1/*
2 *  Copyright (c) 2016 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#include <assert.h>
11#include <stdlib.h>
12#include "./vpx_dsp_rtcd.h"
13#include "vpx_ports/mem.h"
14
15DECLARE_PROTECTED(const int16_t vpx_rv[]) = {
16  8,  5,  2,  2,  8,  12, 4,  9,  8,  3,  0,  3,  9,  0,  0,  0,  8,  3,  14,
17  4,  10, 1,  11, 14, 1,  14, 9,  6,  12, 11, 8,  6,  10, 0,  0,  8,  9,  0,
18  3,  14, 8,  11, 13, 4,  2,  9,  0,  3,  9,  6,  1,  2,  3,  14, 13, 1,  8,
19  2,  9,  7,  3,  3,  1,  13, 13, 6,  6,  5,  2,  7,  11, 9,  11, 8,  7,  3,
20  2,  0,  13, 13, 14, 4,  12, 5,  12, 10, 8,  10, 13, 10, 4,  14, 4,  10, 0,
21  8,  11, 1,  13, 7,  7,  14, 6,  14, 13, 2,  13, 5,  4,  4,  0,  10, 0,  5,
22  13, 2,  12, 7,  11, 13, 8,  0,  4,  10, 7,  2,  7,  2,  2,  5,  3,  4,  7,
23  3,  3,  14, 14, 5,  9,  13, 3,  14, 3,  6,  3,  0,  11, 8,  13, 1,  13, 1,
24  12, 0,  10, 9,  7,  6,  2,  8,  5,  2,  13, 7,  1,  13, 14, 7,  6,  7,  9,
25  6,  10, 11, 7,  8,  7,  5,  14, 8,  4,  4,  0,  8,  7,  10, 0,  8,  14, 11,
26  3,  12, 5,  7,  14, 3,  14, 5,  2,  6,  11, 12, 12, 8,  0,  11, 13, 1,  2,
27  0,  5,  10, 14, 7,  8,  0,  4,  11, 0,  8,  0,  3,  10, 5,  8,  0,  11, 6,
28  7,  8,  10, 7,  13, 9,  2,  5,  1,  5,  10, 2,  4,  3,  5,  6,  10, 8,  9,
29  4,  11, 14, 0,  10, 0,  5,  13, 2,  12, 7,  11, 13, 8,  0,  4,  10, 7,  2,
30  7,  2,  2,  5,  3,  4,  7,  3,  3,  14, 14, 5,  9,  13, 3,  14, 3,  6,  3,
31  0,  11, 8,  13, 1,  13, 1,  12, 0,  10, 9,  7,  6,  2,  8,  5,  2,  13, 7,
32  1,  13, 14, 7,  6,  7,  9,  6,  10, 11, 7,  8,  7,  5,  14, 8,  4,  4,  0,
33  8,  7,  10, 0,  8,  14, 11, 3,  12, 5,  7,  14, 3,  14, 5,  2,  6,  11, 12,
34  12, 8,  0,  11, 13, 1,  2,  0,  5,  10, 14, 7,  8,  0,  4,  11, 0,  8,  0,
35  3,  10, 5,  8,  0,  11, 6,  7,  8,  10, 7,  13, 9,  2,  5,  1,  5,  10, 2,
36  4,  3,  5,  6,  10, 8,  9,  4,  11, 14, 3,  8,  3,  7,  8,  5,  11, 4,  12,
37  3,  11, 9,  14, 8,  14, 13, 4,  3,  1,  2,  14, 6,  5,  4,  4,  11, 4,  6,
38  2,  1,  5,  8,  8,  12, 13, 5,  14, 10, 12, 13, 0,  9,  5,  5,  11, 10, 13,
39  9,  10, 13,
40};
41
42void vpx_post_proc_down_and_across_mb_row_c(unsigned char *src_ptr,
43                                            unsigned char *dst_ptr,
44                                            int src_pixels_per_line,
45                                            int dst_pixels_per_line, int cols,
46                                            unsigned char *f, int size) {
47  unsigned char *p_src, *p_dst;
48  int row;
49  int col;
50  unsigned char v;
51  unsigned char d[4];
52
53  assert(size >= 8);
54  assert(cols >= 8);
55
56  for (row = 0; row < size; row++) {
57    /* post_proc_down for one row */
58    p_src = src_ptr;
59    p_dst = dst_ptr;
60
61    for (col = 0; col < cols; col++) {
62      unsigned char p_above2 = p_src[col - 2 * src_pixels_per_line];
63      unsigned char p_above1 = p_src[col - src_pixels_per_line];
64      unsigned char p_below1 = p_src[col + src_pixels_per_line];
65      unsigned char p_below2 = p_src[col + 2 * src_pixels_per_line];
66
67      v = p_src[col];
68
69      if ((abs(v - p_above2) < f[col]) && (abs(v - p_above1) < f[col]) &&
70          (abs(v - p_below1) < f[col]) && (abs(v - p_below2) < f[col])) {
71        unsigned char k1, k2, k3;
72        k1 = (p_above2 + p_above1 + 1) >> 1;
73        k2 = (p_below2 + p_below1 + 1) >> 1;
