1 /*
2 * Copyright (c) 2015 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
11 #include <stdlib.h>
12
13 #include "./vpx_config.h"
14 #include "./vpx_dsp_rtcd.h"
15
16 #include "vpx/vpx_integer.h"
17 #include "vpx_ports/mem.h"
18
19 /* Sum the difference between every corresponding element of the buffers. */
sad(const uint8_t * src_ptr,int src_stride,const uint8_t * ref_ptr,int ref_stride,int width,int height)20 static INLINE unsigned int sad(const uint8_t *src_ptr, int src_stride,
21 const uint8_t *ref_ptr, int ref_stride,
22 int width, int height) {
23 int y, x;
24 unsigned int sad = 0;
25
26 for (y = 0; y < height; y++) {
27 for (x = 0; x < width; x++) sad += abs(src_ptr[x] - ref_ptr[x]);
28
29 src_ptr += src_stride;
30 ref_ptr += ref_stride;
31 }
32 return sad;
33 }
34
35 #define sadMxN(m, n) \
36 unsigned int vpx_sad##m##x##n##_c(const uint8_t *src_ptr, int src_stride, \
37 const uint8_t *ref_ptr, int ref_stride) { \
38 return sad(src_ptr, src_stride, ref_ptr, ref_stride, m, n); \
39 } \
40 unsigned int vpx_sad##m##x##n##_avg_c( \
41 const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, \
42 int ref_stride, const uint8_t *second_pred) { \
43 DECLARE_ALIGNED(32, uint8_t, comp_pred[m * n]); \
44 vpx_comp_avg_pred_c(comp_pred, second_pred, m, n, ref_ptr, ref_stride); \
45 return sad(src_ptr, src_stride, comp_pred, m, m, n); \
46 } \
47 unsigned int vpx_sad_skip_##m##x##n##_c( \
48 const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, \
49 int ref_stride) { \
50 return 2 * sad(src_ptr, 2 * src_stride, ref_ptr, 2 * ref_stride, (m), \
51 (n / 2)); \
52 }
53
54 // Compare |src_ptr| to 4 distinct references in |ref_array[4]|
55 #define sadMxNx4D(m, n) \
56 void vpx_sad##m##x##n##x4d_c(const uint8_t *src_ptr, int src_stride, \
57 const uint8_t *const ref_array[4], \
58 int ref_stride, uint32_t sad_array[4]) { \
59 int i; \
60 for (i = 0; i < 4; ++i) \
61 sad_array[i] = \
62 vpx_sad##m##x##n##_c(src_ptr, src_stride, ref_array[i], ref_stride); \
63 } \
64 void vpx_sad_skip_##m##x##n##x4d_c(const uint8_t *src_ptr, int src_stride, \
65 const uint8_t *const ref_array[4], \
66 int ref_stride, uint32_t sad_array[4]) { \
67 int i; \
68 for (i = 0; i < 4; ++i) { \
69 sad_array[i] = 2 * sad(src_ptr, 2 * src_stride, ref_array[i], \
70 2 * ref_stride, (m), (n / 2)); \
71 } \
72 }
73
74 /* clang-format off */
75 // 64x64
76 sadMxN(64, 64)
77 sadMxNx4D(64, 64)
78
79 // 64x32
80 sadMxN(64, 32)
81 sadMxNx4D(64, 32)
82
83 // 32x64
84 sadMxN(32, 64)
85 sadMxNx4D(32, 64)
86
87 // 32x32
88 sadMxN(32, 32)
89 sadMxNx4D(32, 32)
90
91 // 32x16
92 sadMxN(32, 16)
93 sadMxNx4D(32, 16)
94
95 // 16x32
96 sadMxN(16, 32)
97 sadMxNx4D(16, 32)
98
99 // 16x16
100 sadMxN(16, 16)
101 sadMxNx4D(16, 16)
102
103 // 16x8
104 sadMxN(16, 8)
105 sadMxNx4D(16, 8)
106
107 // 8x16
108 sadMxN(8, 16)
109 sadMxNx4D(8, 16)
110
111 // 8x8
112 sadMxN(8, 8)
113 sadMxNx4D(8, 8)
114
115 // 8x4
116 sadMxN(8, 4)
117 sadMxNx4D(8, 4)
118
119 // 4x8
120 sadMxN(4, 8)
121 sadMxNx4D(4, 8)
122
123 // 4x4
124 sadMxN(4, 4)
125 sadMxNx4D(4, 4)
126 /* clang-format on */
127
128 #if CONFIG_VP9_HIGHBITDEPTH
129 static INLINE
highbd_sad(const uint8_t * src8_ptr,int src_stride,const uint8_t * ref8_ptr,int ref_stride,int width,int height)130 unsigned int highbd_sad(const uint8_t *src8_ptr, int src_stride,
131 const uint8_t *ref8_ptr, int ref_stride, int width,
132 int height) {
133 int y, x;
134 unsigned int sad = 0;
135 const uint16_t *src = CONVERT_TO_SHORTPTR(src8_ptr);
