FFmpeg
vvcdsp_init.c
Go to the documentation of this file.
1 /*
2  * VVC DSP init for x86
3  *
4  * Copyright (C) 2022-2024 Nuo Mi
5  * Copyright (c) 2023-2024 Wu Jianhua
6  *
7  * This file is part of FFmpeg.
8  *
9  * FFmpeg is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU Lesser General Public
11  * License as published by the Free Software Foundation; either
12  * version 2.1 of the License, or (at your option) any later version.
13  *
14  * FFmpeg is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17  * Lesser General Public License for more details.
18  *
19  * You should have received a copy of the GNU Lesser General Public
20  * License along with FFmpeg; if not, write to the Free Software
21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22  */
23 
24 #include "config.h"
25 
26 #include "libavutil/cpu.h"
27 #include "libavutil/x86/cpu.h"
28 #include "libavcodec/vvc/dec.h"
29 #include "libavcodec/vvc/ctu.h"
30 #include "libavcodec/vvc/dsp.h"
31 #include "libavcodec/x86/h26x/h2656dsp.h"
32 
33 #define PUT_PROTOTYPE(name, depth, opt) \
34 void ff_vvc_put_ ## name ## _ ## depth ## _##opt(int16_t *dst, const uint8_t *src, ptrdiff_t srcstride, int height, const int8_t *hf, const int8_t *vf, int width);
35 
36 #define PUT_PROTOTYPES(name, bitd, opt) \
37  PUT_PROTOTYPE(name##2, bitd, opt) \
38  PUT_PROTOTYPE(name##4, bitd, opt) \
39  PUT_PROTOTYPE(name##8, bitd, opt) \
40  PUT_PROTOTYPE(name##12, bitd, opt) \
41  PUT_PROTOTYPE(name##16, bitd, opt) \
42  PUT_PROTOTYPE(name##24, bitd, opt) \
43  PUT_PROTOTYPE(name##32, bitd, opt) \
44  PUT_PROTOTYPE(name##48, bitd, opt) \
45  PUT_PROTOTYPE(name##64, bitd, opt) \
46  PUT_PROTOTYPE(name##128, bitd, opt)
47 
48 #define PUT_BPC_PROTOTYPES(name, opt) \
49  PUT_PROTOTYPES(name, 8, opt) \
50  PUT_PROTOTYPES(name, 10, opt) \
51  PUT_PROTOTYPES(name, 12, opt)
52 
53 #define PUT_TAP_PROTOTYPES(n, opt) \
54  PUT_BPC_PROTOTYPES(n##tap_h, opt) \
55  PUT_BPC_PROTOTYPES(n##tap_v, opt) \
56  PUT_BPC_PROTOTYPES(n##tap_hv, opt)
57 
58 PUT_BPC_PROTOTYPES(pixels, sse4)
59 PUT_BPC_PROTOTYPES(pixels, avx2)
60 
61 PUT_TAP_PROTOTYPES(4, sse4)
62 PUT_TAP_PROTOTYPES(8, sse4)
63 PUT_TAP_PROTOTYPES(4, avx2)
64 PUT_TAP_PROTOTYPES(8, avx2)
65 
66 #define bf(fn, bd, opt) fn##_##bd##_##opt
67 #define BF(fn, bpc, opt) fn##_##bpc##bpc_##opt
68 
69 #define AVG_BPC_PROTOTYPES(bpc, opt) \
70 void BF(ff_vvc_avg, bpc, opt)(uint8_t *dst, ptrdiff_t dst_stride, \
71  const int16_t *src0, const int16_t *src1, intptr_t width, intptr_t height, intptr_t pixel_max); \
