Installation

zfp consists of three distinct parts: a compression library written in C, a set of C++ header files that implement compressed arrays and corresponding C wrappers, and a set of C and C++ examples. The main compression codec is written in C and should conform to both the ISO C89 and C99 standards. The C++ array classes are implemented entirely in header files and can be included as is, but since they call the compression library, applications must link with libzfp.

On Linux, macOS, and MinGW, zfp is easiest compiled using gcc and gmake. CMake support is also available, e.g., for Windows builds. See below for instructions on GNU and CMake builds.

zfp has successfully been built and tested using these compilers:

  • gcc versions 4.4.7, 4.7.3, 4.8.5, 4.9.4, 5.5.0, 6.1.0, 6.4.0, 7.1.0, 7.3.0, 8.1.0
  • icc versions 15.0.6, 16.0.4, 17.0.2, 18.0.2, 19.0.0
  • clang versions 3.9.1, 4.0.0, 5.0.0, 6.0.0
  • MinGW version 5.3.0
  • Visual Studio versions 14 (2015), 15 (2017)

zfp conforms to various language standards, including C89, C99, C++98, C++11, and C++14.

NOTE: zfp requires 64-bit compiler and operating system support.

GNU Builds

To compile zfp using gcc without OpenMP, type:

make

from the zfp root directory. This builds libzfp as a static library as well as utilities and example programs. To enable OpenMP parallel compression, type:

make ZFP_WITH_OPENMP=1

CMake Builds

To build zfp using CMake on Linux or macOS, start a Unix shell and type:

mkdir build
cd build
cmake ..
make

To also build the examples, replace the cmake line with:

cmake -DBUILD_EXAMPLES=ON ..

By default, CMake builds will attempt to locate and use OpenMP. To disable OpenMP, type:

cmake -DZFP_WITH_OPENMP=OFF ..

To build zfp using Visual Studio on Windows, start a DOS shell, cd to the top-level zfp directory, and type:

mkdir build
cd build
cmake ..
cmake --build . --config Release

This builds zfp in release mode. Replace ‘Release’ with ‘Debug’ to build zfp in debug mode. See the instructions for Linux on how to change the cmake line to also build the example programs.

Testing

To test that zfp is working properly, type:

make test

or using CMake:

ctest

If the compilation or regression tests fail, it is possible that some of the macros in the file Config have to be adjusted. Also, the tests may fail due to minute differences in the computed floating-point fields being compressed, which will be indicated by checksum errors. If most tests succeed and the failures result in byte sizes and error values reasonably close to the expected values, then it is likely that the compressor is working correctly.

Build Targets

To specify which components to build, set the macros below to 1 (GNU make) or ON (CMake), e.g.,

make BUILD_UTILITIES=1 BUILD_EXAMPLES=0

or using CMake

cmake -DBUILD_UTILITIES=ON -DBUILD_EXAMPLES=OFF ..

Regardless of the settings below, libzfp will always be built.

BUILD_CFP

Build libcfp for C bindings to compressed arrays. Default: off.

BUILD_UTILITIES

Build zfp command-line utility for compressing binary files. Default: on.

BUILD_EXAMPLES

Build code examples. Default: off.

BUILD_TESTING

Build testzfp and (when on the develop branch) unit tests. Default: on.

BUILD_SHARED_LIBS

Build libzfp.so and libcfp.so shared objects. On macOS, the SOFLAGS line in the Config file may have to be uncommented when compiling with GNU make. CMake default: on. GNU make default: off.

Compile-Time Macros

The behavior of zfp can be configured at compile time via a set of macros in the same manner that build targets are specified, e.g.,

cmake -DZFP_WITH_OPENMP=OFF ..
ZFP_INT64
ZFP_INT64_SUFFIX
ZFP_UINT64
ZFP_UINT64_SUFFIX

64-bit signed and unsigned integer types and their literal suffixes. Platforms on which long int is 32 bits wide may require long long int as type and ll as suffix. These macros are relevant only when compiling in C89 mode. When compiling in C99 mode, integer types are taken from stdint.h. Defaults: long int, l, unsigned long int, and ul, respectively.

ZFP_WITH_OPENMP

CMake and GNU make macro for enabling or disabling OpenMP support. CMake builds will by default enable OpenMP when available. Set this macro to 0 or OFF to disable OpenMP support. For GNU builds, OpenMP is disabled by default. Set this macro to 1 or ON to enable OpenMP support. See also OMPFLAGS in Config in case the compiler does not recognize -fopenmp. NOTE: clang currently does not support OpenMP on macOS. CMake default: on. GNU make default: off.

ZFP_WITH_CUDA

CMake macro for enabling or disabling CUDA support for GPU compression and decompression. When enabled, CUDA and a compatible host compiler must be installed. For a full list of compatible compilers, please consult the NVIDIA documentation. If a CUDA installation is in the user’s path, it will be automatically found by CMake. Alternatively, the CUDA binary directory can be specified using the CUDA_BIN_DIR environment variable. CMake default: off. GNU make default: off and ignored.

ZFP_WITH_ALIGNED_ALLOC

Use aligned memory allocation in an attempt to align compressed blocks on hardware cache lines. Default: undefined/off.

ZFP_WITH_CACHE_TWOWAY

Use a two-way skew-associative rather than direct-mapped cache. This incurs some overhead that may be offset by better cache utilization. Default: undefined/off.

ZFP_WITH_CACHE_FAST_HASH

Use a simpler hash function for cache line lookup. This is faster but may lead to more collisions. Default: undefined/off.

ZFP_WITH_CACHE_PROFILE

Enable cache profiling to gather and print statistics on cache hit and miss rates. Default: undefined/off.

BIT_STREAM_WORD_TYPE

Unsigned integer type used for buffering bits. Wider types tend to give higher performance at the expense of lower bit rate granularity. For portability of compressed files between little and big endian platforms, BIT_STREAM_WORD_TYPE should be set to uint8. Default: uint64.

ZFP_BIT_STREAM_WORD_SIZE

CMake macro for indirectly setting BIT_STREAM_WORD_TYPE. Valid values are 8, 16, 32, 64. Default: 64.

BIT_STREAM_STRIDED

Enable support for strided bit streams that allow for non-contiguous memory layouts, e.g., to enable progressive access. Default: undefined/off.

CFP_NAMESPACE

Macro for renaming the outermost cfp namespace, e.g., to avoid name clashes. Default: cfp.