| 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 FFmpeg. |
| 10 | * |
| 11 | * FFmpeg 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 | * FFmpeg 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 FFmpeg; 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/attributes.h" |
| 30 | #include "libavutil/x86/asm.h" |
| 31 | #include "libavutil/x86/cpu.h" |
| 32 | #include "libavutil/cpu.h" |
| 33 | #include "libavutil/bswap.h" |
| 34 | #include "libswscale/rgb2rgb.h" |
| 35 | #include "libswscale/swscale.h" |
| 36 | #include "libswscale/swscale_internal.h" |
| 37 | |
| 38 | #if HAVE_INLINE_ASM |
| 39 | |
| 40 | DECLARE_ASM_CONST(8, uint64_t, mmx_ff) = 0x00000000000000FFULL; |
| 41 | DECLARE_ASM_CONST(8, uint64_t, mmx_null) = 0x0000000000000000ULL; |
| 42 | DECLARE_ASM_CONST(8, uint64_t, mmx_one) = 0xFFFFFFFFFFFFFFFFULL; |
| 43 | DECLARE_ASM_CONST(8, uint64_t, mask32b) = 0x000000FF000000FFULL; |
| 44 | DECLARE_ASM_CONST(8, uint64_t, mask32g) = 0x0000FF000000FF00ULL; |
| 45 | DECLARE_ASM_CONST(8, uint64_t, mask32r) = 0x00FF000000FF0000ULL; |
| 46 | DECLARE_ASM_CONST(8, uint64_t, mask32a) = 0xFF000000FF000000ULL; |
| 47 | DECLARE_ASM_CONST(8, uint64_t, mask32) = 0x00FFFFFF00FFFFFFULL; |
| 48 | DECLARE_ASM_CONST(8, uint64_t, mask3216br) = 0x00F800F800F800F8ULL; |
| 49 | DECLARE_ASM_CONST(8, uint64_t, mask3216g) = 0x0000FC000000FC00ULL; |
| 50 | DECLARE_ASM_CONST(8, uint64_t, mask3215g) = 0x0000F8000000F800ULL; |
| 51 | DECLARE_ASM_CONST(8, uint64_t, mul3216) = 0x2000000420000004ULL; |
| 52 | DECLARE_ASM_CONST(8, uint64_t, mul3215) = 0x2000000820000008ULL; |
| 53 | DECLARE_ASM_CONST(8, uint64_t, mask24b) = 0x00FF0000FF0000FFULL; |
| 54 | DECLARE_ASM_CONST(8, uint64_t, mask24g) = 0xFF0000FF0000FF00ULL; |
| 55 | DECLARE_ASM_CONST(8, uint64_t, mask24r) = 0x0000FF0000FF0000ULL; |
| 56 | DECLARE_ASM_CONST(8, uint64_t, mask24l) = 0x0000000000FFFFFFULL; |
| 57 | DECLARE_ASM_CONST(8, uint64_t, mask24h) = 0x0000FFFFFF000000ULL; |
| 58 | DECLARE_ASM_CONST(8, uint64_t, mask24hh) = 0xffff000000000000ULL; |
| 59 | DECLARE_ASM_CONST(8, uint64_t, mask24hhh) = 0xffffffff00000000ULL; |
| 60 | DECLARE_ASM_CONST(8, uint64_t, mask24hhhh) = 0xffffffffffff0000ULL; |
| 61 | DECLARE_ASM_CONST(8, uint64_t, mask15b) = 0x001F001F001F001FULL; /* 00000000 00011111 xxB */ |
| 62 | DECLARE_ASM_CONST(8, uint64_t, mask15rg) = 0x7FE07FE07FE07FE0ULL; /* 01111111 11100000 RGx */ |
| 63 | DECLARE_ASM_CONST(8, uint64_t, mask15s) = 0xFFE0FFE0FFE0FFE0ULL; |
| 64 | DECLARE_ASM_CONST(8, uint64_t, mask15g) = 0x03E003E003E003E0ULL; |
| 65 | DECLARE_ASM_CONST(8, uint64_t, mask15r) = 0x7C007C007C007C00ULL; |
| 66 | #define mask16b mask15b |
| 67 | DECLARE_ASM_CONST(8, uint64_t, mask16g) = 0x07E007E007E007E0ULL; |
| 68 | DECLARE_ASM_CONST(8, uint64_t, mask16r) = 0xF800F800F800F800ULL; |
| 69 | DECLARE_ASM_CONST(8, uint64_t, red_16mask) = 0x0000f8000000f800ULL; |
| 70 | DECLARE_ASM_CONST(8, uint64_t, green_16mask) = 0x000007e0000007e0ULL; |
| 71 | DECLARE_ASM_CONST(8, uint64_t, blue_16mask) = 0x0000001f0000001fULL; |
| 72 | DECLARE_ASM_CONST(8, uint64_t, red_15mask) = 0x00007c0000007c00ULL; |
| 73 | DECLARE_ASM_CONST(8, uint64_t, green_15mask) = 0x000003e0000003e0ULL; |
| 74 | DECLARE_ASM_CONST(8, uint64_t, blue_15mask) = 0x0000001f0000001fULL; |
| 75 | DECLARE_ASM_CONST(8, uint64_t, mul15_mid) = 0x4200420042004200ULL; |
