yading@11: /*
yading@11:  * software RGB to RGB converter
yading@11:  * pluralize by software PAL8 to RGB converter
yading@11:  *              software YUV to YUV converter
yading@11:  *              software YUV to RGB converter
yading@11:  * Written by Nick Kurshev.
yading@11:  * palette & YUV & runtime CPU stuff by Michael (michaelni@gmx.at)
yading@11:  *
yading@11:  * This file is part of FFmpeg.
yading@11:  *
yading@11:  * FFmpeg is free software; you can redistribute it and/or
yading@11:  * modify it under the terms of the GNU Lesser General Public
yading@11:  * License as published by the Free Software Foundation; either
yading@11:  * version 2.1 of the License, or (at your option) any later version.
yading@11:  *
yading@11:  * FFmpeg is distributed in the hope that it will be useful,
yading@11:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@11:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@11:  * Lesser General Public License for more details.
yading@11:  *
yading@11:  * You should have received a copy of the GNU Lesser General Public
yading@11:  * License along with FFmpeg; if not, write to the Free Software
yading@11:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@11:  */
yading@11: 
yading@11: #include <stdint.h>
yading@11: 
yading@11: #include "config.h"
yading@11: #include "libavutil/attributes.h"
yading@11: #include "libavutil/x86/asm.h"
yading@11: #include "libavutil/x86/cpu.h"
yading@11: #include "libavutil/cpu.h"
yading@11: #include "libavutil/bswap.h"
yading@11: #include "libswscale/rgb2rgb.h"
yading@11: #include "libswscale/swscale.h"
yading@11: #include "libswscale/swscale_internal.h"
yading@11: 
yading@11: #if HAVE_INLINE_ASM
yading@11: 
yading@11: DECLARE_ASM_CONST(8, uint64_t, mmx_ff)       = 0x00000000000000FFULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mmx_null)     = 0x0000000000000000ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mmx_one)      = 0xFFFFFFFFFFFFFFFFULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask32b)      = 0x000000FF000000FFULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask32g)      = 0x0000FF000000FF00ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask32r)      = 0x00FF000000FF0000ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask32a)      = 0xFF000000FF000000ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask32)       = 0x00FFFFFF00FFFFFFULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask3216br)   = 0x00F800F800F800F8ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask3216g)    = 0x0000FC000000FC00ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask3215g)    = 0x0000F8000000F800ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mul3216)      = 0x2000000420000004ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mul3215)      = 0x2000000820000008ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask24b)      = 0x00FF0000FF0000FFULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask24g)      = 0xFF0000FF0000FF00ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask24r)      = 0x0000FF0000FF0000ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask24l)      = 0x0000000000FFFFFFULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask24h)      = 0x0000FFFFFF000000ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask24hh)     = 0xffff000000000000ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask24hhh)    = 0xffffffff00000000ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask24hhhh)   = 0xffffffffffff0000ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask15b)      = 0x001F001F001F001FULL; /* 00000000 00011111  xxB */
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask15rg)     = 0x7FE07FE07FE07FE0ULL; /* 01111111 11100000  RGx */
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask15s)      = 0xFFE0FFE0FFE0FFE0ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask15g)      = 0x03E003E003E003E0ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask15r)      = 0x7C007C007C007C00ULL;
yading@11: #define mask16b mask15b
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask16g)      = 0x07E007E007E007E0ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mask16r)      = 0xF800F800F800F800ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, red_16mask)   = 0x0000f8000000f800ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, green_16mask) = 0x000007e0000007e0ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, blue_16mask)  = 0x0000001f0000001fULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, red_15mask)   = 0x00007c0000007c00ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, green_15mask) = 0x000003e0000003e0ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, blue_15mask)  = 0x0000001f0000001fULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mul15_mid)    = 0x4200420042004200ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mul15_hi)     = 0x0210021002100210ULL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, mul16_mid)    = 0x2080208020802080ULL;
yading@11: 
yading@11: #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
yading@11: #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
yading@11: #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
yading@11: #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
yading@11: #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
yading@11: #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
yading@11: #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
yading@11: #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
yading@11: #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
yading@11: 
yading@11: // Note: We have C, MMX, MMXEXT, 3DNOW versions, there is no 3DNOW + MMXEXT one.
yading@11: 
yading@11: #define COMPILE_TEMPLATE_MMXEXT 0
yading@11: #define COMPILE_TEMPLATE_AMD3DNOW 0
yading@11: #define COMPILE_TEMPLATE_SSE2 0
yading@11: 
yading@11: //MMX versions
yading@11: #undef RENAME
yading@11: #define RENAME(a) a ## _MMX
yading@11: #include "rgb2rgb_template.c"
yading@11: 
yading@11: // MMXEXT versions
yading@11: #undef RENAME
yading@11: #undef COMPILE_TEMPLATE_MMXEXT
yading@11: #define COMPILE_TEMPLATE_MMXEXT 1
yading@11: #define RENAME(a) a ## _MMXEXT
yading@11: #include "rgb2rgb_template.c"
yading@11: 
yading@11: //SSE2 versions
yading@11: #undef RENAME
yading@11: #undef COMPILE_TEMPLATE_SSE2
yading@11: #define COMPILE_TEMPLATE_SSE2 1
yading@11: #define RENAME(a) a ## _SSE2
yading@11: #include "rgb2rgb_template.c"
yading@11: 
yading@11: //3DNOW versions
yading@11: #undef RENAME
yading@11: #undef COMPILE_TEMPLATE_MMXEXT
yading@11: #undef COMPILE_TEMPLATE_SSE2
yading@11: #undef COMPILE_TEMPLATE_AMD3DNOW
yading@11: #define COMPILE_TEMPLATE_MMXEXT 0
yading@11: #define COMPILE_TEMPLATE_SSE2 0
yading@11: #define COMPILE_TEMPLATE_AMD3DNOW 1
yading@11: #define RENAME(a) a ## _3DNOW
yading@11: #include "rgb2rgb_template.c"
yading@11: 
yading@11: /*
yading@11:  RGB15->RGB16 original by Strepto/Astral
yading@11:  ported to gcc & bugfixed : A'rpi
yading@11:  MMXEXT, 3DNOW optimization by Nick Kurshev
yading@11:  32-bit C version, and and&add trick by Michael Niedermayer
yading@11: */
yading@11: 
yading@11: #endif /* HAVE_INLINE_ASM */
yading@11: 
yading@11: av_cold void rgb2rgb_init_x86(void)
yading@11: {
yading@11: #if HAVE_INLINE_ASM
yading@11:     int cpu_flags = av_get_cpu_flags();
yading@11: 
yading@11:     if (INLINE_MMX(cpu_flags))
yading@11:         rgb2rgb_init_MMX();
yading@11:     if (INLINE_AMD3DNOW(cpu_flags))
yading@11:         rgb2rgb_init_3DNOW();
yading@11:     if (INLINE_MMXEXT(cpu_flags))
yading@11:         rgb2rgb_init_MMXEXT();
yading@11:     if (INLINE_SSE2(cpu_flags))
yading@11:         rgb2rgb_init_SSE2();
yading@11: #endif /* HAVE_INLINE_ASM */
yading@11: }