136 const uint16_t *ref_ptr = CONVERT_TO_SHORTPTR(ref8_ptr);
137 for (y = 0; y < height; y++) {
138 for (x = 0; x < width; x++) sad += abs(src[x] - ref_ptr[x]);
139
140 src += src_stride;
141 ref_ptr += ref_stride;
142 }
143 return sad;
144 }
145
highbd_sadb(const uint8_t * src8_ptr,int src_stride,const uint16_t * ref_ptr,int ref_stride,int width,int height)146 static INLINE unsigned int highbd_sadb(const uint8_t *src8_ptr, int src_stride,
147 const uint16_t *ref_ptr, int ref_stride,
148 int width, int height) {
149 int y, x;
150 unsigned int sad = 0;
151 const uint16_t *src = CONVERT_TO_SHORTPTR(src8_ptr);
152 for (y = 0; y < height; y++) {
153 for (x = 0; x < width; x++) sad += abs(src[x] - ref_ptr[x]);
154
155 src += src_stride;
156 ref_ptr += ref_stride;
157 }
158 return sad;
159 }
160
161 #define highbd_sadMxN(m, n) \
162 unsigned int vpx_highbd_sad##m##x##n##_c( \
163 const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, \
164 int ref_stride) { \
165 return highbd_sad(src_ptr, src_stride, ref_ptr, ref_stride, m, n); \
166 } \
167 unsigned int vpx_highbd_sad##m##x##n##_avg_c( \
168 const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, \
169 int ref_stride, const uint8_t *second_pred) { \
170 DECLARE_ALIGNED(16, uint16_t, comp_pred[m * n]); \
171 vpx_highbd_comp_avg_pred_c(comp_pred, CONVERT_TO_SHORTPTR(second_pred), m, \
172 n, CONVERT_TO_SHORTPTR(ref_ptr), ref_stride); \
173 return highbd_sadb(src_ptr, src_stride, comp_pred, m, m, n); \
174 } \
175 unsigned int vpx_highbd_sad_skip_##m##x##n##_c( \
176 const uint8_t *src, int src_stride, const uint8_t *ref, \
177 int ref_stride) { \
178 return 2 * \
179 highbd_sad(src, 2 * src_stride, ref, 2 * ref_stride, (m), (n / 2)); \
180 }
181
182 #define highbd_sadMxNx4D(m, n) \
183 void vpx_highbd_sad##m##x##n##x4d_c(const uint8_t *src_ptr, int src_stride, \
184 const uint8_t *const ref_array[4], \
185 int ref_stride, uint32_t sad_array[4]) { \
186 int i; \
187 for (i = 0; i < 4; ++i) { \
188 sad_array[i] = vpx_highbd_sad##m##x##n##_c(src_ptr, src_stride, \
189 ref_array[i], ref_stride); \
190 } \
191 } \
192 void vpx_highbd_sad_skip_##m##x##n##x4d_c( \
193 const uint8_t *src, int src_stride, const uint8_t *const ref_array[4], \
194 int ref_stride, uint32_t sad_array[4]) { \
195 int i; \
196 for (i = 0; i < 4; ++i) { \
197 sad_array[i] = vpx_highbd_sad_skip_##m##x##n##_c( \
198 src, src_stride, ref_array[i], ref_stride); \
199 } \
200 }
201
202 /* clang-format off */
203 // 64x64
204 highbd_sadMxN(64, 64)
205 highbd_sadMxNx4D(64, 64)
206
207 // 64x32
208 highbd_sadMxN(64, 32)
209 highbd_sadMxNx4D(64, 32)
210
211 // 32x64
212 highbd_sadMxN(32, 64)
213 highbd_sadMxNx4D(32, 64)
214
215 // 32x32
216 highbd_sadMxN(32, 32)
217 highbd_sadMxNx4D(32, 32)
218
219 // 32x16
220 highbd_sadMxN(32, 16)
221 highbd_sadMxNx4D(32, 16)
222
223 // 16x32
224 highbd_sadMxN(16, 32)
225 highbd_sadMxNx4D(16, 32)
226
227 // 16x16
228 highbd_sadMxN(16, 16)
229 highbd_sadMxNx4D(16, 16)
230
231 // 16x8
232 highbd_sadMxN(16, 8)
233 highbd_sadMxNx4D(16, 8)
234
235 // 8x16
236 highbd_sadMxN(8, 16)
237 highbd_sadMxNx4D(8, 16)
238
239 // 8x8
240 highbd_sadMxN(8, 8)
241 highbd_sadMxNx4D(8, 8)
242
243 // 8x4
244 highbd_sadMxN(8, 4)
245 highbd_sadMxNx4D(8, 4)
246
247 // 4x8
248 highbd_sadMxN(4, 8)
249 highbd_sadMxNx4D(4, 8)
250
251 // 4x4
252 highbd_sadMxN(4, 4)
253 highbd_sadMxNx4D(4, 4)
254 /* clang-format on */
255
256 #endif // CONFIG_VP9_HIGHBITDEPTH
257