72 void BF(ff_vvc_w_avg, bpc, opt)(uint8_t *dst, ptrdiff_t dst_stride, \
73  const int16_t *src0, const int16_t *src1, intptr_t width, intptr_t height, \
74  intptr_t denom, intptr_t w0, intptr_t w1, intptr_t o0, intptr_t o1, intptr_t pixel_max);
75 
76 #define AVG_PROTOTYPES(bd, opt) \
77 void bf(ff_vvc_avg, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride, \
78  const int16_t *src0, const int16_t *src1, int width, int height); \
79 void bf(ff_vvc_w_avg, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride, \
80  const int16_t *src0, const int16_t *src1, int width, int height, \
81  int denom, int w0, int w1, int o0, int o1);
82 
83 AVG_BPC_PROTOTYPES( 8, avx2)
84 AVG_BPC_PROTOTYPES(16, avx2)
85 
86 AVG_PROTOTYPES( 8, avx2)
87 AVG_PROTOTYPES(10, avx2)
88 AVG_PROTOTYPES(12, avx2)
89 
90 
91 #define DMVR_PROTOTYPES(bd, opt) \
92 void ff_vvc_dmvr_##bd##_##opt(int16_t *dst, const uint8_t *src, ptrdiff_t src_stride, \
93  int height, intptr_t mx, intptr_t my, int width); \
94 void ff_vvc_dmvr_h_##bd##_##opt(int16_t *dst, const uint8_t *src, ptrdiff_t src_stride, \
95  int height, intptr_t mx, intptr_t my, int width); \
96 void ff_vvc_dmvr_v_##bd##_##opt(int16_t *dst, const uint8_t *src, ptrdiff_t src_stride, \
97  int height, intptr_t mx, intptr_t my, int width); \
98 void ff_vvc_dmvr_hv_##bd##_##opt(int16_t *dst, const uint8_t *src, ptrdiff_t src_stride, \
99  int height, intptr_t mx, intptr_t my, int width); \
100 
101 DMVR_PROTOTYPES( 8, avx2)
102 DMVR_PROTOTYPES(10, avx2)
103 DMVR_PROTOTYPES(12, avx2)
104 
105 #define OF_PROTOTYPES(bd, opt) \
106 void ff_vvc_apply_bdof_##bd##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \
107  const int16_t *src0, const int16_t *src1, int w, int h); \
108 
109 OF_PROTOTYPES( 8, avx2)
110 OF_PROTOTYPES(10, avx2)
111 OF_PROTOTYPES(12, avx2)
112 
113 #if ARCH_X86_64 && HAVE_AVX2_EXTERNAL
114 void ff_vvc_apply_bdof_avx2(uint8_t *dst, ptrdiff_t dst_stride, \
115  const int16_t *src0, const int16_t *src1, int w, int h, int pixel_max); \
116 
117 #define OF_FUNC(bd, opt) \
118 void ff_vvc_apply_bdof_##bd##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \
119  const int16_t *src0, const int16_t *src1, int w, int h) \
120 { \
121  ff_vvc_apply_bdof##_##opt(dst, dst_stride, src0, src1, w, h, (1 << bd) - 1); \
122 } \
123 
124 OF_FUNC( 8, avx2)
125 OF_FUNC(10, avx2)
126 OF_FUNC(12, avx2)
127 #endif
128 
129 #define ALF_BPC_PROTOTYPES(bpc, opt) \
130 void BF(ff_vvc_alf_filter_luma, bpc, opt)(uint8_t *dst, ptrdiff_t dst_stride, \
131  const uint8_t *src, ptrdiff_t src_stride, ptrdiff_t width, ptrdiff_t height, \
