Line | Branch | Exec | Source |
---|---|---|---|
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 | #if ARCH_X86_64 | ||
91 | #if HAVE_SSE4_EXTERNAL | ||
92 | #define FW_PUT(name, depth, opt) \ | ||
93 | void ff_vvc_put_ ## name ## _ ## depth ## _##opt(int16_t *dst, const uint8_t *src, ptrdiff_t srcstride, \ | ||
94 | int height, const int8_t *hf, const int8_t *vf, int width) \ | ||
95 | { \ | ||
96 | ff_h2656_put_## name ## _ ## depth ## _##opt(dst, 2 * MAX_PB_SIZE, src, srcstride, height, hf, vf, width); \ | ||
97 | } | ||
98 | |||
99 | #define FW_PUT_TAP(fname, bitd, opt ) \ | ||
100 | FW_PUT(fname##4, bitd, opt ) \ | ||
101 | FW_PUT(fname##8, bitd, opt ) \ | ||
102 | FW_PUT(fname##16, bitd, opt ) \ | ||
103 | FW_PUT(fname##32, bitd, opt ) \ | ||
104 | FW_PUT(fname##64, bitd, opt ) \ | ||
105 | FW_PUT(fname##128, bitd, opt ) \ | ||
106 | |||
107 | #define FW_PUT_4TAP(fname, bitd, opt) \ | ||
108 | FW_PUT(fname ## 2, bitd, opt) \ | ||
109 | FW_PUT_TAP(fname, bitd, opt) | ||
110 | |||
111 | #define FW_PUT_4TAP_SSE4(bitd) \ | ||
112 | FW_PUT_4TAP(pixels, bitd, sse4) \ | ||
113 | FW_PUT_4TAP(4tap_h, bitd, sse4) \ | ||
114 | FW_PUT_4TAP(4tap_v, bitd, sse4) \ | ||
115 | FW_PUT_4TAP(4tap_hv, bitd, sse4) | ||
116 | |||
117 | #define FW_PUT_8TAP_SSE4(bitd) \ | ||
118 | FW_PUT_TAP(8tap_h, bitd, sse4) \ | ||
119 | FW_PUT_TAP(8tap_v, bitd, sse4) \ | ||
120 | FW_PUT_TAP(8tap_hv, bitd, sse4) | ||
121 | |||
122 | #define FW_PUT_SSE4(bitd) \ | ||
123 | FW_PUT_4TAP_SSE4(bitd) \ | ||
124 | FW_PUT_8TAP_SSE4(bitd) | ||
125 | |||
126 | 914 | FW_PUT_SSE4( 8) | |
127 | 1096 | FW_PUT_SSE4(10) | |
128 | 1096 | FW_PUT_SSE4(12) | |
129 | #endif | ||
130 | |||
131 | #if HAVE_AVX2_EXTERNAL | ||
132 | #define FW_PUT_TAP_AVX2(n, bitd) \ | ||
133 | FW_PUT(n ## tap_h32, bitd, avx2) \ | ||
134 | FW_PUT(n ## tap_h64, bitd, avx2) \ | ||
135 | FW_PUT(n ## tap_h128, bitd, avx2) \ | ||
136 | FW_PUT(n ## tap_v32, bitd, avx2) \ | ||
137 | FW_PUT(n ## tap_v64, bitd, avx2) \ | ||
138 | FW_PUT(n ## tap_v128, bitd, avx2) | ||
139 | |||
140 | #define FW_PUT_AVX2(bitd) \ | ||
141 | FW_PUT(pixels32, bitd, avx2) \ | ||
142 | FW_PUT(pixels64, bitd, avx2) \ | ||
143 | FW_PUT(pixels128, bitd, avx2) \ | ||
144 | FW_PUT_TAP_AVX2(4, bitd) \ | ||
145 | FW_PUT_TAP_AVX2(8, bitd) \ | ||
146 | |||
147 | 234 | FW_PUT_AVX2( 8) | |
148 | 234 | FW_PUT_AVX2(10) | |
149 | 234 | FW_PUT_AVX2(12) | |
150 | |||
151 | #define FW_PUT_TAP_16BPC_AVX2(n, bitd) \ | ||
152 | FW_PUT(n ## tap_h16, bitd, avx2) \ | ||
153 | FW_PUT(n ## tap_v16, bitd, avx2) \ | ||
154 | FW_PUT(n ## tap_hv16, bitd, avx2) \ | ||
