swscale.c

/*
 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 *
 * the C code (not assembly, mmx, ...) of this file can be used
 * under the LGPL license too
 */

/*
  supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR32_1, BGR24, BGR16, BGR15, RGB32, RGB32_1, RGB24, Y8/Y800, YVU9/IF09, PAL8
  supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
  {BGR,RGB}{1,4,8,15,16} support dithering

  unscaled special converters (YV12=I420=IYUV, Y800=Y8)
  YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
  x -> x
  YUV9 -> YV12
  YUV9/YV12 -> Y800
  Y800 -> YUV9/YV12
  BGR24 -> BGR32 & RGB24 -> RGB32
  BGR32 -> BGR24 & RGB32 -> RGB24
  BGR15 -> BGR16
*/

/*
tested special converters (most are tested actually, but I did not write it down ...)
 YV12 -> BGR16
 YV12 -> YV12
 BGR15 -> BGR16
 BGR16 -> BGR16
 YVU9 -> YV12

untested special converters
  YV12/I420 -> BGR15/BGR24/BGR32 (it is the yuv2rgb stuff, so it should be ok)
  YV12/I420 -> YV12/I420
  YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
  BGR24 -> BGR32 & RGB24 -> RGB32
  BGR32 -> BGR24 & RGB32 -> RGB24
  BGR24 -> YV12
*/

#define _SVID_SOURCE //needed for MAP_ANONYMOUS
#include <inttypes.h>
#include <string.h>
#include <math.h>
#include <stdio.h>
#include <unistd.h>
#include "config.h"
#include <assert.h>
#ifdef HAVE_SYS_MMAN_H
#include <sys/mman.h>
#if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
#define MAP_ANONYMOUS MAP_ANON
#endif
#endif
#include "swscale.h"
#include "swscale_internal.h"
#include "rgb2rgb.h"
#include "libavutil/x86_cpu.h"
#include "libavutil/bswap.h"

unsigned swscale_version(void)
{
    return LIBSWSCALE_VERSION_INT;
}

#undef MOVNTQ
#undef PAVGB

//#undef HAVE_MMX2
//#define HAVE_3DNOW
//#undef HAVE_MMX
//#undef ARCH_X86
//#define WORDS_BIGENDIAN
#define DITHER1XBPP

#define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit

#define RET 0xC3 //near return opcode for X86

#ifdef M_PI
#define PI M_PI
#else
#define PI 3.14159265358979323846
#endif

#define isSupportedIn(x)    (       \
           (x)==PIX_FMT_YUV420P     \
        || (x)==PIX_FMT_YUVA420P    \
        || (x)==PIX_FMT_YUYV422     \
        || (x)==PIX_FMT_UYVY422     \
        || (x)==PIX_FMT_RGB32       \
        || (x)==PIX_FMT_RGB32_1     \
        || (x)==PIX_FMT_BGR24       \
        || (x)==PIX_FMT_BGR565      \
        || (x)==PIX_FMT_BGR555      \
        || (x)==PIX_FMT_BGR32       \
        || (x)==PIX_FMT_BGR32_1     \
        || (x)==PIX_FMT_RGB24       \
        || (x)==PIX_FMT_RGB565      \
        || (x)==PIX_FMT_RGB555      \
        || (x)==PIX_FMT_GRAY8       \
        || (x)==PIX_FMT_YUV410P     \
        || (x)==PIX_FMT_GRAY16BE    \
        || (x)==PIX_FMT_GRAY16LE    \
        || (x)==PIX_FMT_YUV444P     \
        || (x)==PIX_FMT_YUV422P     \
        || (x)==PIX_FMT_YUV411P     \
        || (x)==PIX_FMT_PAL8        \
        || (x)==PIX_FMT_BGR8        \
        || (x)==PIX_FMT_RGB8        \
        || (x)==PIX_FMT_BGR4_BYTE   \
        || (x)==PIX_FMT_RGB4_BYTE   \
        || (x)==PIX_FMT_YUV440P     \
        || (x)==PIX_FMT_MONOWHITE   \
        || (x)==PIX_FMT_MONOBLACK   \
    )
#define isSupportedOut(x)   (       \
           (x)==PIX_FMT_YUV420P     \
        || (x)==PIX_FMT_YUYV422     \
        || (x)==PIX_FMT_UYVY422     \
        || (x)==PIX_FMT_YUV444P     \
        || (x)==PIX_FMT_YUV422P     \
        || (x)==PIX_FMT_YUV411P     \
        || isRGB(x)                 \
        || isBGR(x)                 \
        || (x)==PIX_FMT_NV12        \
        || (x)==PIX_FMT_NV21        \
        || (x)==PIX_FMT_GRAY16BE    \
        || (x)==PIX_FMT_GRAY16LE    \
        || (x)==PIX_FMT_GRAY8       \
        || (x)==PIX_FMT_YUV410P     \
    )
#define isPacked(x)         (       \
           (x)==PIX_FMT_PAL8        \
        || (x)==PIX_FMT_YUYV422     \
        || (x)==PIX_FMT_UYVY422     \
        || isRGB(x)                 \
        || isBGR(x)                 \
    )