132  const int16_t *filter, const int16_t *clip, ptrdiff_t stride, ptrdiff_t vb_pos, ptrdiff_t pixel_max); \
133 void BF(ff_vvc_alf_filter_chroma, bpc, opt)(uint8_t *dst, ptrdiff_t dst_stride, \
134  const uint8_t *src, ptrdiff_t src_stride, ptrdiff_t width, ptrdiff_t height, \
135  const int16_t *filter, const int16_t *clip, ptrdiff_t stride, ptrdiff_t vb_pos, ptrdiff_t pixel_max); \
136 void BF(ff_vvc_alf_classify_grad, bpc, opt)(int *gradient_sum, \
137  const uint8_t *src, ptrdiff_t src_stride, intptr_t width, intptr_t height, intptr_t vb_pos); \
138 void BF(ff_vvc_alf_classify, bpc, opt)(int *class_idx, int *transpose_idx, const int *gradient_sum, \
139  intptr_t width, intptr_t height, intptr_t vb_pos, intptr_t bit_depth); \
140 
141 #define ALF_PROTOTYPES(bpc, bd, opt) \
142 void bf(ff_vvc_alf_filter_luma, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *src, ptrdiff_t src_stride, \
143  int width, int height, const int16_t *filter, const int16_t *clip, const int vb_pos); \
144 void bf(ff_vvc_alf_filter_chroma, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *src, ptrdiff_t src_stride, \
145  int width, int height, const int16_t *filter, const int16_t *clip, const int vb_pos); \
146 void bf(ff_vvc_alf_classify, bd, opt)(int *class_idx, int *transpose_idx, \
147  const uint8_t *src, ptrdiff_t src_stride, int width, int height, int vb_pos, int *gradient_tmp); \
148 
149 ALF_BPC_PROTOTYPES(8, avx2)
150 ALF_BPC_PROTOTYPES(16, avx2)
151 
152 ALF_PROTOTYPES(8, 8, avx2)
153 ALF_PROTOTYPES(16, 10, avx2)
154 ALF_PROTOTYPES(16, 12, avx2)
155 
156 #if ARCH_X86_64
157 #if HAVE_SSE4_EXTERNAL
158 #define FW_PUT(name, depth, opt) \
159 void ff_vvc_put_ ## name ## _ ## depth ## _##opt(int16_t *dst, const uint8_t *src, ptrdiff_t srcstride, \
160  int height, const int8_t *hf, const int8_t *vf, int width) \
161 { \
162  ff_h2656_put_## name ## _ ## depth ## _##opt(dst, 2 * MAX_PB_SIZE, src, srcstride, height, hf, vf, width); \
163 }
164 
165 #define FW_PUT_TAP(fname, bitd, opt ) \
166  FW_PUT(fname##4, bitd, opt ) \
167  FW_PUT(fname##8, bitd, opt ) \
168  FW_PUT(fname##16, bitd, opt ) \
169  FW_PUT(fname##32, bitd, opt ) \
170  FW_PUT(fname##64, bitd, opt ) \
171  FW_PUT(fname##128, bitd, opt ) \
172 
173 #define FW_PUT_4TAP(fname, bitd, opt) \
174  FW_PUT(fname ## 2, bitd, opt) \
175  FW_PUT_TAP(fname, bitd, opt)
176 
177 #define FW_PUT_4TAP_SSE4(bitd) \
178  FW_PUT_4TAP(pixels, bitd, sse4) \
179  FW_PUT_4TAP(4tap_h, bitd, sse4) \
180  FW_PUT_4TAP(4tap_v, bitd, sse4) \
