comparing the c ffi overhead on various programming languages
Build requirements:
- gcc
- tup
Requirements on the exec path:
- nimrod
- javac, java
- go
- rustc
- luajit
To initialize, exec:
- tup init
Workaround for go (I've to figure out how to pass rpath to go)
- tup upd
- cd newplus
- sudo ln -s $PWD/libnewplus.so /usr/lib/libnewplus.so
To compile, exec:
- ./compile-all.sh
Compile opts:
- -O2 (gcc - applies to c/jni)
- --opt-level=2 (rust)
To run, exec:
- ./run-all.sh
The main files:
- newplus/plus.c
- hello.c
- hello.nim
- jhello/Hello.java
- hello.go
- hello.rs
- hello.lua
My environment:
- Intel i7-3630QM laptop (4cores, HT) with 16g ram
- Ubuntu 13.04 x64
- gcc 4.7.3
- tup 0.6.5-2
- nimrod 0.9.3
- java 1.7.0_25
- go 1.1.2
- rust 0.8
- luajit 2.0.2
Measurement:
- call the c function "plusone" 2 billion times and print out the elapsed time in millis.
int x = 0;
while (x < 2000000000) x = plusone(x);- 2 samples/runs
##Results
The results are elapsed time in milliseconds
============================================
c:
4839
4822
luajit:
3612
3616
nimrod:
4218
4222
rust:
5436
5427
java:
18111
18131
go:
95536
95556
For enhancements, pull requests are welcome.