#define RGB2YUV_SHIFT 15
#define BY ( (int)(0.114*219/255*(1<<RGB2YUV_SHIFT)+0.5))
#define BV (-(int)(0.081*224/255*(1<<RGB2YUV_SHIFT)+0.5))
#define BU ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
#define GY ( (int)(0.587*219/255*(1<<RGB2YUV_SHIFT)+0.5))
#define GV (-(int)(0.419*224/255*(1<<RGB2YUV_SHIFT)+0.5))
#define GU (-(int)(0.331*224/255*(1<<RGB2YUV_SHIFT)+0.5))
#define RY ( (int)(0.299*219/255*(1<<RGB2YUV_SHIFT)+0.5))
#define RV ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
#define RU (-(int)(0.169*224/255*(1<<RGB2YUV_SHIFT)+0.5))

extern const int32_t Inverse_Table_6_9[8][4];

static const double rgb2yuv_table[8][9]={
    {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
    {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
    {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
    {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
    {0.59  , 0.11  , 0.30  , -0.331, 0.5, -0.169, -0.421, -0.079, 0.5}, //FCC
    {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
    {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5}, //SMPTE 170M
    {0.701 , 0.087 , 0.212 , -0.384, 0.5  -0.116, -0.445, -0.055, 0.5}, //SMPTE 240M
};

/*
NOTES
Special versions: fast Y 1:1 scaling (no interpolation in y direction)

TODO
more intelligent misalignment avoidance for the horizontal scaler
write special vertical cubic upscale version
Optimize C code (yv12 / minmax)
add support for packed pixel yuv input & output
add support for Y8 output
optimize bgr24 & bgr32
add BGR4 output support
write special BGR->BGR scaler
*/

#if defined(ARCH_X86) && defined (CONFIG_GPL)
DECLARE_ASM_CONST(8, uint64_t, bF8)=       0xF8F8F8F8F8F8F8F8LL;
DECLARE_ASM_CONST(8, uint64_t, bFC)=       0xFCFCFCFCFCFCFCFCLL;
DECLARE_ASM_CONST(8, uint64_t, w10)=       0x0010001000100010LL;
DECLARE_ASM_CONST(8, uint64_t, w02)=       0x0002000200020002LL;
DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;

static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;

const DECLARE_ALIGNED(8, uint64_t, ff_dither4[2]) = {
        0x0103010301030103LL,
        0x0200020002000200LL,};

const DECLARE_ALIGNED(8, uint64_t, ff_dither8[2]) = {
        0x0602060206020602LL,
        0x0004000400040004LL,};

DECLARE_ASM_CONST(8, uint64_t, b16Mask)=   0x001F001F001F001FLL;
DECLARE_ASM_CONST(8, uint64_t, g16Mask)=   0x07E007E007E007E0LL;
DECLARE_ASM_CONST(8, uint64_t, r16Mask)=   0xF800F800F800F800LL;
DECLARE_ASM_CONST(8, uint64_t, b15Mask)=   0x001F001F001F001FLL;
DECLARE_ASM_CONST(8, uint64_t, g15Mask)=   0x03E003E003E003E0LL;
DECLARE_ASM_CONST(8, uint64_t, r15Mask)=   0x7C007C007C007C00LL;

