FFmpeg
dsp_init.c
Go to the documentation of this file.
1 /*
2  * VVC filters DSP
3  *
4  * Copyright (C) 2024 Zhao Zhili
5  *
6  * This file is part of FFmpeg.
7  *
8  * FFmpeg is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * FFmpeg is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with FFmpeg; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22 
23 #include "libavutil/cpu.h"
24 #include "libavutil/aarch64/cpu.h"
25 #include "libavcodec/aarch64/h26x/dsp.h"
26 #include "libavcodec/vvc/dsp.h"
27 #include "libavcodec/vvc/dec.h"
28 #include "libavcodec/vvc/ctu.h"
29 
30 #define BDOF_BLOCK_SIZE 16
31 #define BDOF_MIN_BLOCK_SIZE 4
32 
33 void ff_vvc_put_luma_h8_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
34  const int height, const int8_t *hf, const int8_t *vf, const int width);
35 void ff_vvc_put_luma_h16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
36  const int height, const int8_t *hf, const int8_t *vf, const int width);
37 void ff_vvc_put_luma_h_x16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
38  const int height, const int8_t *hf, const int8_t *vf, const int width);
39 void ff_vvc_put_luma_h8_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
40  const int height, const int8_t *hf, const int8_t *vf, const int width);
41 void ff_vvc_put_luma_h16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
42  const int height, const int8_t *hf, const int8_t *vf, const int width);
43 void ff_vvc_put_luma_h_x16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
44  const int height, const int8_t *hf, const int8_t *vf, const int width);
45 
46 void ff_vvc_put_chroma_h8_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
47  const int height, const int8_t *hf, const int8_t *vf, const int width);
48 void ff_vvc_put_chroma_h16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
49  const int height, const int8_t *hf, const int8_t *vf, const int width);
50 void ff_vvc_put_chroma_h_x16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
51  const int height, const int8_t *hf, const int8_t *vf, const int width);
52 void ff_vvc_put_chroma_h8_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
53  const int height, const int8_t *hf, const int8_t *vf, const int width);
54 void ff_vvc_put_chroma_h16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
55  const int height, const int8_t *hf, const int8_t *vf, const int width);
56 void ff_vvc_put_chroma_h_x16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
57  const int height, const int8_t *hf, const int8_t *vf, const int width);
58 
59 void ff_vvc_put_chroma_v4_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
60  const int height, const int8_t *hf, const int8_t *vf, const int width);
61 void ff_vvc_put_chroma_v8_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
62  const int height, const int8_t *hf, const int8_t *vf, const int width);
63 void ff_vvc_put_chroma_v16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
64  const int height, const int8_t *hf, const int8_t *vf, const int width);
65 void ff_vvc_put_chroma_v_x16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
66  const int height, const int8_t *hf, const int8_t *vf, const int width);
67 void ff_vvc_put_chroma_v4_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
68  const int height, const int8_t *hf, const int8_t *vf, const int width);
69 void ff_vvc_put_chroma_v8_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
