unity/crnlib/crn_ryg_dxt.cpp

// File: crn_ryg_dxt.cpp
// RYG's real-time DXT compressor - Public domain.
#include "crn_core.h"
#include "crn_ryg_types.hpp"
#include "crn_ryg_dxt.hpp"

#ifdef _MSC_VER
#pragma warning(disable : 4244)  // conversion from 'a' to 'b', possible loss of data
#endif

namespace ryg_dxt {
// Couple of tables...
sU8 Expand5[32];
sU8 Expand6[64];
sU8 OMatch5[256][2];
sU8 OMatch6[256][2];
sU8 OMatch5_3[256][2];
sU8 OMatch6_3[256][2];
sU8 QuantRBTab[256 + 16];
sU8 QuantGTab[256 + 16];

static sInt Mul8Bit(sInt a, sInt b) {
  sInt t = a * b + 128;
  return (t + (t >> 8)) >> 8;
}

union Pixel {
  struct
  {
    sU8 b, g, r, a;
  };
  sU32 v;

  void From16Bit(sU16 v) {
    sInt rv = (v & 0xf800) >> 11;
    sInt gv = (v & 0x07e0) >> 5;
    sInt bv = (v & 0x001f) >> 0;

    a = 0;
    r = Expand5[rv];
    g = Expand6[gv];
    b = Expand5[bv];
  }

  sU16 As16Bit() const {
    return (Mul8Bit(r, 31) << 11) + (Mul8Bit(g, 63) << 5) + Mul8Bit(b, 31);
  }

  void LerpRGB(const Pixel& p1, const Pixel& p2, sInt f) {
    r = p1.r + Mul8Bit(p2.r - p1.r, f);
    g = p1.g + Mul8Bit(p2.g - p1.g, f);
    b = p1.b + Mul8Bit(p2.b - p1.b, f);
  }
};

/****************************************************************************/

static void PrepareOptTable4(sU8* Table, const sU8* expand, sInt size) {
  for (sInt i = 0; i < 256; i++) {
    sInt bestErr = 256;

    for (sInt min = 0; min < size; min++) {
      for (sInt max = 0; max < size; max++) {
        sInt mine = expand[min];
        sInt maxe = expand[max];
        //sInt err = sAbs(maxe + Mul8Bit(mine-maxe,0x55) - i);
        sInt err = sAbs(((maxe * 2 + mine) / 3) - i);
        err += ((sAbs(maxe - mine) * 8) >> 8);  // approx. .03f

        if (err < bestErr) {
          Table[i * 2 + 0] = max;
          Table[i * 2 + 1] = min;
          bestErr = err;
        }
      }
    }
  }
}

static void PrepareOptTable3(sU8* Table, const sU8* expand, sInt size) {
  for (sInt i = 0; i < 256; i++) {
    sInt bestErr = 256;

    for (sInt min = 0; min < size; min++) {
      for (sInt max = 0; max < size; max++) {
        sInt mine = expand[min];
        sInt maxe = expand[max];
        sInt err = sAbs(((mine + maxe) >> 1) - i);
        err += ((sAbs(maxe - mine) * 8) >> 8);  // approx. .03f

        if (err < bestErr) {
          Table[i * 2 + 0] = max;
          Table[i * 2 + 1] = min;
          bestErr = err;
        }
      }
    }
  }
}

static inline void EvalColors(Pixel* color, sU16 c0, sU16 c1) {
  color[0].From16Bit(c0);
  color[1].From16Bit(c1);
  color[2].LerpRGB(color[0], color[1], 0x55);
  color[3].LerpRGB(color[0], color[1], 0xaa);
}

// Block dithering function. Simply dithers a block to 565 RGB.
// (Floyd-Steinberg)
static void DitherBlock(Pixel* dest, const Pixel* block) {
  sInt err[8], *ep1 = err, *ep2 = err + 4;

  // process channels seperately
  for (sInt ch = 0; ch < 3; ch++) {
    sU8* bp = (sU8*)block;
    sU8* dp = (sU8*)dest;
    sU8* quant = (ch == 1) ? QuantGTab + 8 : QuantRBTab + 8;

    bp += ch;
    dp += ch;
    sSetMem(err, 0, sizeof(err));

    for (sInt y = 0; y < 4; y++) {
      // pixel 0
      dp[0] = quant[bp[0] + ((3 * ep2[1] + 5 * ep2[0]) >> 4)];
      ep1[0] = bp[0] - dp[0];

      // pixel 1
      dp[4] = quant[bp[4] + ((7 * ep1[0] + 3 * ep2[2] + 5 * ep2[1] + ep2[0]) >> 4)];
      ep1[1] = bp[4] - dp[4];