DECLARE_ALIGNED(8, const uint64_t, ff_M24A)         = 0x00FF0000FF0000FFLL;
DECLARE_ALIGNED(8, const uint64_t, ff_M24B)         = 0xFF0000FF0000FF00LL;
DECLARE_ALIGNED(8, const uint64_t, ff_M24C)         = 0x0000FF0000FF0000LL;

#ifdef FAST_BGR2YV12
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff)   = 0x000000210041000DULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff)   = 0x0000FFEEFFDC0038ULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff)   = 0x00000038FFD2FFF8ULL;
#else
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff)   = 0x000020E540830C8BULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff)   = 0x0000ED0FDAC23831ULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff)   = 0x00003831D0E6F6EAULL;
#endif /* FAST_BGR2YV12 */
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset)  = 0x1010101010101010ULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
DECLARE_ALIGNED(8, const uint64_t, ff_w1111)        = 0x0001000100010001ULL;

DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toY1Coeff) = 0x0C88000040870C88ULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toY2Coeff) = 0x20DE4087000020DEULL;
DECLARE_ALIGNED(8, const uint64_t, ff_rgb24toY1Coeff) = 0x20DE0000408720DEULL;
DECLARE_ALIGNED(8, const uint64_t, ff_rgb24toY2Coeff) = 0x0C88408700000C88ULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toYOffset) = 0x0008400000084000ULL;

DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toUV[2][4]) = {
    {0x38380000DAC83838ULL, 0xECFFDAC80000ECFFULL, 0xF6E40000D0E3F6E4ULL, 0x3838D0E300003838ULL},
    {0xECFF0000DAC8ECFFULL, 0x3838DAC800003838ULL, 0x38380000D0E33838ULL, 0xF6E4D0E30000F6E4ULL},
};

DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toUVOffset)= 0x0040400000404000ULL;

#endif /* defined(ARCH_X86) */

// clipping helper table for C implementations:
static unsigned char clip_table[768];

static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);

extern const uint8_t dither_2x2_4[2][8];
extern const uint8_t dither_2x2_8[2][8];
extern const uint8_t dither_8x8_32[8][8];
extern const uint8_t dither_8x8_73[8][8];
extern const uint8_t dither_8x8_220[8][8];

const char *sws_format_name(enum PixelFormat format)
{
    switch (format) {
        case PIX_FMT_YUV420P:
            return "yuv420p";
        case PIX_FMT_YUVA420P:
            return "yuva420p";
        case PIX_FMT_YUYV422:
            return "yuyv422";
        case PIX_FMT_RGB24:
            return "rgb24";
        case PIX_FMT_BGR24:
            return "bgr24";
        case PIX_FMT_YUV422P:
            return "yuv422p";
        case PIX_FMT_YUV444P:
            return "yuv444p";
        case PIX_FMT_RGB32:
            return "rgb32";
        case PIX_FMT_YUV410P:
            return "yuv410p";
        case PIX_FMT_YUV411P:
            return "yuv411p";
        case PIX_FMT_RGB565:
            return "rgb565";
        case PIX_FMT_RGB555:
            return "rgb555";
        case PIX_FMT_GRAY16BE:
            return "gray16be";
        case PIX_FMT_GRAY16LE:
            return "gray16le";
        case PIX_FMT_GRAY8:
            return "gray8";
        case PIX_FMT_MONOWHITE:
            return "mono white";
        case PIX_FMT_MONOBLACK:
            return "mono black";
        case PIX_FMT_PAL8:
            return "Palette";
        case PIX_FMT_YUVJ420P:
            return "yuvj420p";
        case PIX_FMT_YUVJ422P:
            return "yuvj422p";
        case PIX_FMT_YUVJ444P:
            return "yuvj444p";
        case PIX_FMT_XVMC_MPEG2_MC:
            return "xvmc_mpeg2_mc";
        case PIX_FMT_XVMC_MPEG2_IDCT:
            return "xvmc_mpeg2_idct";
        case PIX_FMT_UYVY422:
            return "uyvy422";
        case PIX_FMT_UYYVYY411:
            return "uyyvyy411";
        case PIX_FMT_RGB32_1:
            return "rgb32x";
        case PIX_FMT_BGR32_1:
            return "bgr32x";
        case PIX_FMT_BGR32:
            return "bgr32";
        case PIX_FMT_BGR565:
            return "bgr565";
        case PIX_FMT_BGR555:
            return "bgr555";
        case PIX_FMT_BGR8:
            return "bgr8";
        case PIX_FMT_BGR4:
            return "bgr4";
        case PIX_FMT_BGR4_BYTE:
            return "bgr4 byte";
        case PIX_FMT_RGB8:
            return "rgb8";
        case PIX_FMT_RGB4:
            return "rgb4";
        case PIX_FMT_RGB4_BYTE:
            return "rgb4 byte";
        case PIX_FMT_NV12:
            return "nv12";
        case PIX_FMT_NV21:
            return "nv21";
        case PIX_FMT_YUV440P:
            return "yuv440p";
        default:
            return "Unknown format";
    }
}

