FFmpeg coverage


Directory: ../../../ffmpeg/
File: src/libavcodec/x86/videodsp_init.c
Date: 2025-12-13 00:37:54
Exec Total Coverage
Lines: 53 54 98.1%
Functions: 4 4 100.0%
Branches: 30 32 93.8%

Line Branch Exec Source
1 /*
2 * Copyright (C) 2002-2012 Michael Niedermayer
3 * Copyright (C) 2012 Ronald S. Bultje
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 #include "config.h"
23 #include "libavutil/attributes.h"
24 #include "libavutil/avassert.h"
25 #include "libavutil/common.h"
26 #include "libavutil/cpu.h"
27 #include "libavutil/x86/asm.h"
28 #include "libavutil/x86/cpu.h"
29 #include "libavcodec/videodsp.h"
30
31 typedef void emu_edge_vfix_func(uint8_t *dst, x86_reg dst_stride,
32 const uint8_t *src, x86_reg src_stride,
33 x86_reg start_y, x86_reg end_y, x86_reg bh);
34 typedef void emu_edge_vvar_func(uint8_t *dst, x86_reg dst_stride,
35 const uint8_t *src, x86_reg src_stride,
36 x86_reg start_y, x86_reg end_y, x86_reg bh,
37 x86_reg w);
38
39 extern emu_edge_vfix_func ff_emu_edge_vfix1_sse2;
40 extern emu_edge_vfix_func ff_emu_edge_vfix2_sse2;
41 extern emu_edge_vfix_func ff_emu_edge_vfix3_sse2;
42 extern emu_edge_vfix_func ff_emu_edge_vfix4_sse2;
43 extern emu_edge_vfix_func ff_emu_edge_vfix5_sse2;
44 extern emu_edge_vfix_func ff_emu_edge_vfix6_sse2;
45 extern emu_edge_vfix_func ff_emu_edge_vfix7_sse2;
46 extern emu_edge_vfix_func ff_emu_edge_vfix8_sse2;
47 extern emu_edge_vfix_func ff_emu_edge_vfix9_sse2;
48 extern emu_edge_vfix_func ff_emu_edge_vfix10_sse2;
49 extern emu_edge_vfix_func ff_emu_edge_vfix11_sse2;
50 extern emu_edge_vfix_func ff_emu_edge_vfix12_sse2;
51 extern emu_edge_vfix_func ff_emu_edge_vfix13_sse2;
52 extern emu_edge_vfix_func ff_emu_edge_vfix14_sse2;
53 extern emu_edge_vfix_func ff_emu_edge_vfix15_sse2;
54 extern emu_edge_vfix_func ff_emu_edge_vfix16_sse2;
55 extern emu_edge_vfix_func ff_emu_edge_vfix17_sse2;
56 extern emu_edge_vfix_func ff_emu_edge_vfix18_sse2;
57 extern emu_edge_vfix_func ff_emu_edge_vfix19_sse2;
58 extern emu_edge_vfix_func ff_emu_edge_vfix20_sse2;
59 extern emu_edge_vfix_func ff_emu_edge_vfix21_sse2;
60 extern emu_edge_vfix_func ff_emu_edge_vfix22_sse2;
61 static emu_edge_vfix_func * const vfixtbl_sse2[22] = {
62 ff_emu_edge_vfix1_sse2, ff_emu_edge_vfix2_sse2, ff_emu_edge_vfix3_sse2,
63 ff_emu_edge_vfix4_sse2, ff_emu_edge_vfix5_sse2, ff_emu_edge_vfix6_sse2,
64 ff_emu_edge_vfix7_sse2, ff_emu_edge_vfix8_sse2, ff_emu_edge_vfix9_sse2,
65 ff_emu_edge_vfix10_sse2, ff_emu_edge_vfix11_sse2, ff_emu_edge_vfix12_sse2,
66 ff_emu_edge_vfix13_sse2, ff_emu_edge_vfix14_sse2, ff_emu_edge_vfix15_sse2,
67 ff_emu_edge_vfix16_sse2, ff_emu_edge_vfix17_sse2, ff_emu_edge_vfix18_sse2,
68 ff_emu_edge_vfix19_sse2, ff_emu_edge_vfix20_sse2, ff_emu_edge_vfix21_sse2,
69 ff_emu_edge_vfix22_sse2
70 };
71 extern emu_edge_vvar_func ff_emu_edge_vvar_sse;
72
73 typedef void emu_edge_hfix_func(uint8_t *dst, x86_reg dst_stride,
74 x86_reg start_x, x86_reg bh);
75 typedef void emu_edge_hvar_func(uint8_t *dst, x86_reg dst_stride,
76 x86_reg start_x, x86_reg n_words, x86_reg bh);
77
78 extern emu_edge_hfix_func ff_emu_edge_hfix2_sse2;