181  FW_PUT_4TAP(4tap_hv, bitd, sse4)
182 
183 #define FW_PUT_8TAP_SSE4(bitd) \
184  FW_PUT_TAP(8tap_h, bitd, sse4) \
185  FW_PUT_TAP(8tap_v, bitd, sse4) \
186  FW_PUT_TAP(8tap_hv, bitd, sse4)
187 
188 #define FW_PUT_SSE4(bitd) \
189  FW_PUT_4TAP_SSE4(bitd) \
190  FW_PUT_8TAP_SSE4(bitd)
191 
192 FW_PUT_SSE4( 8)
193 FW_PUT_SSE4(10)
194 FW_PUT_SSE4(12)
195 #endif
196 
197 #if HAVE_AVX2_EXTERNAL
198 #define FW_PUT_TAP_AVX2(n, bitd) \
199  FW_PUT(n ## tap_h32, bitd, avx2) \
200  FW_PUT(n ## tap_h64, bitd, avx2) \
201  FW_PUT(n ## tap_h128, bitd, avx2) \
202  FW_PUT(n ## tap_v32, bitd, avx2) \
203  FW_PUT(n ## tap_v64, bitd, avx2) \
204  FW_PUT(n ## tap_v128, bitd, avx2)
205 
206 #define FW_PUT_AVX2(bitd) \
207  FW_PUT(pixels32, bitd, avx2) \
208  FW_PUT(pixels64, bitd, avx2) \
209  FW_PUT(pixels128, bitd, avx2) \
210  FW_PUT_TAP_AVX2(4, bitd) \
211  FW_PUT_TAP_AVX2(8, bitd) \
212 
213 FW_PUT_AVX2( 8)
214 FW_PUT_AVX2(10)
215 FW_PUT_AVX2(12)
216 
217 #define FW_PUT_TAP_16BPC_AVX2(n, bitd) \
218  FW_PUT(n ## tap_h16, bitd, avx2) \
219  FW_PUT(n ## tap_v16, bitd, avx2) \
220  FW_PUT(n ## tap_hv16, bitd, avx2) \
221  FW_PUT(n ## tap_hv32, bitd, avx2) \
222  FW_PUT(n ## tap_hv64, bitd, avx2) \
223  FW_PUT(n ## tap_hv128, bitd, avx2)
224 
225 #define FW_PUT_16BPC_AVX2(bitd) \
226  FW_PUT(pixels16, bitd, avx2) \
227  FW_PUT_TAP_16BPC_AVX2(4, bitd) \
228  FW_PUT_TAP_16BPC_AVX2(8, bitd)
229 
230 FW_PUT_16BPC_AVX2(10)
231 FW_PUT_16BPC_AVX2(12)
232 
233 #define AVG_FUNCS(bpc, bd, opt) \
234 void bf(ff_vvc_avg, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride, \
235  const int16_t *src0, const int16_t *src1, int width, int height) \
236 { \
237  BF(ff_vvc_avg, bpc, opt)(dst, dst_stride, src0, src1, width, height, (1 << bd) - 1); \
238 } \
239 void bf(ff_vvc_w_avg, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride, \
240  const int16_t *src0, const int16_t *src1, int width, int height, \
241  int denom, int w0, int w1, int o0, int o1) \
242 { \
243  BF(ff_vvc_w_avg, bpc, opt)(dst, dst_stride, src0, src1, width, height, \
244  denom, w0, w1, o0, o1, (1 << bd) - 1); \
245 }
246 
247 AVG_FUNCS(8, 8, avx2)
248 AVG_FUNCS(16, 10, avx2)
249 AVG_FUNCS(16, 12, avx2)
250 
251 #define ALF_FUNCS(bpc, bd, opt) \
252 void bf(ff_vvc_alf_filter_luma, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *src, ptrdiff_t src_stride, \
253  int width, int height, const int16_t *filter, const int16_t *clip, const int vb_pos) \
254 { \
255  const int param_stride = (width >> 2) * ALF_NUM_COEFF_LUMA; \
256  BF(ff_vvc_alf_filter_luma, bpc, opt)(dst, dst_stride, src, src_stride, width, height, \