static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
                               int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
                               uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
{
    //FIXME Optimize (just quickly writen not opti..)
    int i;
    for (i=0; i<dstW; i++)
    {
        int val=1<<18;
        int j;
        for (j=0; j<lumFilterSize; j++)
            val += lumSrc[j][i] * lumFilter[j];

        dest[i]= av_clip_uint8(val>>19);
    }

    if (uDest)
        for (i=0; i<chrDstW; i++)
        {
            int u=1<<18;
            int v=1<<18;
            int j;
            for (j=0; j<chrFilterSize; j++)
            {
                u += chrSrc[j][i] * chrFilter[j];
                v += chrSrc[j][i + VOFW] * chrFilter[j];
            }

            uDest[i]= av_clip_uint8(u>>19);
            vDest[i]= av_clip_uint8(v>>19);
        }
}

static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
                                int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
                                uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
{
    //FIXME Optimize (just quickly writen not opti..)
    int i;
    for (i=0; i<dstW; i++)
    {
        int val=1<<18;
        int j;
        for (j=0; j<lumFilterSize; j++)
            val += lumSrc[j][i] * lumFilter[j];

        dest[i]= av_clip_uint8(val>>19);
    }

    if (!uDest)
        return;

    if (dstFormat == PIX_FMT_NV12)
        for (i=0; i<chrDstW; i++)
        {
            int u=1<<18;
            int v=1<<18;
            int j;
            for (j=0; j<chrFilterSize; j++)
            {
                u += chrSrc[j][i] * chrFilter[j];
                v += chrSrc[j][i + VOFW] * chrFilter[j];
            }

            uDest[2*i]= av_clip_uint8(u>>19);
            uDest[2*i+1]= av_clip_uint8(v>>19);
        }
    else
        for (i=0; i<chrDstW; i++)
        {
            int u=1<<18;
            int v=1<<18;
            int j;
            for (j=0; j<chrFilterSize; j++)
            {
                u += chrSrc[j][i] * chrFilter[j];
                v += chrSrc[j][i + VOFW] * chrFilter[j];
            }

            uDest[2*i]= av_clip_uint8(v>>19);
            uDest[2*i+1]= av_clip_uint8(u>>19);
        }
}

#define YSCALE_YUV_2_PACKEDX_NOCLIP_C(type) \
    for (i=0; i<(dstW>>1); i++){\
        int j;\
        int Y1 = 1<<18;\
        int Y2 = 1<<18;\
        int U  = 1<<18;\
        int V  = 1<<18;\
        type av_unused *r, *b, *g;\
        const int i2= 2*i;\
        \
        for (j=0; j<lumFilterSize; j++)\
        {\
            Y1 += lumSrc[j][i2] * lumFilter[j];\
            Y2 += lumSrc[j][i2+1] * lumFilter[j];\
        }\
        for (j=0; j<chrFilterSize; j++)\
        {\
            U += chrSrc[j][i] * chrFilter[j];\
            V += chrSrc[j][i+VOFW] * chrFilter[j];\
        }\
        Y1>>=19;\
        Y2>>=19;\
        U >>=19;\
        V >>=19;\

#define YSCALE_YUV_2_PACKEDX_C(type) \
        YSCALE_YUV_2_PACKEDX_NOCLIP_C(type)\
        if ((Y1|Y2|U|V)&256)\
        {\
            if (Y1>255)   Y1=255; \
            else if (Y1<0)Y1=0;   \
            if (Y2>255)   Y2=255; \
            else if (Y2<0)Y2=0;   \
            if (U>255)    U=255;  \
            else if (U<0) U=0;    \
            if (V>255)    V=255;  \
            else if (V<0) V=0;    \
        }

