[6] = c1 - c5;
  coeff[1] = c2 - c6;
  coeff[5] = c3 - c7;
}

Johann's avatar
Johann committed
67 68
// The order of the output coeff of the hadamard is not important. For
// optimization purposes the final transpose may be skipped.
69
void vpx_hadamard_8x8_c(const int16_t *src_diff, ptrdiff_t src_stride,
70
                        tran_low_t *coeff) {
Jingning Han's avatar
Jingning Han committed
71 72
  int idx;
  int16_t buffer[64];
73
  int16_t buffer2[64];
Jingning Han's avatar
Jingning Han committed
74 75 76 77 78 79 80 81 82 83
  int16_t *tmp_buf = &buffer[0];
  for (idx = 0; idx < 8; ++idx) {
    hadamard_col8(src_diff, src_stride, tmp_buf);  // src_diff: 9 bit
                                                   // dynamic range [-255, 255]
    tmp_buf += 8;
    ++src_diff;
  }

  tmp_buf = &buffer[0];
  for (idx = 0; idx < 8; ++idx) {
84 85 86 87
    hadamard_col8(tmp_buf, 8, buffer2 + 8 * idx);  // tmp_buf: 12 bit
    // dynamic range [-2040, 2040]
    // buffer2: 15 bit
    // dynamic range [-16320, 16320]
Jingning Han's avatar
Jingning Han committed
88 89
    ++tmp_buf;
  }
90 91

  for (idx = 0; idx < 64; ++idx) coeff[idx] = (tran_low_t)buffer2[idx];
Jingning Han's avatar
Jingning Han committed
92 93 94
}

// In place 16x16 2D Hadamard transform
95
void vpx_hadamard_16x16_c(const int16_t *src_diff, ptrdiff_t src_stride,
96
                          tran_low_t *coeff) {
Jingning Han's avatar
Jingning Han committed
97 98 99
  int idx;
  for (idx = 0; idx < 4; ++idx) {
    // src_diff: 9 bit, dynamic range [-255, 255]
clang-format's avatar
clang-format committed
100 101
    const int16_t *src_ptr =
        src_diff + (idx >> 1) * 8 * src_stride + (idx & 0x01) * 8;
James Zern's avatar
James Zern committed
102
    vpx_hadamard_8x8_c(src_ptr, src_stride, coeff + idx * 64);
Jingning Han's avatar
Jingning Han committed
103 104 105 106
  }

  // coeff: 15 bit, dynamic range [-16320, 16320]
  for (idx = 0; idx < 64; ++idx) {
107 108 109 110
    tran_low_t a0 = coeff[0];
    tran_low_t a1 = coeff[64];
    tran_low_t a2 = coeff[128];
    tran_low_t a3 = coeff[192];
Jingning Han's avatar
Jingning Han committed
111

112 113 114 115
    tran_low_t b0 = (a0 + a1) >> 1;  // (a0 + a1): 16 bit, [-32640, 32640]
    tran_low_t b1 = (a0 - a1) >> 1;  // b0-b3: 15 bit, dynamic range
    tran_low_t b2 = (a2 + a3) >> 1;  // [-16320, 16320]
    tran_low_t b3 = (a2 - a3) >> 1;
Jingning Han's avatar
Jingning Han committed
116

clang-format's avatar
clang-format committed
117 118
    coeff[0] = b0 + b2;  // 16 bit, [-32640, 32640]
    coeff[64] = b1 + b3;
Jingning Han's avatar
Jingning Han committed
119 120 121 122 123 124 125 126 127
    coeff[128] = b0 - b2;
    coeff[192] = b1 - b3;

    ++coeff;
  }
}

// coeff: 16 bits, dynamic range [-32640, 32640].
// length: value range {16, 64, 256, 1024}.
128
int vpx_satd_c(const tran_low_t *coeff, int length) {
Jingning Han's avatar
Jingning Han committed
129 130
  int i;
  int satd = 0;
clang-format's avatar
clang-format committed
131
  for (i = 0; i < length; ++i) satd += abs(coeff[i]);
Jingning Han's avatar
Jingning Han committed
132 133

  // satd: 26 bits, dynamic range [-32640 * 1024, 32640 * 1024]
James Zern's avatar
James Zern committed
134
  return satd;
Jingning Han's avatar
Jingning Han committed
135 136 137 138
}

// Integer projection onto row vectors.
// height: value range {16, 32, 64}.
Johann's avatar
Johann committed
139
void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t *ref,
Jingning Han's avatar
Jingning Han committed
140 141 142 143 144 145 146
                       const int ref_stride, const int height) {
  int idx;
  const int norm_factor = height >> 1;
  for (idx = 0; idx < 16; ++idx) {
    int i;
    hbuf[idx] = 0;
    // hbuf[idx]: 14 bit, dynamic range [0, 16320].
clang-format's avatar
clang-format committed
147
    for (i = 0; i < height; ++i) hbuf[idx] += ref[i * ref_stride];
Jingning Han's avatar
Jingning Han committed
148 149 150 151 152 153 154
    // hbuf[idx]: 9 bit, dynamic range [0, 510].
    hbuf[idx] /= norm_factor;
    ++ref;
  }
}