257  filter, clip, param_stride, vb_pos, (1 << bd) - 1); \
258 } \
259 void bf(ff_vvc_alf_filter_chroma, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *src, ptrdiff_t src_stride, \
260  int width, int height, const int16_t *filter, const int16_t *clip, const int vb_pos) \
261 { \
262  BF(ff_vvc_alf_filter_chroma, bpc, opt)(dst, dst_stride, src, src_stride, width, height, \
263  filter, clip, 0, vb_pos,(1 << bd) - 1); \
264 } \
265 void bf(ff_vvc_alf_classify, bd, opt)(int *class_idx, int *transpose_idx, \
266  const uint8_t *src, ptrdiff_t src_stride, int width, int height, int vb_pos, int *gradient_tmp) \
267 { \
268  BF(ff_vvc_alf_classify_grad, bpc, opt)(gradient_tmp, src, src_stride, width, height, vb_pos); \
269  BF(ff_vvc_alf_classify, bpc, opt)(class_idx, transpose_idx, gradient_tmp, width, height, vb_pos, bd); \
270 } \
271 
272 ALF_FUNCS(8, 8, avx2)
273 ALF_FUNCS(16, 10, avx2)
274 ALF_FUNCS(16, 12, avx2)
275 
276 #endif
277 
278 #define PEL_LINK(dst, C, W, idx1, idx2, name, D, opt) \
279  dst[C][W][idx1][idx2] = ff_vvc_put_## name ## _ ## D ## _##opt; \
280  dst ## _uni[C][W][idx1][idx2] = ff_h2656_put_uni_ ## name ## _ ## D ## _##opt; \
281 
282 #define MC_TAP_LINKS(pointer, C, my, mx, fname, bitd, opt ) \
283  PEL_LINK(pointer, C, 1, my , mx , fname##4 , bitd, opt ); \
284  PEL_LINK(pointer, C, 2, my , mx , fname##8 , bitd, opt ); \
285  PEL_LINK(pointer, C, 3, my , mx , fname##16, bitd, opt ); \
286  PEL_LINK(pointer, C, 4, my , mx , fname##32, bitd, opt ); \
287  PEL_LINK(pointer, C, 5, my , mx , fname##64, bitd, opt ); \
288  PEL_LINK(pointer, C, 6, my , mx , fname##128, bitd, opt );
289 
290 #define MC_8TAP_LINKS(pointer, my, mx, fname, bitd, opt) \
291  MC_TAP_LINKS(pointer, LUMA, my, mx, fname, bitd, opt)
292 
293 #define MC_8TAP_LINKS_SSE4(bd) \
294  MC_8TAP_LINKS(c->inter.put, 0, 0, pixels, bd, sse4); \
295  MC_8TAP_LINKS(c->inter.put, 0, 1, 8tap_h, bd, sse4); \
296  MC_8TAP_LINKS(c->inter.put, 1, 0, 8tap_v, bd, sse4); \
297  MC_8TAP_LINKS(c->inter.put, 1, 1, 8tap_hv, bd, sse4)
298 
299 #define MC_4TAP_LINKS(pointer, my, mx, fname, bitd, opt) \
300  PEL_LINK(pointer, CHROMA, 0, my , mx , fname##2 , bitd, opt ); \
301  MC_TAP_LINKS(pointer, CHROMA, my, mx, fname, bitd, opt) \
302 
303 #define MC_4TAP_LINKS_SSE4(bd) \
304  MC_4TAP_LINKS(c->inter.put, 0, 0, pixels, bd, sse4); \
305  MC_4TAP_LINKS(c->inter.put, 0, 1, 4tap_h, bd, sse4); \
306  MC_4TAP_LINKS(c->inter.put, 1, 0, 4tap_v, bd, sse4); \
307  MC_4TAP_LINKS(c->inter.put, 1, 1, 4tap_hv, bd, sse4)