79 extern emu_edge_hfix_func ff_emu_edge_hfix4_sse2;
80 extern emu_edge_hfix_func ff_emu_edge_hfix6_sse2;
81 extern emu_edge_hfix_func ff_emu_edge_hfix8_sse2;
82 extern emu_edge_hfix_func ff_emu_edge_hfix10_sse2;
83 extern emu_edge_hfix_func ff_emu_edge_hfix12_sse2;
84 extern emu_edge_hfix_func ff_emu_edge_hfix14_sse2;
85 extern emu_edge_hfix_func ff_emu_edge_hfix16_sse2;
86 extern emu_edge_hfix_func ff_emu_edge_hfix18_sse2;
87 extern emu_edge_hfix_func ff_emu_edge_hfix20_sse2;
88 extern emu_edge_hfix_func ff_emu_edge_hfix22_sse2;
89 static emu_edge_hfix_func * const hfixtbl_sse2[11] = {
90 ff_emu_edge_hfix2_sse2, ff_emu_edge_hfix4_sse2, ff_emu_edge_hfix6_sse2,
91 ff_emu_edge_hfix8_sse2, ff_emu_edge_hfix10_sse2, ff_emu_edge_hfix12_sse2,
92 ff_emu_edge_hfix14_sse2, ff_emu_edge_hfix16_sse2, ff_emu_edge_hfix18_sse2,
93 ff_emu_edge_hfix20_sse2, ff_emu_edge_hfix22_sse2
94 };
95 extern emu_edge_hvar_func ff_emu_edge_hvar_sse2;
96 #if HAVE_AVX2_EXTERNAL
97 extern emu_edge_hfix_func ff_emu_edge_hfix8_avx2;
98 extern emu_edge_hfix_func ff_emu_edge_hfix10_avx2;
99 extern emu_edge_hfix_func ff_emu_edge_hfix12_avx2;
100 extern emu_edge_hfix_func ff_emu_edge_hfix14_avx2;
101 extern emu_edge_hfix_func ff_emu_edge_hfix16_avx2;
102 extern emu_edge_hfix_func ff_emu_edge_hfix18_avx2;
103 extern emu_edge_hfix_func ff_emu_edge_hfix20_avx2;
104 extern emu_edge_hfix_func ff_emu_edge_hfix22_avx2;
105 static emu_edge_hfix_func * const hfixtbl_avx2[11] = {
106 ff_emu_edge_hfix2_sse2, ff_emu_edge_hfix4_sse2, ff_emu_edge_hfix6_sse2,
107 ff_emu_edge_hfix8_avx2, ff_emu_edge_hfix10_avx2, ff_emu_edge_hfix12_avx2,
108 ff_emu_edge_hfix14_avx2, ff_emu_edge_hfix16_avx2, ff_emu_edge_hfix18_avx2,
109 ff_emu_edge_hfix20_avx2, ff_emu_edge_hfix22_avx2
110 };
111 extern emu_edge_hvar_func ff_emu_edge_hvar_avx2;
112 #endif
113
114 409204 static av_always_inline void emulated_edge_mc(uint8_t *dst, const uint8_t *src,
115 ptrdiff_t dst_stride,
116 ptrdiff_t src_stride,
117 x86_reg block_w, x86_reg block_h,
118 x86_reg src_x, x86_reg src_y,
119 x86_reg w, x86_reg h,
120 emu_edge_vfix_func * const *vfix_tbl,
121 emu_edge_vvar_func *v_extend_var,
122 emu_edge_hfix_func * const *hfix_tbl,
123 emu_edge_hvar_func *h_extend_var)
124 {
125 409204 x86_reg start_y, start_x, end_y, end_x, src_y_add = 0, p;
126
127
2/4
✓ Branch 0 taken 409204 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 409204 times.
409204 if (!w || !h)
128 return;
129
130 av_assert2(block_w <= FFABS(dst_stride));
131
132
2/2
✓ Branch 0 taken 2231 times.
✓ Branch 1 taken 406973 times.
409204 if (src_y >= h) {
133 2231 src -= src_y*src_stride;
134 2231 src_y_add = h - 1;
135 2231 src_y = h - 1;
136
2/2
✓ Branch 0 taken 403 times.
✓ Branch 1 taken 406570 times.
406973 } else if (src_y <= -block_h) {
137 403 src -= src_y*src_stride;
138 403 src_y_add = 1 - block_h;
139 403 src_y = 1 - block_h;
140 }
141
2/2
✓ Branch 0 taken 758 times.
✓ Branch 1 taken 408446 times.
409204 if (src_x >= w) {
142 758 src += w - 1 - src_x;
143 758 src_x = w - 1;
144
2/2
✓ Branch 0 taken 423 times.
✓ Branch 1 taken 408023 times.
