FFmpeg coverage

Directory: ../../../ffmpeg/
File: src/libavcodec/x86/vp9dsp_init_16bpp_template.c
Date: 2026-05-03 08:24:11
Exec Total Coverage
Lines: 60 62 96.8%
Functions: 390 390 100.0%
Branches: 15 20 75.0%
Line Branch Exec Source
1 /*
2 * VP9 SIMD optimizations
3 *
4 * Copyright (c) 2013 Ronald S. Bultje <rsbultje gmail com>
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/attributes.h"
24 #include "libavutil/cpu.h"
25 #include "libavutil/x86/cpu.h"
26 #include "libavcodec/vp9dsp.h"
27 #include "libavcodec/x86/vp9dsp_init.h"
28
29 extern const int16_t ff_filters_16bpp[3][15][4][16];
30
31 decl_mc_funcs(4, sse2, int16_t, 16, BPC);
32 decl_mc_funcs(8, sse2, int16_t, 16, BPC);
33 decl_mc_funcs(16, avx2, int16_t, 16, BPC);
34
35 1344 mc_rep_funcs(16, 8, 16, sse2, int16_t, 16, BPC)
36 576 mc_rep_funcs(32, 16, 32, sse2, int16_t, 16, BPC)
37 192 mc_rep_funcs(64, 32, 64, sse2, int16_t, 16, BPC)
38 #if HAVE_AVX2_EXTERNAL
39 308 mc_rep_funcs(32, 16, 32, avx2, int16_t, 16, BPC)
40 96 mc_rep_funcs(64, 32, 64, avx2, int16_t, 16, BPC)
41 #endif
42
43 3566 filters_8tap_2d_fn3(put, 16, BPC, 2, sse2, 16bpp)
44 760 filters_8tap_2d_fn3(avg, 16, BPC, 2, sse2, 16bpp)
45 #if HAVE_AVX2_EXTERNAL
46 12 filters_8tap_2d_fn(put, 64, 32, BPC, 2, avx2, 16bpp)
47 12 filters_8tap_2d_fn(avg, 64, 32, BPC, 2, avx2, 16bpp)
48 16 filters_8tap_2d_fn(put, 32, 32, BPC, 2, avx2, 16bpp)
49 16 filters_8tap_2d_fn(avg, 32, 32, BPC, 2, avx2, 16bpp)
50 192 filters_8tap_2d_fn(put, 16, 32, BPC, 2, avx2, 16bpp)
51 62 filters_8tap_2d_fn(avg, 16, 32, BPC, 2, avx2, 16bpp)
52 #endif
53
54 5590 filters_8tap_1d_fn4(put, BPC, sse2, 16bpp)
55 588 filters_8tap_1d_fn4(avg, BPC, sse2, 16bpp)
56 #if HAVE_AVX2_EXTERNAL
57 24 filters_8tap_1d_fn2(put, 64, BPC, avx2, 16bpp)
58 24 filters_8tap_1d_fn2(avg, 64, BPC, avx2, 16bpp)
59 26 filters_8tap_1d_fn2(put, 32, BPC, avx2, 16bpp)
60 26 filters_8tap_1d_fn2(avg, 32, BPC, avx2, 16bpp)
61 324 filters_8tap_1d_fn2(put, 16, BPC, avx2, 16bpp)
62 58 filters_8tap_1d_fn2(avg, 16, BPC, avx2, 16bpp)
63 #endif
64
65 #define decl_lpf_func(dir, wd, bpp, opt) \
66 void ff_vp9_loop_filter_##dir##_##wd##_##bpp##_##opt(uint8_t *dst, ptrdiff_t stride, \
67 int E, int I, int H)
68
69 #define decl_lpf_funcs(dir, wd, bpp) \
70 decl_lpf_func(dir, wd, bpp, sse2); \
71 decl_lpf_func(dir, wd, bpp, ssse3); \
72 decl_lpf_func(dir, wd, bpp, avx)
73
74 #define decl_lpf_funcs_wd(dir) \
75 decl_lpf_funcs(dir, 4, BPC); \
76 decl_lpf_funcs(dir, 8, BPC); \
77 decl_lpf_funcs(dir, 16, BPC)
78
79 decl_lpf_funcs_wd(h);
80 decl_lpf_funcs_wd(v);
81
82 #define lpf_16_wrapper(dir, off, bpp, opt) \
83 static void loop_filter_##dir##_16_##bpp##_##opt(uint8_t *dst, ptrdiff_t stride, \
84 int E, int I, int H) \
85 { \
86 ff_vp9_loop_filter_##dir##_16_##bpp##_##opt(dst, stride, E, I, H); \
87 ff_vp9_loop_filter_##dir##_16_##bpp##_##opt(dst + off, stride, E, I, H); \
88 }
89
90 #define lpf_16_wrappers(bpp, opt) \
91 lpf_16_wrapper(h, 8 * stride, bpp, opt) \
92 lpf_16_wrapper(v, 16, bpp, opt)
93
94 16 lpf_16_wrappers(BPC, sse2)
95 16 lpf_16_wrappers(BPC, ssse3)
96 20458 lpf_16_wrappers(BPC, avx)
97
98 #define lpf_mix2_wrapper(dir, off, wd1, wd2, bpp, opt) \
99 static void loop_filter_##dir##_##wd1##wd2##_##bpp##_##opt(uint8_t *dst, ptrdiff_t stride, \
100 int E, int I, int H) \
101 { \
102 ff_vp9_loop_filter_##dir##_##wd1##_##bpp##_##opt(dst, stride, \
103 E & 0xff, I & 0xff, H & 0xff); \