308 
309 #define MC_LINK_SSE4(bd) \
310  MC_4TAP_LINKS_SSE4(bd) \
311  MC_8TAP_LINKS_SSE4(bd)
312 
313 #define MC_TAP_LINKS_AVX2(C,tap,bd) do { \
314  PEL_LINK(c->inter.put, C, 4, 0, 0, pixels32, bd, avx2) \
315  PEL_LINK(c->inter.put, C, 5, 0, 0, pixels64, bd, avx2) \
316  PEL_LINK(c->inter.put, C, 6, 0, 0, pixels128, bd, avx2) \
317  PEL_LINK(c->inter.put, C, 4, 0, 1, tap##tap_h32, bd, avx2) \
318  PEL_LINK(c->inter.put, C, 5, 0, 1, tap##tap_h64, bd, avx2) \
319  PEL_LINK(c->inter.put, C, 6, 0, 1, tap##tap_h128, bd, avx2) \
320  PEL_LINK(c->inter.put, C, 4, 1, 0, tap##tap_v32, bd, avx2) \
321  PEL_LINK(c->inter.put, C, 5, 1, 0, tap##tap_v64, bd, avx2) \
322  PEL_LINK(c->inter.put, C, 6, 1, 0, tap##tap_v128, bd, avx2) \
323  } while (0)
324 
325 #define MC_LINKS_AVX2(bd) \
326  MC_TAP_LINKS_AVX2(LUMA, 8, bd); \
327  MC_TAP_LINKS_AVX2(CHROMA, 4, bd);
328 
329 #define MC_TAP_LINKS_16BPC_AVX2(C, tap, bd) do { \
330  PEL_LINK(c->inter.put, C, 3, 0, 0, pixels16, bd, avx2) \
331  PEL_LINK(c->inter.put, C, 3, 0, 1, tap##tap_h16, bd, avx2) \
332  PEL_LINK(c->inter.put, C, 3, 1, 0, tap##tap_v16, bd, avx2) \
333  PEL_LINK(c->inter.put, C, 3, 1, 1, tap##tap_hv16, bd, avx2) \
334  PEL_LINK(c->inter.put, C, 4, 1, 1, tap##tap_hv32, bd, avx2) \
335  PEL_LINK(c->inter.put, C, 5, 1, 1, tap##tap_hv64, bd, avx2) \
336  PEL_LINK(c->inter.put, C, 6, 1, 1, tap##tap_hv128, bd, avx2) \
337  } while (0)
338 
339 #define MC_LINKS_16BPC_AVX2(bd) \
340  MC_TAP_LINKS_16BPC_AVX2(LUMA, 8, bd); \
341  MC_TAP_LINKS_16BPC_AVX2(CHROMA, 4, bd);
342 
343 #define AVG_INIT(bd, opt) do { \
344  c->inter.avg = bf(ff_vvc_avg, bd, opt); \
345  c->inter.w_avg = bf(ff_vvc_w_avg, bd, opt); \
346 } while (0)
347 
348 #define DMVR_INIT(bd) do { \
349  c->inter.dmvr[0][0] = ff_vvc_dmvr_##bd##_avx2; \
350  c->inter.dmvr[0][1] = ff_vvc_dmvr_h_##bd##_avx2; \
351  c->inter.dmvr[1][0] = ff_vvc_dmvr_v_##bd##_avx2; \
352  c->inter.dmvr[1][1] = ff_vvc_dmvr_hv_##bd##_avx2; \
353 } while (0)
354 
355 #define OF_INIT(bd) do { \
356  c->inter.apply_bdof = ff_vvc_apply_bdof_##bd##_avx2; \
357 } while (0)
358 
359 #define ALF_INIT(bd) do { \
360  c->alf.filter[LUMA] = ff_vvc_alf_filter_luma_##bd##_avx2; \
361  c->alf.filter[CHROMA] = ff_vvc_alf_filter_chroma_##bd##_avx2; \
362  c->alf.classify = ff_vvc_alf_classify_##bd##_avx2; \
363 } while (0)
364 
365 int ff_vvc_sad_avx2(const int16_t *src0, const int16_t *src1, int dx, int dy, int block_w, int block_h);
366 #define SAD_INIT() c->inter.sad = ff_vvc_sad_avx2
367 #endif
368 
369 void ff_vvc_dsp_init_x86(VVCDSPContext *const c, const int bd)
