slightly higher quality vaguely fast DXT5 encoder.

/*
Copyright (C) 2013 by Brendan G Bohannon
Email: [email protected]
Copying: http://pastebin.com/iJxtZHm6
 */

/*
Tries to deliver moderate quality, while still trying to allow for reasonably fast encoding.
Generally handles interpolation by treating it like a fixed-point fraction, and using this fraction to index into a table.

This version also adds fixed-pattern dither and some tweaks to try to improve image quality.
*/

void BGBBTJ_BCn_EncodeBlockDXT5(byte *block,
    byte *rgba, int xstride, int ystride)
{
    static const char idxtab[16]=
        { 0,0,0,0, 0,0,2,2, 3,3,1,1, 1,1,1,1 };
    static const char idxtab2[16]=
        { 0,0,0,0, 2,2,3,3, 4,4,5,5, 1,1,1,1 };
    byte pxa[16], pxy[16];
    int p0, p1, p2, p3, p4, p5, p6, p7;
    int l0, l1, l2, l3a, l3b;
    int mcr, mcg, mcb, mca, mcy;
    int ncr, ncg, ncb, nca, ncy;
    int acr, acg, acb, aca, acy;
    int cr, cg, cb, ca, cy;
    int i, j, k, l;

    mcr=255; mcg=255; mcb=255; mca=255; mcy=255;
    ncr=  0; ncg=  0; ncb=  0; nca=  0; ncy=  0;
    acr=  0; acg=  0; acb=  0; aca=  0; acy=  0;

    for(i=0; i<4; i++)
        for(j=0; j<4; j++)
    {
        k=(i*ystride)+(j*xstride);
        cr=rgba[k+0];   cg=rgba[k+1];
        cb=rgba[k+2];   ca=rgba[k+3];
        cy=(cr+(2*cg)+cb)>>2;       //Luma

        pxa[i*4+j]=ca;
        pxy[i*4+j]=cy;

        //Find Mins/Maxs
        if(cy<mcy) { mcr=cr; mcg=cg; mcb=cb; mcy=cy; }
        if(cy>ncy) { ncr=cr; ncg=cg; ncb=cb; ncy=cy; }
        if(ca<mca) { mca=ca; }
        if(ca>nca) { nca=ca; }

        acr+=cr; acg+=cg;
        acb+=cb; acy+=cy;
    }

    acr=(acr+8)>>4;
    acg=(acg+8)>>4;
    acb=(acb+8)>>4;
    acy=(acr+(2*acg)+acb)>>2;

    l=(256*mcy)/(acy+1);
    cr=(acr*l+128)>>8;
    cg=(acg*l+128)>>8;
    cb=(acb*l+128)>>8;

    mcr=(mcr+cr)>>1;
    mcg=(mcg+cg)>>1;
    mcb=(mcb+cb)>>1;


    l=(256*ncy)/(acy+1);
    cr=(acr*l+128)>>8;
    cg=(acg*l+128)>>8;
    cb=(acb*l+128)>>8;

    ncr=(ncr+cr)>>1;
    ncg=(ncg+cg)>>1;
    ncb=(ncb+cb)>>1;

    aca=(mca+nca)>>1;           //Median Alpha
    acy=(mcy+ncy)>>1;           //Median Luma
    l0=49152/(nca-aca+1);       //Fix-Point Scale (Alpha)
    l1=32768/(ncy-acy+1);       //Fix-Point Scale (Luma)
    l2=65536-2048;

    if(mca==nca)
    {
        block[0]=mca; block[1]=nca;
        block[2]=0; block[3]=0;
        block[4]=0; block[5]=0;
        block[6]=0; block[7]=0;
    }else
    {
        //Emit Alpha Block
        block[0]=mca;
        block[1]=nca;

        p0=idxtab2[((pxa[ 0]-aca)*l0+l2)>>13];
        p1=idxtab2[((pxa[ 1]-aca)*l0+l2)>>13];
        p2=idxtab2[((pxa[ 2]-aca)*l0+l2)>>13];
        p3=idxtab2[((pxa[ 3]-aca)*l0+l2)>>13];
        p4=idxtab2[((pxa[ 4]-aca)*l0+l2)>>13];
        p5=idxtab2[((pxa[ 5]-aca)*l0+l2)>>13];
        p6=idxtab2[((pxa[ 6]-aca)*l0+l2)>>13];
        p7=idxtab2[((pxa[ 7]-aca)*l0+l2)>>13];
        block[2]=p0|(p1<<3)|(p2<<6);
        block[3]=(p2>>2)|(p3<<1)|(p4<<4)|(p5<<7);
        block[4]=(p5>>1)|(p6<<2)|(p7<<5);
        p0=idxtab2[((pxa[ 8]-aca)*l0+l2)>>13];
        p1=idxtab2[((pxa[ 9]-aca)*l0+l2)>>13];
        p2=idxtab2[((pxa[10]-aca)*l0+l2)>>13];
        p3=idxtab2[((pxa[11]-aca)*l0+l2)>>13];
        p4=idxtab2[((pxa[12]-aca)*l0+l2)>>13];
        p5=idxtab2[((pxa[13]-aca)*l0+l2)>>13];
        p6=idxtab2[((pxa[14]-aca)*l0+l2)>>13];
        p7=idxtab2[((pxa[15]-aca)*l0+l2)>>13];
        block[5]=p0|(p1<<3)|(p2<<6);
        block[6]=(p2>>2)|(p3<<1)|(p4<<4)|(p5<<7);
        block[7]=(p5>>1)|(p6<<2)|(p7<<5);
    }