70  const int height, const int8_t *hf, const int8_t *vf, const int width);
71 void ff_vvc_put_chroma_v16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
72  const int height, const int8_t *hf, const int8_t *vf, const int width);
73 void ff_vvc_put_chroma_v_x16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
74  const int height, const int8_t *hf, const int8_t *vf, const int width);
75 
76 void ff_vvc_put_luma_v4_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
77  const int height, const int8_t *hf, const int8_t *vf, const int width);
78 void ff_vvc_put_luma_v8_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
79  const int height, const int8_t *hf, const int8_t *vf, const int width);
80 void ff_vvc_put_luma_v16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
81  const int height, const int8_t *hf, const int8_t *vf, const int width);
82 void ff_vvc_put_luma_v_x16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
83  const int height, const int8_t *hf, const int8_t *vf, const int width);
84 void ff_vvc_put_luma_v4_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
85  const int height, const int8_t *hf, const int8_t *vf, const int width);
86 void ff_vvc_put_luma_v8_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
87  const int height, const int8_t *hf, const int8_t *vf, const int width);
88 void ff_vvc_put_luma_v16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
89  const int height, const int8_t *hf, const int8_t *vf, const int width);
90 void ff_vvc_put_luma_v_x16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
91  const int height, const int8_t *hf, const int8_t *vf, const int width);
92 
93 void ff_vvc_put_luma_hv8_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
94  const int height, const int8_t *hf, const int8_t *vf, const int width);
95 void ff_vvc_put_luma_hv16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
96  const int height, const int8_t *hf, const int8_t *vf, const int width);
97 void ff_vvc_put_luma_hv_x16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
98  const int height, const int8_t *hf, const int8_t *vf, const int width);
99 void ff_vvc_put_luma_hv8_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
100  const int height, const int8_t *hf, const int8_t *vf, const int width);
101 void ff_vvc_put_luma_hv16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
102  const int height, const int8_t *hf, const int8_t *vf, const int width);
103 void ff_vvc_put_luma_hv_x16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
104  const int height, const int8_t *hf, const int8_t *vf, const int width);
105 
106 void ff_alf_classify_sum_neon(int *sum0, int *sum1, int16_t *grad, uint32_t gshift, uint32_t steps);
107 
108 #define BIT_DEPTH 8
109 #include "alf_template.c"
110 #undef BIT_DEPTH
111 
112 #define BIT_DEPTH 10
113 #include "alf_template.c"
114 #undef BIT_DEPTH
115 
116 #define BIT_DEPTH 12
117 #include "alf_template.c"
118 #undef BIT_DEPTH
119 
120 int ff_vvc_sad_neon(const int16_t *src0, const int16_t *src1, int dx, int dy,
121  const int block_w, const int block_h);
122 
123 void ff_vvc_avg_8_neon(uint8_t *dst, ptrdiff_t dst_stride,
124  const int16_t *src0, const int16_t *src1, int width,
125  int height);
126 void ff_vvc_avg_10_neon(uint8_t *dst, ptrdiff_t dst_stride,
127  const int16_t *src0, const int16_t *src1, int width,
128  int height);
129 void ff_vvc_avg_12_neon(uint8_t *dst, ptrdiff_t dst_stride,