      // pixel 2
      dp[8] = quant[bp[8] + ((7 * ep1[1] + 3 * ep2[3] + 5 * ep2[2] + ep2[1]) >> 4)];
      ep1[2] = bp[8] - dp[8];

      // pixel 3
      dp[12] = quant[bp[12] + ((7 * ep1[2] + 5 * ep2[3] + ep2[2]) >> 4)];
      ep1[3] = bp[12] - dp[12];

      // advance to next line
      sSwap(ep1, ep2);
      bp += 16;
      dp += 16;
    }
  }
}

// The color matching function
static sU32 MatchColorsBlock(const Pixel* block, const Pixel* color, sBool dither) {
  sU32 mask = 0;
  sInt dirr = color[0].r - color[1].r;
  sInt dirg = color[0].g - color[1].g;
  sInt dirb = color[0].b - color[1].b;

  sInt dots[16];
  for (sInt i = 0; i < 16; i++)
    dots[i] = block[i].r * dirr + block[i].g * dirg + block[i].b * dirb;

  sInt stops[4];
  for (sInt i = 0; i < 4; i++)
    stops[i] = color[i].r * dirr + color[i].g * dirg + color[i].b * dirb;

  sInt c0Point = (stops[1] + stops[3]) >> 1;
  sInt halfPoint = (stops[3] + stops[2]) >> 1;
  sInt c3Point = (stops[2] + stops[0]) >> 1;

  if (!dither) {
    // the version without dithering is straightforward
    for (sInt i = 15; i >= 0; i--) {
      mask <<= 2;
      sInt dot = dots[i];

      if (dot < halfPoint)
        mask |= (dot < c0Point) ? 1 : 3;
      else
        mask |= (dot < c3Point) ? 2 : 0;
    }
  } else {
    // with floyd-steinberg dithering (see above)
    sInt err[8], *ep1 = err, *ep2 = err + 4;
    sInt* dp = dots;

    c0Point <<= 4;
    halfPoint <<= 4;
    c3Point <<= 4;
    for (sInt i = 0; i < 8; i++)
      err[i] = 0;

    for (sInt y = 0; y < 4; y++) {
      sInt dot, lmask, step;

      // pixel 0
      dot = (dp[0] << 4) + (3 * ep2[1] + 5 * ep2[0]);
      if (dot < halfPoint)
        step = (dot < c0Point) ? 1 : 3;
      else
        step = (dot < c3Point) ? 2 : 0;

      ep1[0] = dp[0] - stops[step];
      lmask = step;

      // pixel 1
      dot = (dp[1] << 4) + (7 * ep1[0] + 3 * ep2[2] + 5 * ep2[1] + ep2[0]);
      if (dot < halfPoint)
        step = (dot < c0Point) ? 1 : 3;
      else
        step = (dot < c3Point) ? 2 : 0;

      ep1[1] = dp[1] - stops[step];
      lmask |= step << 2;

      // pixel 2
      dot = (dp[2] << 4) + (7 * ep1[1] + 3 * ep2[3] + 5 * ep2[2] + ep2[1]);
      if (dot < halfPoint)
        step = (dot < c0Point) ? 1 : 3;
      else
        step = (dot < c3Point) ? 2 : 0;

      ep1[2] = dp[2] - stops[step];
      lmask |= step << 4;

      // pixel 3
      dot = (dp[3] << 4) + (7 * ep1[2] + 5 * ep2[3] + ep2[2]);
      if (dot < halfPoint)
        step = (dot < c0Point) ? 1 : 3;
      else
        step = (dot < c3Point) ? 2 : 0;

      ep1[3] = dp[3] - stops[step];
      lmask |= step << 6;

      // advance to next line
      sSwap(ep1, ep2);
      dp += 4;
      mask |= lmask << (y * 8);
    }
  }

  return mask;
}

// The color optimization function. (Clever code, part 1)
static void OptimizeColorsBlock(const Pixel* block, sU16& max16, sU16& min16) {
  static const sInt nIterPower = 4;

  // determine color distribution
  sInt mu[3], min[3], max[3];

  for (sInt ch = 0; ch < 3; ch++) {
    const sU8* bp = ((const sU8*)block) + ch;
    sInt muv, minv, maxv;

    muv = minv = maxv = bp[0];
    for (sInt i = 4; i < 64; i += 4) {
      muv += bp[i];
      minv = sMin<sInt>(minv, bp[i]);
      maxv = sMax<sInt>(maxv, bp[i]);
    }

    mu[ch] = (muv + 8) >> 4;
    min[ch] = minv;
    max[ch] = maxv;
  }

  // determine covariance matrix
  sInt cov[6];
  for (sInt i = 0; i < 6; i++)
    cov[i] = 0;

  for (sInt i = 0; i < 16; i++) {
    sInt r = block[i].r - mu[2];
    sInt g = block[i].g - mu[1];
    sInt b = block[i].b - mu[0];

    cov[0] += r * r;
    cov[1] += r * g;
    cov[2] += r * b;
    cov[3] += g * g;
    cov[4] += g * b;
    cov[5] += b * b;
  }