408446 } else if (src_x <= -block_w) {
145 423 src += 1 - block_w - src_x;
146 423 src_x = 1 - block_w;
147 }
148
149 409204 start_y = FFMAX(0, -src_y);
150 409204 start_x = FFMAX(0, -src_x);
151 409204 end_y = FFMIN(block_h, h-src_y);
152 409204 end_x = FFMIN(block_w, w-src_x);
153 av_assert2(start_x < end_x && block_w > 0);
154 av_assert2(start_y < end_y && block_h > 0);
155
156 // fill in the to-be-copied part plus all above/below
157 409204 src += (src_y_add + start_y) * src_stride + start_x;
158 409204 w = end_x - start_x;
159
2/2
✓ Branch 0 taken 403544 times.
✓ Branch 1 taken 5660 times.
409204 if (w <= 22) {
160 403544 vfix_tbl[w - 1](dst + start_x, dst_stride, src, src_stride,
161 start_y, end_y, block_h);
162 } else {
163 5660 v_extend_var(dst + start_x, dst_stride, src, src_stride,
164 start_y, end_y, block_h, w);
165 }
166
167 // fill left
168
2/2
✓ Branch 0 taken 33855 times.
✓ Branch 1 taken 375349 times.
409204 if (start_x) {
169
2/2
✓ Branch 0 taken 33216 times.
✓ Branch 1 taken 639 times.
33855 if (start_x <= 22) {
170 33216 hfix_tbl[(start_x - 1) >> 1](dst, dst_stride, start_x, block_h);
171 } else {
172 639 h_extend_var(dst, dst_stride,
173 639 start_x, (start_x + 1) >> 1, block_h);
174 }
175 }
176
177 // fill right
178 409204 p = block_w - end_x;
179
2/2
✓ Branch 0 taken 110990 times.
✓ Branch 1 taken 298214 times.
409204 if (p) {
180
2/2
✓ Branch 0 taken 109734 times.
✓ Branch 1 taken 1256 times.
110990 if (p <= 22) {
181 109734 hfix_tbl[(p - 1) >> 1](dst + end_x - (p & 1), dst_stride,
182 109734 -!(p & 1), block_h);
183 } else {
184 1256 h_extend_var(dst + end_x - (p & 1), dst_stride,
185 1256 -!(p & 1), (p + 1) >> 1, block_h);
186 }
187 }
188 }
189
190 1440 static av_noinline void emulated_edge_mc_sse2(uint8_t *buf, const uint8_t *src,
191 ptrdiff_t buf_stride,
192 ptrdiff_t src_stride,
193 int block_w, int block_h,
194 int src_x, int src_y, int w,
195 int h)
196 {
197 1440 emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
198 src_x, src_y, w, h, vfixtbl_sse2, &ff_emu_edge_vvar_sse,
199 hfixtbl_sse2, &ff_emu_edge_hvar_sse2);
200 1440 }
201
202 #if HAVE_AVX2_EXTERNAL
203 407764 static av_noinline void emulated_edge_mc_avx2(uint8_t *buf, const uint8_t *src,
204 ptrdiff_t buf_stride,
205 ptrdiff_t src_stride,
206 int block_w, int block_h,
207 int src_x, int src_y, int w,
208 int h)
209 {
210 407764 emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
211 src_x, src_y, w, h, vfixtbl_sse2, &ff_emu_edge_vvar_sse,
212 hfixtbl_avx2, &ff_emu_edge_hvar_avx2);
213 407764 }
214 #endif /* HAVE_AVX2_EXTERNAL */
215
216 void ff_prefetch_mmxext(const uint8_t *buf, ptrdiff_t stride, int h);
217
218 3711 av_cold void ff_videodsp_init_x86(VideoDSPContext *ctx, int bpc)
219 {
220 3711 int cpu_flags = av_get_cpu_flags();
221
222
2/2
✓ Branch 0 taken 261 times.
✓ Branch 1 taken 3450 times.
3711 if (EXTERNAL_MMXEXT(cpu_flags)) {
223 261 ctx->prefetch = ff_prefetch_mmxext;
224 }
225
4/4
✓ Branch 0 taken 259 times.
✓ Branch 1 taken 3452 times.
✓ Branch 2 taken 232 times.
✓ Branch 3 taken 27 times.
3711 if (EXTERNAL_SSE2(cpu_flags) && bpc <= 8) {
226 232 ctx->emulated_edge_mc = emulated_edge_mc_sse2;
227 }
228 #if HAVE_AVX2_EXTERNAL
229
4/4
✓ Branch 0 taken 251 times.
✓ Branch 1 taken 3460 times.
✓ Branch 2 taken 224 times.
✓ Branch 3 taken 27 times.
3711 if (EXTERNAL_AVX2(cpu_flags) && bpc <= 8) {
230 224 ctx->emulated_edge_mc = emulated_edge_mc_avx2;
231 }
232 #endif
233 3711 }
234