FFmpeg
Loading...
Searching...
No Matches
sao.c
Go to the documentation of this file.
1/*
2 * Copyright (c) 2025 Zhao Zhili
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21#include "sao.h"
22
23#include <wasm_simd128.h>
24
25#include "libavcodec/defs.h"
26
27#define HEVC_MAX_PB_SIZE 64
28
29void ff_hevc_sao_band_filter_8x8_8_simd128(uint8_t *dst, const uint8_t *src,
30 ptrdiff_t stride_dst,
31 ptrdiff_t stride_src,
32 const int16_t *sao_offset_val,
33 int sao_left_class, int width,
34 int height)
35{
36 int8_t offset_table[32] = {0};
37 v128_t offset_low, offset_high;
38
39 for (int k = 0; k < 4; k++)
40 offset_table[(k + sao_left_class) & 31] = (int8_t)sao_offset_val[k + 1];
41
42 offset_low = wasm_v128_load(offset_table);
43 offset_high = wasm_v128_load(&offset_table[16]);
44
45 for (int y = height; y > 0; y -= 2) {
46 v128_t src_v, src_high;
47 v128_t v0, v1;
48
49 src_v = wasm_v128_load64_zero(src);
50 src += stride_src;
51 src_v = wasm_v128_load64_lane(src, src_v, 1);
52 src += stride_src;
53
54 v0 = wasm_u8x16_shr(src_v, 3);
55 v1 = wasm_i8x16_sub(v0, wasm_i8x16_const_splat(16));
56 v0 = wasm_i8x16_swizzle(offset_low, v0);
57 v1 = wasm_i8x16_swizzle(offset_high, v1);
58 v0 = wasm_v128_or(v0, v1);
59 src_high = wasm_u16x8_extend_high_u8x16(src_v);
60 v1 = wasm_i16x8_extend_high_i8x16(v0);
61 src_v = wasm_u16x8_extend_low_u8x16(src_v);
62 v0 = wasm_i16x8_extend_low_i8x16(v0);
63
64 v0 = wasm_i16x8_add_sat(src_v, v0);
65 v1 = wasm_i16x8_add_sat(src_high, v1);
66 v0 = wasm_u8x16_narrow_i16x8(v0, v1);
67
68 wasm_v128_store64_lane(dst, v0, 0);
69 dst += stride_dst;
70 wasm_v128_store64_lane(dst, v0, 1);
71 dst += stride_dst;
72 }
73}
74
75void ff_hevc_sao_band_filter_16x16_8_simd128(uint8_t *dst, const uint8_t *src,
76 ptrdiff_t stride_dst,
77 ptrdiff_t stride_src,
78 const int16_t *sao_offset_val,
79 int sao_left_class, int width,
80 int height)
81{
82 int8_t offset_table[32] = {0};
83 v128_t offset_low, offset_high;
84
85 for (int k = 0; k < 4; k++)
86 offset_table[(k + sao_left_class) & 31] = (int8_t)sao_offset_val[k + 1];
87
88 offset_low = wasm_v128_load(offset_table);
89 offset_high = wasm_v128_load(&offset_table[16]);
90
91 for (int y = height; y > 0; y--) {
92 for (int x = 0; x < width; x += 16) {
93 v128_t src_v, src_high;
94 v128_t v0, v1;
95
96 src_v = wasm_v128_load(&src[x]);
97
98 v0 = wasm_u8x16_shr(src_v, 3);
99 v1 = wasm_i8x16_sub(v0, wasm_i8x16_const_splat(16));
100 v0 = wasm_i8x16_swizzle(offset_low, v0);
101 v1 = wasm_i8x16_swizzle(offset_high, v1);
102 v0 = wasm_v128_or(v0, v1);
103 src_high = wasm_u16x8_extend_high_u8x16(src_v);
104 v1 = wasm_i16x8_extend_high_i8x16(v0);
105 src_v = wasm_u16x8_extend_low_u8x16(src_v);
106 v0 = wasm_i16x8_extend_low_i8x16(v0);
107
108 v0 = wasm_i16x8_add_sat(src_v, v0);
109 v1 = wasm_i16x8_add_sat(src_high, v1);
110 v0 = wasm_u8x16_narrow_i16x8(v0, v1);
111 wasm_v128_store(&dst[x], v0);
112 }
113
114 dst += stride_dst;
115 src += stride_src;
116 }
117}
118
119void ff_hevc_sao_edge_filter_8x8_8_simd128(uint8_t *dst, const uint8_t *src,
120 ptrdiff_t stride_dst,
121 const int16_t *sao_offset_val,