| 76 | DECLARE_ASM_CONST(8, uint64_t, mul15_hi) = 0x0210021002100210ULL; |
| 77 | DECLARE_ASM_CONST(8, uint64_t, mul16_mid) = 0x2080208020802080ULL; |
| 78 | |
| 79 | DECLARE_ALIGNED(8, extern const uint64_t, ff_bgr2YOffset); |
| 80 | DECLARE_ALIGNED(8, extern const uint64_t, ff_w1111); |
| 81 | DECLARE_ALIGNED(8, extern const uint64_t, ff_bgr2UVOffset); |
| 82 | |
| 83 | #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5)) |
| 84 | #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5)) |
| 85 | #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5)) |
| 86 | #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5)) |
| 87 | #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5)) |
| 88 | #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5)) |
| 89 | #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5)) |
| 90 | #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5)) |
| 91 | #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5)) |
| 92 | |
| 93 | // Note: We have C, MMX, MMXEXT, 3DNOW versions, there is no 3DNOW + MMXEXT one. |
| 94 | |
| 95 | #define COMPILE_TEMPLATE_MMXEXT 0 |
| 96 | #define COMPILE_TEMPLATE_AMD3DNOW 0 |
| 97 | #define COMPILE_TEMPLATE_SSE2 0 |
| 98 | #define COMPILE_TEMPLATE_AVX 0 |
| 99 | |
| 100 | //MMX versions |
| 101 | #undef RENAME |
| 102 | #define RENAME(a) a ## _mmx |
| 103 | #include "rgb2rgb_template.c" |
| 104 | |
| 105 | // MMXEXT versions |
| 106 | #undef RENAME |
| 107 | #undef COMPILE_TEMPLATE_MMXEXT |
| 108 | #define COMPILE_TEMPLATE_MMXEXT 1 |
| 109 | #define RENAME(a) a ## _mmxext |
| 110 | #include "rgb2rgb_template.c" |
| 111 | |
| 112 | //SSE2 versions |
| 113 | #undef RENAME |
| 114 | #undef COMPILE_TEMPLATE_SSE2 |
| 115 | #define COMPILE_TEMPLATE_SSE2 1 |
| 116 | #define RENAME(a) a ## _sse2 |
| 117 | #include "rgb2rgb_template.c" |
| 118 | |
| 119 | //AVX versions |
| 120 | #undef RENAME |
| 121 | #undef COMPILE_TEMPLATE_AVX |
| 122 | #define COMPILE_TEMPLATE_AVX 1 |
| 123 | #define RENAME(a) a ## _avx |
| 124 | #include "rgb2rgb_template.c" |
| 125 | |
| 126 | //3DNOW versions |
| 127 | #undef RENAME |
| 128 | #undef COMPILE_TEMPLATE_MMXEXT |
| 129 | #undef COMPILE_TEMPLATE_SSE2 |
| 130 | #undef COMPILE_TEMPLATE_AVX |
| 131 | #undef COMPILE_TEMPLATE_AMD3DNOW |
| 132 | #define COMPILE_TEMPLATE_MMXEXT 0 |
| 133 | #define COMPILE_TEMPLATE_SSE2 0 |
| 134 | #define COMPILE_TEMPLATE_AVX 0 |
| 135 | #define COMPILE_TEMPLATE_AMD3DNOW 1 |
| 136 | #define RENAME(a) a ## _3dnow |
| 137 | #include "rgb2rgb_template.c" |
| 138 | |
| 139 | /* |
| 140 | RGB15->RGB16 original by Strepto/Astral |
| 141 | ported to gcc & bugfixed : A'rpi |
| 142 | MMXEXT, 3DNOW optimization by Nick Kurshev |
| 143 | 32-bit C version, and and&add trick by Michael Niedermayer |
| 144 | */ |
| 145 | |
| 146 | #endif /* HAVE_INLINE_ASM */ |
| 147 | |
| 148 | av_cold void rgb2rgb_init_x86(void) |
| 149 | { |
| 150 | #if HAVE_INLINE_ASM |
| 151 | int cpu_flags = av_get_cpu_flags(); |
| 152 | |
| 153 | if (INLINE_MMX(cpu_flags)) |
| 154 | rgb2rgb_init_mmx(); |
| 155 | if (INLINE_AMD3DNOW(cpu_flags)) |
| 156 | rgb2rgb_init_3dnow(); |
| 157 | if (INLINE_MMXEXT(cpu_flags)) |
| 158 | rgb2rgb_init_mmxext(); |
| 159 | if (INLINE_SSE2(cpu_flags)) |
| 160 | rgb2rgb_init_sse2(); |
| 161 | if (INLINE_AVX(cpu_flags)) |
| 162 | rgb2rgb_init_avx(); |
| 163 | #endif /* HAVE_INLINE_ASM */ |
| 164 | } |