130  const int16_t *src0, const int16_t *src1, int width,
131  int height);
132 
133 void ff_vvc_w_avg_8_neon(uint8_t *_dst, ptrdiff_t _dst_stride,
134  const int16_t *src0, const int16_t *src1,
135  int width, int height,
136  uintptr_t w0_w1, uintptr_t offset_shift);
137 void ff_vvc_w_avg_10_neon(uint8_t *_dst, ptrdiff_t _dst_stride,
138  const int16_t *src0, const int16_t *src1,
139  int width, int height,
140  uintptr_t w0_w1, uintptr_t offset_shift);
141 void ff_vvc_w_avg_12_neon(uint8_t *_dst, ptrdiff_t _dst_stride,
142  const int16_t *src0, const int16_t *src1,
143  int width, int height,
144  uintptr_t w0_w1, uintptr_t offset_shift);
145 /* When passing arguments to functions, Apple platforms diverge from the ARM64
146  * standard ABI for functions that require passing arguments on the stack. To
147  * simplify portability in the assembly function interface, use a different
148  * function signature that doesn't require passing arguments on the stack.
149  */
150 #define W_AVG_FUN(bit_depth) \
151 static void vvc_w_avg_ ## bit_depth(uint8_t *dst, ptrdiff_t dst_stride, \
152  const int16_t *src0, const int16_t *src1, int width, int height, \
153  int denom, int w0, int w1, int o) \
154 { \
155  int shift = denom + FFMAX(3, 15 - bit_depth); \
156  int offset = (o * (1 << (bit_depth - 8)) + 1) * (1 << (shift - 1)); \
157  uintptr_t w0_w1 = ((uintptr_t)w0 << 32) | (uint32_t)w1; \
158  uintptr_t offset_shift = ((uintptr_t)offset << 32) | (uint32_t)shift; \
159  ff_vvc_w_avg_ ## bit_depth ## _neon(dst, dst_stride, src0, src1, width, height, w0_w1, offset_shift); \
160 }
161 
162 W_AVG_FUN(8)
163 W_AVG_FUN(10)
164 W_AVG_FUN(12)
165 
166 #define DMVR_FUN(fn, bd) \
167  void ff_vvc_dmvr_ ## fn ## bd ## _neon(int16_t *dst, \
168  const uint8_t *_src, ptrdiff_t _src_stride, int height, \
169  intptr_t mx, intptr_t my, int width);
170 
171 DMVR_FUN(, 8)
172 DMVR_FUN(, 12)
173 DMVR_FUN(h_, 8)
174 DMVR_FUN(h_, 10)
175 DMVR_FUN(h_, 12)
176 DMVR_FUN(v_, 8)
177 DMVR_FUN(hv_, 8)
178 DMVR_FUN(hv_, 10)
179 DMVR_FUN(hv_, 12)
180 
181 #define APPLY_BDOF_FUNC(bd) \
182  void ff_vvc_apply_bdof_ ## bd ## _neon(uint8_t *_dst, ptrdiff_t _dst_stride, \
183  const int16_t *_src0, const int16_t *_src1, \
184  int block_w, int block_h);
185 
186 APPLY_BDOF_FUNC(8)
187 APPLY_BDOF_FUNC(10)
188 APPLY_BDOF_FUNC(12)
189 
190 void ff_vvc_dsp_init_aarch64(VVCDSPContext *const c, const int bd)
191 {
192  int cpu_flags = av_get_cpu_flags();
193  if (!have_neon(cpu_flags))
194  return;
195 
196  if (bd == 8) {
197  c->inter.put[0][1][0][0] = ff_vvc_put_pel_pixels4_8_neon;
198  c->inter.put[0][2][0][0] = ff_vvc_put_pel_pixels8_8_neon;
199  c->inter.put[0][3][0][0] = ff_vvc_put_pel_pixels16_8_neon;
200  c->inter.put[0][4][0][0] = ff_vvc_put_pel_pixels32_8_neon;
201  c->inter.put[0][5][0][0] = ff_vvc_put_pel_pixels64_8_neon;
202  c->inter.put[0][6][0][0] = ff_vvc_put_pel_pixels128_8_neon;
203 
204  c->inter.put[0][1][0][1] = ff_vvc_put_qpel_h4_8_neon;
205  c->inter.put[0][2][0][1] = ff_vvc_put_qpel_h8_8_neon;
206  c->inter.put[0][3][0][1] = ff_vvc_put_qpel_h16_8_neon;
207  c->inter.put[0][4][0][1] =
208  c->inter.put[0][5][0][1] =
209  c->inter.put[0][6][0][1] = ff_vvc_put_qpel_h32_8_neon;
210 
211  c->inter.put[0][1][1][0] = ff_vvc_put_qpel_v4_8_neon;
212  c->inter.put[0][2][1][0] =
213  c->inter.put[0][3][1][0] =
214  c->inter.put[0][4][1][0] =
215  c->inter.put[0][5][1][0] =
