swscale: move away x86 specific code from rgb2rgb
[libav.git] / libswscale / x86 / rgb2rgb.c
CommitLineData
c0038328
LB
1/*
2 * software RGB to RGB converter
3 * pluralize by software PAL8 to RGB converter
4 * software YUV to YUV converter
5 * software YUV to RGB converter
6 * Written by Nick Kurshev.
7 * palette & YUV & runtime CPU stuff by Michael (michaelni@gmx.at)
8 *
9 * This file is part of Libav.
10 *
11 * Libav is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public
13 * License as published by the Free Software Foundation; either
14 * version 2.1 of the License, or (at your option) any later version.
15 *
16 * Libav is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * Lesser General Public License for more details.
20 *
21 * You should have received a copy of the GNU Lesser General Public
22 * License along with Libav; if not, write to the Free Software
23 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 */
25
26#include <stdint.h>
27
28#include "config.h"
29#include "libavutil/x86_cpu.h"
30#include "libavutil/bswap.h"
31#include "libswscale/rgb2rgb.h"
32#include "libswscale/swscale.h"
33#include "libswscale/swscale_internal.h"
34
35DECLARE_ASM_CONST(8, uint64_t, mmx_ff) = 0x00000000000000FFULL;
36DECLARE_ASM_CONST(8, uint64_t, mmx_null) = 0x0000000000000000ULL;
37DECLARE_ASM_CONST(8, uint64_t, mmx_one) = 0xFFFFFFFFFFFFFFFFULL;
38DECLARE_ASM_CONST(8, uint64_t, mask32b) = 0x000000FF000000FFULL;
39DECLARE_ASM_CONST(8, uint64_t, mask32g) = 0x0000FF000000FF00ULL;
40DECLARE_ASM_CONST(8, uint64_t, mask32r) = 0x00FF000000FF0000ULL;
41DECLARE_ASM_CONST(8, uint64_t, mask32a) = 0xFF000000FF000000ULL;
42DECLARE_ASM_CONST(8, uint64_t, mask32) = 0x00FFFFFF00FFFFFFULL;
43DECLARE_ASM_CONST(8, uint64_t, mask3216br) = 0x00F800F800F800F8ULL;
44DECLARE_ASM_CONST(8, uint64_t, mask3216g) = 0x0000FC000000FC00ULL;
45DECLARE_ASM_CONST(8, uint64_t, mask3215g) = 0x0000F8000000F800ULL;
46DECLARE_ASM_CONST(8, uint64_t, mul3216) = 0x2000000420000004ULL;
47DECLARE_ASM_CONST(8, uint64_t, mul3215) = 0x2000000820000008ULL;
48DECLARE_ASM_CONST(8, uint64_t, mask24b) = 0x00FF0000FF0000FFULL;
49DECLARE_ASM_CONST(8, uint64_t, mask24g) = 0xFF0000FF0000FF00ULL;
50DECLARE_ASM_CONST(8, uint64_t, mask24r) = 0x0000FF0000FF0000ULL;
51DECLARE_ASM_CONST(8, uint64_t, mask24l) = 0x0000000000FFFFFFULL;
52DECLARE_ASM_CONST(8, uint64_t, mask24h) = 0x0000FFFFFF000000ULL;
53DECLARE_ASM_CONST(8, uint64_t, mask24hh) = 0xffff000000000000ULL;
54DECLARE_ASM_CONST(8, uint64_t, mask24hhh) = 0xffffffff00000000ULL;
55DECLARE_ASM_CONST(8, uint64_t, mask24hhhh) = 0xffffffffffff0000ULL;
56DECLARE_ASM_CONST(8, uint64_t, mask15b) = 0x001F001F001F001FULL; /* 00000000 00011111 xxB */
57DECLARE_ASM_CONST(8, uint64_t, mask15rg) = 0x7FE07FE07FE07FE0ULL; /* 01111111 11100000 RGx */
58DECLARE_ASM_CONST(8, uint64_t, mask15s) = 0xFFE0FFE0FFE0FFE0ULL;
59DECLARE_ASM_CONST(8, uint64_t, mask15g) = 0x03E003E003E003E0ULL;
