From: Paul Lutus <nospam@...>
Date: 2006-10-26T15:25:04+09:00
Subject: Re: simple math question

Phrogz wrote:

> Rick DeNatale wrote:
>> (int & 1).zero?  is probably faster than
>> int[0].zero?
>>
>> Since the latter generates the mask using the bit position.
>>
>> and both should be faster than
>>
>> (int % 2).zero?
>>
>> But it it's performance critical, benchmarking is recommended to
>> verify. One should never simply take 'rules of thumb' for granted.
> 
> Let's just do it rather than speculating:
> 
> require 'benchmark'
> 
> ITERATIONS = 1_000_000
> MAX_INT    = 2 ** 30
> NUMS = (1..ITERATIONS).map{ rand(MAX_INT) }
> Benchmark.bmbm{ |x|
>   x.report '(n % 2).zero?' do
>     NUMS.each{ |n|
>       (n % 2).zero?
>     }
>   end
> 
>   x.report 'n[0].zero?' do
>     NUMS.each{ |n|
>       n[0].zero?
>     }
>   end
> 
>   x.report '(n & 1).zero?' do
>     NUMS.each{ |n|
>       (n & 1).zero?
>     }
>   end
> }
> 
> #=> Rehearsal -------------------------------------------------
> #=> (n % 2).zero?   2.060000   0.020000   2.080000 (  2.244595)
> #=> n[0].zero?      1.960000   0.010000   1.970000 (  2.145238)
> #=> (n & 1).zero?   1.990000   0.020000   2.010000 (  2.275232)
> #=> ---------------------------------------- total: 6.060000sec
> #=>
> #=>                     user     system      total        real
> #=> (n % 2).zero?   2.070000   0.010000   2.080000 (  2.315105)
> #=> n[0].zero?      1.960000   0.020000   1.980000 (  2.143038)
> #=> (n & 1).zero?   1.990000   0.010000   2.000000 (  2.173120)
> 
> There is a very, very, very small difference in speed.

I feel like Alice falling into the rabbit hole. I never thought I would see
the day when the difference between a division and a bit test would be
reduced to insignificance by other factors. Obviously I started thinking
about this (and developing my prejudices) before there were numeric
coprocessors.

Nice demonstration, BTW.

-- 
Paul Lutus
http://www.arachnoid.com