FFmpeg
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 
28 #include "libavutil/attributes.h"
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 
36 void ff_put_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
37  ptrdiff_t line_size, int h);
38 void ff_put_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels,
39  ptrdiff_t line_size, int h);
40 void ff_avg_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels,
41  ptrdiff_t line_size, int h);
42 void ff_put_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
43  ptrdiff_t line_size, int h);
44 void ff_avg_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
45  ptrdiff_t line_size, int h);
46 void ff_put_no_rnd_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
47  ptrdiff_t line_size, int h);
48 void ff_put_no_rnd_pixels8_x2_exact_mmxext(uint8_t *block,
49  const uint8_t *pixels,
50  ptrdiff_t line_size, int h);
51 void ff_put_no_rnd_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels,
52  ptrdiff_t line_size, int h);
53 void ff_avg_no_rnd_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels,
54  ptrdiff_t line_size, int h);
55 void ff_put_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
56  ptrdiff_t line_size, int h);
57 void ff_put_no_rnd_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
58  ptrdiff_t line_size, int h);
59 void ff_put_no_rnd_pixels8_y2_exact_mmxext(uint8_t *block,
60  const uint8_t *pixels,
61  ptrdiff_t line_size, int h);
62 void ff_put_no_rnd_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
63  ptrdiff_t line_size, int h);
64 void ff_avg_no_rnd_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
65  ptrdiff_t line_size, int h);
66 void ff_put_no_rnd_pixels8_xy2_ssse3(uint8_t *block, const uint8_t *pixels,
67  ptrdiff_t line_size, int h);
68 void ff_put_no_rnd_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels,
69  ptrdiff_t line_size, int h);
70 void ff_avg_no_rnd_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels,
71  ptrdiff_t line_size, int h);
72 void ff_avg_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
73  ptrdiff_t line_size, int h);
74 void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
75  ptrdiff_t line_size, int h);
76 
77 static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags)
78 {
79 #if HAVE_MMXEXT_EXTERNAL
80  c->put_pixels_tab[1][1] = ff_put_pixels8_x2_mmxext;
81  c->put_pixels_tab[1][2] = ff_put_pixels8_y2_mmxext;
82 
83  c->avg_pixels_tab[1][0] = ff_avg_pixels8_mmxext;
84  c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_mmxext;
85  c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_mmxext;
86 
87  c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_exact_mmxext;
88  c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_exact_mmxext;
89 
90  if (!(flags & AV_CODEC_FLAG_BITEXACT)) {
91  c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_mmxext;
92  c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_mmxext;
93  }
94 #endif /* HAVE_MMXEXT_EXTERNAL */
95 }
96 
97 static void hpeldsp_init_sse2(HpelDSPContext *c, int flags)
98 {
99 #if HAVE_SSE2_EXTERNAL
100  c->put_pixels_tab[0][0] = ff_put_pixels16_sse2;
101  c->put_pixels_tab[0][1] = ff_put_pixels16_x2_sse2;
102  c->put_pixels_tab[0][2] = ff_put_pixels16_y2_sse2;
103  c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_sse2;
104 
105  c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2;
106  c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_sse2;
107  c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_sse2;
108  c->put_no_rnd_pixels_tab[0][3] = ff_put_no_rnd_pixels16_xy2_sse2;
109 
110  c->put_no_rnd_pixels_tab[1][0] =
111  c->put_pixels_tab[1][0] = ff_put_pixels8_sse2;
112 
113  c->avg_pixels_tab[0][0] = ff_avg_pixels16_sse2;
114  c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_sse2;
115  c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_sse2;
116  c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_sse2;
117 
118  c->avg_no_rnd_pixels_tab[0] = ff_avg_pixels16_sse2;
119  c->avg_no_rnd_pixels_tab[1] = ff_avg_no_rnd_pixels16_x2_sse2;
120  c->avg_no_rnd_pixels_tab[2] = ff_avg_no_rnd_pixels16_y2_sse2;
121  c->avg_no_rnd_pixels_tab[3] = ff_avg_no_rnd_pixels16_xy2_sse2;
122 #endif /* HAVE_SSE2_EXTERNAL */
123 }
124 
125 static void hpeldsp_init_ssse3(HpelDSPContext *c, int flags)
126 {
127 #if HAVE_SSSE3_EXTERNAL
128  c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_ssse3;
129  c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_ssse3;
130  c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_ssse3;
131  c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_ssse3;
132 