370 {
371 #if ARCH_X86_64
372  const int cpu_flags = av_get_cpu_flags();
373 
374  switch (bd) {
375  case 8:
376  if (EXTERNAL_SSE4(cpu_flags)) {
377  MC_LINK_SSE4(8);
378  }
379  if (EXTERNAL_AVX2_FAST(cpu_flags)) {
380  ALF_INIT(8);
381  AVG_INIT(8, avx2);
382  MC_LINKS_AVX2(8);
383  OF_INIT(8);
384  DMVR_INIT(8);
385  SAD_INIT();
386  }
387  break;
388  case 10:
389  if (EXTERNAL_SSE4(cpu_flags)) {
390  MC_LINK_SSE4(10);
391  }
392  if (EXTERNAL_AVX2_FAST(cpu_flags)) {
393  ALF_INIT(10);
394  AVG_INIT(10, avx2);
395  MC_LINKS_AVX2(10);
396  MC_LINKS_16BPC_AVX2(10);
397  OF_INIT(10);
398  DMVR_INIT(10);
399  SAD_INIT();
400  }
401  break;
402  case 12:
403  if (EXTERNAL_SSE4(cpu_flags)) {
404  MC_LINK_SSE4(12);
405  }
406  if (EXTERNAL_AVX2_FAST(cpu_flags)) {
407  ALF_INIT(12);
408  AVG_INIT(12, avx2);
409  MC_LINKS_AVX2(12);
410  MC_LINKS_16BPC_AVX2(12);
411  OF_INIT(12);
412  DMVR_INIT(12);
413  SAD_INIT();
414  }
415  break;
416  default:
417  break;
418  }
419 #endif
420 }
ff_vvc_dsp_init_x86
void ff_vvc_dsp_init_x86(VVCDSPContext *const c, const int bd)
Definition: vvcdsp_init.c:369
cpu.h
ALF_PROTOTYPES
#define ALF_PROTOTYPES(bpc, bd, opt)
Definition: vvcdsp_init.c:141
src1
const pixel * src1
Definition: h264pred_template.c:421
EXTERNAL_AVX2_FAST
#define EXTERNAL_AVX2_FAST(flags)
Definition: cpu.h:79
w
uint8_t w
Definition: llviddspenc.c:38
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:107
cpu_flags
static atomic_int cpu_flags
Definition: cpu.c:56
ALF_BPC_PROTOTYPES
#define ALF_BPC_PROTOTYPES(bpc, opt)
Definition: vvcdsp_init.c:129
AVG_PROTOTYPES
#define AVG_PROTOTYPES(bd, opt)
Definition: vvcdsp_init.c:76
dsp.h
OF_PROTOTYPES
#define OF_PROTOTYPES(bd, opt)
Definition: vvcdsp_init.c:105
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
h2656dsp.h
dst
uint8_t ptrdiff_t const uint8_t ptrdiff_t int intptr_t intptr_t int int16_t * dst
Definition: dsp.h:83
cpu.h
PUT_BPC_PROTOTYPES
#define PUT_BPC_PROTOTYPES(name, opt)
Definition: vvcdsp_init.c:48
DMVR_PROTOTYPES
#define DMVR_PROTOTYPES(bd, opt)
Definition: vvcdsp_init.c:91
EXTERNAL_SSE4
#define EXTERNAL_SSE4(flags)
Definition: cpu.h:68
src0
const pixel *const src0
Definition: h264pred_template.c:420
AVG_BPC_PROTOTYPES
#define AVG_BPC_PROTOTYPES(bpc, opt)
Definition: vvcdsp_init.c:69
h
h
Definition: vp9dsp_template.c:2070
ctu.h
PUT_TAP_PROTOTYPES
#define PUT_TAP_PROTOTYPES(n, opt)
Definition: vvcdsp_init.c:53
dec.h
VVCDSPContext
Definition: dsp.h:169
Generated on Fri Aug 22 2025 13:59:01 for FFmpeg by   doxygen 1.8.17