  // convert covariance matrix to float, find principal axis via power iter
  sF32 covf[6], vfr, vfg, vfb;
  for (sInt i = 0; i < 6; i++)
    covf[i] = cov[i] / 255.0f;

  vfr = max[2] - min[2];
  vfg = max[1] - min[1];
  vfb = max[0] - min[0];

  for (sInt iter = 0; iter < nIterPower; iter++) {
    sF32 r = vfr * covf[0] + vfg * covf[1] + vfb * covf[2];
    sF32 g = vfr * covf[1] + vfg * covf[3] + vfb * covf[4];
    sF32 b = vfr * covf[2] + vfg * covf[4] + vfb * covf[5];

    vfr = r;
    vfg = g;
    vfb = b;
  }

  sF32 magn = sMax(sMax(sFAbs(vfr), sFAbs(vfg)), sFAbs(vfb));
  sInt v_r, v_g, v_b;

  if (magn < 4.0f)  // too small, default to luminance
  {
    v_r = 148;
    v_g = 300;
    v_b = 58;
  } else {
    magn = 512.0f / magn;
    v_r = vfr * magn;
    v_g = vfg * magn;
    v_b = vfb * magn;
  }

  // Pick colors at extreme points
  sInt mind = 0x7fffffff, maxd = -0x7fffffff;
  Pixel minp, maxp;

  for (sInt i = 0; i < 16; i++) {
    sInt dot = block[i].r * v_r + block[i].g * v_g + block[i].b * v_b;

    if (dot < mind) {
      mind = dot;
      minp = block[i];
    }

    if (dot > maxd) {
      maxd = dot;
      maxp = block[i];
    }
  }

  // Reduce to 16 bit colors
  max16 = maxp.As16Bit();
  min16 = minp.As16Bit();
}

// The refinement function. (Clever code, part 2)
// Tries to optimize colors to suit block contents better.
// (By solving a least squares system via normal equations+Cramer's rule)
static sBool RefineBlock(const Pixel* block, sU16& max16, sU16& min16, sU32 mask) {
  static const sInt w1Tab[4] = {3, 0, 2, 1};
  static const sInt prods[4] = {0x090000, 0x000900, 0x040102, 0x010402};
  // ^some magic to save a lot of multiplies in the accumulating loop...

  sInt akku = 0;
  sInt At1_r, At1_g, At1_b;
  sInt At2_r, At2_g, At2_b;
  sU32 cm = mask;

  At1_r = At1_g = At1_b = 0;
  At2_r = At2_g = At2_b = 0;
  for (sInt i = 0; i < 16; i++, cm >>= 2) {
    sInt step = cm & 3;
    sInt w1 = w1Tab[step];
    sInt r = block[i].r;
    sInt g = block[i].g;
    sInt b = block[i].b;

    akku += prods[step];
    At1_r += w1 * r;
    At1_g += w1 * g;
    At1_b += w1 * b;
    At2_r += r;
    At2_g += g;
    At2_b += b;
  }

  At2_r = 3 * At2_r - At1_r;
  At2_g = 3 * At2_g - At1_g;
  At2_b = 3 * At2_b - At1_b;

  // extract solutions and decide solvability
  sInt xx = akku >> 16;
  sInt yy = (akku >> 8) & 0xff;
  sInt xy = (akku >> 0) & 0xff;

  if (!yy || !xx || xx * yy == xy * xy)
    return sFALSE;

  sF32 frb = 3.0f * 31.0f / 255.0f / (xx * yy - xy * xy);
  sF32 fg = frb * 63.0f / 31.0f;

  sU16 oldMin = min16;
  sU16 oldMax = max16;

  // solve.
  max16 = sClamp<sInt>((At1_r * yy - At2_r * xy) * frb + 0.5f, 0, 31) << 11;
  max16 |= sClamp<sInt>((At1_g * yy - At2_g * xy) * fg + 0.5f, 0, 63) << 5;
  max16 |= sClamp<sInt>((At1_b * yy - At2_b * xy) * frb + 0.5f, 0, 31) << 0;

  min16 = sClamp<sInt>((At2_r * xx - At1_r * xy) * frb + 0.5f, 0, 31) << 11;
  min16 |= sClamp<sInt>((At2_g * xx - At1_g * xy) * fg + 0.5f, 0, 63) << 5;
  min16 |= sClamp<sInt>((At2_b * xx - At1_b * xy) * frb + 0.5f, 0, 31) << 0;

  return oldMin != min16 || oldMax != max16;
}

// Color block compression
static void CompressColorBlock(sU8* dest, const sU32* src, sInt quality) {
  const Pixel* block = (const Pixel*)src;
  Pixel dblock[16], color[4];

