Skip to content

miloyip/itoa-benchmark

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

93 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

itoa Benchmark

Copyright(c) 2014-2016 Milo Yip (miloyip@gmail.com)

Introduction

This benchmark evaluates the performance of conversion from 32-bit/64-bit integer to ASCII string in decimal. The function prototypes are:

void u32toa(uint32_t value, char* buffer);
void i32toa(int32_t value, char* buffer);
void u64toa(uint64_t value, char* buffer);
void i64toa(int64_t value, char* buffer);

Note that itoa() is not a standard function in C and C++, but provided by some compilers.

Procedure

Firstly the program verifies the correctness of implementations.

Then, two cases for benchmark are carried out:

  1. Sequential: Converts consecutive values in same number of decimal digits.

    For u32toa(), the tested values are { 1, 2, ..., 9 }, {10, 11, ..., 99 }, ... { 4000000000, 4000000001, ..., 4294967296}, i.e., groups of 1 to 10 decimal digits.

    For signed versions, use alternate signs, e.g. { 1, -2, 3, -4, ... 9 }.

    For 64-bit integer, there are groups of 1 to 20 decimal digits.

  2. Random: Converts the shuffled sequence of the first case.

Each digit group is run for 100000 times. The minimum time duration is measured for 10 trials.

Build and Run

  1. Obtain premake5.
  2. make

Results

The following are sequential results measured on a PC (Core i7 920 @2.67Ghz), where u32toa() is compiled by Visual C++ 2013 and run on Windows 64-bit. The speedup is based on sprintf().

Function Time (ns) Speedup
sprintf 194.225 1.00x
vc 61.522 3.16x
naive 26.743 7.26x
count 20.552 9.45x
lut 17.810 10.91x
countlut 9.926 19.57x
branchlut 8.430 23.04x
sse2 7.614 25.51x
null 2.230 87.09x

corei7920@2.67_win64_vc2013_u32toa_sequential_time

corei7920@2.67_win64_vc2013_u32toa_sequential_timedigit

Note that the null implementation does nothing. It measures the overheads of looping and function call.

Since the C++ standard library implementations (ostringstream, ostrstream, to_string) are slow, they are turned off by default. User can re-enable them by defining RUN_CPPITOA macro.

Some results of various configurations are located at itoa-benchmark/result. They can be accessed online, with interactivity provided by Google Charts:

Implementations

Function  Description
ostringstream std::ostringstream in C++ standard library.
ostrstream std::ostrstream in C++ standard library.
to_string std::to_string() in C++11 standard library.
sprintf sprintf() in C standard library
vc Visual C++'s _itoa(), _i64toa(), _ui64toa()
naive Compute division/modulo of 10 for each digit, store digits in temp array and copy to buffer in reverse order.
unnamed Compute division/modulo of 10 for each digit, store directly in buffer
count Count number of decimal digits first, using technique from [1].
lut Uses lookup table (LUT) of digit pairs for division/modulo of 100. Mentioned in [2]
countlut Combines count and lut.
branchlut Use branching to divide-and-conquer the range of value, make computation more parallel.
sse2 Based on branchlut scheme, use SSE2 SIMD instructions to convert 8 digits in parallel. The algorithm is designed by Wojciech Muła [3]. (Experiment shows it is useful for values equal to or more than 9 digits)
null Do nothing.

FAQ

  1. How to add an implementation?

    You may clone an existing implementation file (e.g. naive.cpp). And then modify it. Re-run premake to add it to project or makefile. Note that it will automatically register to the benchmark by macro REGISTER_TEST(name).

    Making pull request of new implementations is welcome.

  2. Why not converting integers to std::string?

    It may introduce heap allocation, which is a big overhead. User can easily wrap these low-level functions to return std::string, if needed.

  3. Why fast itoa() functions is needed?

    They are a very common operations in writing data in text format. The standard way of sprintf(), std::stringstream, std::to_string(int) (C++11) often provides poor performance. The author of this benchmark would optimize the "naive" implementation in RapidJSON, thus he creates this project.

References

[1] Anderson, Bit Twiddling Hacks, 1997.

[2] Alexandrescu, Three Optimization Tips for C++, 2012.

[3] Muła, SSE: conversion integers to decimal representation, 2011.

Related Benchmarks and Discussions