/// <summary>Compress a buffer in native memory</summary>
/// <param name="buffer">Pointer to the native memory buffer to compress</param>
/// <param name="count">Number of bytes to compress</param>
/// <param name="output">Where to write the compressed words</param>
/// <returns>Number of extra bits that where output in the last literal word (or none if 0)</returns>
internal static unsafe int CompressToUnsafe(byte *buffer, int count, CompressedBitmapWriter output)
{
// Simplified algorithm:
// 1) read 31 bits from input (BE)
// 2) if not all 0 (or all 1), then output a literal word (MSB set to 0), and jump back to step 1)
// 3) set LENGTH = 1, FILL_BIT = 0 (or 1)
// 4) Peek at next 31 bits, and if they are still all 0 (or 1), increment N, and jump back to step 4)
// 5) output a repeat word, with MSB set to 1, followed by FILL_BIT, and then LENGTH-1 (30 bit), and jump back to step 1)
// Optimizations:
// - for very small inputs (3 bytes or less) we return a single literal word
// - we read 64 bits at a time in the buffer, because it fits nicely in an UInt64 register
var bucket = new UncompressedWordReader(buffer, count);
uint word;
while ((word = bucket.Read()) != UncompressedWordReader.NotEnough)
{
output.Write(word);
}
// if there are remaining bits, they are padded with 0 and written as a literal
int bits = bucket.Bits;
if (bits > 0)
{
//note: MSB will already be 0
word = bucket.ReadLast();
output.Write(word);
}
// write the header
output.Pack();
return(bits);
}
/// <summary>Compress a buffer in native memory</summary>
/// <param name="buffer">Pointer to the native memory buffer to compress</param>
/// <param name="count">Number of bytes to compress</param>
/// <param name="output">Where to write the compressed words</param>
/// <returns>Number of extra bits that where output in the last literal word (or none if 0)</returns>
internal static unsafe int CompressToUnsafe(byte* buffer, int count, CompressedBitmapWriter output)
{
// Simplified algorithm:
// 1) read 31 bits from input (BE)
// 2) if not all 0 (or all 1), then output a literal word (MSB set to 0), and jump back to step 1)
// 3) set LENGTH = 1, FILL_BIT = 0 (or 1)
// 4) Peek at next 31 bits, and if they are still all 0 (or 1), increment N, and jump back to step 4)
// 5) output a repeat word, with MSB set to 1, followed by FILL_BIT, and then LENGTH-1 (30 bit), and jump back to step 1)
// Optimisations:
// - for very small inputs (3 bytes or less) we return a single literal word
// - we read 64 bits at a time in the buffer, because it fits nicely in an UInt64 register
var bucket = new UncompressedWordReader(buffer, count);
uint word;
while ((word = bucket.Read()) != UncompressedWordReader.NotEnough)
{
output.Write(word);
}
// if there are reamining bits, they are padded with 0 and written as a literal
int bits = bucket.Bits;
if (bits > 0)
{
//note: MSB will already be 0
word = bucket.ReadLast();
output.Write(word);
}
// write the header
output.Pack();
return bits;
}