155 | FW_PUT(n ## tap_hv32, bitd, avx2) \ | ||
156 | FW_PUT(n ## tap_hv64, bitd, avx2) \ | ||
157 | FW_PUT(n ## tap_hv128, bitd, avx2) | ||
158 | |||
159 | #define FW_PUT_16BPC_AVX2(bitd) \ | ||
160 | FW_PUT(pixels16, bitd, avx2) \ | ||
161 | FW_PUT_TAP_16BPC_AVX2(4, bitd) \ | ||
162 | FW_PUT_TAP_16BPC_AVX2(8, bitd) | ||
163 | |||
164 | 182 | FW_PUT_16BPC_AVX2(10) | |
165 | 182 | FW_PUT_16BPC_AVX2(12) | |
166 | |||
167 | #define AVG_FUNCS(bpc, bd, opt) \ | ||
168 | void bf(ff_vvc_avg, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride, \ | ||
169 | const int16_t *src0, const int16_t *src1, int width, int height) \ | ||
170 | { \ | ||
171 | BF(ff_vvc_avg, bpc, opt)(dst, dst_stride, src0, src1, width, height, (1 << bd) - 1); \ | ||
172 | } \ | ||
173 | void bf(ff_vvc_w_avg, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride, \ | ||
174 | const int16_t *src0, const int16_t *src1, int width, int height, \ | ||
175 | int denom, int w0, int w1, int o0, int o1) \ | ||
176 | { \ | ||
177 | BF(ff_vvc_w_avg, bpc, opt)(dst, dst_stride, src0, src1, width, height, \ | ||
178 | denom, w0, w1, o0, o1, (1 << bd) - 1); \ | ||
179 | } | ||
180 | |||
181 | 196 | AVG_FUNCS(8, 8, avx2) | |
182 | 196 | AVG_FUNCS(16, 10, avx2) | |
183 | 196 | AVG_FUNCS(16, 12, avx2) | |
184 | #endif | ||
185 | |||
186 | #define PEL_LINK(dst, C, W, idx1, idx2, name, D, opt) \ | ||
187 | dst[C][W][idx1][idx2] = ff_vvc_put_## name ## _ ## D ## _##opt; \ | ||
188 | dst ## _uni[C][W][idx1][idx2] = ff_h2656_put_uni_ ## name ## _ ## D ## _##opt; \ | ||
189 | |||
190 | #define MC_TAP_LINKS(pointer, C, my, mx, fname, bitd, opt ) \ | ||
191 | PEL_LINK(pointer, C, 1, my , mx , fname##4 , bitd, opt ); \ | ||
192 | PEL_LINK(pointer, C, 2, my , mx , fname##8 , bitd, opt ); \ | ||
193 | PEL_LINK(pointer, C, 3, my , mx , fname##16, bitd, opt ); \ | ||
194 | PEL_LINK(pointer, C, 4, my , mx , fname##32, bitd, opt ); \ | ||
195 | PEL_LINK(pointer, C, 5, my , mx , fname##64, bitd, opt ); \ | ||
196 | PEL_LINK(pointer, C, 6, my , mx , fname##128, bitd, opt ); | ||
197 | |||
198 | #define MC_8TAP_LINKS(pointer, my, mx, fname, bitd, opt) \ | ||
199 | MC_TAP_LINKS(pointer, LUMA, my, mx, fname, bitd, opt) | ||
200 | |||
201 | #define MC_8TAP_LINKS_SSE4(bd) \ | ||
202 | MC_8TAP_LINKS(c->inter.put, 0, 0, pixels, bd, sse4); \ | ||
203 | MC_8TAP_LINKS(c->inter.put, 0, 1, 8tap_h, bd, sse4); \ | ||
204 | MC_8TAP_LINKS(c->inter.put, 1, 0, 8tap_v, bd, sse4); \ | ||
205 | MC_8TAP_LINKS(c->inter.put, 1, 1, 8tap_hv, bd, sse4) | ||
206 | |||
207 | #define MC_4TAP_LINKS(pointer, my, mx, fname, bitd, opt) \ | ||
