Overview

libdeflate is a library for fast, whole-buffer DEFLATE-based compression and decompression.

The supported formats are:

• DEFLATE (raw)
• zlib (a.k.a. DEFLATE with a zlib wrapper)
• gzip (a.k.a. DEFLATE with a gzip wrapper)

libdeflate is heavily optimized. It is significantly faster than the zlib library, both for compression and decompression, and especially on x86 and ARM processors. In addition, libdeflate provides optional high compression modes that provide a better compression ratio than the zlib's "level 9".

libdeflate itself is a library. The following command-line programs which use this library are also included:

• libdeflate-gzip, a program which can be a drop-in replacement for standard gzip under some circumstances. Note that libdeflate-gzip has some limitations; it is provided for convenience and is not meant to be the main use case of libdeflate. It needs a lot of memory to process large files, and it omits support for some infrequently-used options of GNU gzip.

• benchmark, a test program that does round-trip compression and decompression of the provided data, and measures the compression and decompression speed. It can use libdeflate, zlib, or a combination of the two.

• checksum, a test program that checksums the provided data with Adler-32 or CRC-32, and optionally measures the speed. It can use libdeflate or zlib.

For the release notes, see the NEWS file.

Table of Contents

• Building
  • Using CMake
  • Directly integrating the library sources
  • Supported compilers
• API
• Bindings for other programming languages
• DEFLATE vs. zlib vs. gzip
• Compression levels
• Motivation
• License

Building

Using CMake

libdeflate uses CMake. It can be built just like any other CMake project, e.g. with:

cmake -B build && cmake --build build

By default the following targets are built:

• The static library (normally called libdeflate.a)
• The shared library (normally called libdeflate.so)
• The libdeflate-gzip program, including its alias libdeflate-gunzip

Besides the standard CMake build and installation options, there are some libdeflate-specific build options. See CMakeLists.txt for the list of these options. To set an option, add -DOPTION=VALUE to the cmake command.

Prebuilt Windows binaries can be downloaded from https://github.com/ebiggers/libdeflate/releases.

Directly integrating the library sources

Although the official build system is CMake, care has been taken to keep the library source files compilable directly, without a prerequisite configuration step. Therefore, it is also fine to just add the library source files directly to your application, without using CMake.

You should compile both lib/*.c and lib/*/*.c. You don't need to worry about excluding irrelevant architecture-specific code, as this is already handled in the source files themselves using #ifdefs.

If you are doing a freestanding build with -ffreestanding, you must add -DFREESTANDING as well (matching what the CMakeLists.txt does).

Supported compilers

• gcc: v4.9 and later
• clang: v3.9 and later (upstream), Xcode 8 and later (Apple)
• MSVC: Visual Studio 2015 and later
• Other compilers: any other C99-compatible compiler should work, though if your compiler pretends to be gcc, clang, or MSVC, it needs to be sufficiently compatible with the compiler it pretends to be.

The above are the minimums, but using a newer compiler allows more of the architecture-optimized code to be built. libdeflate is most heavily optimized for gcc and clang, but MSVC is supported fairly well now too.

The recommended optimization flag is -O2, and the CMakeLists.txt sets this for release builds. -O3 is fine too, but often -O2 actually gives better results. It's unnecessary to add flags such as -mavx2 or /arch:AVX2, though you can do so if you want to. Most of the relevant optimized functions are built regardless of such flags, and appropriate ones are selected at runtime. For the same reason, flags like -mno-avx2 do not cause all code using the corresponding instruction set extension to be omitted from the binary; this is working as intended due to the use of runtime CPU feature detection.

If using gcc, your gcc should always be paired with a binutils version that is not much older than itself, to avoid problems where the compiler generates instructions the assembler cannot assemble. Usually systems have their gcc and binutils paired properly, but rarely a mismatch can arise in cases such as the user installing a newer gcc version without a proper binutils alongside it. Since libdeflate v1.22, the CMake-based build system will detect incompatible binutils versions and disable some optimized code accordingly. In older versions of libdeflate, or if CMake is not being used, a too-old binutils can cause build errors like "no such instruction" from the assembler.

API

libdeflate has a simple API that is not zlib-compatible. You can create compressors and decompressors and use them to compress or decompress buffers. See libdeflate.h for details.

There is currently no support for streaming. This has been considered, but it always significantly increases complexity and slows down fast paths. Unfortunately, at this point it remains a future TODO. So: if your application compresses data in "chunks", say, less than 1 MB in size, then libdeflate is a great choice for you; that's what it's designed to do. This is perfect for certain use cases such as transparent filesystem compression. But if your application compresses large files as a single compressed stream, similarly to the gzip program, then libdeflate isn't for you.