#define YSCALE_YUV_2_PACKEDX_FULL_C \
    for (i=0; i<dstW; i++){\
        int j;\
        int Y = 0;\
        int U = -128<<19;\
        int V = -128<<19;\
        int R,G,B;\
        \
        for (j=0; j<lumFilterSize; j++){\
            Y += lumSrc[j][i     ] * lumFilter[j];\
        }\
        for (j=0; j<chrFilterSize; j++){\
            U += chrSrc[j][i     ] * chrFilter[j];\
            V += chrSrc[j][i+VOFW] * chrFilter[j];\
        }\
        Y >>=10;\
        U >>=10;\
        V >>=10;\

#define YSCALE_YUV_2_RGBX_FULL_C(rnd) \
    YSCALE_YUV_2_PACKEDX_FULL_C\
        Y-= c->oy;\
        Y*= c->cy;\
        Y+= rnd;\
        R= Y + V*c->cvr;\
        G= Y + V*c->cvg + U*c->cug;\
        B= Y +            U*c->cub;\
        if ((R|G|B)&(0xC0000000)){\
            if (R>=(256<<22))   R=(256<<22)-1; \
            else if (R<0)R=0;   \
            if (G>=(256<<22))   G=(256<<22)-1; \
            else if (G<0)G=0;   \
            if (B>=(256<<22))   B=(256<<22)-1; \
            else if (B<0)B=0;   \
        }\


#define YSCALE_YUV_2_GRAY16_C \
    for (i=0; i<(dstW>>1); i++){\
        int j;\
        int Y1 = 1<<18;\
        int Y2 = 1<<18;\
        int U  = 1<<18;\
        int V  = 1<<18;\
        \
        const int i2= 2*i;\
        \
        for (j=0; j<lumFilterSize; j++)\
        {\
            Y1 += lumSrc[j][i2] * lumFilter[j];\
            Y2 += lumSrc[j][i2+1] * lumFilter[j];\
        }\
        Y1>>=11;\
        Y2>>=11;\
        if ((Y1|Y2|U|V)&65536)\
        {\
            if (Y1>65535)   Y1=65535; \
            else if (Y1<0)Y1=0;   \
            if (Y2>65535)   Y2=65535; \
            else if (Y2<0)Y2=0;   \
        }

#define YSCALE_YUV_2_RGBX_C(type) \
    YSCALE_YUV_2_PACKEDX_NOCLIP_C(type)  \
    r = (type *)c->table_rV[V];   \
    g = (type *)(c->table_gU[U] + c->table_gV[V]); \
    b = (type *)c->table_bU[U];   \

#define YSCALE_YUV_2_PACKED2_C   \
    for (i=0; i<(dstW>>1); i++){ \
        const int i2= 2*i;       \
        int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;           \
        int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;           \
        int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;  \
        int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19;  \

#define YSCALE_YUV_2_GRAY16_2_C   \
    for (i=0; i<(dstW>>1); i++){ \
        const int i2= 2*i;       \
        int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>11;           \
        int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>11;           \

#define YSCALE_YUV_2_RGB2_C(type) \
    YSCALE_YUV_2_PACKED2_C\
    type *r, *b, *g;\
    r = (type *)c->table_rV[V];\
    g = (type *)(c->table_gU[U] + c->table_gV[V]);\
    b = (type *)c->table_bU[U];\

#define YSCALE_YUV_2_PACKED1_C \
    for (i=0; i<(dstW>>1); i++){\
        const int i2= 2*i;\
        int Y1= buf0[i2  ]>>7;\
        int Y2= buf0[i2+1]>>7;\
        int U= (uvbuf1[i     ])>>7;\
        int V= (uvbuf1[i+VOFW])>>7;\

#define YSCALE_YUV_2_GRAY16_1_C \
    for (i=0; i<(dstW>>1); i++){\
        const int i2= 2*i;\
        int Y1= buf0[i2  ]<<1;\
        int Y2= buf0[i2+1]<<1;\

#define YSCALE_YUV_2_RGB1_C(type) \
    YSCALE_YUV_2_PACKED1_C\
    type *r, *b, *g;\
    r = (type *)c->table_rV[V];\
    g = (type *)(c->table_gU[U] + c->table_gV[V]);\
    b = (type *)c->table_bU[U];\

