1.0.0 (August 1, 2022)

This release is not ABI compatible with prior releases due to numerous changes to function signatures and data structures like zfp_field. However, few of the API changes, other than to the cfp C API for compressed arrays, should impact existing code. Note that numerous header files have been renamed or moved relative to prior versions.

Added

• zfp::const_array: read-only variable-rate array that supports fixed-precision, fixed-accuracy, and reversible modes.
• Compressed-array classes for 4D data.
• const versions of array references, pointers, and iterators.
• A more complete API for pointers and iterators.
• cfp support for proxy references and pointers, iterators, and (de)serialization.
• Support for pointers and iterators into array views.
• zfp::array::size_bytes() allows querying the size of different components of an array object (e.g., payload, cache, index, metadata, ...).
• Templated C++ wrappers around the low-level C API.
• A generic codec for storing blocks of uncompressed scalars in zfp's C++ arrays.
• Additional functions for querying zfp_field and zfp_stream structs.
• zfp_config: struct that encapsulates compression mode and parameters.
• Rounding modes for reducing bias in compression errors.
• New examples: array, iteratorC, and ppm.

Changed

• Headers from array/, cfp/include/, and include/ have been renamed and reorganized into a common include/ directory.
  • The libzfp API is now confined to zfp.h, zfp.hpp, and zfp.mod for C, C++, and Fortran bindings, respectively. These all appear in the top-level include/ directory upon installation.
  • C++ headers now use a .hpp suffix; C headers use a .h suffix.
  • C++ headers like array/zfparray.h have been renamed zfp/array.hpp.
  • C headers like cfp/include/cfparrays.h have been renamed zfp/array.h.
• size_t and ptrdiff_t replace uint and int for array sizes and strides in the array classes and C/Fortran APIs.
• zfp_bool replaces int as Boolean type in the C API.
• bitstream_offset and bitstream_size replace size_t to ensure support for 64-bit offsets into and lengths of bit streams. Consequently, the bitstream API has changed accordingly.
• All array and view iterators are now random-access iterators.
• Array inspectors now return const_reference rather than a scalar type like float to allow obtaining a const_pointer to an element of an immutable array.
• zfp::array::compressed_data() now returns void* instead of uchar*.
• The array (de)serialization API has been revised, resulting in new zfp::array::header and zfp::exception classes with new exception messages.
• The array codec class is now responsible for all details regarding compression.
• The compressed-array C++ implementation has been completely refactored to make it more modular, extensible, and reusable across array types.
• Array block shapes are now computed on the fly rather than stored.
• The cfp C API now wraps array objects in structs.
• The zfpy Python API now supports the more general memoryview over bytes objects for decompression.
• The zFORp Fortran module name is now zfp instead of zforp_module.
• Some command-line options for the diffusion example have changed.
• CMake 3.9 or later is now required for CMake builds.

Removed

• zfp::array::get_header() has been replaced with a zfp::array::header constructor that accepts an array object.
• ZFP_VERSION_RELEASE is no longer defined (use ZFP_VERSION_PATCH).

Fixed

• #66: make install overwrites googletest.
• #84: Incorrect order of parameters in CUDA memset().
• #86: C++ compiler warns when __STDC_VERSION__ is undefined.
• #87: CXXFLAGS is misspelled in cfp/src/Makefile.
• #98: zfp_stream_maximum_size() underestimates size in reversible mode.
• #99: Incorrect `pr

0.5.5 (May 5, 2019)

• Added support for reversible (lossless) compression of floating-point and integer data.

• Added methods for serializing and deserializing zfp's compressed arrays.

• Added Python bindings for compressing NumPy arrays.

• Added Fortran bindings to zfp's high-level C API.

• Change:

  • The default compressed-array cache size is now a function of the total number of array elements, irrespective of array shape.

• Bug fixes:

  • Incorrect handling of execution policy in zfp utility.
  • Incorrect handling of decompression via header in zfp utility.
  • Incorrect cleanup of device memory in CUDA decompress.
  • Tests for failing mallocs.
  • CMake installation of CFP when built.
  • zfp_write_header and zfp_field_metadata now fail if array dimensions are too large to fit in header.

0.5.4 (October 1, 2018)

• Added support for CUDA fixed-rate compression and decompression.

• Added views into compressed arrays for thread safety, nested array indexing, slicing, and array subsetting.

• Added C language bindings for compressed arrays.

• Added support for compressing and decompressing 4D data.

• Changes:

  • Execution policy now applies to both compression and decompression.
  • Compressed array accessors now return Scalar type instead of const Scalar& to avoid stale references to evicted cache lines.

• Bug fixes:

  • Handling of negative strides.
  • Command line tool handling of arrays with more than 2^32 elements.
  • bitstream C++ compatibility.
  • Respect minimum cache size request.

0.5.3 (March 28, 2018)

• Added support for OpenMP multithreaded compression (but not decompression).

• Added options for OpenMP execution to zfp command-line tool.

• Changed return value of zfp_decompress to indicate the number of compressed bytes processed so far (now returns same value as zfp_compress on success).

• Added compressed array support for copy construction and assignment via deep copies.

• Added virtual destructors to enable inheritance from zfp arrays.