    if((mcr==ncr) && (mcg==ncg) && (mcb==ncb))
    {
        //Flat Color Block
        mcr=((mcr+3)>>3)<<3;
        mcg=((mcg+1)>>2)<<2;
        mcb=((mcb+3)>>3)<<3;
        ncr=((ncr+5)>>3)<<3;
        ncg=((ncg+3)>>2)<<2;
        ncb=((ncb+5)>>3)<<3;
        mcr=(mcr<0)?0:((mcr>255)?255:mcr);
        mcg=(mcg<0)?0:((mcg>255)?255:mcg);
        mcb=(mcb<0)?0:((mcb>255)?255:mcb);
        ncr=(ncr<0)?0:((ncr>255)?255:ncr);
        ncg=(ncg<0)?0:((ncg>255)?255:ncg);
        ncb=(ncb<0)?0:((ncb>255)?255:ncb);
        i=((mcr<<8)&0xF800) | ((mcg<<3)&0x07E0) | ((mcb>>3)&0x001F);
        j=((ncr<<8)&0xF800) | ((ncg<<3)&0x07E0) | ((ncb>>3)&0x001F);
        block[ 8]=i;    block[ 9]=i>>8;
        block[10]=j;    block[11]=j>>8;
        block[12]=(2<<6)|(3<<4)|(2<<2)|3;
        block[13]=(3<<6)|(2<<4)|(3<<2)|2;
        block[14]=(2<<6)|(3<<4)|(2<<2)|3;
        block[15]=(3<<6)|(2<<4)|(3<<2)|2;
    }else
    {
        //Emit RGB Block

        mcr=((mcr+3)>>3)<<3;
        mcg=((mcg+1)>>2)<<2;
        mcb=((mcb+3)>>3)<<3;
        ncr=((ncr+5)>>3)<<3;
        ncg=((ncg+3)>>2)<<2;
        ncb=((ncb+5)>>3)<<3;
        mcr=(mcr<0)?0:((mcr>255)?255:mcr);
        mcg=(mcg<0)?0:((mcg>255)?255:mcg);
        mcb=(mcb<0)?0:((mcb>255)?255:mcb);
        ncr=(ncr<0)?0:((ncr>255)?255:ncr);
        ncg=(ncg<0)?0:((ncg>255)?255:ncg);
        ncb=(ncb<0)?0:((ncb>255)?255:ncb);

        l3a=65536-1024;
        l3b=65536+1024;

        i=((mcr<<8)&0xF800) | ((mcg<<3)&0x07E0) | ((mcb>>3)&0x001F);
        j=((ncr<<8)&0xF800) | ((ncg<<3)&0x07E0) | ((ncb>>3)&0x001F);
        block[ 8]=i;    block[ 9]=i>>8;
        block[10]=j;    block[11]=j>>8;

        p0=idxtab[((pxy[ 0]-acy)*l1+l3a)>>13];
        p1=idxtab[((pxy[ 1]-acy)*l1+l3b)>>13];
        p2=idxtab[((pxy[ 2]-acy)*l1+l3a)>>13];
        p3=idxtab[((pxy[ 3]-acy)*l1+l3b)>>13];
        block[12]=(p3<<6)|(p2<<4)|(p1<<2)|p0;
        p0=idxtab[((pxy[ 4]-acy)*l1+l3b)>>13];
        p1=idxtab[((pxy[ 5]-acy)*l1+l3a)>>13];
        p2=idxtab[((pxy[ 6]-acy)*l1+l3b)>>13];
        p3=idxtab[((pxy[ 7]-acy)*l1+l3a)>>13];
        block[13]=(p3<<6)|(p2<<4)|(p1<<2)|p0;
        p0=idxtab[((pxy[ 8]-acy)*l1+l3a)>>13];
        p1=idxtab[((pxy[ 9]-acy)*l1+l3b)>>13];
        p2=idxtab[((pxy[10]-acy)*l1+l3a)>>13];
        p3=idxtab[((pxy[11]-acy)*l1+l3b)>>13];
        block[14]=(p3<<6)|(p2<<4)|(p1<<2)|p0;
        p0=idxtab[((pxy[12]-acy)*l1+l3b)>>13];
        p1=idxtab[((pxy[13]-acy)*l1+l3a)>>13];
        p2=idxtab[((pxy[14]-acy)*l1+l3b)>>13];
        p3=idxtab[((pxy[15]-acy)*l1+l3a)>>13];
        block[15]=(p3<<6)|(p2<<4)|(p1<<2)|p0;
    }
}

BGBBTJ_API void BGBBTJ_BCn_EncodeImageDXT5(byte *block,
    byte *rgba, int xs, int ys, int stride)
{
    int xs1, ys1;
    int i, j;

    xs1=xs>>2; ys1=ys>>2;
    for(i=0; i<ys1; i++)
        for(j=0; j<xs1; j++)
    {
        BGBBTJ_BCn_EncodeBlockDXT5(
            block+(i*xs1+j)*16,
            rgba+(i*4*xs+j*4)*stride,
            stride, xs*stride);
    }
}