216  c->inter.put[0][6][1][0] = ff_vvc_put_qpel_v8_8_neon;
217 
218  c->inter.put[0][1][1][1] = ff_vvc_put_qpel_hv4_8_neon;
219  c->inter.put[0][2][1][1] = ff_vvc_put_qpel_hv8_8_neon;
220  c->inter.put[0][3][1][1] = ff_vvc_put_qpel_hv16_8_neon;
221  c->inter.put[0][4][1][1] = ff_vvc_put_qpel_hv32_8_neon;
222  c->inter.put[0][5][1][1] = ff_vvc_put_qpel_hv64_8_neon;
223  c->inter.put[0][6][1][1] = ff_vvc_put_qpel_hv128_8_neon;
224 
225  c->inter.put[1][1][0][0] = ff_vvc_put_pel_pixels4_8_neon;
226  c->inter.put[1][2][0][0] = ff_vvc_put_pel_pixels8_8_neon;
227  c->inter.put[1][3][0][0] = ff_vvc_put_pel_pixels16_8_neon;
228  c->inter.put[1][4][0][0] = ff_vvc_put_pel_pixels32_8_neon;
229  c->inter.put[1][5][0][0] = ff_vvc_put_pel_pixels64_8_neon;
230  c->inter.put[1][6][0][0] = ff_vvc_put_pel_pixels128_8_neon;
231 
232  c->inter.put[1][1][0][1] = ff_vvc_put_epel_h4_8_neon;
233  c->inter.put[1][2][0][1] = ff_vvc_put_epel_h8_8_neon;
234  c->inter.put[1][3][0][1] = ff_vvc_put_epel_h16_8_neon;
235  c->inter.put[1][4][0][1] =
236  c->inter.put[1][5][0][1] =
237  c->inter.put[1][6][0][1] = ff_vvc_put_epel_h32_8_neon;
238 
239  c->inter.put[1][1][1][1] = ff_vvc_put_epel_hv4_8_neon;
240  c->inter.put[1][2][1][1] = ff_vvc_put_epel_hv8_8_neon;
241  c->inter.put[1][3][1][1] = ff_vvc_put_epel_hv16_8_neon;
242  c->inter.put[1][4][1][1] = ff_vvc_put_epel_hv32_8_neon;
243  c->inter.put[1][5][1][1] = ff_vvc_put_epel_hv64_8_neon;
244  c->inter.put[1][6][1][1] = ff_vvc_put_epel_hv128_8_neon;
245 
246  c->inter.put_uni[0][1][0][0] = ff_vvc_put_pel_uni_pixels4_8_neon;
247  c->inter.put_uni[0][2][0][0] = ff_vvc_put_pel_uni_pixels8_8_neon;
248  c->inter.put_uni[0][3][0][0] = ff_vvc_put_pel_uni_pixels16_8_neon;
249  c->inter.put_uni[0][4][0][0] = ff_vvc_put_pel_uni_pixels32_8_neon;
250  c->inter.put_uni[0][5][0][0] = ff_vvc_put_pel_uni_pixels64_8_neon;
251  c->inter.put_uni[0][6][0][0] = ff_vvc_put_pel_uni_pixels128_8_neon;
252 
253  c->inter.put_uni[0][1][0][1] = ff_vvc_put_qpel_uni_h4_8_neon;
254  c->inter.put_uni[0][2][0][1] = ff_vvc_put_qpel_uni_h8_8_neon;
255  c->inter.put_uni[0][3][0][1] = ff_vvc_put_qpel_uni_h16_8_neon;
256  c->inter.put_uni[0][4][0][1] =
257  c->inter.put_uni[0][5][0][1] =
258  c->inter.put_uni[0][6][0][1] = ff_vvc_put_qpel_uni_h32_8_neon;
259 
260  c->inter.put_uni_w[0][1][0][0] = ff_vvc_put_pel_uni_w_pixels4_8_neon;
261  c->inter.put_uni_w[0][2][0][0] = ff_vvc_put_pel_uni_w_pixels8_8_neon;
262  c->inter.put_uni_w[0][3][0][0] = ff_vvc_put_pel_uni_w_pixels16_8_neon;
263  c->inter.put_uni_w[0][4][0][0] = ff_vvc_put_pel_uni_w_pixels32_8_neon;
264  c->inter.put_uni_w[0][5][0][0] = ff_vvc_put_pel_uni_w_pixels64_8_neon;
265  c->inter.put_uni_w[0][6][0][0] = ff_vvc_put_pel_uni_w_pixels128_8_neon;
266 
267  c->inter.avg = ff_vvc_avg_8_neon;
268  c->inter.w_avg = vvc_w_avg_8;
269  c->inter.dmvr[0][0] = ff_vvc_dmvr_8_neon;
270  c->inter.dmvr[0][1] = ff_vvc_dmvr_h_8_neon;
271  c->inter.dmvr[1][0] = ff_vvc_dmvr_v_8_neon;
272  c->inter.dmvr[1][1] = ff_vvc_dmvr_hv_8_neon;
273  c->inter.apply_bdof = ff_vvc_apply_bdof_8_neon;
274 
275  c->sao.band_filter[0] = ff_h26x_sao_band_filter_8x8_8_neon;
276  for (int i = 1; i < FF_ARRAY_ELEMS(c->sao.band_filter); i++)
277  c->sao.band_filter[i] = ff_h26x_sao_band_filter_16x16_8_neon;
278  c->sao.edge_filter[0] = ff_vvc_sao_edge_filter_8x8_8_neon;
279  for (int i = 1; i < FF_ARRAY_ELEMS(c->sao.edge_filter); i++)
280  c->sao.edge_filter[i] = ff_vvc_sao_edge_filter_16x16_8_neon;
