swscale.c

ffmpeg的完整源代码和作者自己写的文档。不但有在Linux的工程哦

    switch(c->dstFormat)
    {
    case PIX_FMT_BGR32:
    case PIX_FMT_RGB32:
        YSCALE_YUV_2_RGBX_C(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:
        YSCALE_YUV_2_RGBX_C(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:
        YSCALE_YUV_2_RGBX_C(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];
            YSCALE_YUV_2_RGBX_C(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];
            YSCALE_YUV_2_RGBX_C(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];
            YSCALE_YUV_2_RGBX_C(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];
            YSCALE_YUV_2_RGBX_C(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];
            YSCALE_YUV_2_RGBX_C(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:
        {
            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]= acc;
                    dest++;
                }
            }
        }
        break;
    case PIX_FMT_YUYV422:
        YSCALE_YUV_2_PACKEDX_C(void)
            ((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:
        YSCALE_YUV_2_PACKEDX_C(void)
            ((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;
    }
}


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

#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;
#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++)
        {
            int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
            int j;

            (*filterPos)[i]= xx;
                //Bilinear upscale / linear interpolate / Area averaging
                for (j=0; j<filterSize; j++)
                {
                    double d= FFABS((xx<<16) - xDstInSrc)/(double)(1<<16);
                    double coeff= 1.0 - d;
                    if (coeff<0) coeff=0;
                    filter[i*filterSize + j]= coeff;
                    xx++;
                }
            xDstInSrc+= xInc;
        }
    }
    else
    {
        double xDstInSrc;
        double sizeFactor, filterSizeInSrc;
        const double xInc1= (double)xInc / (double)(1<<16);

        if      (flags&SWS_BICUBIC)      sizeFactor=  4.0;
        else if (flags&SWS_X)            sizeFactor=  8.0;
        else if (flags&SWS_AREA)         sizeFactor=  1.0; //downscale only, for upscale it is bilinear
        else if (flags&SWS_GAUSS)        sizeFactor=  8.0;   // infinite ;)
        else if (flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
        else if (flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
        else if (flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
        else if (flags&SWS_BILINEAR)     sizeFactor=  2.0;
        else {
            sizeFactor= 0.0; //GCC warning killer
            ASSERT(0)
        }

        if (xInc1 <= 1.0)       filterSizeInSrc= sizeFactor; // upscale
        else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;

        filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
        if (filterSize > srcW-2) filterSize=srcW-2;

        filter= av_malloc(dstW*sizeof(double)*filterSize);

        xDstInSrc= xInc1 / 2.0 - 0.5;
        for (i=0; i<dstW; i++)
        {
            int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
            int j;
            (*filterPos)[i]= xx;
            for (j=0; j<filterSize; j++)