float_serialization_benchmark

This repository contains benchmarking code for floating-point serialization. The goal is to compare different approaches to serializing floating-point numbers, i.e., converting them from an IEEE 754 binary representation to a string decimal representation.

Currently, the following approaches are compared:

• dragon4
• std::to_string
• fmt::format
• netlib
• sprintf
• grisu2
• grisu-exact
• Errol
• std::to_chars
• schubfach
• Dragonbox
• Ryu
• double-conversion
• Abseil
• Teju Jagua
• yy_double
• SwiftDtoa

If you have a recent version of CMake under linux or macOS, you can simply go in the directory and type the following commands:

cmake -B build .
cmake --build build
./build/benchmarks/benchmark 

Some functionality might be disabled under some systems. For example, under macOS with libc++, some tests might be omitted because it does not yet fully support C++17.

We also support Visual Studio, please refer to the cmake documentation.

You may use docker to run these benchmarks easily on a variety of platforms: see https://github.com/lemire/docker_programming_station

Windows

Usage under Windows is similar. After installing cmake and Visual Studio 2019, one might type in the appropriate shell:

cmake -B build .
cmake --build build --config Release
.\build\benchmarks\Release\benchmark.exe

Advanced Usage

Serialize the strings (one per line) included in a text file:

./build/benchmarks/benchmark -f data/canada.txt

Serialize strings generated from floats in (0,1):

./build/benchmarks/benchmark

Exhaustive 32-bit check

We also include an exhaustive check of all 32-bit floats, to verify that we can produce the shortest string representation (measured by the number of significant digits).

Under Linux, you may run the following check:

cmake -B build .
cmake --build build --config Release
./build/benchmarks/exhaustivefloat32

Thorough 64-bit check

We also include a thorough check of many 64-bit floats, to verify that we can produce the shortest string representation (measured by the number of significant digits).

Under Linux, you may run the following check:

cmake -B build .
cmake --build build --config Release
./build/benchmarks/thoroughfloat64

Fixed-precision evaluation

By default, we compare algorithms that output shortest-significand representation which can round-trip to the original floating-point value. We can also compare algorithms that output fixed-precision representation of a given precision:

./build/benchmarks/benchmark -f data/canada.txt -F [precision]

Note that this only works when we are comparing speeds, not measuring properties of the algorithms, i.e., we can't use both -F/--fixed and -t/--test at the same time.

Other existing benchmarks

• dtoa Benchmark
• parse-bench
• Drackennest