122 int eo, int width, int height)
123{
124 static const int8_t pos[4][2][2] = {
125 { { -1, 0 }, { 1, 0 } }, // horizontal
126 { { 0, -1 }, { 0, 1 } }, // vertical
127 { { -1, -1 }, { 1, 1 } }, // 45 degree
128 { { 1, -1 }, { -1, 1 } }, // 135 degree
129 };
130 int a_stride, b_stride;
131 ptrdiff_t stride_src = (2 * HEVC_MAX_PB_SIZE + AV_INPUT_BUFFER_PADDING_SIZE);
132 const v128_t edge_idx = wasm_u8x16_make(1, 2, 0, 3,
133 4, 0, 0, 0,
134 0, 0, 0, 0,
135 0, 0, 0, 0);
136 v128_t sao_offset = wasm_v128_load(sao_offset_val);
137 v128_t one = wasm_i8x16_const_splat(1);
138 v128_t two = wasm_i8x16_const_splat(2);
139
140 a_stride = pos[eo][0][0] + pos[eo][0][1] * stride_src;
141 b_stride = pos[eo][1][0] + pos[eo][1][1] * stride_src;
142 for (int y = height; y > 0; y -= 2) {
143 v128_t v0, v1, v2;
144 v128_t diff0, diff1;
145
146 v0 = wasm_v128_load64_zero(src);
147 v1 = wasm_v128_load64_zero(src + a_stride);
148 v2 = wasm_v128_load64_zero(src + b_stride);
149 src += stride_src;
150 v0 = wasm_v128_load64_lane(src, v0, 1);
151 v1 = wasm_v128_load64_lane(src + a_stride, v1, 1);
152 v2 = wasm_v128_load64_lane(src + b_stride, v2, 1);
153 src += stride_src;
154
155 diff0 = wasm_u8x16_gt(v0, v1);
156 v1 = wasm_u8x16_lt(v0, v1);
157 diff0 = wasm_i8x16_sub(v1, diff0);
158
159 diff1 = wasm_u8x16_gt(v0, v2);
160 v2 = wasm_u8x16_lt(v0, v2);
161 diff1 = wasm_i8x16_sub(v2, diff1);
162
163 v1 = wasm_i8x16_add(diff0, two);
164 v1 = wasm_i8x16_add(v1, diff1);
165
166 v2 = wasm_i8x16_swizzle(edge_idx, v1); // offset_val
167 v1 = wasm_i8x16_shl(v2, 1); // Access int16_t
168 v2 = wasm_i8x16_add(v1, one); // Access upper half of int16_t
169 diff0 = wasm_i8x16_shuffle(v1, v2, 0, 16, 1, 17, 2, 18, 3, 19, 4,
170 20, 5, 21, 6, 22, 7, 23);
171 diff1 = wasm_i8x16_shuffle(v1, v2, 8, 24, 9, 25, 10, 26, 11, 27,
172 12, 28, 13, 29, 14, 30, 15, 31);
173 v1 = wasm_u16x8_extend_high_u8x16(v0);
174 v0 = wasm_u16x8_extend_low_u8x16(v0);
175 diff0 = wasm_i8x16_swizzle(sao_offset, diff0);
176 diff1 = wasm_i8x16_swizzle(sao_offset, diff1);
177
178 v0 = wasm_i16x8_add_sat(v0, diff0);
179 v1 = wasm_i16x8_add_sat(v1, diff1);
180 v0 = wasm_u8x16_narrow_i16x8(v0, v1);
181
182 wasm_v128_store64_lane(dst, v0, 0);
183 dst += stride_dst;
184 wasm_v128_store64_lane(dst, v0, 1);
185 dst += stride_dst;
186 }
187}
188
189void ff_hevc_sao_edge_filter_16x16_8_simd128(uint8_t *dst, const uint8_t *src,
190 ptrdiff_t stride_dst,
191 const int16_t *sao_offset_val,
192 int eo, int width, int height)
193{
194 static const int8_t pos[4][2][2] = {
195 { { -1, 0 }, { 1, 0 } }, // horizontal
196 { { 0, -1 }, { 0, 1 } }, // vertical
197 { { -1, -1 }, { 1, 1 } }, // 45 degree
198 { { 1, -1 }, { -1, 1 } }, // 135 degree
199 };
200 int a_stride, b_stride;
201 ptrdiff_t stride_src = (2 * HEVC_MAX_PB_SIZE + AV_INPUT_BUFFER_PADDING_SIZE);
202 const v128_t edge_idx = wasm_u8x16_make(1, 2, 0, 3,
203 4, 0, 0, 0,
204 0, 0, 0, 0,
205 0, 0, 0, 0);
206 v128_t sao_offset = wasm_v128_load(sao_offset_val);
207 v128_t one = wasm_i8x16_const_splat(1);
208 v128_t two = wasm_i8x16_const_splat(2);
209
210 a_stride = pos[eo][0][0] + pos[eo][0][1] * stride_src;
