private void SpanValues(out RgbF32 r0, out RgbF32 r1)
{
bool hasAlpha = alphaMask != 0;
int cSteps = hasAlpha ? 3 : 4;
var pC = hasAlpha ? pC3 : pC4;
var pD = hasAlpha ? pD3 : pD4;
var values = this.qvalues;
// Find Min and Max points, as starting point
RgbF32 X = UseUniformWeighting ? new RgbF32(1, 1, 1) :
new RgbF32(RWeight, GWeight, BWeight);
RgbF32 Y = new RgbF32(0, 0, 0);
for (int i = 0; i < values.Length; i++)
{
var v = values[i];
#if COLOR_WEIGHTS
if ((alphaMask & (1 << i)) != 0)
#endif
{
if (v.R < X.R)
{
X.R = v.R;
}
if (v.G < X.G)
{
X.G = v.G;
}
if (v.B < X.B)
{
X.B = v.B;
}
if (v.R > Y.R)
{
Y.R = v.R;
}
if (v.G > Y.G)
{
Y.G = v.G;
}
if (v.B > Y.B)
{
Y.B = v.B;
}
}
}
// Diagonal axis
RgbF32 AB;
AB.R = Y.R - X.R;
AB.G = Y.G - X.G;
AB.B = Y.B - X.B;
float fAB = AB.R * AB.R + AB.G * AB.G + AB.B * AB.B;
// Single color block.. no need to root-find
if (fAB < float.Epsilon)
{
r0 = X;
r1 = Y;
return;
}
// Try all four axis directions, to determine which diagonal best fits data
float fABInv = 1.0f / fAB;
RgbF32 Dir;
Dir.R = AB.R * fABInv;
Dir.G = AB.G * fABInv;
Dir.B = AB.B * fABInv;
RgbF32 Mid;
Mid.R = (X.R + Y.R) * 0.5f;
Mid.G = (X.G + Y.G) * 0.5f;
Mid.B = (X.B + Y.B) * 0.5f;
fDir[0] = fDir[1] = fDir[2] = fDir[3] = 0.0F;
for (int i = 0; i < values.Length; i++)
{
var v = values[i];
RgbF32 Pt;
Pt.R = (v.R - Mid.R) * Dir.R;
Pt.G = (v.G - Mid.G) * Dir.G;
Pt.B = (v.B - Mid.B) * Dir.B;
float f;
#if COLOR_WEIGHTS
f = Pt.R + Pt.G + Pt.B;
fDir[0] += v.a * f * f;
f = Pt.R + Pt.G - Pt.B;
fDir[1] += v.a * f * f;
f = Pt.R - Pt.G + Pt.B;
fDir[2] += v.a * f * f;
f = Pt.R - Pt.G - Pt.B;
fDir[3] += v.a * f * f;
#else
f = Pt.R + Pt.G + Pt.B;
fDir[0] += f * f;
f = Pt.R + Pt.G - Pt.B;
fDir[1] += f * f;
f = Pt.R - Pt.G + Pt.B;
fDir[2] += f * f;
f = Pt.R - Pt.G - Pt.B;
fDir[3] += f * f;
#endif
}
float fDirMax = fDir[0];
int iDirMax = 0;
for (int iDir = 1; iDir < fDir.Length; iDir++)
{
var d = fDir[iDir];
if (d > fDirMax)
{
fDirMax = d;
iDirMax = iDir;
}
}
if ((iDirMax & 2) != 0)
{
float f = X.G; X.G = Y.G; Y.G = f;
}
if ((iDirMax & 1) != 0)
{
float f = X.B; X.B = Y.B; Y.B = f;
}
// Two color block.. no need to root-find
if (fAB < 1.0f / 4096.0f)
{
r0 = X;
r1 = Y;
return;
}
// Use Newton's Method to find local minima of sum-of-squares error.