281  c->alf.filter[LUMA] = alf_filter_luma_8_neon;
282  c->alf.filter[CHROMA] = alf_filter_chroma_8_neon;
283  c->alf.classify = alf_classify_8_neon;
284 
285  if (have_i8mm(cpu_flags)) {
286  c->inter.put[0][1][0][1] = ff_vvc_put_qpel_h4_8_neon_i8mm;
287  c->inter.put[0][2][0][1] = ff_vvc_put_qpel_h8_8_neon_i8mm;
288  c->inter.put[0][3][0][1] = ff_vvc_put_qpel_h16_8_neon_i8mm;
289  c->inter.put[0][4][0][1] = ff_vvc_put_qpel_h32_8_neon_i8mm;
290  c->inter.put[0][5][0][1] = ff_vvc_put_qpel_h64_8_neon_i8mm;
291  c->inter.put[0][6][0][1] = ff_vvc_put_qpel_h128_8_neon_i8mm;
292 
293  c->inter.put[0][1][1][1] = ff_vvc_put_qpel_hv4_8_neon_i8mm;
294  c->inter.put[0][2][1][1] = ff_vvc_put_qpel_hv8_8_neon_i8mm;
295  c->inter.put[0][3][1][1] = ff_vvc_put_qpel_hv16_8_neon_i8mm;
296  c->inter.put[0][4][1][1] = ff_vvc_put_qpel_hv32_8_neon_i8mm;
297  c->inter.put[0][5][1][1] = ff_vvc_put_qpel_hv64_8_neon_i8mm;
298  c->inter.put[0][6][1][1] = ff_vvc_put_qpel_hv128_8_neon_i8mm;
299 
300  c->inter.put[1][1][0][1] = ff_vvc_put_epel_h4_8_neon_i8mm;
301  c->inter.put[1][2][0][1] = ff_vvc_put_epel_h8_8_neon_i8mm;
302  c->inter.put[1][3][0][1] = ff_vvc_put_epel_h16_8_neon_i8mm;
303  c->inter.put[1][4][0][1] = ff_vvc_put_epel_h32_8_neon_i8mm;
304  c->inter.put[1][5][0][1] = ff_vvc_put_epel_h64_8_neon_i8mm;
305  c->inter.put[1][6][0][1] = ff_vvc_put_epel_h128_8_neon_i8mm;
306 
307  c->inter.put[1][1][1][1] = ff_vvc_put_epel_hv4_8_neon_i8mm;
308  c->inter.put[1][2][1][1] = ff_vvc_put_epel_hv8_8_neon_i8mm;
309  c->inter.put[1][3][1][1] = ff_vvc_put_epel_hv16_8_neon_i8mm;
310  c->inter.put[1][4][1][1] = ff_vvc_put_epel_hv32_8_neon_i8mm;
311  c->inter.put[1][5][1][1] = ff_vvc_put_epel_hv64_8_neon_i8mm;
312  c->inter.put[1][6][1][1] = ff_vvc_put_epel_hv128_8_neon_i8mm;
313  }
314  if (have_sme2(cpu_flags) && have_sme_i16i64(cpu_flags)) {
315  c->alf.filter[LUMA] = alf_filter_luma_8_sme2;
316  }
317  } else if (bd == 10) {
318  c->inter.avg = ff_vvc_avg_10_neon;
319  c->inter.w_avg = vvc_w_avg_10;
320  c->inter.dmvr[0][1] = ff_vvc_dmvr_h_10_neon;
321  c->inter.dmvr[1][1] = ff_vvc_dmvr_hv_10_neon;
322  c->inter.apply_bdof = ff_vvc_apply_bdof_10_neon;
323 
324  c->inter.put[1][2][0][1] = ff_vvc_put_chroma_h8_10_neon;
325  c->inter.put[1][3][0][1] = ff_vvc_put_chroma_h16_10_neon;
326  c->inter.put[1][4][0][1] =
327  c->inter.put[1][5][0][1] =
328  c->inter.put[1][6][0][1] = ff_vvc_put_chroma_h_x16_10_neon;
329 
330  c->inter.put[0][2][0][1] = ff_vvc_put_luma_h8_10_neon;
331  c->inter.put[0][3][0][1] = ff_vvc_put_luma_h16_10_neon;
332  c->inter.put[0][4][0][1] =
333  c->inter.put[0][5][0][1] =
334  c->inter.put[0][6][0][1] = ff_vvc_put_luma_h_x16_10_neon;
335 
336  c->inter.put[1][1][1][0] = ff_vvc_put_chroma_v4_10_neon;
337  c->inter.put[1][2][1][0] = ff_vvc_put_chroma_v8_10_neon;
338  c->inter.put[1][3][1][0] = ff_vvc_put_chroma_v16_10_neon;
339  c->inter.put[1][4][1][0] =
340  c->inter.put[1][5][1][0] =
341  c->inter.put[1][6][1][0] = ff_vvc_put_chroma_v_x16_10_neon;
342 
343  c->inter.put[0][1][1][0] = ff_vvc_put_luma_v4_10_neon;
344  c->inter.put[0][2][1][0] = ff_vvc_put_luma_v8_10_neon;
345  c->inter.put[0][3][1][0] = ff_vvc_put_luma_v16_10_neon;
346  c->inter.put[0][4][1][0] =
347  c->inter.put[0][5][1][0] =
348  c->inter.put[0][6][1][0] = ff_vvc_put_luma_v_x16_10_neon;
349 
350  c->inter.put[0][2][1][1] = ff_vvc_put_luma_hv8_10_neon;
351  c->inter.put[0][3][1][1] = ff_vvc_put_luma_hv16_10_neon;