211 b_stride = pos[eo][1][0] + pos[eo][1][1] * stride_src;
212 for (int y = height; y > 0; y--) {
213 for (int x = 0; x < width; x += 16) {
214 v128_t v0, v1, v2;
215 v128_t diff0, diff1;
216
217 v0 = wasm_v128_load(&src[x]);
218 v1 = wasm_v128_load(&src[x + a_stride]);
219 v2 = wasm_v128_load(&src[x + b_stride]);
220
221 diff0 = wasm_u8x16_gt(v0, v1);
222 v1 = wasm_u8x16_lt(v0, v1);
223 diff0 = wasm_i8x16_sub(v1, diff0);
224
225 diff1 = wasm_u8x16_gt(v0, v2);
226 v2 = wasm_u8x16_lt(v0, v2);
227 diff1 = wasm_i8x16_sub(v2, diff1);
228
229 v1 = wasm_i8x16_add(diff0, two);
230 v1 = wasm_i8x16_add(v1, diff1);
231
232 v2 = wasm_i8x16_swizzle(edge_idx, v1); // offset_val
233 v1 = wasm_i8x16_shl(v2, 1); // Access int16_t
234 v2 = wasm_i8x16_add(v1, one); // Access upper half of int16_t
235 diff0 = wasm_i8x16_shuffle(v1, v2, 0, 16, 1, 17, 2, 18, 3, 19, 4,
236 20, 5, 21, 6, 22, 7, 23);
237 diff1 = wasm_i8x16_shuffle(v1, v2, 8, 24, 9, 25, 10, 26, 11, 27,
238 12, 28, 13, 29, 14, 30, 15, 31);
239 v1 = wasm_u16x8_extend_high_u8x16(v0);
240 v0 = wasm_u16x8_extend_low_u8x16(v0);
241 diff0 = wasm_i8x16_swizzle(sao_offset, diff0);
242 diff1 = wasm_i8x16_swizzle(sao_offset, diff1);
243
244 v0 = wasm_i16x8_add_sat(v0, diff0);
245 v1 = wasm_i16x8_add_sat(v1, diff1);
246 v0 = wasm_u8x16_narrow_i16x8(v0, v1);
247 wasm_v128_store(&dst[x], v0);
248 }
249
250 src += stride_src;
251 dst += stride_dst;
252 }
253}
dst
uint8_t ptrdiff_t const uint8_t ptrdiff_t int intptr_t intptr_t int int16_t * dst
Definition dsp.h:87
defs.h
Misc types and constants that do not belong anywhere else.
offset_table
static const uint8_t offset_table[]
Definition escape130.c:42
AV_INPUT_BUFFER_PADDING_SIZE
#define AV_INPUT_BUFFER_PADDING_SIZE
Required number of additionally allocated bytes at the end of the input bitstream for decoding.
Definition defs.h:40
ff_hevc_sao_edge_filter_16x16_8_simd128
void ff_hevc_sao_edge_filter_16x16_8_simd128(uint8_t *dst, const uint8_t *src, ptrdiff_t stride_dst, const int16_t *sao_offset_val, int eo, int width, int height)
Definition sao.c:189
HEVC_MAX_PB_SIZE
#define HEVC_MAX_PB_SIZE
Definition sao.c:27
ff_hevc_sao_edge_filter_8x8_8_simd128
void ff_hevc_sao_edge_filter_8x8_8_simd128(uint8_t *dst, const uint8_t *src, ptrdiff_t stride_dst, const int16_t *sao_offset_val, int eo, int width, int height)
Definition sao.c:119
ff_hevc_sao_band_filter_16x16_8_simd128
void ff_hevc_sao_band_filter_16x16_8_simd128(uint8_t *dst, const uint8_t *src, ptrdiff_t stride_dst, ptrdiff_t stride_src, const int16_t *sao_offset_val, int sao_left_class, int width, int height)
Definition sao.c:75
ff_hevc_sao_band_filter_8x8_8_simd128
void ff_hevc_sao_band_filter_8x8_8_simd128(uint8_t *dst, const uint8_t *src, ptrdiff_t stride_dst, ptrdiff_t stride_src, const int16_t *sao_offset_val, int sao_left_class, int width, int height)
Definition sao.c:29
pos
unsigned int pos
Definition spdifenc.c:431
src
#define src
Definition vp8dsp.c:248
height
#define height
Definition dsp.h:89
width
#define width
Definition dsp.h:89
Generated on Sun Aug 16 2026 20:20:54 for FFmpeg by doxygen 1.13.2