float fSteps = (float)(cSteps - 1);
for (int iIteration = 0; iIteration < 8; iIteration++)
{
// Calculate new steps
for (int iStep = 0; iStep < cSteps; iStep++)
{
interpValues[iStep].R = X.R * pC[iStep] + Y.R * pD[iStep];
interpValues[iStep].G = X.G * pC[iStep] + Y.G * pD[iStep];
interpValues[iStep].B = X.B * pC[iStep] + Y.B * pD[iStep];
}
// Calculate color direction
Dir.R = Y.R - X.R;
Dir.G = Y.G - X.G;
Dir.B = Y.B - X.B;
float fLen = (Dir.R * Dir.R + Dir.G * Dir.G + Dir.B * Dir.B);
if (fLen < (1.0f / 4096.0f))
{
break;
}
float fScale = fSteps / fLen;
Dir.R *= fScale;
Dir.G *= fScale;
Dir.B *= fScale;
// Evaluate function, and derivatives
float d2X, d2Y;
RgbF32 dX, dY;
d2X = d2Y = dX.R = dX.G = dX.B = dY.R = dY.G = dY.B = 0.0f;
for (int i = 0; i < values.Length; i++)
{
var v = values[i];
float fDot = (v.R - X.R) * Dir.R +
(v.G - X.G) * Dir.G +
(v.B - X.B) * Dir.B;
int iStep;
if (fDot <= 0.0f)
{
iStep = 0;
}
else if (fDot >= fSteps)
{
iStep = cSteps - 1;
}
else
{
iStep = (int)(fDot + 0.5f);
}
RgbF32 Diff;
Diff.R = interpValues[iStep].R - v.R;
Diff.G = interpValues[iStep].G - v.G;
Diff.B = interpValues[iStep].B - v.B;
#if COLOR_WEIGHTS
float fC = pC[iStep] * v.a * (1.0f / 8.0f);
float fD = pD[iStep] * v.a * (1.0f / 8.0f);
#else
float fC = pC[iStep] * (1.0f / 8.0f);
float fD = pD[iStep] * (1.0f / 8.0f);
#endif // COLOR_WEIGHTS
d2X += fC * pC[iStep];
dX.R += fC * Diff.R;
dX.G += fC * Diff.G;
dX.B += fC * Diff.B;
d2Y += fD * pD[iStep];
dY.R += fD * Diff.R;
dY.G += fD * Diff.G;
dY.B += fD * Diff.B;
}
// Move endpoints
if (d2X > 0.0f)
{
float f = -1.0f / d2X;
X.R += dX.R * f;
X.G += dX.G * f;
X.B += dX.B * f;
}
if (d2Y > 0.0f)
{
float f = -1.0f / d2Y;
Y.R += dY.R * f;
Y.G += dY.G * f;
Y.B += dY.B * f;
}
if ((dX.R * dX.R < fEpsilon) && (dX.G * dX.G < fEpsilon) && (dX.B * dX.B < fEpsilon) &&
(dY.R * dY.R < fEpsilon) && (dY.G * dY.G < fEpsilon) && (dY.B * dY.B < fEpsilon))
{
break;
}
}
r0 = X;
r1 = Y;
}
public BC1Block Encode()
{
BC1Block ret;
if (alphaMask == 0xFFFF)
{
ret.PackedValue = BC1Block.TransparentValue;
return(ret);
}
QuantizeValues();
RgbF32 r0, r1;
SpanValues(out r0, out r1);
//quantize the endpoints
bool weightValues = !UseUniformWeighting;
if (weightValues)
{
r0.R *= RInvWeight;
r0.G *= GInvWeight;
r0.B *= BInvWeight;
r1.R *= RInvWeight;
r1.G *= GInvWeight;
r1.B *= BInvWeight;
}
var pr0 = Rgb565.Pack(r0);
var pr1 = Rgb565.Pack(r1);
if (alphaMask == 0 && pr0.PackedValue == pr1.PackedValue)
{
return(new BC1Block(pr0, pr1));
}
pr0.Unpack(out r0);
pr1.Unpack(out r1);
if (weightValues)
{
r0.R *= RWeight;
r0.G *= GWeight;
r0.B *= BWeight;
r1.R *= RWeight;
r1.G *= GWeight;
r1.B *= BWeight;
}
//interp out the steps
RgbF32 s0;
if ((alphaMask != 0) == (pr0.PackedValue <= pr1.PackedValue))
{
ret = new BC1Block(pr0, pr1);
interpValues[0] = s0 = r0;
interpValues[1] = r1;
}
else
{
ret = new BC1Block(pr1, pr0);
interpValues[0] = s0 = r1;
interpValues[1] = r0;
}
uint[] pSteps;
if (alphaMask != 0)
{
pSteps = pSteps3;
RgbF32.Lerp(out interpValues[2], interpValues[0], interpValues[1], 0.5F);
}
else
{
pSteps = pSteps4;
RgbF32.Lerp(out interpValues[2], interpValues[0], interpValues[1], 1.0F / 3.0F);
RgbF32.Lerp(out interpValues[3], interpValues[0], interpValues[1], 2.0F / 3.0F);
}
//find the best values
RgbF32 dir;
dir.R = interpValues[1].R - s0.R;
dir.G = interpValues[1].G - s0.G;
dir.B = interpValues[1].B - s0.B;
float fSteps = alphaMask != 0 ? 2 : 3;
float fScale = (pr0.PackedValue != pr1.PackedValue) ?