// width: value range {16, 32, 64}.
Johann's avatar
Johann committed
155
int16_t vpx_int_pro_col_c(const uint8_t *ref, const int width) {
Jingning Han's avatar
Jingning Han committed
156 157 158
  int idx;
  int16_t sum = 0;
  // sum: 14 bit, dynamic range [0, 16320]
clang-format's avatar
clang-format committed
159
  for (idx = 0; idx < width; ++idx) sum += ref[idx];
Jingning Han's avatar
Jingning Han committed
160 161 162 163 164 165
  return sum;
}

// ref: [0 - 510]
// src: [0 - 510]
// bwl: {2, 3, 4}
clang-format's avatar
clang-format committed
166
int vpx_vector_var_c(const int16_t *ref, const int16_t *src, const int bwl) {
Jingning Han's avatar
Jingning Han committed
167 168 169 170 171 172 173 174 175 176 177 178 179 180 181
  int i;
  int width = 4 << bwl;
  int sse = 0, mean = 0, var;

  for (i = 0; i < width; ++i) {
    int diff = ref[i] - src[i];  // diff: dynamic range [-510, 510], 10 bits.
    mean += diff;                // mean: dynamic range 16 bits.
    sse += diff * diff;          // sse:  dynamic range 26 bits.
  }

  // (mean * mean): dynamic range 31 bits.
  var = sse - ((mean * mean) >> (bwl + 2));
  return var;
}

James Zern's avatar
James Zern committed
182
void vpx_minmax_8x8_c(const uint8_t *s, int p, const uint8_t *d, int dp,
Jingning Han's avatar
Jingning Han committed
183 184 185 186 187 188
                      int *min, int *max) {
  int i, j;
  *min = 255;
  *max = 0;
  for (i = 0; i < 8; ++i, s += p, d += dp) {
    for (j = 0; j < 8; ++j) {
clang-format's avatar
clang-format committed
189
      int diff = abs(s[j] - d[j]);
Jingning Han's avatar
Jingning Han committed
190 191 192 193 194 195 196
      *min = diff < *min ? diff : *min;
      *max = diff > *max ? diff : *max;
    }
  }
}

#if CONFIG_VP9_HIGHBITDEPTH
James Zern's avatar
James Zern committed
197
unsigned int vpx_highbd_avg_8x8_c(const uint8_t *s8, int p) {
Jingning Han's avatar
Jingning Han committed
198 199
  int i, j;
  int sum = 0;
clang-format's avatar
clang-format committed
200 201 202 203
  const uint16_t *s = CONVERT_TO_SHORTPTR(s8);
  for (i = 0; i < 8; ++i, s += p)
    for (j = 0; j < 8; sum += s[j], ++j) {
    }
Jingning Han's avatar
Jingning Han committed
204 205 206 207

  return (sum + 32) >> 6;
}

James Zern's avatar
James Zern committed
208
unsigned int vpx_highbd_avg_4x4_c(const uint8_t *s8, int p) {
Jingning Han's avatar
Jingning Han committed
209 210
  int i, j;
  int sum = 0;
clang-format's avatar
clang-format committed
211 212 213 214
  const uint16_t *s = CONVERT_TO_SHORTPTR(s8);
  for (i = 0; i < 4; ++i, s += p)
    for (j = 0; j < 4; sum += s[j], ++j) {
    }
Jingning Han's avatar
Jingning Han committed
215 216 217 218

  return (sum + 8) >> 4;
}

James Zern's avatar
James Zern committed
219
void vpx_highbd_minmax_8x8_c(const uint8_t *s8, int p, const uint8_t *d8,
Jingning Han's avatar
Jingning Han committed
220 221
                             int dp, int *min, int *max) {
  int i, j;
clang-format's avatar
clang-format committed
222 223
  const uint16_t *s = CONVERT_TO_SHORTPTR(s8);
  const uint16_t *d = CONVERT_TO_SHORTPTR(d8);
Jingning Han's avatar
Jingning Han committed
224 225 226 227
  *min = 255;
  *max = 0;
  for (i = 0; i < 8; ++i, s += p, d += dp) {
    for (j = 0; j < 8; ++j) {
clang-format's avatar
clang-format committed
228
      int diff = abs(s[j] - d[j]);
Jingning Han's avatar
Jingning Han committed
229 230 231 232 233 234
      *min = diff < *min ? diff : *min;
      *max = diff > *max ? diff : *max;
    }
  }
}
#endif  // CONFIG_VP9_HIGHBITDEPTH