#define YSCALE_YUV_2_PACKED1B_C \
    for (i=0; i<(dstW>>1); i++){\
        const int i2= 2*i;\
        int Y1= buf0[i2  ]>>7;\
        int Y2= buf0[i2+1]>>7;\
        int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
        int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\

#define YSCALE_YUV_2_RGB1B_C(type) \
    YSCALE_YUV_2_PACKED1B_C\
    type *r, *b, *g;\
    r = (type *)c->table_rV[V];\
    g = (type *)(c->table_gU[U] + c->table_gV[V]);\
    b = (type *)c->table_bU[U];\

#define YSCALE_YUV_2_MONO2_C \
    const uint8_t * const d128=dither_8x8_220[y&7];\
    uint8_t *g= c->table_gU[128] + c->table_gV[128];\
    for (i=0; i<dstW-7; i+=8){\
        int acc;\
        acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
        acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
        acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
        acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
        acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
        acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
        acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
        acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
        ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
        dest++;\
    }\


#define YSCALE_YUV_2_MONOX_C \
    const uint8_t * const d128=dither_8x8_220[y&7];\
    uint8_t *g= c->table_gU[128] + c->table_gV[128];\
    int acc=0;\
    for (i=0; i<dstW-1; i+=2){\
        int j;\
        int Y1=1<<18;\
        int Y2=1<<18;\
\
        for (j=0; j<lumFilterSize; j++)\
        {\
            Y1 += lumSrc[j][i] * lumFilter[j];\
            Y2 += lumSrc[j][i+1] * lumFilter[j];\
        }\
        Y1>>=19;\
        Y2>>=19;\
        if ((Y1|Y2)&256)\
        {\
            if (Y1>255)   Y1=255;\
            else if (Y1<0)Y1=0;\
            if (Y2>255)   Y2=255;\
            else if (Y2<0)Y2=0;\
        }\
        acc+= acc + g[Y1+d128[(i+0)&7]];\
        acc+= acc + g[Y2+d128[(i+1)&7]];\
        if ((i&7)==6){\
            ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
            dest++;\
        }\
    }


