swscale/x86/rgb2rgb: add support for AVX
[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"
5a6e3c03 29#include "libavutil/attributes.h"
c318626c 30#include "libavutil/x86/asm.h"
e0c6cce4 31#include "libavutil/x86/cpu.h"
e66149e7 32#include "libavutil/cpu.h"
c0038328
LB
33#include "libavutil/bswap.h"
34#include "libswscale/rgb2rgb.h"
35#include "libswscale/swscale.h"
36#include "libswscale/swscale_internal.h"
37
b2668c85
RB
38#if HAVE_INLINE_ASM
39
c0038328
LB
40DECLARE_ASM_CONST(8, uint64_t, mmx_ff) = 0x00000000000000FFULL;
41DECLARE_ASM_CONST(8, uint64_t, mmx_null) = 0x0000000000000000ULL;
42DECLARE_ASM_CONST(8, uint64_t, mmx_one) = 0xFFFFFFFFFFFFFFFFULL;
43DECLARE_ASM_CONST(8, uint64_t, mask32b) = 0x000000FF000000FFULL;
44DECLARE_ASM_CONST(8, uint64_t, mask32g) = 0x0000FF000000FF00ULL;
45DECLARE_ASM_CONST(8, uint64_t, mask32r) = 0x00FF000000FF0000ULL;
46DECLARE_ASM_CONST(8, uint64_t, mask32a) = 0xFF000000FF000000ULL;
47DECLARE_ASM_CONST(8, uint64_t, mask32) = 0x00FFFFFF00FFFFFFULL;
48DECLARE_ASM_CONST(8, uint64_t, mask3216br) = 0x00F800F800F800F8ULL;
49DECLARE_ASM_CONST(8, uint64_t, mask3216g) = 0x0000FC000000FC00ULL;
50DECLARE_ASM_CONST(8, uint64_t, mask3215g) = 0x0000F8000000F800ULL;
51DECLARE_ASM_CONST(8, uint64_t, mul3216) = 0x2000000420000004ULL;
52DECLARE_ASM_CONST(8, uint64_t, mul3215) = 0x2000000820000008ULL;
53DECLARE_ASM_CONST(8, uint64_t, mask24b) = 0x00FF0000FF0000FFULL;
54DECLARE_ASM_CONST(8, uint64_t, mask24g) = 0xFF0000FF0000FF00ULL;
55DECLARE_ASM_CONST(8, uint64_t, mask24r) = 0x0000FF0000FF0000ULL;
56DECLARE_ASM_CONST(8, uint64_t, mask24l) = 0x0000000000FFFFFFULL;
57DECLARE_ASM_CONST(8, uint64_t, mask24h) = 0x0000FFFFFF000000ULL;
58DECLARE_ASM_CONST(8, uint64_t, mask24hh) = 0xffff000000000000ULL;
59DECLARE_ASM_CONST(8, uint64_t, mask24hhh) = 0xffffffff00000000ULL;
60DECLARE_ASM_CONST(8, uint64_t, mask24hhhh) = 0xffffffffffff0000ULL;
61DECLARE_ASM_CONST(8, uint64_t, mask15b) = 0x001F001F001F001FULL; /* 00000000 00011111 xxB */
62DECLARE_ASM_CONST(8, uint64_t, mask15rg) = 0x7FE07FE07FE07FE0ULL; /* 01111111 11100000 RGx */
63DECLARE_ASM_CONST(8, uint64_t, mask15s) = 0xFFE0FFE0FFE0FFE0ULL;
64DECLARE_ASM_CONST(8, uint64_t, mask15g) = 0x03E003E003E003E0ULL;
65DECLARE_ASM_CONST(8, uint64_t, mask15r) = 0x7C007C007C007C00ULL;
66#define mask16b mask15b
67DECLARE_ASM_CONST(8, uint64_t, mask16g) = 0x07E007E007E007E0ULL;
68DECLARE_ASM_CONST(8, uint64_t, mask16r) = 0xF800F800F800F800ULL;
69DECLARE_ASM_CONST(8, uint64_t, red_16mask) = 0x0000f8000000f800ULL;
70DECLARE_ASM_CONST(8, uint64_t, green_16mask) = 0x000007e0000007e0ULL;
71DECLARE_ASM_CONST(8, uint64_t, blue_16mask) = 0x0000001f0000001fULL;
72DECLARE_ASM_CONST(8, uint64_t, red_15mask) = 0x00007c0000007c00ULL;
73DECLARE_ASM_CONST(8, uint64_t, green_15mask) = 0x000003e0000003e0ULL;
74DECLARE_ASM_CONST(8, uint64_t, blue_15mask) = 0x0000001f0000001fULL;
75
76#define RGB2YUV_SHIFT 8
77#define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
78#define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
79#define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
80#define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
81#define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
82#define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
83#define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
84#define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
85#define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
86
652f5185 87// Note: We have C, MMX, MMXEXT, 3DNOW versions, there is no 3DNOW + MMXEXT one.
