FFmpeg
Loading...
Searching...
No Matches
videodsp_init.c
Go to the documentation of this file.
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"
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#include "videodsp.h"
31
32typedef void emu_edge_vfix_func(uint8_t *dst, x86_reg dst_stride,
33 const uint8_t *src, x86_reg src_stride,
34 x86_reg start_y, x86_reg end_y, x86_reg bh);
35typedef void emu_edge_vvar_func(uint8_t *dst, x86_reg dst_stride,
36 const uint8_t *src, x86_reg src_stride,
37 x86_reg start_y, x86_reg end_y, x86_reg bh,
38 x86_reg w);
39
73
74typedef void emu_edge_hfix_func(uint8_t *dst, x86_reg dst_stride,
75 x86_reg start_x, x86_reg bh);
76typedef void emu_edge_hvar_func(uint8_t *dst, x86_reg dst_stride,
77 x86_reg start_x, x86_reg n_words, x86_reg bh);
78
97#if HAVE_AVX2_EXTERNAL
98extern emu_edge_hfix_func ff_emu_edge_hfix8_avx2;
99extern emu_edge_hfix_func ff_emu_edge_hfix10_avx2;
100extern emu_edge_hfix_func ff_emu_edge_hfix12_avx2;
101extern emu_edge_hfix_func ff_emu_edge_hfix14_avx2;
102extern emu_edge_hfix_func ff_emu_edge_hfix16_avx2;
103extern emu_edge_hfix_func ff_emu_edge_hfix18_avx2;
104extern emu_edge_hfix_func ff_emu_edge_hfix20_avx2;
105extern emu_edge_hfix_func ff_emu_edge_hfix22_avx2;
106static emu_edge_hfix_func * const hfixtbl_avx2[11] = {
108 ff_emu_edge_hfix8_avx2, ff_emu_edge_hfix10_avx2, ff_emu_edge_hfix12_avx2,
109 ff_emu_edge_hfix14_avx2, ff_emu_edge_hfix16_avx2, ff_emu_edge_hfix18_avx2,
110 ff_emu_edge_hfix20_avx2, ff_emu_edge_hfix22_avx2
111};
112extern emu_edge_hvar_func ff_emu_edge_hvar_avx2;
113#endif
114
115static av_always_inline void emulated_edge_mc(uint8_t *dst, const uint8_t *src,
116 ptrdiff_t dst_stride,
117 ptrdiff_t src_stride,
118 x86_reg block_w, x86_reg block_h,
119 x86_reg src_x, x86_reg src_y,
121 emu_edge_vfix_func * const *vfix_tbl,
122 emu_edge_vvar_func *v_extend_var,
123 emu_edge_hfix_func * const *hfix_tbl,
124 emu_edge_hvar_func *h_extend_var)
125{
126 x86_reg start_y, start_x, end_y, end_x, src_y_add = 0, p;
127
128 if (!w || !h)
129 return;
130
131 av_assert2(block_w <= FFABS(dst_stride));
132
133 if (src_y >= h) {
134 src -= src_y*src_stride;
135 src_y_add = h - 1;
136 src_y = h - 1;
137 } else if (src_y <= -block_h) {
138 src -= src_y*src_stride;
139 src_y_add = 1 - block_h;
140 src_y = 1 - block_h;
141 }
142 if (src_x >= w) {
143 src += w - 1 - src_x;
144 src_x = w - 1;
145 } else if (src_x <= -block_w) {
146 src += 1 - block_w - src_x;
147 src_x = 1 - block_w;
148 }
149
150 start_y = FFMAX(0, -src_y);
151 start_x = FFMAX(0, -src_x);
152 end_y = FFMIN(block_h, h-src_y);
153 end_x = FFMIN(block_w, w-src_x);
154 av_assert2(start_x < end_x && block_w > 0);
155 av_assert2(start_y < end_y && block_h > 0);
156
157 // fill in the to-be-copied part plus all above/below
158 src += (src_y_add + start_y) * src_stride + start_x;
159 w = end_x - start_x;
160 if (w <= 22) {
161 vfix_tbl[w - 1](dst + start_x, dst_stride, src, src_stride,
162 start_y, end_y, block_h);
163 } else {