60DECLARE_ASM_CONST(8, uint64_t, mask15r) = 0x7C007C007C007C00ULL;
61#define mask16b mask15b
62DECLARE_ASM_CONST(8, uint64_t, mask16g) = 0x07E007E007E007E0ULL;
63DECLARE_ASM_CONST(8, uint64_t, mask16r) = 0xF800F800F800F800ULL;
64DECLARE_ASM_CONST(8, uint64_t, red_16mask) = 0x0000f8000000f800ULL;
65DECLARE_ASM_CONST(8, uint64_t, green_16mask) = 0x000007e0000007e0ULL;
66DECLARE_ASM_CONST(8, uint64_t, blue_16mask) = 0x0000001f0000001fULL;
67DECLARE_ASM_CONST(8, uint64_t, red_15mask) = 0x00007c0000007c00ULL;
68DECLARE_ASM_CONST(8, uint64_t, green_15mask) = 0x000003e0000003e0ULL;
69DECLARE_ASM_CONST(8, uint64_t, blue_15mask) = 0x0000001f0000001fULL;
70
71#define RGB2YUV_SHIFT 8
72#define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
73#define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
74#define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
75#define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
76#define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
77#define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
78#define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
79#define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
80#define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
81
82//Note: We have C, MMX, MMX2, 3DNOW versions, there is no 3DNOW + MMX2 one.
83
84#define COMPILE_TEMPLATE_MMX 0
85#define COMPILE_TEMPLATE_MMX2 0
86#define COMPILE_TEMPLATE_AMD3DNOW 0
87#define COMPILE_TEMPLATE_SSE2 0
88
89//MMX versions
90#undef RENAME
91#undef COMPILE_TEMPLATE_MMX
92#define COMPILE_TEMPLATE_MMX 1
93#define RENAME(a) a ## _MMX
94#include "rgb2rgb_template.c"
95
96//MMX2 versions
97#undef RENAME
98#undef COMPILE_TEMPLATE_MMX2
99#define COMPILE_TEMPLATE_MMX2 1
100#define RENAME(a) a ## _MMX2
101#include "rgb2rgb_template.c"
102
103//SSE2 versions
104#undef RENAME
105#undef COMPILE_TEMPLATE_SSE2
106#define COMPILE_TEMPLATE_SSE2 1
107#define RENAME(a) a ## _SSE2
108#include "rgb2rgb_template.c"
109
110//3DNOW versions
111#undef RENAME
112#undef COMPILE_TEMPLATE_MMX2
113#undef COMPILE_TEMPLATE_SSE2
114#undef COMPILE_TEMPLATE_AMD3DNOW
115#define COMPILE_TEMPLATE_MMX2 0
116#define COMPILE_TEMPLATE_SSE2 1
117#define COMPILE_TEMPLATE_AMD3DNOW 1
118#define RENAME(a) a ## _3DNOW
119#include "rgb2rgb_template.c"
120
121/*
122 RGB15->RGB16 original by Strepto/Astral
123 ported to gcc & bugfixed : A'rpi
124 MMX2, 3DNOW optimization by Nick Kurshev
125 32-bit C version, and and&add trick by Michael Niedermayer
126*/
127
128void rgb2rgb_init_x86(int flags)
129{
130#if HAVE_MMX2 || HAVE_AMD3DNOW || HAVE_MMX
131 if (flags & SWS_CPU_CAPS_SSE2)
132 rgb2rgb_init_SSE2();
133 else if (flags & SWS_CPU_CAPS_MMX2)
134 rgb2rgb_init_MMX2();
135 else if (flags & SWS_CPU_CAPS_3DNOW)
136 rgb2rgb_init_3DNOW();
137 else if (flags & SWS_CPU_CAPS_MMX)
138 rgb2rgb_init_MMX();
139#endif /* HAVE_MMX2 || HAVE_AMD3DNOW || HAVE_MMX */
140}