pre-release 0.7-rc4

• fix thrust inclusion

release 0.6.0

• improve end-to-end performance
  • revert to serial CPU-based process of building Huffman codebook (with canonization)
• source code management
  • pszmem now a portable module with no dependency

release 0.5.0

• source code management
  • remove unused submodules: nvcomp, cub
• performance improvement
  • enable prediction-quantization with compaction
  • enable histsp, a sparsity-aware/special version of histogram kernel

release 0.4.2

• add prototype memory pool (in the interface) on top of pszmem-based memory management

release 0.4.1

• hotfix in memory management on top of v0.4

pre-release 0.8-rc2

• disable thrust-based error-boundness check to speed up the build process

pre-release 0.8-rc1

This release focuses on source code organization and speeding up the build process.

• disable some research-oriented methods (e.g., building Huffman codebook on GPU) by default
• THRUST now a backend explicitly; less thrust-based methods are used to speed up the build process.
• retire CUB in CMakeLists.txt

pre-release 0.7-rc1

• change top-level APIs: explicitly separate init (with malloc), compress/decompress, and release (free)

pre-release 0.4.0

• improve the build process
  • improve the compiler (combination) compatibility
  • improve compiler firewall to link using host compiler
  • parallelize template instantiation
• improve memory management
  • wrap into structure pszmem
• modularize kernels
  • create variants of Lorenzo prediction kernels for maintainability
  • use shfl in Lorenzo 3D kernel
  • explicitly isolate architecture-dependent parallel primitives
• declutter source code
  • decrease wrapper/abstraction levels
  • use less OOD

release 0.3.1

• high-level compressor API in C
• retire old build script in favor of cmake use
• dynamic link by default
• simplify gather-scatter (outlier handling) using thrust
• initial test case and example in favor of portability
• modularization and compiler firewall
• document moved to repo wiki