133  c->put_no_rnd_pixels_tab[1][3] = ff_put_no_rnd_pixels8_xy2_ssse3;
134 #endif
135 }
136 
137 av_cold void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)
138 {
139  int cpu_flags = av_get_cpu_flags();
140 
141  if (EXTERNAL_MMXEXT(cpu_flags))
142  hpeldsp_init_mmxext(c, flags);
143 
144  if (EXTERNAL_SSE2(cpu_flags))
145  hpeldsp_init_sse2(c, flags);
146 
147  if (EXTERNAL_SSSE3(cpu_flags))
148  hpeldsp_init_ssse3(c, flags);
149 }
flags
const SwsFlags flags[]
Definition: swscale.c:61
cpu.h
ff_put_pixels16_y2_sse2
void ff_put_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_put_no_rnd_pixels8_x2_exact_mmxext
void ff_put_no_rnd_pixels8_x2_exact_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_put_pixels8_xy2_ssse3
void ff_put_pixels8_xy2_ssse3(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:109
cpu_flags
static atomic_int cpu_flags
Definition: cpu.c:56
hpeldsp_init_ssse3
static void hpeldsp_init_ssse3(HpelDSPContext *c, int flags)
Definition: hpeldsp_init.c:125
ff_put_no_rnd_pixels8_y2_exact_mmxext
void ff_put_no_rnd_pixels8_y2_exact_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
hpeldsp_init_sse2
static void hpeldsp_init_sse2(HpelDSPContext *c, int flags)
Definition: hpeldsp_init.c:97
ff_hpeldsp_init_x86
av_cold void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)
Definition: hpeldsp_init.c:137
ff_avg_pixels8_y2_mmxext
void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_put_no_rnd_pixels16_xy2_sse2
void ff_put_no_rnd_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
av_cold
#define av_cold
Definition: attributes.h:106
ff_avg_pixels8_x2_mmxext
void ff_avg_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
hpeldsp_init_mmxext
static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags)
Definition: hpeldsp_init.c:77
ff_avg_pixels8_mmxext
void ff_avg_pixels8_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_put_pixels8_sse2
void ff_put_pixels8_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_put_pixels16_sse2
void ff_put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_put_no_rnd_pixels16_y2_sse2
void ff_put_no_rnd_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_avg_pixels16_y2_sse2
void ff_avg_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_avg_no_rnd_pixels16_x2_sse2
void ff_avg_no_rnd_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_avg_pixels16_xy2_sse2
void ff_avg_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
c
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Definition: undefined.txt:32
ff_put_pixels16_xy2_ssse3
void ff_put_pixels16_xy2_ssse3(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
HpelDSPContext
Half-pel DSP context.
Definition: hpeldsp.h:46
ff_put_pixels16_x2_sse2
void ff_put_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_avg_pixels16_sse2
void ff_avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
cpu.h
ff_put_no_rnd_pixels16_x2_sse2
void ff_put_no_rnd_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
attributes.h
EXTERNAL_SSE2
#define EXTERNAL_SSE2(flags)
Definition: cpu.h:59
ff_put_pixels16_xy2_sse2
void ff_put_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_put_pixels8_x2_mmxext
void ff_put_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
avcodec.h
ff_avg_pixels16_xy2_ssse3
void ff_avg_pixels16_xy2_ssse3(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_avg_no_rnd_pixels16_xy2_sse2
void ff_avg_no_rnd_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_avg_pixels16_x2_sse2
void ff_avg_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_put_pixels8_y2_mmxext
void ff_put_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
fpel.h
AV_CODEC_FLAG_BITEXACT
#define AV_CODEC_FLAG_BITEXACT
Use only bitexact stuff (except (I)DCT).
Definition: avcodec.h:322
ff_put_no_rnd_pixels8_y2_mmxext
void ff_put_no_rnd_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_avg_no_rnd_pixels16_y2_sse2
void ff_avg_no_rnd_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
hpeldsp.h
block
The exact code depends on how similar the blocks are and how related they are to the block
Definition: filter_design.txt:207
ff_put_no_rnd_pixels8_xy2_ssse3
void ff_put_no_rnd_pixels8_xy2_ssse3(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_avg_pixels8_xy2_ssse3
void ff_avg_pixels8_xy2_ssse3(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
h
h
Definition: vp9dsp_template.c:2070
EXTERNAL_SSSE3
#define EXTERNAL_SSSE3(flags)
Definition: cpu.h:65
ff_put_no_rnd_pixels8_x2_mmxext
void ff_put_no_rnd_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
EXTERNAL_MMXEXT
#define EXTERNAL_MMXEXT(flags)
Definition: cpu.h:57
Generated on Wed Oct 22 2025 19:22:42 for FFmpeg by   doxygen 1.8.17