FFmpeg
Loading...
Searching...
No Matches
hpeldsp_init.c
Go to the documentation of this file.
1/*
2 * SIMD-optimized halfpel functions
3 * Copyright (c) 2000, 2001 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
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 * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
23 */
24
25#include <stddef.h>
26#include <stdint.h>
27
29#include "libavutil/cpu.h"
30#include "libavutil/x86/cpu.h"
31#include "libavcodec/avcodec.h"
32#include "libavcodec/hpeldsp.h"
33#include "fpel.h"
34#include "hpeldsp.h"
35
36void ff_put_pixels8_x2_sse2(uint8_t *block, const uint8_t *pixels,
37 ptrdiff_t line_size, int h);
38void ff_put_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels,
39 ptrdiff_t line_size, int h);
40void ff_avg_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels,
41 ptrdiff_t line_size, int h);
42void ff_put_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
43 ptrdiff_t line_size, int h);
44void ff_avg_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
45 ptrdiff_t line_size, int h);
46void ff_put_no_rnd_pixels8_x2_approx_sse2(uint8_t *block, const uint8_t *pixels,
47 ptrdiff_t line_size, int h);
48void ff_put_no_rnd_pixels8_x2_sse2(uint8_t *block, const uint8_t *pixels,
49 ptrdiff_t line_size, int h);
50void ff_put_no_rnd_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels,
51 ptrdiff_t line_size, int h);
52void ff_avg_no_rnd_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels,
53 ptrdiff_t line_size, int h);
54void ff_put_pixels8_y2_sse2(uint8_t *block, const uint8_t *pixels,
55 ptrdiff_t line_size, int h);
56void ff_put_no_rnd_pixels8_y2_approx_sse2(uint8_t *block, const uint8_t *pixels,
57 ptrdiff_t line_size, int h);
58void ff_put_no_rnd_pixels8_y2_sse2(uint8_t *block, const uint8_t *pixels,
59 ptrdiff_t line_size, int h);
60void ff_put_no_rnd_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
61 ptrdiff_t line_size, int h);
62void ff_avg_no_rnd_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
63 ptrdiff_t line_size, int h);
64void ff_put_no_rnd_pixels8_xy2_ssse3(uint8_t *block, const uint8_t *pixels,
65 ptrdiff_t line_size, int h);
66void ff_put_no_rnd_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels,
67 ptrdiff_t line_size, int h);
68void ff_avg_no_rnd_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels,
69 ptrdiff_t line_size, int h);
70void ff_avg_pixels8_x2_sse2(uint8_t *block, const uint8_t *pixels,
71 ptrdiff_t line_size, int h);
72void ff_avg_pixels8_y2_sse2(uint8_t *block, const uint8_t *pixels,
73 ptrdiff_t line_size, int h);
74
76{
77#if HAVE_SSE2_EXTERNAL
78 c->put_pixels_tab[0][0] = ff_put_pixels16_sse2;
79 c->put_pixels_tab[0][1] = ff_put_pixels16_x2_sse2;
80 c->put_pixels_tab[0][2] = ff_put_pixels16_y2_sse2;
81 c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_sse2;
82
83 c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2;
84 c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_sse2;
85 c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_sse2;
86 c->put_no_rnd_pixels_tab[0][3] = ff_put_no_rnd_pixels16_xy2_sse2;
87
88 c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_sse2;
89 c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_sse2;
90 c->put_no_rnd_pixels_tab[1][0] =
91 c->put_pixels_tab[1][0] = ff_put_pixels8_sse2;
92 c->put_pixels_tab[1][1] = ff_put_pixels8_x2_sse2;
93 c->put_pixels_tab[1][2] = ff_put_pixels8_y2_sse2;
94
95 c->avg_pixels_tab[0][0] = ff_avg_pixels16_sse2;
96 c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_sse2;
97 c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_sse2;
98 c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_sse2;
99 c->avg_pixels_tab[1][0] = ff_avg_pixels8_sse2;
100 c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_sse2;
101 c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_sse2;
102
103 c->avg_no_rnd_pixels_tab[0] = ff_avg_pixels16_sse2;
104 c->avg_no_rnd_pixels_tab[1] = ff_avg_no_rnd_pixels16_x2_sse2;
105 c->avg_no_rnd_pixels_tab[2] = ff_avg_no_rnd_pixels16_y2_sse2;
106 c->avg_no_rnd_pixels_tab[3] = ff_avg_no_rnd_pixels16_xy2_sse2;
107
108 if (!(flags & AV_CODEC_FLAG_BITEXACT)) {
109 c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_approx_sse2;
110 c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_approx_sse2;
111 }
112#endif /* HAVE_SSE2_EXTERNAL */
113}
114
116{
117#if HAVE_SSSE3_EXTERNAL
118 c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_ssse3;
119 c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_ssse3;
120 c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_ssse3;
121 c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_ssse3;
122
123 c->put_no_rnd_pixels_tab[1][3] = ff_put_no_rnd_pixels8_xy2_ssse3;
124#endif
125}
126
Libavcodec external API header.
#define flags(name, subs,...)
Definition cbs_h264.c:74
static int16_t block[64]
Definition dct.c:125
void ff_avg_pixels8_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_pixels8_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
#define AV_CODEC_FLAG_BITEXACT
Use only bitexact stuff (except (I)DCT).
Definition avcodec.h:322
Half-pel DSP functions.
static void hpeldsp_init_sse2(HpelDSPContext *c, int flags)
void ff_put_no_rnd_pixels8_x2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
av_cold void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)
static void hpeldsp_init_ssse3(HpelDSPContext *c, int flags)
void ff_avg_pixels8_y2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_no_rnd_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_avg_no_rnd_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_no_rnd_pixels8_xy2_ssse3(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_avg_pixels8_x2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_no_rnd_pixels8_x2_approx_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_avg_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_no_rnd_pixels8_y2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_pixels8_x2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_no_rnd_pixels8_y2_approx_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_pixels8_y2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_avg_no_rnd_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_avg_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_no_rnd_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_no_rnd_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_avg_no_rnd_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
Macro definitions for various function/variable attributes.
#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_SSSE3(flags)
Definition cpu.h:59
#define EXTERNAL_SSE2(flags)
Definition cpu.h:53
Half-pel DSP context.
Definition hpeldsp.h:46
static double c[64]
void ff_avg_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_pixels16_xy2_ssse3(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_put_pixels8_xy2_ssse3(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_avg_pixels16_xy2_ssse3(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
void ff_avg_pixels8_xy2_ssse3(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)