Note that with chunk-based compression, you generally should have the uncompressed size of each chunk stored outside of the compressed data itself. This enables you to allocate an output buffer of the correct size without guessing. However, libdeflate's decompression routines do optionally provide the actual number of output bytes in case you need it.

Windows developers: note that the calling convention of libdeflate.dll is "cdecl". (libdeflate v1.4 through v1.12 used "stdcall" instead.)

Bindings for other programming languages

The libdeflate project itself only provides a C library. If you need to use libdeflate from a programming language other than C or C++, consider using the following bindings:

• C#: LibDeflate.NET
• Go: go-libdeflate
• Java: libdeflate-java
• Julia: LibDeflate.jl
• Nim: libdeflate-nim
• Perl: Gzip::Libdeflate
• PHP: ext-libdeflate
• Python: deflate
• Ruby: libdeflate-ruby
• Rust: libdeflater

Note: these are third-party projects which haven't necessarily been vetted by the authors of libdeflate. Please direct all questions, bugs, and improvements for these bindings to their authors.

Also, unfortunately many of these bindings bundle or pin an old version of libdeflate. To avoid known issues in old versions and to improve performance, before using any of these bindings please ensure that the bundled or pinned version of libdeflate has been upgraded to the latest release.

DEFLATE vs. zlib vs. gzip

The DEFLATE format (rfc1951), the zlib format (rfc1950), and the gzip format (rfc1952) are commonly confused with each other as well as with the zlib software library, which actually supports all three formats. libdeflate (this library) also supports all three formats.

Briefly, DEFLATE is a raw compressed stream, whereas zlib and gzip are different wrappers for this stream. Both zlib and gzip include checksums, but gzip can include extra information such as the original filename. Generally, you should choose a format as follows:

• If you are compressing whole files with no subdivisions, similar to the gzip program, you probably should use the gzip format.
• Otherwise, if you don't need the features of the gzip header and footer but do still want a checksum for corruption detection, you probably should use the zlib format.
• Otherwise, you probably should use raw DEFLATE. This is ideal if you don't need checksums, e.g. because they're simply not needed for your use case or because you already compute your own checksums that are stored separately from the compressed stream.

Note that gzip and zlib streams can be distinguished from each other based on their starting bytes, but this is not necessarily true of raw DEFLATE streams.

Compression levels

An often-underappreciated fact of compression formats such as DEFLATE is that there are an enormous number of different ways that a given input could be compressed. Different algorithms and different amounts of computation time will result in different compression ratios, while remaining equally compatible with the decompressor.

For this reason, the commonly used zlib library provides nine compression levels. Level 1 is the fastest but provides the worst compression; level 9 provides the best compression but is the slowest. It defaults to level 6. libdeflate uses this same design but is designed to improve on both zlib's performance and compression ratio at every compression level. In addition, libdeflate's levels go up to 12 to make room for a minimum-cost-path based algorithm (sometimes called "optimal parsing") that can significantly improve on zlib's compression ratio.

If you are using DEFLATE (or zlib, or gzip) in your application, you should test different levels to see which works best for your application.

Motivation

Despite DEFLATE's widespread use mainly through the zlib library, in the compression community this format from the early 1990s is often considered obsolete. And in a few significant ways, it is.

So why implement DEFLATE at all, instead of focusing entirely on bzip2/LZMA/xz/LZ4/LZX/ZSTD/Brotli/LZHAM/LZFSE/[insert cool new format here]?

To do something better, you need to understand what came before. And it turns out that most ideas from DEFLATE are still relevant. Many of the newer formats share a similar structure as DEFLATE, with different tweaks. The effects of trivial but very useful tweaks, such as increasing the sliding window size, are often confused with the effects of nontrivial but less useful tweaks. And actually, many of these formats are similar enough that common algorithms and optimizations (e.g. those dealing with LZ77 matchfinding) can be reused.

In addition, comparing compressors fairly is difficult because the performance of a compressor depends heavily on optimizations which are not intrinsic to the compression format itself. In this respect, the zlib library sometimes compares poorly to certain newer code because zlib is not well optimized for modern processors. libdeflate addresses this by providing an optimized DEFLATE implementation which can be used for benchmarking purposes. And, of course, real applications can use it as well.

License

libdeflate is MIT-licensed.

I am not aware of any patents or patent applications relevant to libdeflate.