164 v_extend_var(dst + start_x, dst_stride, src, src_stride,
165 start_y, end_y, block_h, w);
166 }
167
168 // fill left
169 if (start_x) {
170 if (start_x <= 22) {
171 hfix_tbl[(start_x - 1) >> 1](dst, dst_stride, start_x, block_h);
172 } else {
173 h_extend_var(dst, dst_stride,
174 start_x, (start_x + 1) >> 1, block_h);
175 }
176 }
177
178 // fill right
179 p = block_w - end_x;
180 if (p) {
181 if (p <= 22) {
182 hfix_tbl[(p - 1) >> 1](dst + end_x - (p & 1), dst_stride,
183 -!(p & 1), block_h);
184 } else {
185 h_extend_var(dst + end_x - (p & 1), dst_stride,
186 -!(p & 1), (p + 1) >> 1, block_h);
187 }
188 }
189}
190
191void ff_emulated_edge_mc_sse2(uint8_t *buf, const uint8_t *src,
192 ptrdiff_t buf_stride,
193 ptrdiff_t src_stride,
194 int block_w, int block_h,
195 int src_x, int src_y, int w,
196 int h)
197{
198 emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
199 src_x, src_y, w, h, vfixtbl_sse2, &ff_emu_edge_vvar_sse,
201}
202
203#if HAVE_AVX2_EXTERNAL
204static av_noinline void emulated_edge_mc_avx2(uint8_t *buf, const uint8_t *src,
205 ptrdiff_t buf_stride,
206 ptrdiff_t src_stride,
207 int block_w, int block_h,
208 int src_x, int src_y, int w,
209 int h)
210{
211 emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
212 src_x, src_y, w, h, vfixtbl_sse2, &ff_emu_edge_vvar_sse,
213 hfixtbl_avx2, &ff_emu_edge_hvar_avx2);
214}
215#endif /* HAVE_AVX2_EXTERNAL */
216
217void ff_prefetch_mmxext(const uint8_t *buf, ptrdiff_t stride, int h);
218
220{
222
224 ctx->prefetch = ff_prefetch_mmxext;
225 }
226 if (EXTERNAL_SSE2(cpu_flags) && bpc <= 8) {
227 ctx->emulated_edge_mc = ff_emulated_edge_mc_sse2;
228 }
229#if HAVE_AVX2_EXTERNAL
230 if (EXTERNAL_AVX2(cpu_flags) && bpc <= 8) {
231 ctx->emulated_edge_mc = emulated_edge_mc_avx2;
232 }
233#endif
234}
uint8_t ptrdiff_t const uint8_t ptrdiff_t int intptr_t intptr_t int int16_t * dst
Definition dsp.h:87
static AVFormatContext * ctx
int x86_reg
Definition asm.h:71
simple assert() macros that are a bit more flexible than ISO C assert().
#define av_assert2(cond)
assert() equivalent, that does lie in speed critical code.
Definition avassert.h:68
common internal and external API header
#define FFABS(a)
Absolute value, Note, INT_MIN / INT64_MIN result in undefined behavior as they are not representable ...
Definition common.h:74
Macro definitions for various function/variable attributes.
#define av_always_inline
Definition attributes.h:72
#define av_noinline
Definition attributes.h:97
#define av_cold
Definition attributes.h:117
static atomic_int cpu_flags
Definition cpu.c:56
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition cpu.c:109
#define EXTERNAL_MMXEXT(flags)
Definition cpu.h:51
#define EXTERNAL_AVX2(flags)
Definition cpu.h:72
#define EXTERNAL_SSE2(flags)
Definition cpu.h:53
uint8_t w
Definition llvidencdsp.c:39
#define FFMIN(a, b)
Definition macros.h:49
#define FFMAX(a, b)
Definition macros.h:47
#define stride
#define src
Definition vp8dsp.c:248
Core video DSP helper functions.
