1 /*
2  * Copyright (c) 2018, Alliance for Open Media. All rights reserved
3  *
4  * This source code is subject to the terms of the BSD 2 Clause License and
5  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6  * was not distributed with this source code in the LICENSE file, you can
7  * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8  * Media Patent License 1.0 was not distributed with this source code in the
9  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10  */
11 
12 #include <assert.h>
13 #include <stdlib.h>
14 #include <math.h>
15 
16 #include "config/av1_rtcd.h"
17 #include "av1/encoder/dwt.h"
18 
19 // Note: block length must be even for this implementation
analysis_53_row(int length,tran_low_t * x,tran_low_t * lowpass,tran_low_t * highpass)20 static void analysis_53_row(int length, tran_low_t *x, tran_low_t *lowpass,
21                             tran_low_t *highpass) {
22   int n;
23   tran_low_t r, *a, *b;
24 
25   n = length >> 1;
26   b = highpass;
27   a = lowpass;
28   while (--n) {
29     *a++ = (r = *x++) * 2;
30     *b++ = *x - ((r + x[1] + 1) >> 1);
31     x++;
32   }
33   *a = (r = *x++) * 2;
34   *b = *x - r;
35 
36   n = length >> 1;
37   b = highpass;
38   a = lowpass;
39   r = *highpass;
40   while (n--) {
41     *a++ += (r + (*b) + 1) >> 1;
42     r = *b++;
43   }
44 }
45 
analysis_53_col(int length,tran_low_t * x,tran_low_t * lowpass,tran_low_t * highpass)46 static void analysis_53_col(int length, tran_low_t *x, tran_low_t *lowpass,
47                             tran_low_t *highpass) {
48   int n;
49   tran_low_t r, *a, *b;
50 
51   n = length >> 1;
52   b = highpass;
53   a = lowpass;
54   while (--n) {
55     *a++ = (r = *x++);
56     *b++ = (((*x) * 2) - (r + x[1]) + 2) >> 2;
57     x++;
58   }
59   *a = (r = *x++);
60   *b = (*x - r + 1) >> 1;
61 
62   n = length >> 1;
63   b = highpass;
64   a = lowpass;
65   r = *highpass;
66   while (n--) {
67     *a++ += (r + (*b) + 1) >> 1;
68     r = *b++;
69   }
70 }
71 
dyadic_analyze_53_uint8_input(int levels,int width,int height,uint8_t * x,int pitch_x,tran_low_t * c,int pitch_c,int dwt_scale_bits,int hbd)72 static void dyadic_analyze_53_uint8_input(int levels, int width, int height,
73                                           uint8_t *x, int pitch_x,
74                                           tran_low_t *c, int pitch_c,
75                                           int dwt_scale_bits, int hbd) {
76   int lv, i, j, nh, nw, hh = height, hw = width;
77   tran_low_t buffer[2 * DWT_MAX_LENGTH];
78 
79   if (hbd) {
80     uint16_t *x16 = CONVERT_TO_SHORTPTR(x);
81     for (i = 0; i < height; i++) {
82       for (j = 0; j < width; j++) {
83         c[i * pitch_c + j] = x16[i * pitch_x + j] << dwt_scale_bits;
84       }
85     }
86   } else {
87     for (i = 0; i < height; i++) {
88       for (j = 0; j < width; j++) {
89         c[i * pitch_c + j] = x[i * pitch_x + j] << dwt_scale_bits;
90       }
91     }
92   }
93 
94   for (lv = 0; lv < levels; lv++) {
95     nh = hh;
96     hh = (hh + 1) >> 1;
97     nw = hw;
98     hw = (hw + 1) >> 1;
99     if ((nh < 2) || (nw < 2)) return;
100     for (i = 0; i < nh; i++) {
101       memcpy(buffer, &c[i * pitch_c], nw * sizeof(tran_low_t));
102       analysis_53_row(nw, buffer, &c[i * pitch_c], &c[i * pitch_c] + hw);
103     }
104     for (j = 0; j < nw; j++) {
105       for (i = 0; i < nh; i++) buffer[i + nh] = c[i * pitch_c + j];
106       analysis_53_col(nh, buffer + nh, buffer, buffer + hh);
107       for (i = 0; i < nh; i++) c[i * pitch_c + j] = buffer[i];
108     }
109   }
110 }
111 
av1_fdwt8x8_uint8_input_c(uint8_t * input,tran_low_t * output,int stride,int hbd)112 void av1_fdwt8x8_uint8_input_c(uint8_t *input, tran_low_t *output, int stride,
113                                int hbd) {
114   dyadic_analyze_53_uint8_input(4, 8, 8, input, stride, output, 8, 2, hbd);
115 }
116 
av1_haar_ac_sad(tran_low_t * output,int bw,int bh,int stride)117 int av1_haar_ac_sad(tran_low_t *output, int bw, int bh, int stride) {
118   int acsad = 0;
119 
120   for (int r = 0; r < bh; ++r)
121     for (int c = 0; c < bw; ++c) {
122       if (r >= bh / 2 || c >= bw / 2) acsad += abs(output[r * stride + c]);
123     }
124   return acsad;
125 }
126 
av1_dct_ac_sad(tran_low_t * output,int bw,int bh,int stride)127 uint64_t av1_dct_ac_sad(tran_low_t *output, int bw, int bh, int stride) {
128   uint64_t acsad = 0;
129 
130   for (int r = 0; r < bh; ++r)
131     for (int c = 0; c < bw; ++c) {
132       if (r > 0 || c > 0) acsad += abs(output[r * stride + c]);
133     }
134 
135   return acsad;
136 }
137 
av1_variance(uint8_t * input,int bw,int bh,int stride)138 uint32_t av1_variance(uint8_t *input, int bw, int bh, int stride) {
139   int sum = 0;
140   uint32_t sse = 0;
141 
142   for (int r = 0; r < bh; ++r)
143     for (int c = 0; c < bw; ++c) {
144       sum += input[r * stride + c];
145       sse += input[r * stride + c] * input[r * stride + c];
146     }
147   return sse - (uint32_t)(((int64_t)sum * sum) / (bw * bh));
148 }
149 
av1_haar_ac_sad_8x8_uint8_input(uint8_t * input,int stride,int hbd)150 int av1_haar_ac_sad_8x8_uint8_input(uint8_t *input, int stride, int hbd) {
151   tran_low_t output[64];
152 
153   av1_fdwt8x8_uint8_input_c(input, output, stride, hbd);
154   return av1_haar_ac_sad(output, 8, 8, 8);
155 }
156