352  c->inter.put[0][4][1][1] =
353  c->inter.put[0][5][1][1] =
354  c->inter.put[0][6][1][1] = ff_vvc_put_luma_hv_x16_10_neon;
355 
356  c->alf.filter[LUMA] = alf_filter_luma_10_neon;
357  c->alf.filter[CHROMA] = alf_filter_chroma_10_neon;
358  c->alf.classify = alf_classify_10_neon;
359  if (have_sme2(cpu_flags) && have_sme_i16i64(cpu_flags)) {
360  c->alf.filter[LUMA] = alf_filter_luma_10_sme2;
361  }
362  } else if (bd == 12) {
363  c->inter.avg = ff_vvc_avg_12_neon;
364  c->inter.w_avg = vvc_w_avg_12;
365  c->inter.dmvr[0][0] = ff_vvc_dmvr_12_neon;
366  c->inter.dmvr[0][1] = ff_vvc_dmvr_h_12_neon;
367  c->inter.dmvr[1][1] = ff_vvc_dmvr_hv_12_neon;
368  c->inter.apply_bdof = ff_vvc_apply_bdof_12_neon;
369 
370  c->inter.put[1][2][0][1] = ff_vvc_put_chroma_h8_12_neon;
371  c->inter.put[1][3][0][1] = ff_vvc_put_chroma_h16_12_neon;
372  c->inter.put[1][4][0][1] =
373  c->inter.put[1][5][0][1] =
374  c->inter.put[1][6][0][1] = ff_vvc_put_chroma_h_x16_12_neon;
375 
376  c->inter.put[0][2][0][1] = ff_vvc_put_luma_h8_12_neon;
377  c->inter.put[0][3][0][1] = ff_vvc_put_luma_h16_12_neon;
378  c->inter.put[0][4][0][1] =
379  c->inter.put[0][5][0][1] =
380  c->inter.put[0][6][0][1] = ff_vvc_put_luma_h_x16_12_neon;
381 
382  c->inter.put[0][2][1][1] = ff_vvc_put_luma_hv8_12_neon;
383  c->inter.put[0][3][1][1] = ff_vvc_put_luma_hv16_12_neon;
384  c->inter.put[0][4][1][1] =
385  c->inter.put[0][5][1][1] =
386  c->inter.put[0][6][1][1] = ff_vvc_put_luma_hv_x16_12_neon;
387 
388  c->inter.put[0][1][1][0] = ff_vvc_put_luma_v4_12_neon;
389  c->inter.put[0][2][1][0] = ff_vvc_put_luma_v8_12_neon;
390  c->inter.put[0][3][1][0] = ff_vvc_put_luma_v16_12_neon;
391  c->inter.put[0][4][1][0] =
392  c->inter.put[0][5][1][0] =
393  c->inter.put[0][6][1][0] = ff_vvc_put_luma_v_x16_12_neon;
394 
395  c->inter.put[1][1][1][0] = ff_vvc_put_chroma_v4_12_neon;
396  c->inter.put[1][2][1][0] = ff_vvc_put_chroma_v8_12_neon;
397  c->inter.put[1][3][1][0] = ff_vvc_put_chroma_v16_12_neon;
398  c->inter.put[1][4][1][0] =
399  c->inter.put[1][5][1][0] =
400  c->inter.put[1][6][1][0] = ff_vvc_put_chroma_v_x16_12_neon;
401 
402  c->alf.filter[LUMA] = alf_filter_luma_12_neon;
403  c->alf.filter[CHROMA] = alf_filter_chroma_12_neon;
404  c->alf.classify = alf_classify_12_neon;
405  if (have_sme2(cpu_flags) && have_sme_i16i64(cpu_flags)) {
406  c->alf.filter[LUMA] = alf_filter_luma_12_sme2;
407  }
408  }
409 
410  c->inter.sad = ff_vvc_sad_neon;
411 }
_dst
uint8_t * _dst
Definition: dsp.h:56
ff_vvc_put_chroma_v4_10_neon
void ff_vvc_put_chroma_v4_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
LUMA
#define LUMA
Definition: filter.c:31
ff_vvc_put_chroma_h16_12_neon
void ff_vvc_put_chroma_h16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
src1
const pixel * src1
Definition: h264pred_template.c:420
ff_vvc_put_qpel_v8_8_neon
void ff_vvc_put_qpel_v8_8_neon(int16_t *dst, const uint8_t *_src, ptrdiff_t _srcstride, int height, const int8_t *hf, const int8_t *vf, int width)
ff_vvc_put_luma_h_x16_12_neon
void ff_vvc_put_luma_h_x16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_put_luma_h16_10_neon
void ff_vvc_put_luma_h16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_put_chroma_h_x16_12_neon
void ff_vvc_put_chroma_h_x16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_put_chroma_h8_10_neon
void ff_vvc_put_chroma_h8_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_put_luma_v8_10_neon