#define YSCALE_YUV_2_ANYRGB_C(func, func2, func_g16, func_monoblack)\
    switch(c->dstFormat)\
    {\
    case PIX_FMT_RGB32:\
    case PIX_FMT_BGR32:\
    case PIX_FMT_RGB32_1:\
    case PIX_FMT_BGR32_1:\
        func(uint32_t)\
            ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
            ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
        }                \
        break;\
    case PIX_FMT_RGB24:\
        func(uint8_t)\
            ((uint8_t*)dest)[0]= r[Y1];\
            ((uint8_t*)dest)[1]= g[Y1];\
            ((uint8_t*)dest)[2]= b[Y1];\
            ((uint8_t*)dest)[3]= r[Y2];\
            ((uint8_t*)dest)[4]= g[Y2];\
            ((uint8_t*)dest)[5]= b[Y2];\
            dest+=6;\
        }\
        break;\
    case PIX_FMT_BGR24:\
        func(uint8_t)\
            ((uint8_t*)dest)[0]= b[Y1];\
            ((uint8_t*)dest)[1]= g[Y1];\
            ((uint8_t*)dest)[2]= r[Y1];\
            ((uint8_t*)dest)[3]= b[Y2];\
            ((uint8_t*)dest)[4]= g[Y2];\
            ((uint8_t*)dest)[5]= r[Y2];\
            dest+=6;\
        }\
        break;\
    case PIX_FMT_RGB565:\
    case PIX_FMT_BGR565:\
        {\
            const int dr1= dither_2x2_8[y&1    ][0];\
            const int dg1= dither_2x2_4[y&1    ][0];\
            const int db1= dither_2x2_8[(y&1)^1][0];\
            const int dr2= dither_2x2_8[y&1    ][1];\
            const int dg2= dither_2x2_4[y&1    ][1];\
            const int db2= dither_2x2_8[(y&1)^1][1];\
            func(uint16_t)\
                ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
                ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
            }\
        }\
        break;\
    case PIX_FMT_RGB555:\
    case PIX_FMT_BGR555:\
        {\
            const int dr1= dither_2x2_8[y&1    ][0];\
            const int dg1= dither_2x2_8[y&1    ][1];\
            const int db1= dither_2x2_8[(y&1)^1][0];\
            const int dr2= dither_2x2_8[y&1    ][1];\
            const int dg2= dither_2x2_8[y&1    ][0];\
            const int db2= dither_2x2_8[(y&1)^1][1];\
            func(uint16_t)\
                ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
                ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
            }\
        }\
        break;\
    case PIX_FMT_RGB8:\
    case PIX_FMT_BGR8:\
        {\
            const uint8_t * const d64= dither_8x8_73[y&7];\
            const uint8_t * const d32= dither_8x8_32[y&7];\
            func(uint8_t)\
                ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
                ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
            }\
        }\
        break;\
    case PIX_FMT_RGB4:\
    case PIX_FMT_BGR4:\
        {\
            const uint8_t * const d64= dither_8x8_73 [y&7];\
            const uint8_t * const d128=dither_8x8_220[y&7];\
            func(uint8_t)\
                ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
                                 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
            }\
        }\
        break;\
    case PIX_FMT_RGB4_BYTE:\
    case PIX_FMT_BGR4_BYTE:\
        {\
            const uint8_t * const d64= dither_8x8_73 [y&7];\
            const uint8_t * const d128=dither_8x8_220[y&7];\
            func(uint8_t)\
                ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
                ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
            }\
        }\
        break;\
    case PIX_FMT_MONOBLACK:\
    case PIX_FMT_MONOWHITE:\
        {\
            func_monoblack\
        }\
        break;\
    case PIX_FMT_YUYV422:\
        func2\
            ((uint8_t*)dest)[2*i2+0]= Y1;\
            ((uint8_t*)dest)[2*i2+1]= U;\
            ((uint8_t*)dest)[2*i2+2]= Y2;\
            ((uint8_t*)dest)[2*i2+3]= V;\
        }                \
        break;\
    case PIX_FMT_UYVY422:\
        func2\
            ((uint8_t*)dest)[2*i2+0]= U;\
            ((uint8_t*)dest)[2*i2+1]= Y1;\
            ((uint8_t*)dest)[2*i2+2]= V;\
            ((uint8_t*)dest)[2*i2+3]= Y2;\
        }                \
        break;\
    case PIX_FMT_GRAY16BE:\
        func_g16\
            ((uint8_t*)dest)[2*i2+0]= Y1>>8;\
            ((uint8_t*)dest)[2*i2+1]= Y1;\
            ((uint8_t*)dest)[2*i2+2]= Y2>>8;\
            ((uint8_t*)dest)[2*i2+3]= Y2;\
        }                \
        break;\
    case PIX_FMT_GRAY16LE:\
        func_g16\
            ((uint8_t*)dest)[2*i2+0]= Y1;\
            ((uint8_t*)dest)[2*i2+1]= Y1>>8;\
            ((uint8_t*)dest)[2*i2+2]= Y2;\
            ((uint8_t*)dest)[2*i2+3]= Y2>>8;\
        }                \
        break;\
    }\


static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
                                  int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
                                  uint8_t *dest, int dstW, int y)
{
    int i;
    YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGBX_C, YSCALE_YUV_2_PACKEDX_C(void), YSCALE_YUV_2_GRAY16_C, YSCALE_YUV_2_MONOX_C)
}

static inline void yuv2rgbXinC_full(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
                                    int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
                                    uint8_t *dest, int dstW, int y)
{
    int i;
    int step= fmt_depth(c->dstFormat)/8;

    switch(c->dstFormat){
    case PIX_FMT_ARGB:
        dest++;
    case PIX_FMT_RGB24:
    case PIX_FMT_RGBA:
        YSCALE_YUV_2_RGBX_FULL_C(1<<21)
            dest[0]= R>>22;
            dest[1]= G>>22;
            dest[2]= B>>22;
            dest[3]= 0;
            dest+= step;
        }
        break;
    case PIX_FMT_ABGR:
        dest++;
    case PIX_FMT_BGR24:
    case PIX_FMT_BGRA:
        YSCALE_YUV_2_RGBX_FULL_C(1<<21)
            dest[0]= B>>22;
            dest[1]= G>>22;
            dest[2]= R>>22;
            dest[3]= 0;
            dest+= step;
        }
        break;
    default:
        assert(0);
    }
}

//Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
//Plain C versions
#if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) || !defined(CONFIG_GPL)
#define COMPILE_C
#endif

#ifdef ARCH_POWERPC
#if (defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
#define COMPILE_ALTIVEC
#endif //HAVE_ALTIVEC
#endif //ARCH_POWERPC

#if defined(ARCH_X86)

#if ((defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
#define COMPILE_MMX
#endif

#if (defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
#define COMPILE_MMX2
#endif

#if ((defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
#define COMPILE_3DNOW
#endif
#endif //ARCH_X86 || ARCH_X86_64

#undef HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_3DNOW

#ifdef COMPILE_C
#undef HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_3DNOW
#undef HAVE_ALTIVEC
#define RENAME(a) a ## _C
#include "swscale_template.c"
#endif

#ifdef COMPILE_ALTIVEC
#undef RENAME
#define HAVE_ALTIVEC
#define RENAME(a) a ## _altivec
#include "swscale_template.c"
#endif

#if defined(ARCH_X86)

//X86 versions
/*
#undef RENAME
#undef HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_3DNOW
#define ARCH_X86
#define RENAME(a) a ## _X86
#include "swscale_template.c"
*/
//MMX versions
#ifdef COMPILE_MMX
#undef RENAME
#define HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_3DNOW
#define RENAME(a) a ## _MMX
#include "swscale_template.c"
#endif

//MMX2 versions
#ifdef COMPILE_MMX2
#undef RENAME
#define HAVE_MMX
#define HAVE_MMX2
#undef HAVE_3DNOW
#define RENAME(a) a ## _MMX2
#include "swscale_template.c"
#endif

//3DNOW versions
#ifdef COMPILE_3DNOW
#undef RENAME
#define HAVE_MMX
#undef HAVE_MMX2
#define HAVE_3DNOW
#define RENAME(a) a ## _3DNow
#include "swscale_template.c"
#endif

#endif //ARCH_X86 || ARCH_X86_64

// minor note: the HAVE_xyz is messed up after that line so don't use it

static double getSplineCoeff(double a, double b, double c, double d, double dist)
{
//    printf("%f %f %f %f %f\n", a,b,c,d,dist);
    if (dist<=1.0)      return ((d*dist + c)*dist + b)*dist +a;
    else                return getSplineCoeff(        0.0,
                                             b+ 2.0*c + 3.0*d,
                                                    c + 3.0*d,
                                            -b- 3.0*c - 6.0*d,
                                            dist-1.0);
}

static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
                             int srcW, int dstW, int filterAlign, int one, int flags,
                             SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
{
    int i;
    int filterSize;
    int filter2Size;
    int minFilterSize;
    double *filter=NULL;
    double *filter2=NULL;
    int ret= -1;
#if defined(ARCH_X86)
    if (flags & SWS_CPU_CAPS_MMX)
        asm volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
#endif

    // Note the +1 is for the MMXscaler which reads over the end
    *filterPos = av_malloc((dstW+1)*sizeof(int16_t));

    if (FFABS(xInc - 0x10000) <10) // unscaled
    {
        int i;
        filterSize= 1;
        filter= av_malloc(dstW*sizeof(double)*filterSize);
        for (i=0; i<dstW*filterSize; i++) filter[i]=0;

        for (i=0; i<dstW; i++)
        {
            filter[i*filterSize]=1;
            (*filterPos)[i]=i;
        }

    }
    else if (flags&SWS_POINT) // lame looking point sampling mode
    {
        int i;
        int xDstInSrc;
        filterSize= 1;
        filter= av_malloc(dstW*sizeof(double)*filterSize);

        xDstInSrc= xInc/2 - 0x8000;
        for (i=0; i<dstW; i++)
        {
            int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;

            (*filterPos)[i]= xx;
            filter[i]= 1.0;
            xDstInSrc+= xInc;
        }
    }
    else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
    {
        int i;
        int xDstInSrc;
        if      (flags&SWS_BICUBIC) filterSize= 4;
        else if (flags&SWS_X      ) filterSize= 4;
        else                        filterSize= 2; // SWS_BILINEAR / SWS_AREA
        filter= av_malloc(dstW*sizeof(double)*filterSize);

        xDstInSrc= xInc/2 - 0x8000;
        for (i=0; i<dstW; i++)