void ff_emulated_edge_mc_sse2(uint8_t *buf, const uint8_t *src, ptrdiff_t buf_stride, ptrdiff_t src_stride, int block_w, int block_h, int src_x, int src_y, int w, int h)
emu_edge_vfix_func ff_emu_edge_vfix7_sse2
emu_edge_hfix_func ff_emu_edge_hfix10_sse2
emu_edge_vfix_func ff_emu_edge_vfix6_sse2
emu_edge_hfix_func ff_emu_edge_hfix18_sse2
emu_edge_vfix_func ff_emu_edge_vfix11_sse2
emu_edge_hfix_func ff_emu_edge_hfix8_sse2
void emu_edge_hvar_func(uint8_t *dst, x86_reg dst_stride, x86_reg start_x, x86_reg n_words, x86_reg bh)
emu_edge_hfix_func ff_emu_edge_hfix16_sse2
void emu_edge_vfix_func(uint8_t *dst, x86_reg dst_stride, const uint8_t *src, x86_reg src_stride, x86_reg start_y, x86_reg end_y, x86_reg bh)
void ff_emulated_edge_mc_sse2(uint8_t *buf, const uint8_t *src, ptrdiff_t buf_stride, ptrdiff_t src_stride, int block_w, int block_h, int src_x, int src_y, int w, int h)
emu_edge_vfix_func ff_emu_edge_vfix21_sse2
emu_edge_hfix_func ff_emu_edge_hfix14_sse2
static emu_edge_hfix_func *const hfixtbl_sse2[11]
void ff_prefetch_mmxext(const uint8_t *buf, ptrdiff_t stride, int h)
emu_edge_vfix_func ff_emu_edge_vfix5_sse2
emu_edge_vfix_func ff_emu_edge_vfix20_sse2
emu_edge_vfix_func ff_emu_edge_vfix10_sse2
emu_edge_vfix_func ff_emu_edge_vfix15_sse2
emu_edge_hfix_func ff_emu_edge_hfix2_sse2
emu_edge_vfix_func ff_emu_edge_vfix13_sse2
emu_edge_hfix_func ff_emu_edge_hfix4_sse2
emu_edge_vfix_func ff_emu_edge_vfix22_sse2
emu_edge_hfix_func ff_emu_edge_hfix6_sse2
emu_edge_vvar_func ff_emu_edge_vvar_sse
emu_edge_hfix_func ff_emu_edge_hfix12_sse2
emu_edge_vfix_func ff_emu_edge_vfix1_sse2
emu_edge_vfix_func ff_emu_edge_vfix19_sse2
emu_edge_vfix_func ff_emu_edge_vfix8_sse2
emu_edge_hvar_func ff_emu_edge_hvar_sse2
emu_edge_vfix_func ff_emu_edge_vfix4_sse2
emu_edge_hfix_func ff_emu_edge_hfix22_sse2
static av_always_inline void emulated_edge_mc(uint8_t *dst, const uint8_t *src, ptrdiff_t dst_stride, ptrdiff_t src_stride, x86_reg block_w, x86_reg block_h, x86_reg src_x, x86_reg src_y, x86_reg w, x86_reg h, emu_edge_vfix_func *const *vfix_tbl, emu_edge_vvar_func *v_extend_var, emu_edge_hfix_func *const *hfix_tbl, emu_edge_hvar_func *h_extend_var)
static emu_edge_vfix_func *const vfixtbl_sse2[22]
emu_edge_vfix_func ff_emu_edge_vfix2_sse2
emu_edge_vfix_func ff_emu_edge_vfix3_sse2
emu_edge_hfix_func ff_emu_edge_hfix20_sse2
emu_edge_vfix_func ff_emu_edge_vfix17_sse2
void emu_edge_hfix_func(uint8_t *dst, x86_reg dst_stride, x86_reg start_x, x86_reg bh)
emu_edge_vfix_func ff_emu_edge_vfix16_sse2
emu_edge_vfix_func ff_emu_edge_vfix18_sse2
emu_edge_vfix_func ff_emu_edge_vfix12_sse2
emu_edge_vfix_func ff_emu_edge_vfix9_sse2
void emu_edge_vvar_func(uint8_t *dst, x86_reg dst_stride, const uint8_t *src, x86_reg src_stride, x86_reg start_y, x86_reg end_y, x86_reg bh, x86_reg w)
emu_edge_vfix_func ff_emu_edge_vfix14_sse2
av_cold void ff_videodsp_init_x86(VideoDSPContext *ctx, int bpc)