void ff_vvc_put_luma_v8_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_put_luma_hv16_12_neon
void ff_vvc_put_luma_hv16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:109
ff_vvc_dsp_init_aarch64
void ff_vvc_dsp_init_aarch64(VVCDSPContext *const c, const int bd)
Definition: dsp_init.c:190
have_sme2
#define have_sme2(flags)
Definition: cpu.h:37
cpu_flags
static atomic_int cpu_flags
Definition: cpu.c:56
_src
uint8_t ptrdiff_t const uint8_t * _src
Definition: dsp.h:56
ff_vvc_put_luma_h8_10_neon
void ff_vvc_put_luma_h8_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_put_luma_h16_12_neon
void ff_vvc_put_luma_h16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
dsp.h
W_AVG_FUN
#define W_AVG_FUN(bit_depth)
Definition: dsp_init.c:150
ff_vvc_put_chroma_h8_12_neon
void ff_vvc_put_chroma_h8_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_w_avg_10_neon
void ff_vvc_w_avg_10_neon(uint8_t *_dst, ptrdiff_t _dst_stride, const int16_t *src0, const int16_t *src1, int width, int height, uintptr_t w0_w1, uintptr_t offset_shift)
ff_vvc_put_qpel_v4_8_neon
void ff_vvc_put_qpel_v4_8_neon(int16_t *dst, const uint8_t *_src, ptrdiff_t _srcstride, int height, const int8_t *hf, const int8_t *vf, int width)
ff_vvc_put_luma_hv_x16_12_neon
void ff_vvc_put_luma_hv_x16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
FF_ARRAY_ELEMS
#define FF_ARRAY_ELEMS(a)
Definition: sinewin_tablegen.c:29
dsp.h
ff_vvc_put_luma_hv16_10_neon
void ff_vvc_put_luma_hv16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_put_luma_hv8_10_neon
void ff_vvc_put_luma_hv8_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_put_luma_hv_x16_10_neon
void ff_vvc_put_luma_hv_x16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_put_chroma_v_x16_10_neon
void ff_vvc_put_chroma_v_x16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
DMVR_FUN
#define DMVR_FUN(fn, bd)
Definition: dsp_init.c:166
ff_h26x_sao_band_filter_16x16_8_neon
void ff_h26x_sao_band_filter_16x16_8_neon(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)
ff_vvc_put_chroma_v16_10_neon
void ff_vvc_put_chroma_v16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
if
if(ret)
Definition: filter_design.txt:179
ff_vvc_put_chroma_v16_12_neon
void ff_vvc_put_chroma_v16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
grad
static double grad(int hash, double x, double y, double z)
Definition: perlin.c:42
ff_vvc_put_luma_hv8_12_neon
void ff_vvc_put_luma_hv8_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_sao_edge_filter_8x8_8_neon
void ff_vvc_sao_edge_filter_8x8_8_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride_dst, const int16_t *sao_offset_val, int eo, int width, int height)
ff_vvc_put_chroma_v8_10_neon
void ff_vvc_put_chroma_v8_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
c
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Definition: undefined.txt:32
have_i8mm
#define have_i8mm(flags)
Definition: cpu.h:32
ff_vvc_put_chroma_v_x16_12_neon
void ff_vvc_put_chroma_v_x16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
height
#define height
Definition: dsp.h:89
ff_vvc_put_luma_v_x16_10_neon
void ff_vvc_put_luma_v_x16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
dst
uint8_t ptrdiff_t const uint8_t ptrdiff_t int intptr_t intptr_t int int16_t * dst
Definition: dsp.h:87
ff_vvc_put_chroma_h16_10_neon
void ff_vvc_put_chroma_h16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_put_luma_h_x16_10_neon