(fSteps / (dir.R * dir.R + dir.G * dir.G + dir.B * dir.B)) : 0.0F;
dir.R *= fScale;
dir.G *= fScale;
dir.B *= fScale;
bool dither = DitherRgb;
if (dither)
{
if (error == null)
{
error = new RgbF32[16];
}
else
{
Array.Clear(error, 0, 16);
}
}
for (int i = 0; i < values.Length; i++)
{
if ((alphaMask & (1 << i)) != 0)
{
ret.PackedValue |= 3U << (32 + i * 2);
}
else
{
var cl = values[i];
if (weightValues)
{
cl.R *= RWeight;
cl.G *= GWeight;
cl.B *= BWeight;
}
if (dither)
{
var e = error[i];
cl.R += e.R;
cl.G += e.G;
cl.B += e.B;
}
float fDot =
(cl.R - s0.R) * dir.R +
(cl.G - s0.G) * dir.G +
(cl.B - s0.B) * dir.B;
uint iStep;
if (fDot <= 0)
{
iStep = 0;
}
else if (fDot >= fSteps)
{
iStep = 1;
}
else
{
iStep = pSteps[(int)(fDot + 0.5F)];
}
ret.PackedValue |= (ulong)iStep << (32 + i * 2);
if (dither)
{
RgbF32 e, d, interp = interpValues[iStep];
d.R = cl.R - interp.R;
d.G = cl.G - interp.G;
d.B = cl.B - interp.B;
if ((i & 3) != 3)
{
e = error[i + 1];
e.R += d.R * (7.0F / 16.0F);
e.G += d.G * (7.0F / 16.0F);
e.B += d.B * (7.0F / 16.0F);
error[i + 1] = e;
}
if (i < 12)
{
if ((i & 3) != 0)
{
e = error[i + 3];
e.R += d.R * (3.0F / 16.0F);
e.G += d.G * (3.0F / 16.0F);
e.B += d.B * (3.0F / 16.0F);
error[i + 3] = e;
}
e = error[i + 4];
e.R += d.R * (5.0F / 16.0F);
e.G += d.G * (5.0F / 16.0F);
e.B += d.B * (5.0F / 16.0F);
error[i + 4] = e;
if (3 != (i & 3))
{
e = error[i + 5];
e.R += d.R * (1.0F / 16.0F);
e.G += d.G * (1.0F / 16.0F);
e.B += d.B * (1.0F / 16.0F);
error[i + 5] = e;
}
}
}
}
}
return(ret);
}
internal void Unpack(out RgbF32 cl)
{
cl.R = RF;
cl.G = GF;
cl.B = BF;
}
public static void Lerp(out RgbF32 o, RgbF32 a, RgbF32 b, float t)
{
o.R = a.R + t * (b.R - a.R);
o.G = a.G + t * (b.G - a.G);
o.B = a.B + t * (b.B - a.B);
}
internal static Rgb565 Pack(RgbF32 cl)
{
return(Pack(cl.R, cl.G, cl.B));
}