104 ff_vp9_loop_filter_##dir##_##wd2##_##bpp##_##opt(dst + off, stride, \
105 E >> 8, I >> 8, H >> 8); \
106 }
107
108 #define lpf_mix2_wrappers(wd1, wd2, bpp, opt) \
109 lpf_mix2_wrapper(h, 8 * stride, wd1, wd2, bpp, opt) \
110 lpf_mix2_wrapper(v, 16, wd1, wd2, bpp, opt)
111
112 #define lpf_mix2_wrappers_set(bpp, opt) \
113 lpf_mix2_wrappers(4, 4, bpp, opt) \
114 lpf_mix2_wrappers(4, 8, bpp, opt) \
115 lpf_mix2_wrappers(8, 4, bpp, opt) \
116 lpf_mix2_wrappers(8, 8, bpp, opt) \
117
118 64 lpf_mix2_wrappers_set(BPC, sse2)
119 64 lpf_mix2_wrappers_set(BPC, ssse3)
120 37966 lpf_mix2_wrappers_set(BPC, avx)
121
122 decl_ipred_fns(tm, BPC, mmxext, sse2);
123
124 decl_itxfm_func(iwht, iwht, 4, BPC, mmxext);
125 #if BPC == 10
126 decl_itxfm_funcs(4, BPC, ssse3);
127 decl_itxfm_funcs(16, BPC, avx512icl);
128 decl_itxfm_func(idct, idct, 32, BPC, avx512icl);
129 #else
130 decl_itxfm_func(idct, idct, 4, BPC, sse2);
131 #endif
132 decl_itxfm_func(idct, iadst, 4, BPC, sse2);
133 decl_itxfm_func(iadst, idct, 4, BPC, sse2);
134 decl_itxfm_func(iadst, iadst, 4, BPC, sse2);
135 decl_itxfm_funcs(8, BPC, sse2);
136 decl_itxfm_funcs(16, BPC, sse2);
137 decl_itxfm_func(idct, idct, 32, BPC, sse2);
138
139 166 av_cold void INIT_FUNC(VP9DSPContext *dsp, int bitexact)
140 {
141 166 int cpu_flags = av_get_cpu_flags();
142
143 #define init_lpf_8_func(idx1, idx2, dir, wd, bpp, opt) \
144 dsp->loop_filter_8[idx1][idx2] = ff_vp9_loop_filter_##dir##_##wd##_##bpp##_##opt
145 #define init_lpf_16_func(idx, dir, bpp, opt) \
146 dsp->loop_filter_16[idx] = loop_filter_##dir##_16_##bpp##_##opt
147 #define init_lpf_mix2_func(idx1, idx2, idx3, dir, wd1, wd2, bpp, opt) \
148 dsp->loop_filter_mix2[idx1][idx2][idx3] = loop_filter_##dir##_##wd1##wd2##_##bpp##_##opt
149
150 #define init_lpf_funcs(bpp, opt) \
151 init_lpf_8_func(0, 0, h, 4, bpp, opt); \
152 init_lpf_8_func(0, 1, v, 4, bpp, opt); \
153 init_lpf_8_func(1, 0, h, 8, bpp, opt); \
154 init_lpf_8_func(1, 1, v, 8, bpp, opt); \
155 init_lpf_8_func(2, 0, h, 16, bpp, opt); \
156 init_lpf_8_func(2, 1, v, 16, bpp, opt); \
157 init_lpf_16_func(0, h, bpp, opt); \
158 init_lpf_16_func(1, v, bpp, opt); \
159 init_lpf_mix2_func(0, 0, 0, h, 4, 4, bpp, opt); \
160 init_lpf_mix2_func(0, 1, 0, h, 4, 8, bpp, opt); \
161 init_lpf_mix2_func(1, 0, 0, h, 8, 4, bpp, opt); \
162 init_lpf_mix2_func(1, 1, 0, h, 8, 8, bpp, opt); \
163 init_lpf_mix2_func(0, 0, 1, v, 4, 4, bpp, opt); \
164 init_lpf_mix2_func(0, 1, 1, v, 4, 8, bpp, opt); \
165 init_lpf_mix2_func(1, 0, 1, v, 8, 4, bpp, opt); \
166 init_lpf_mix2_func(1, 1, 1, v, 8, 8, bpp, opt)
167
168 #define init_itx_func(idxa, idxb, typea, typeb, size, bpp, opt) \
169 dsp->itxfm_add[idxa][idxb] = \
170 cat(ff_vp9_##typea##_##typeb##_##size##x##size##_add_, bpp, _##opt);
171 #define init_itx_func_one(idx, typea, typeb, size, bpp, opt) \
172 init_itx_func(idx, DCT_DCT, typea, typeb, size, bpp, opt); \
173 init_itx_func(idx, ADST_DCT, typea, typeb, size, bpp, opt); \
174 init_itx_func(idx, DCT_ADST, typea, typeb, size, bpp, opt); \
175 init_itx_func(idx, ADST_ADST, typea, typeb, size, bpp, opt)
176 #define init_itx_funcs(idx, size, bpp, opt) \
177 init_itx_func(idx, DCT_DCT, idct, idct, size, bpp, opt); \
178 init_itx_func(idx, ADST_DCT, idct, iadst, size, bpp, opt); \
179 init_itx_func(idx, DCT_ADST, iadst, idct, size, bpp, opt); \
180 init_itx_func(idx, ADST_ADST, iadst, iadst, size, bpp, opt); \
181
182
2/2
✓ Branch 0 taken 123 times.