void ff_vvc_put_luma_h_x16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
cpu.h
i
#define i(width, name, range_min, range_max)
Definition: cbs_h264.c:63
for
for(k=2;k<=8;++k)
Definition: h264pred_template.c:424
ff_vvc_put_chroma_v8_12_neon
void ff_vvc_put_chroma_v8_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_avg_8_neon
void ff_vvc_avg_8_neon(uint8_t *dst, ptrdiff_t dst_stride, const int16_t *src0, const int16_t *src1, int width, int height)
have_neon
#define have_neon(flags)
Definition: cpu.h:26
ff_h26x_sao_band_filter_8x8_8_neon
void ff_h26x_sao_band_filter_8x8_8_neon(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)
ff_vvc_put_luma_v16_10_neon
void ff_vvc_put_luma_v16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_avg_10_neon
void ff_vvc_avg_10_neon(uint8_t *dst, ptrdiff_t dst_stride, const int16_t *src0, const int16_t *src1, int width, int height)
ff_alf_classify_sum_neon
void ff_alf_classify_sum_neon(int *sum0, int *sum1, int16_t *grad, uint32_t gshift, uint32_t steps)
ff_vvc_put_chroma_v4_12_neon
void ff_vvc_put_chroma_v4_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
hf
uint8_t ptrdiff_t const uint8_t ptrdiff_t int const int8_t * hf
Definition: dsp.h:262
ff_vvc_put_luma_v4_10_neon
void ff_vvc_put_luma_v4_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ff_vvc_put_luma_v8_12_neon
void ff_vvc_put_luma_v8_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
CHROMA
@ CHROMA
Definition: vf_waveform.c:49
ff_vvc_put_luma_v4_12_neon
void ff_vvc_put_luma_v4_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
steps
static const int16_t steps[16]
Definition: misc4.c:30
ff_vvc_w_avg_8_neon
void ff_vvc_w_avg_8_neon(uint8_t *_dst, ptrdiff_t _dst_stride, const int16_t *src0, const int16_t *src1, int width, int height, uintptr_t w0_w1, uintptr_t offset_shift)
ff_vvc_w_avg_12_neon
void ff_vvc_w_avg_12_neon(uint8_t *_dst, ptrdiff_t _dst_stride, const int16_t *src0, const int16_t *src1, int width, int height, uintptr_t w0_w1, uintptr_t offset_shift)
src0
const pixel *const src0
Definition: h264pred_template.c:419
ff_vvc_avg_12_neon
void ff_vvc_avg_12_neon(uint8_t *dst, ptrdiff_t dst_stride, const int16_t *src0, const int16_t *src1, int width, int height)
ff_vvc_sao_edge_filter_16x16_8_neon
void ff_vvc_sao_edge_filter_16x16_8_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride_dst, const int16_t *sao_offset_val, int eo, int width, int height)
alf_template.c
ff_vvc_put_chroma_h_x16_10_neon
void ff_vvc_put_chroma_h_x16_10_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
have_sme_i16i64
#define have_sme_i16i64(flags)
Definition: cpu.h:36
ff_vvc_sad_neon
int ff_vvc_sad_neon(const int16_t *src0, const int16_t *src1, int dx, int dy, const int block_w, const int block_h)
APPLY_BDOF_FUNC
#define APPLY_BDOF_FUNC(bd)
Definition: dsp_init.c:181
ff_vvc_put_luma_v16_12_neon
void ff_vvc_put_luma_v16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
ctu.h
ff_vvc_put_luma_h8_12_neon
void ff_vvc_put_luma_h8_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
width
#define width
Definition: dsp.h:89
cpu.h
vf
uint8_t ptrdiff_t const uint8_t ptrdiff_t int const int8_t const int8_t * vf
Definition: dsp.h:262
ff_vvc_put_luma_v_x16_12_neon
void ff_vvc_put_luma_v_x16_12_neon(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int8_t *hf, const int8_t *vf, const int width)
dec.h
VVCDSPContext
Definition: dsp.h:170
Generated on Mon Apr 20 2026 19:21:59 for FFmpeg by   doxygen 1.8.17