private static unsafe int GetEndPointSelectionErrorFast(ReadOnlySpan <uint> tile, int subsetCount, int partition, int w, int h, int maxError)
{
byte[] partitionTable = BC67Tables.PartitionTable[subsetCount - 1][partition];
Span <RgbaColor8> minColors = stackalloc RgbaColor8[subsetCount];
Span <RgbaColor8> maxColors = stackalloc RgbaColor8[subsetCount];
BC67Utils.GetMinMaxColors(partitionTable, tile, w, h, minColors, maxColors, subsetCount);
Span <uint> endPoints0 = stackalloc uint[subsetCount];
Span <uint> endPoints1 = stackalloc uint[subsetCount];
SelectEndPointsFast(partitionTable, tile, w, h, subsetCount, minColors, maxColors, endPoints0, endPoints1, uint.MaxValue);
Span <RgbaColor32> palette = stackalloc RgbaColor32[8];
int errorSum = 0;
for (int subset = 0; subset < subsetCount; subset++)
{
RgbaColor32 blockDir = maxColors[subset].GetColor32() - minColors[subset].GetColor32();
int sum = blockDir.R + blockDir.G + blockDir.B + blockDir.A;
if (sum != 0)
{
blockDir = (blockDir << 6) / new RgbaColor32(sum);
}
uint c0 = endPoints0[subset];
uint c1 = endPoints1[subset];
int pBit0 = GetPBit(c0, 6, 0);
int pBit1 = GetPBit(c1, 6, 0);
c0 = BC67Utils.Quantize(RgbaColor8.FromUInt32(c0), 6, 0, pBit0).ToUInt32();
c1 = BC67Utils.Quantize(RgbaColor8.FromUInt32(c1), 6, 0, pBit1).ToUInt32();
if (Sse41.IsSupported)
{
Vector128 <byte> c0Rep = Vector128.Create(c0).AsByte();
Vector128 <byte> c1Rep = Vector128.Create(c1).AsByte();
Vector128 <byte> c0c1 = Sse2.UnpackLow(c0Rep, c1Rep);
Vector128 <byte> rWeights;
Vector128 <byte> lWeights;
fixed(byte *pWeights = BC67Tables.Weights[1], pInvWeights = BC67Tables.InverseWeights[1])
{
rWeights = Sse2.LoadScalarVector128((ulong *)pWeights).AsByte();
lWeights = Sse2.LoadScalarVector128((ulong *)pInvWeights).AsByte();
}
Vector128 <byte> iWeights = Sse2.UnpackLow(rWeights, lWeights);
Vector128 <byte> iWeights01 = Sse2.UnpackLow(iWeights.AsInt16(), iWeights.AsInt16()).AsByte();
Vector128 <byte> iWeights23 = Sse2.UnpackHigh(iWeights.AsInt16(), iWeights.AsInt16()).AsByte();
Vector128 <byte> iWeights0 = Sse2.UnpackLow(iWeights01.AsInt16(), iWeights01.AsInt16()).AsByte();
Vector128 <byte> iWeights1 = Sse2.UnpackHigh(iWeights01.AsInt16(), iWeights01.AsInt16()).AsByte();
Vector128 <byte> iWeights2 = Sse2.UnpackLow(iWeights23.AsInt16(), iWeights23.AsInt16()).AsByte();
Vector128 <byte> iWeights3 = Sse2.UnpackHigh(iWeights23.AsInt16(), iWeights23.AsInt16()).AsByte();
private static Block EncodeFast(ReadOnlySpan <uint> tile, int w, int h)
{
(RgbaColor8 minColor, RgbaColor8 maxColor) = BC67Utils.GetMinMaxColors(tile, w, h);
bool alphaNotOne = minColor.A != 255 || maxColor.A != 255;
int variance = BC67Utils.SquaredDifference(minColor.GetColor32(), maxColor.GetColor32());
int selectedMode;
int indexMode = 0;
if (alphaNotOne)
{
bool constantAlpha = minColor.A == maxColor.A;
if (constantAlpha)
{
selectedMode = variance > MinColorVarianceForModeChange ? 7 : 6;
}
else
{
if (variance > MinColorVarianceForModeChange)
{
Span <uint> uniqueRGB = stackalloc uint[16];
Span <uint> uniqueAlpha = stackalloc uint[16];
int uniqueRGBCount = 0;
int uniqueAlphaCount = 0;
uint rgbMask = new RgbaColor8(255, 255, 255, 0).ToUInt32();
uint alphaMask = new RgbaColor8(0, 0, 0, 255).ToUInt32();
for (int i = 0; i < tile.Length; i++)
{
uint c = tile[i];
if (!uniqueRGB.Slice(0, uniqueRGBCount).Contains(c & rgbMask))
{
uniqueRGB[uniqueRGBCount++] = c & rgbMask;
}
if (!uniqueAlpha.Slice(0, uniqueAlphaCount).Contains(c & alphaMask))
{
uniqueAlpha[uniqueAlphaCount++] = c & alphaMask;
}
}
selectedMode = 4;
indexMode = uniqueRGBCount > uniqueAlphaCount ? 1 : 0;
}
else
{
selectedMode = 5;
}
}
}
else
{
if (variance > MinColorVarianceForModeChange)
{
selectedMode = 1;
}
else
{
selectedMode = 6;
}
}
int selectedPartition = 0;
if (selectedMode == 1 || selectedMode == 7)
{
int partitionSelectionLowestError = int.MaxValue;
for (int i = 0; i < _mostFrequentPartitions.Length; i++)
{
int p = _mostFrequentPartitions[i];
int error = GetEndPointSelectionErrorFast(tile, 2, p, w, h, partitionSelectionLowestError);
if (error < partitionSelectionLowestError)
{
partitionSelectionLowestError = error;
selectedPartition = p;
}
}
}
return(Encode(selectedMode, selectedPartition, 0, indexMode, fastMode: true, tile, w, h, out _));
}