ryu_fortran

A Fortran implementation of Ryu algorithm which converts floating point numbers to decimal strings. It is more effective than internal file approach.

This implementation is based on the Scala version of Ryu.

Install

Fortran package manager

Add fpm dependency declaration in the fpm.toml file of your project.

[dependencies]
ryu_fortran = { git = "https://github.com/St-Maxwell/ryu_fortran" }

Usage

use ryu, only: f2shortest, d2shortest, d2fixed, d2exp
use iso_fortran_env, only: real32, real64

write (*, "(A)") f2shortest(3.14159_real32)
write (*, "(A)") d2shortest(2.718281828_real64)
write (*, "(A)") d2fixed(1.2345678987654321_real64, 10)
write (*, "(A)") d2exp(299792458._real64, 5)

! 3.14159
! 2.718281828
! 1.2345678988
! 2.99792E+08

Test

Ryu algorithm is meant to generate the shortest decimal representaion of a floating point number, and is able to preserve the information after conversion. That is, if we convert the produced string back to a floating point number, we should obtain the same binary representation comparing to the original number.

f2shortest, d2shortest, d2fixed and d2exp have been fully tested using the test cases from ulfjack/ryu.

To perform the tests, run

fpm test [test-item]

where test-item can be test-f2shortest, test-d2shortest, test-d2fixed and test-d2exp.

Benchmark

• Compiler: gfortran version 9.3.0 (Ubuntu-20.04 WSL2)
• Command: fpm test benchmark --profile release

Results

Benchmark for f2shortest
f2shortest Time (us): 0.0785894   Std Dev:  0.0118
internal IO Time (us): 0.7998928   Std Dev:  0.0778

Benchmark for d2shortest
d2shortest Time (us): 0.1047708   Std Dev:  0.0110
internal IO Time (us): 0.9942683   Std Dev:  0.1660

Benchmark for d2exp
d2exp Time (us): 0.1041636   Std Dev:  0.0167
internal IO Time (us): 0.9326320   Std Dev:  0.1252

Benchmark for d2fixed
d2fixed Time (us): 0.2314018   Std Dev:  0.2261
internal IO Time (us): 2.3488336   Std Dev:  2.3387