208 | PEL_LINK(pointer, CHROMA, 0, my , mx , fname##2 , bitd, opt ); \ | ||
209 | MC_TAP_LINKS(pointer, CHROMA, my, mx, fname, bitd, opt) \ | ||
210 | |||
211 | #define MC_4TAP_LINKS_SSE4(bd) \ | ||
212 | MC_4TAP_LINKS(c->inter.put, 0, 0, pixels, bd, sse4); \ | ||
213 | MC_4TAP_LINKS(c->inter.put, 0, 1, 4tap_h, bd, sse4); \ | ||
214 | MC_4TAP_LINKS(c->inter.put, 1, 0, 4tap_v, bd, sse4); \ | ||
215 | MC_4TAP_LINKS(c->inter.put, 1, 1, 4tap_hv, bd, sse4) | ||
216 | |||
217 | #define MC_LINK_SSE4(bd) \ | ||
218 | MC_4TAP_LINKS_SSE4(bd) \ | ||
219 | MC_8TAP_LINKS_SSE4(bd) | ||
220 | |||
221 | #define MC_TAP_LINKS_AVX2(C,tap,bd) do { \ | ||
222 | PEL_LINK(c->inter.put, C, 4, 0, 0, pixels32, bd, avx2) \ | ||
223 | PEL_LINK(c->inter.put, C, 5, 0, 0, pixels64, bd, avx2) \ | ||
224 | PEL_LINK(c->inter.put, C, 6, 0, 0, pixels128, bd, avx2) \ | ||
225 | PEL_LINK(c->inter.put, C, 4, 0, 1, tap##tap_h32, bd, avx2) \ | ||
226 | PEL_LINK(c->inter.put, C, 5, 0, 1, tap##tap_h64, bd, avx2) \ | ||
227 | PEL_LINK(c->inter.put, C, 6, 0, 1, tap##tap_h128, bd, avx2) \ | ||
228 | PEL_LINK(c->inter.put, C, 4, 1, 0, tap##tap_v32, bd, avx2) \ | ||
229 | PEL_LINK(c->inter.put, C, 5, 1, 0, tap##tap_v64, bd, avx2) \ | ||
230 | PEL_LINK(c->inter.put, C, 6, 1, 0, tap##tap_v128, bd, avx2) \ | ||
231 | } while (0) | ||
232 | |||
233 | #define MC_LINKS_AVX2(bd) \ | ||
234 | MC_TAP_LINKS_AVX2(LUMA, 8, bd); \ | ||
235 | MC_TAP_LINKS_AVX2(CHROMA, 4, bd); | ||
236 | |||
237 | #define MC_TAP_LINKS_16BPC_AVX2(C, tap, bd) do { \ | ||
238 | PEL_LINK(c->inter.put, C, 3, 0, 0, pixels16, bd, avx2) \ | ||
239 | PEL_LINK(c->inter.put, C, 3, 0, 1, tap##tap_h16, bd, avx2) \ | ||
240 | PEL_LINK(c->inter.put, C, 3, 1, 0, tap##tap_v16, bd, avx2) \ | ||
241 | PEL_LINK(c->inter.put, C, 3, 1, 1, tap##tap_hv16, bd, avx2) \ | ||
242 | PEL_LINK(c->inter.put, C, 4, 1, 1, tap##tap_hv32, bd, avx2) \ | ||
243 | PEL_LINK(c->inter.put, C, 5, 1, 1, tap##tap_hv64, bd, avx2) \ | ||
244 | PEL_LINK(c->inter.put, C, 6, 1, 1, tap##tap_hv128, bd, avx2) \ | ||
245 | } while (0) | ||
246 | |||
247 | #define MC_LINKS_16BPC_AVX2(bd) \ | ||
248 | MC_TAP_LINKS_16BPC_AVX2(LUMA, 8, bd); \ | ||
249 | MC_TAP_LINKS_16BPC_AVX2(CHROMA, 4, bd); | ||
250 | |||
251 | #define AVG_INIT(bd, opt) do { \ | ||
252 | c->inter.avg = bf(ff_vvc_avg, bd, opt); \ | ||
253 | c->inter.w_avg = bf(ff_vvc_w_avg, bd, opt); \ | ||
254 | } while (0) | ||
255 | #endif | ||
256 | |||
257 | 1067 | void ff_vvc_dsp_init_x86(VVCDSPContext *const c, const int bd) | |
258 | { | ||
259 | #if ARCH_X86_64 | ||
260 | 1067 | const int cpu_flags = av_get_cpu_flags(); | |
261 | |||
262 |
2/2✓ Branch 0 taken 68 times.
