ANTLR Benchmarks

The repository contains benchmarks of different grammars over different runtimes:

• Java Standard
• Java Optimized
• C# Standard
• C# Optimized
• Python 3
• JavaScript (nodejs)
• Go
• Php

The integration of other runtimes (C++, Swift) is under development.

How to run

Just run the script ./benchmark.ps1 in Benchmark directory

Requirements

• Powershell
• Installed runtimes that you want to benchmark (.NET Core, Java, Python 3, NodeJS, Go, Php)
• Installed antlr4 runtimes on the runtimes

Method of benchmarking

The following libraries are used for benchmarking:

• Java: JMH
• C#: BenchmarkDotNet
• Php: phpbench
• Python: timeit function
• JavaScript, Go: manual measurement using warmup and many iterations

Results

Here is an example of NotLeft vs Left Recursion comparison. The main parameter is Ratio, the Mean parameter can be different from machine to machine.

Not Left Recursion

Runtime	Mean [ns]	Ratio
JavaStandard_NotLeftRecursion	150702	1,00
JavaOptimized_NotLeftRecursion	147858	0,98
CSharpOptimized_NotLeftRecursion	322393	2,14
Python3Standard_NotLeftRecursion	31555147	209,39
JavaScriptStandard_NotLeftRecursion	594170	3,94
GoStandard_NotLeftRecursion	789169	5,24
PhpStandard_NotLeftRecursion	166762375	1106,57

Left Recursion

Runtime	Mean [ns]	Ratio
JavaStandard_LeftRecursion	173714	1,15
JavaOptimized_LeftRecursion	415581	2,76
CSharpOptimized_LeftRecursion	1302527	8,64
Python3Standard_LeftRecursion	41263081	273,81
JavaScriptStandard_LeftRecursion	618468	4,10
GoStandard_LeftRecursion	1023692	6,79
PhpStandard_LeftRecursion	258153250	1713,00

As we see NotLeftRecursion grammar is faster than LeftRecursion one on ~15% in the worst case.

Java runtime is the fastest and PHP runtime is the slowest: more than 1000 times slower than Java.