c0038328 88
239fdf1b 89#define COMPILE_TEMPLATE_MMXEXT 0
c0038328
LB
90#define COMPILE_TEMPLATE_AMD3DNOW 0
91#define COMPILE_TEMPLATE_SSE2 0
7597e6ef 92#define COMPILE_TEMPLATE_AVX 0
c0038328
LB
93
94//MMX versions
95#undef RENAME
c16bfb14 96#define RENAME(a) a ## _mmx
c0038328
LB
97#include "rgb2rgb_template.c"
98
652f5185 99// MMXEXT versions
c0038328 100#undef RENAME
239fdf1b
DB
101#undef COMPILE_TEMPLATE_MMXEXT
102#define COMPILE_TEMPLATE_MMXEXT 1
c16bfb14 103#define RENAME(a) a ## _mmxext
c0038328
LB
104#include "rgb2rgb_template.c"
105
106//SSE2 versions
107#undef RENAME
108#undef COMPILE_TEMPLATE_SSE2
109#define COMPILE_TEMPLATE_SSE2 1
c16bfb14 110#define RENAME(a) a ## _sse2
c0038328
LB
111#include "rgb2rgb_template.c"
112
7597e6ef
MN
113//AVX versions
114#undef RENAME
115#undef COMPILE_TEMPLATE_AVX
116#define COMPILE_TEMPLATE_AVX 1
117#define RENAME(a) a ## _avx
118#include "rgb2rgb_template.c"
119
c0038328
LB
120//3DNOW versions
121#undef RENAME
239fdf1b 122#undef COMPILE_TEMPLATE_MMXEXT
c0038328 123#undef COMPILE_TEMPLATE_SSE2
7597e6ef 124#undef COMPILE_TEMPLATE_AVX
c0038328 125#undef COMPILE_TEMPLATE_AMD3DNOW
239fdf1b 126#define COMPILE_TEMPLATE_MMXEXT 0
522d65ba 127#define COMPILE_TEMPLATE_SSE2 0
c0038328 128#define COMPILE_TEMPLATE_AMD3DNOW 1
c16bfb14 129#define RENAME(a) a ## _3dnow
c0038328
LB
130#include "rgb2rgb_template.c"
131
132/*
133 RGB15->RGB16 original by Strepto/Astral
134 ported to gcc & bugfixed : A'rpi
652f5185 135 MMXEXT, 3DNOW optimization by Nick Kurshev
c0038328
LB
136 32-bit C version, and and&add trick by Michael Niedermayer
137*/
138
b2668c85
RB
139#endif /* HAVE_INLINE_ASM */
140
5a6e3c03 141av_cold void rgb2rgb_init_x86(void)
c0038328 142{
b2668c85 143#if HAVE_INLINE_ASM
e66149e7
RB
144 int cpu_flags = av_get_cpu_flags();
145
e0c6cce4 146 if (INLINE_MMX(cpu_flags))
c16bfb14 147 rgb2rgb_init_mmx();
e0c6cce4 148 if (INLINE_AMD3DNOW(cpu_flags))
c16bfb14 149 rgb2rgb_init_3dnow();
e0c6cce4 150 if (INLINE_MMXEXT(cpu_flags))
c16bfb14 151 rgb2rgb_init_mmxext();
e0c6cce4 152 if (INLINE_SSE2(cpu_flags))
c16bfb14 153 rgb2rgb_init_sse2();
7597e6ef
MN
154 if (INLINE_AVX(cpu_flags))
155 rgb2rgb_init_avx();
b2668c85 156#endif /* HAVE_INLINE_ASM */
c0038328 157}