✓ Branch 1 taken 999 times.
|
1067 | if (bd == 8) { |
263 |
2/2✓ Branch 0 taken 30 times.
✓ Branch 1 taken 38 times.
|
68 | if (EXTERNAL_SSE4(cpu_flags)) { |
264 | 30 | MC_LINK_SSE4(8); | |
265 | } | ||
266 |
3/4✓ Branch 0 taken 5 times.
✓ Branch 1 taken 63 times.
✓ Branch 2 taken 5 times.
✗ Branch 3 not taken.
|
68 | if (EXTERNAL_AVX2_FAST(cpu_flags)) { |
267 | 5 | MC_LINKS_AVX2(8); | |
268 | } | ||
269 |
2/2✓ Branch 0 taken 934 times.
✓ Branch 1 taken 65 times.
|
999 | } else if (bd == 10) { |
270 |
2/2✓ Branch 0 taken 30 times.
✓ Branch 1 taken 904 times.
|
934 | if (EXTERNAL_SSE4(cpu_flags)) { |
271 | 30 | MC_LINK_SSE4(10); | |
272 | } | ||
273 |
3/4✓ Branch 0 taken 5 times.
✓ Branch 1 taken 929 times.
✓ Branch 2 taken 5 times.
✗ Branch 3 not taken.
|
934 | if (EXTERNAL_AVX2_FAST(cpu_flags)) { |
274 | 5 | MC_LINKS_AVX2(10); | |
275 | 5 | MC_LINKS_16BPC_AVX2(10); | |
276 | } | ||
277 |
1/2✓ Branch 0 taken 65 times.
✗ Branch 1 not taken.
|
65 | } else if (bd == 12) { |
278 |
2/2✓ Branch 0 taken 30 times.
✓ Branch 1 taken 35 times.
|
65 | if (EXTERNAL_SSE4(cpu_flags)) { |
279 | 30 | MC_LINK_SSE4(12); | |
280 | } | ||
281 |
3/4✓ Branch 0 taken 5 times.
✓ Branch 1 taken 60 times.
✓ Branch 2 taken 5 times.
✗ Branch 3 not taken.
|
65 | if (EXTERNAL_AVX2_FAST(cpu_flags)) { |
282 | 5 | MC_LINKS_AVX2(12); | |
283 | 5 | MC_LINKS_16BPC_AVX2(12); | |
284 | } | ||
285 | } | ||
286 | |||
287 |
2/2✓ Branch 0 taken 15 times.
✓ Branch 1 taken 1052 times.
|
1067 | if (EXTERNAL_AVX2(cpu_flags)) { |
288 |
3/4✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
✓ Branch 2 taken 5 times.
✗ Branch 3 not taken.
|
15 | switch (bd) { |
289 | 5 | case 8: | |
290 | 5 | AVG_INIT(8, avx2); | |
291 | 5 | break; | |
292 | 5 | case 10: | |
293 | 5 | AVG_INIT(10, avx2); | |
294 | 5 | break; | |
295 | 5 | case 12: | |
296 | 5 | AVG_INIT(12, avx2); | |
297 | 5 | break; | |
298 | ✗ | default: | |
299 | ✗ | break; | |
300 | } | ||
301 | } | ||
302 | #endif | ||
303 | 1067 | } | |
304 |