From: Ryan Davis <ryand-ruby@...>
Date: 2011-04-19T04:15:19+09:00
Subject: Re: Need for speed -> a C extension?


On Apr 18, 2011, at 11:30 , Martin Hansen wrote:

> WJ wrote in post #993576:
>> Martin Hansen wrote:
>> 
>>> The below code is too slow for practical use. I need it to run at least
>>> 20 times faster. Perhaps that is possible with some C code? I have no
>>> experience with writing Ruby extensions. What are the pitfalls? Which
>>> part of the code should be ported? Any pointers to get me started?
>> 
>> Please give a clear description of the algorithm, and then
>> give some sample input and output.
> 
> 
> Here is a working version of the code that can be profiled (though it 
> will take forever with 20M iterations):
> 
> http://pastie.org/1808127
> 
> The slow part according to profiler is:

Profiling is great, but you also need to write a benchmark in order to objectively measure the work you're doing. You might also want to write microbenchmarks against the bottlenecks you identify in your profiling work. Eg:

>  def match?(char1, char2)
>    (EQUAL[char1.upcase.ord] & EQUAL[char2.upcase.ord]) != 0
>  end

That would be really easy to write tests and a benchmark against and then rewrite in C using RubyInline... without tests and a benchmark tho, you won't know that you've done it correctly and provides a measurable benefit.