void DecodeARW(ImageBinaryReader input, long w, long h)
{
BitPumpMSB bits = new BitPumpMSB(input);
int sum = 0;
for (long x = w; (x--) != 0;)
{
for (int y = 0; y < h + 1; y += 2)
{
bits.Fill();
if (y == h)
{
y = 1;
}
int len = 4 - (int)bits.GetBits(2);
if (len == 3 && bits.GetBit() != 0)
{
len = 0;
}
if (len == 4)
{
while (len < 17 && bits.GetBit() == 0)
{
len++;
}
}
int diff = (int)bits.GetBits(len);
if (len != 0 && (diff & (1 << (len - 1))) == 0)
{
diff -= (1 << len) - 1;
}
sum += diff;
Debug.Assert((sum >> 12) == 0);
if (y < h)
{
rawImage.fullSize.rawView[x + y * rawImage.fullSize.dim.width] = (ushort)sum;
}
}
}
}
/* This is probably the slowest decoder of them all.
* I cannot see any way to effectively speed up the prediction
* phase, which is by far the slowest part of this algorithm.
* Also there is no way to multithread this code, since prediction
* is based on the output of all previous pixel (bar the first four)
*/
private void DecodeCompressed(ImageBinaryReader s, uint width, uint height)
{
int nbits;
long left0, nw0, left1, nw1;
long sign, low, high;
long[] acarry0 = new long[3], acarry1 = new long[3];
long pred, diff;
//uint pitch = rawImage.pitch;
/* Build a table to quickly look up "high" value */
byte[] bittable = new byte[4096];
for (int i = 0; i < 4096; i++)
{
int b = i;
for (high = 0; high < 12; high++)
{
if (((b >> (11 - (int)high)) & 1) != 0)
{
break;
}
}
bittable[i] = (byte)Math.Min(12, high);
}
left0 = nw0 = left1 = nw1 = 0;
s.ReadBytes(7);
BitPumpMSB bits = new BitPumpMSB(s);
for (int y = 0; y < height; y++)
{
var pos = y * rawImage.fullSize.UncroppedDim.width;
acarry0 = new long[3];
acarry1 = new long[3];
bool y_border = y < 2;
bool border = true;
for (int x = 0; x < width; x++)
{
bits.Fill();
int i = 0;
if (acarry0[2] < 3)
{
i = 2;
}
for (nbits = 2 + i; acarry0[0] >> (nbits + i) != 0; nbits++)
{
;
}
uint b = bits.PeekBits(15);
sign = (b >> 14) * -1;
low = (b >> 12) & 3;
high = bittable[b & 4095];
// Skip bytes used above or read bits
if (high == 12)
{
bits.SkipBits(15);
high = bits.GetBits(16 - nbits) >> 1;
}
else
{
bits.SkipBits((int)high + 1 + 3);
}
acarry0[0] = (high << nbits) | bits.GetBits(nbits);
diff = (acarry0[0] ^ sign) + acarry0[1];
acarry0[1] = (diff * 3 + acarry0[1]) >> 5;
acarry0[2] = acarry0[0] > 16 ? 0 : acarry0[2] + 1;
if (border)
{
if (y_border && x < 2)
{
pred = 0;
}
else if (y_border)
{
pred = left0;
}
else
{
pred = nw0 = rawImage.fullSize.rawView[pos - rawImage.fullSize.UncroppedDim.width + x];
}
rawImage.fullSize.rawView[pos + x] = (ushort)(pred + ((diff << 2) | low));
// Set predictor
left0 = rawImage.fullSize.rawView[pos + x];
}
else
{
// Have local variables for values used several tiles
// (having a "UInt16 *dst_up" that caches dest[-pitch+((int)x)] is actually slower, probably stack spill or aliasing)
int up = rawImage.fullSize.rawView[pos - rawImage.fullSize.UncroppedDim.width + x];
long leftMinusNw = left0 - nw0;
long upMinusNw = up - nw0;
// Check if sign is different, and one is not zero
if (leftMinusNw * upMinusNw < 0)
{
if (Other_abs(leftMinusNw) > 32 || Other_abs(upMinusNw) > 32)
{
pred = left0 + upMinusNw;
}
else
{
pred = (left0 + up) >> 1;
}
}
else
{
pred = Other_abs(leftMinusNw) > Other_abs(upMinusNw) ? left0 : up;
}
rawImage.fullSize.rawView[pos + x] = (ushort)(pred + ((diff << 2) | low));
// Set predictors
left0 = rawImage.fullSize.rawView[pos + x];
nw0 = up;
}
// ODD PIXELS
x += 1;
bits.Fill();
i = 0;
if (acarry1[2] < 3)
{
i = 2;
}
for (nbits = 2 + i; acarry1[0] >> (nbits + i) != 0; nbits++)
{
;
}
b = bits.PeekBits(15);
sign = (b >> 14) * -1;
low = (b >> 12) & 3;
high = bittable[b & 4095];
// Skip bytes used above or read bits
if (high == 12)
{
bits.SkipBits(15);
high = bits.GetBits(16 - nbits) >> 1;
}
else
{
bits.SkipBits((int)high + 1 + 3);
}
acarry1[0] = (high << nbits) | bits.GetBits(nbits);
diff = (acarry1[0] ^ sign) + acarry1[1];
acarry1[1] = (diff * 3 + acarry1[1]) >> 5;
acarry1[2] = acarry1[0] > 16 ? 0 : acarry1[2] + 1;
if (border)
{
if (y_border && x < 2)
{
pred = 0;
}
else if (y_border)
{
pred = left1;
}
else
{
pred = nw1 = rawImage.fullSize.rawView[pos - rawImage.fullSize.UncroppedDim.width + x];
}
rawImage.fullSize.rawView[pos + x] = (ushort)(left1 = pred + ((diff << 2) | low));
}
else
{
int up = rawImage.fullSize.rawView[pos - rawImage.fullSize.UncroppedDim.width + x];
long leftminusNw = left1 - nw1;
long upminusNw = up - nw1;
// Check if sign is different, and one is not zero
if (leftminusNw * upminusNw < 0)
{
if (Other_abs(leftminusNw) > 32 || Other_abs(upminusNw) > 32)
{
pred = left1 + upminusNw;
}
else
{
pred = (left1 + up) >> 1;
}
}
else
{
pred = Other_abs(leftminusNw) > Other_abs(upminusNw) ? left1 : up;
}
rawImage.fullSize.rawView[pos + x] = (ushort)(left1 = pred + ((diff << 2) | low));
nw1 = up;
}
border = y_border;
}
}
}