74        k3 = (k1 + k2 + 1) >> 1;
75        v = (k3 + v + 1) >> 1;
76      }
77
78      p_dst[col] = v;
79    }
80
81    /* now post_proc_across */
82    p_src = dst_ptr;
83    p_dst = dst_ptr;
84
85    p_src[-2] = p_src[-1] = p_src[0];
86    p_src[cols] = p_src[cols + 1] = p_src[cols - 1];
87
88    for (col = 0; col < cols; col++) {
89      v = p_src[col];
90
91      if ((abs(v - p_src[col - 2]) < f[col]) &&
92          (abs(v - p_src[col - 1]) < f[col]) &&
93          (abs(v - p_src[col + 1]) < f[col]) &&
94          (abs(v - p_src[col + 2]) < f[col])) {
95        unsigned char k1, k2, k3;
96        k1 = (p_src[col - 2] + p_src[col - 1] + 1) >> 1;
97        k2 = (p_src[col + 2] + p_src[col + 1] + 1) >> 1;
98        k3 = (k1 + k2 + 1) >> 1;
99        v = (k3 + v + 1) >> 1;
100      }
101
102      d[col & 3] = v;
103
104      if (col >= 2) p_dst[col - 2] = d[(col - 2) & 3];
105    }
106
107    /* handle the last two pixels */
108    p_dst[col - 2] = d[(col - 2) & 3];
109    p_dst[col - 1] = d[(col - 1) & 3];
110
111    /* next row */
112    src_ptr += src_pixels_per_line;
113    dst_ptr += dst_pixels_per_line;
114  }
115}
116
117void vpx_mbpost_proc_across_ip_c(unsigned char *src, int pitch, int rows,
118                                 int cols, int flimit) {
119  int r, c, i;
120
121  unsigned char *s = src;
122  unsigned char d[16];
123
124  for (r = 0; r < rows; r++) {
125    int sumsq = 16;
126    int sum = 0;
127
128    for (i = -8; i < 0; i++) s[i] = s[0];
129
130    /* 17 avoids valgrind warning - we buffer values in c in d
131     * and only write them when we've read 8 ahead...
132     */
133    for (i = 0; i < 17; i++) s[i + cols] = s[cols - 1];
134
135    for (i = -8; i <= 6; i++) {
136      sumsq += s[i] * s[i];
137      sum += s[i];
138      d[i + 8] = 0;
139    }
140
141    for (c = 0; c < cols + 8; c++) {
142      int x = s[c + 7] - s[c - 8];
143      int y = s[c + 7] + s[c - 8];
144
145      sum += x;
146      sumsq += x * y;
147
148      d[c & 15] = s[c];
149
150      if (sumsq * 15 - sum * sum < flimit) {
151        d[c & 15] = (8 + sum + s[c]) >> 4;
152      }
153
154      s[c - 8] = d[(c - 8) & 15];
155    }
156
157    s += pitch;
158  }
159}
160
161void vpx_mbpost_proc_down_c(unsigned char *dst, int pitch, int rows, int cols,
162                            int flimit) {
163  int r, c, i;
164
165  for (c = 0; c < cols; c++) {
166    unsigned char *s = &dst[c];
167    int sumsq = 0;
168    int sum = 0;
169    unsigned char d[16];
170
171    for (i = -8; i < 0; i++) s[i * pitch] = s[0];
172
173    /* 17 avoids valgrind warning - we buffer values in c in d
174     * and only write them when we've read 8 ahead...
175     */
176    for (i = 0; i < 17; i++) s[(i + rows) * pitch] = s[(rows - 1) * pitch];
177
178    for (i = -8; i <= 6; i++) {
179      sumsq += s[i * pitch] * s[i * pitch];
180      sum += s[i * pitch];
181    }
182
183    for (r = 0; r < rows + 8; r++) {
184      sumsq += s[7 * pitch] * s[7 * pitch] - s[-8 * pitch] * s[-8 * pitch];
185      sum += s[7 * pitch] - s[-8 * pitch];
186      d[r & 15] = s[0];
187
188      if (sumsq * 15 - sum * sum < flimit) {
189        d[r & 15] = (vpx_rv[(r & 127) + (c & 7)] + sum + s[0]) >> 4;
190      }
191      if (r >= 8) s[-8 * pitch] = d[(r - 8) & 15];
192      s += pitch;
193    }
194  }
195}
196