- Force macro expansion in version namespace identifier
- Update to RTM v2.1.5
- Update sjson-cpp to v0.8.3
- Add support for clang 12, 13, and 14
- Add support for GCC 11
- Add support for XCode 12 and 13
- Add support for Arm64 development on OS X and Linux
- Misc CI improvements
- Update to RTM v2.1.4
- Update to Catch2 v2.13.7
- Disable versioned namespace by default to avoid breaking ABI in patch release
- Update sjson-cpp to v0.8.2
- Update rtm to v2.1.3
- Add versioned namespace to allow multiple versions to coexist within a binary
- Fix database sampling interpolation when using a rounding mode other than
none
- Other minor fixes
- Fix potential heap corruption when stripping a database that isn't split
- Fix database tier bulk data size to include its padding
- Fix incorrect seek offset when seeking past 0.0 in a single frame clip
- Unified and cleaned up APIs
- Cleaned up naming convention to match C++ stdlib, boost
- Introduced streaming database support
- Decompression profiling now uses Google Benchmark
- Decompression has been heavily optimized
- Compression has been heavily optimized
- First release to support backwards compatibility going forward
- Migrated all math to Realtime Math
- Clips now support 4 billion samples/tracks
- WebAssembly support added through emscripten
- Many other improvements
- Gracefully fail compression when we have more than 50000 samples
- Update Catch2 to 2.13.1 to work with latest CMake
- Avoid assert when using an additive base with a static pose/single frame
- Fix single track decompression when scale is present with more than one segment
- Gracefully fail compression when we have more than 65535 samples
- Fix crash when compressing with an empty track array
- Strip unused code when stat logging isn't required
- Fix CompressedClip hash to be deterministic
- Fix bug with scalar track decompression where garbage could be returned
- Fix scalar track quantization to properly check the resulting error
- Fix scalar track creation and properly copy the sample size
- Other minor fixes and improvements
- Added support for VS2019, GCC 9, clang7, and Xcode 11
- Updated sjson-cpp and added a dependency on Realtime Math (RTM)
- Optimized compression and decompression significantly
- Added support for multiple root bones
- Added support for scalar track compression
- Many bug fixes and improvements
- Silence SSE floating point exceptions during compression
- Minor fixes
- Added support for GCC 8, clang 6, Xcode 10, and Windows ARM64
- Updated catch2 and sjson-cpp
- Integrated SonarCloud
- Added a compression level setting and changed default to Medium
- Various bug fixes, minor optimizations, and cleanup
- Added proper ARM NEON support
- Properly detect SSE2 with MSVC if AVX isn't used
- Lots of decompression performance optimizations
- Minor fixes and cleanup
- Minor additions to fully support UE4
- Minor cleanup
- Improved error handling
- Added additive clip support
- Added
acl_decompressor
tool to profile and test decompression - Increased warning levels to highest possible
- Many more improvements and fixes
- Added full support for Android and iOS
- Added support for GCC6 and GCC7 on Linux
- Downgraded C++ version to from 14 to 11
- Added regression tests
- Added lots of unit tests for core and math headers
- Many more improvements and fixes
- Hooked up continuous build integration
- Added support for VS2017 on Windows
- Added support for GCC5, Clang4, and Clang5 on Linux
- Added support for Xcode 8.3 and Xcode 9.2 on OS X
- Added support for x86 on Windows, Linux, and OS X
- Better handle scale with built in error metrics
- Many more improvements and fixes
- Added support for 3D scale
- Added partial support for Android (works in Unreal 4.15)
- A fix to the variable bit rate optimization algorithm
- Added a CLA system
- Refactoring to support multiple algorithms better
- More changes and additions to stat logging
- Many more improvements and fixes
- Lots of math performance, accuracy, and consistency improvements
- Implemented segmenting support in uniformly sampled algorithm
- Range reduction per segment support added
- Minor fixes to fbx2acl
- Optimized variable quantization considerably
- Major changes to which stats are dumped and how they are processed
- Many more improvements and fixes
- Added CMake support
- Improved error measuring and accuracy
- Improved variable quantization
- Convert most math to float32 to improve accuracy and performance
- Many more improvements and fixes
- Added clip_writer to create ACL files from a game integration
- Added some unit tests and moved them into their own project
- Added basic per track variable quantization
- Added CMake support
- Lots of cleanup, minor changes, and fixes
Initial release!
- Uniformly sampled algorithm
- Various rotation and vector formats
- Clip range reduction
- ACL SJSON file format
- Custom allocator interface
- Assert overrides
- Custom math types and functions
- Various tools to test the library
- Visual Studio 2015 supported, x86 and x64