  // check if block is constant
  sU32 min, max;
  min = max = block[0].v;

  for (sInt i = 1; i < 16; i++) {
    min = sMin(min, block[i].v);
    max = sMax(max, block[i].v);
  }

  // perform block compression
  sU16 min16, max16;
  sU32 mask;

  if (min != max)  // no constant color
  {
    // first step: compute dithered version for PCA if desired
    if (quality)
      DitherBlock(dblock, block);

    // second step: pca+map along principal axis
    OptimizeColorsBlock(quality ? dblock : block, max16, min16);
    if (max16 != min16) {
      EvalColors(color, max16, min16);
      mask = MatchColorsBlock(block, color, quality != 0);
    } else
      mask = 0;

    // third step: refine
    if (RefineBlock(quality ? dblock : block, max16, min16, mask)) {
      if (max16 != min16) {
        EvalColors(color, max16, min16);
        mask = MatchColorsBlock(block, color, quality != 0);
      } else
        mask = 0;
    }

  } else  // constant color
  {
    sInt r = block[0].r;
    sInt g = block[0].g;
    sInt b = block[0].b;

    mask = 0xaaaaaaaa;
    max16 = (OMatch5[r][0] << 11) | (OMatch6[g][0] << 5) | OMatch5[b][0];
    min16 = (OMatch5[r][1] << 11) | (OMatch6[g][1] << 5) | OMatch5[b][1];
  }

  // write the color block
  if (max16 < min16) {
    sSwap(max16, min16);
    mask ^= 0x55555555;
  }

  ((sU16*)dest)[0] = max16;
  ((sU16*)dest)[1] = min16;
  ((sU32*)dest)[1] = mask;
}

// Alpha block compression (this is easy for a change)
static void CompressAlphaBlock(sU8* dest, const sU32* src) {
  const Pixel* block = (const Pixel*)src;

  // find min/max color
  sInt min, max;
  min = max = block[0].a;

  for (sInt i = 1; i < 16; i++) {
    min = sMin<sInt>(min, block[i].a);
    max = sMax<sInt>(max, block[i].a);
  }

  // encode them
  *dest++ = max;
  *dest++ = min;

  // determine bias and emit color indices
  sInt dist = max - min;
  sInt bias = min * 7 - (dist >> 1);
  sInt dist4 = dist * 4;
  sInt dist2 = dist * 2;
  sInt bits = 0, mask = 0;

  for (sInt i = 0; i < 16; i++) {
    sInt a = block[i].a * 7 - bias;
    sInt ind, t;

    // select index (hooray for bit magic)
    t = (dist4 - a) >> 31;
    ind = t & 4;
    a -= dist4 & t;
    t = (dist2 - a) >> 31;
    ind += t & 2;
    a -= dist2 & t;
    t = (dist - a) >> 31;
    ind += t & 1;

    ind = -ind & 7;
    ind ^= (2 > ind);

    // write index
    mask |= ind << bits;
    if ((bits += 3) >= 8) {
      *dest++ = mask;
      mask >>= 8;
      bits -= 8;
    }
  }
}

/****************************************************************************/

void sInitDXT() {
  for (sInt i = 0; i < 32; i++)
    Expand5[i] = (i << 3) | (i >> 2);

  for (sInt i = 0; i < 64; i++)
    Expand6[i] = (i << 2) | (i >> 4);

  for (sInt i = 0; i < 256 + 16; i++) {
    sInt v = sClamp(i - 8, 0, 255);
    QuantRBTab[i] = Expand5[Mul8Bit(v, 31)];
    QuantGTab[i] = Expand6[Mul8Bit(v, 63)];
  }

  PrepareOptTable4(&OMatch5[0][0], Expand5, 32);
  PrepareOptTable4(&OMatch6[0][0], Expand6, 64);

  PrepareOptTable3(&OMatch5_3[0][0], Expand5, 32);
  PrepareOptTable3(&OMatch6_3[0][0], Expand6, 64);
}

void sCompressDXTBlock(sU8* dest, const sU32* src, sBool alpha, sInt quality) {
  CRNLIB_ASSERT(Expand5[1]);

  // if alpha specified, compress alpha as well
  if (alpha) {
    CompressAlphaBlock(dest, src);
    dest += 8;
  }

  // compress the color part
  CompressColorBlock(dest, src, quality);
}

void sCompressDXT5ABlock(sU8* dest, const sU32* src) {
  CRNLIB_ASSERT(Expand5[1]);

  CompressAlphaBlock(dest, src);
}

}  // namespace ryg_dxt