✓ Branch 1 taken 43 times.
 166 if (EXTERNAL_MMXEXT(cpu_flags)) {
183 123 init_ipred_func(tm, TM_VP8, 4, BPC, mmxext);
184
1/2
✓ Branch 0 taken 123 times.
✗ Branch 1 not taken.
 123 if (!bitexact) {
185 123 init_itx_func_one(4 /* lossless */, iwht, iwht, 4, BPC, mmxext);
186 }
187 }
188
189
2/2
✓ Branch 0 taken 103 times.
✓ Branch 1 taken 63 times.
 166 if (EXTERNAL_SSE2(cpu_flags)) {
190 103 init_subpel3(0, put, BPC, sse2);
191 103 init_subpel3(1, avg, BPC, sse2);
192 103 init_lpf_funcs(BPC, sse2);
193 103 init_8_16_32_ipred_funcs(tm, TM_VP8, BPC, sse2);
194 #if BPC == 10
195
1/2
✓ Branch 0 taken 52 times.
✗ Branch 1 not taken.
 52 if (!bitexact) {
196 52 init_itx_func(TX_4X4, ADST_DCT, idct, iadst, 4, 10, sse2);
197 52 init_itx_func(TX_4X4, DCT_ADST, iadst, idct, 4, 10, sse2);
198 52 init_itx_func(TX_4X4, ADST_ADST, iadst, iadst, 4, 10, sse2);
199 }
200 #else
201 51 init_itx_funcs(TX_4X4, 4, 12, sse2);
202 #endif
203 103 init_itx_funcs(TX_8X8, 8, BPC, sse2);
204 103 init_itx_funcs(TX_16X16, 16, BPC, sse2);
205 103 init_itx_func_one(TX_32X32, idct, idct, 32, BPC, sse2);
206 }
207
208
2/2
✓ Branch 0 taken 83 times.
✓ Branch 1 taken 83 times.
 166 if (EXTERNAL_SSSE3(cpu_flags)) {
209 83 init_lpf_funcs(BPC, ssse3);
210 #if BPC == 10
211
1/2
✓ Branch 0 taken 42 times.
✗ Branch 1 not taken.
 42 if (!bitexact) {
212 42 init_itx_funcs(TX_4X4, 4, BPC, ssse3);
213 }
214 #endif
215 }
216
217
2/2
✓ Branch 0 taken 33 times.
✓ Branch 1 taken 133 times.
 166 if (EXTERNAL_AVX(cpu_flags)) {
218 33 init_lpf_funcs(BPC, avx);
219 }
220
221
3/4
✓ Branch 0 taken 13 times.
✓ Branch 1 taken 153 times.
✓ Branch 2 taken 13 times.
✗ Branch 3 not taken.
 166 if (EXTERNAL_AVX2_FAST(cpu_flags)) {
222 #if HAVE_AVX2_EXTERNAL
223 13 init_subpel3_32_64(0, put, BPC, avx2);
224 13 init_subpel3_32_64(1, avg, BPC, avx2);
225 13 init_subpel2(2, 0, 16, put, BPC, avx2);
226 13 init_subpel2(2, 1, 16, avg, BPC, avx2);
227 #endif
228 }
229
230 #if ARCH_X86_64 && BPC == 10
231
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 86 times.
 86 if (EXTERNAL_AVX512ICL(cpu_flags)) {
232 ✗ init_itx_funcs(TX_16X16, 16, BPC, avx512icl);
233 ✗ init_itx_func_one(TX_32X32, idct, idct, 32, BPC, avx512icl);
234 }
235 #endif
236
237 166 ff_vp9dsp_init_16bpp_x86(dsp);
238 166 }
239