1.5.2

New features
Added compression status return for high level API
Added build example with ADIOS2 operators
Enable autotuning for kernels in Huffman encoding
Enabled async LZ4 and Zstd compression
Added pipeline optimization for compression time series data
Added compressor cache optimization
Added functions for explicitly initialize and finalize RuntimeX
Improved memory usage estimation
Added a function to estimate max output size of MDR-X
Added readme for build scripts

Bug fixes
Fixed a bug in Huffman codebook generation
Fixed a bug when using LZ4 compression
Fixed a bug in domain decomposer
Fixed a bug when using reduce memory footprint mode
Fixed a bug in MDR-X reconstruction
Fixed a bug when calculating sign-bit using GCC

1.5.0

New features

• New MGARD-lambda compression pipeline for presenting non-linear QoIs in XGC data
• New MGARD-RoI compression pipeline for preserving Region-of-Interest (RoI)
• New GPU compression pipeline for out-of-core large-scale data compression
• Support Apple silicon ARM architecture

Bug fixes

• Fixed an issue with older CUDA versions
• Fixed an issue with CUDA 12+
• Fixed an issue with NVIDIA HPC SDK
• Fixed a compilation bug in MDR-X
• Fixed an issue with linear quantization overflow
• Fixed an issue with using device buffers as input for compression

MGARD 1.4.0

• Added multi-device support for high-level compression/decompression
• Added multi-device support in RuntimeX and Array
• Added workspace pre-allocation functionality for low-level compression/decompression
• Fixed an issue with GCC 9
• Added logging functions
• Fixed synchronization issue with Huffman encoding
• Added a build option to enable/disable uniform coordinates normalization
• Added a new OpenMP backend
• Fixed ambiguous overloads for Xcode
• Added a new DeviceLauncher
• Greatly reduced build time by allowing disabling autotuning
• Modularized compression workflow and multi-precision refactoring workflow
• Added high-level APIs for MDR-X
• Added block-based domain decomposition method
• Improved quantization performance
• Fixed a build issue when multiple backends are enabled
• Fixed a bug in the 3D correction calculation
• Fixed an issue with collective operations in the SYCL backend
• Updated build scripts for Ampere architecture and integration with ADIOS2

MGARD 1.3.0

Added MGARD-X for portable compression for CPU and GPU

• Include support for CPU (serial), CPU (multi-threaded), NVIDIA GPUs, AMD GPUs, and Intel GPUs
• Include self-describing format
• Include automatic domain decomposition for handling large datasets
• Include automatic parallel compression across multiple GPUs
• Include high-level compression API for automatic metadata management and domain decomposition
• Include low-level compression API for full control over the compression process on devices

Added MDR-X for portable multi-precision data refactoring on CPU and GPU

• Modular design to allow flexible control over data refactoring and reconstruction process
• Has the same portability as MGARD-X

MGARD 1.0.0

• Migrate continuous integration setup
• Improve CPU compression and decompression speed
  • iterator optimizations
  • index precomputations
  • shuffle values to improve memory access patterns
• Added self-describing command line executable
• Add support for 'flat' datasets
• Parallelize operator application with OpenMP
• Improve quantization speed (improve quantum computation speed)
• Fixed several bugs in the multilevel decomposition
• Improved the performance of the multilevel decomposition
• Added a command line interface
• Added high-level APIs for easier integration with user’s program and I/O libraries
• Enabling users to choose different kinds of lossless compressors
• Added self-describing decompression API
• Added examples

MGARD 0.1.0

MGARD 0.1.0 includes the following new features

• Initial support for unstructured data
• re-structured code for expandability
• GPU support for 2D and 3D
• Entropy encoding using Huffman
• Integration with ZSTD
• Continuous integration using Travis CI

MGARD quantities of interest release

This is the initial MGARD version with support for lossy compression while preserving quantities of interest. This version also adds single precision support.