V 2.1.0

• 60% compression speed improvement

V 2.0.0

• A small change in the encoding tables for 8 bits and under results in improved compression with no speed penalty
• Inputs with very small dimensions are allowed. Previously each dimension had a minimum of 4 pixels, now even 1x1 images are properly handled. Images with 16 pixels and under are stored while larger inputs with one dimension under 4 pixels are reordered and encoded
• Incompatible with V 1.x

V 1.3.2

Speed improvements, especially for 8 bit data types

V 1.3.1

• Made QB3M_FTL the default mode. The output is barely larger than QB3M_BASE
  while being 25% faster for both encoding and decoding, a significant advantage
• Refactored RLE
• Add comparison with PNG document
• Update QB3 algorithm description to better reflect current code

V1.3.0: Merge pull request #21 from lucianpls/one-three

• Stride encode and decode, to allow tiled or parallel encoding and decoding of large rasters
• Single public header file, QB3.h
• Performance tweaks and bug fixes
• Improved testing

V 1.2.1

• Bug fix, decoding of best mode could fail for 32 and 64 bit integers, due to
  overflow of an internal accumulator used to decode index groups.
• Removed encoding dependency on STL::sort by including an implementation
  of bubble sort. Since the sorted vector had at most 8 values, the use of STL sort
  was overkill and increased size of the library code without a noticeable
  performance benefit. A side benefit is that porting to C is now easier.

V 1.2

Version 1.2.0

Faster code
    Faster encoding by replacing the max value search with a faster way to determine top set bit
    Faster decoding by using inline constant lookup tables and skipping temporary buffer for rungs 1 and 2

New mode, QB3M_FTL
    25% faster with a tiny (1%) compression loss
    Removes the top bit step compression
    Test availability of QB3_HAS_FTL macro definition

Web Assembly
    Modified code to make WASM compilation possible
    Test encoding and decoding, about 25% performance loss via native compilation on x64

V 1.1

Version 1.1.0

New scan order, second order Hilbert curve is now the default
    Better compression at the same or better speed
    The old scan order is available
Small performance improvements
Bug fixes
Stride decoding, allowind decoding to non-contiguous line buffers
Build system improvements
    Removed MSVC project, CMake is now the only build system
    Default target is now the library, not the conversion program
    cqb3 and the test program are not built by default

1.0

Initial release, using Morton curve by default