Path: csiph.com!x330-a1.tempe.blueboxinc.net!usenet.pasdenom.info!aioe.org!eternal-september.org!feeder.eternal-september.org!.POSTED!not-for-mail
From: markspace <-@.>
Newsgroups: comp.lang.java.programmer
Subject: Re: Using Enumerated Types as Array Indexes
Date: Thu, 18 Aug 2011 11:50:11 -0700
Organization: A noiseless patient Spider
Lines: 32
Message-ID: <j2jmtd$v0o$1@dont-email.me>
References: <b9ad6662-3240-4754-86a0-b715e8cdbc48@z7g2000vbp.googlegroups.com> <4e4b0d81$0$314$14726298@news.sunsite.dk> <j2f3ci$1lh$1@dont-email.me> <4e4b1fe4$0$314$14726298@news.sunsite.dk> <j2fc1g$b9j$1@dont-email.me> <6d418cda-dab3-43df-a9ec-293b43f2bbd8@glegroupsg2000goo.googlegroups.com> <vTL2q.38278$tp.27319@newsfe06.iad> <slrnj4ngfi.6gl.avl@gamma.logic.tuwien.ac.at> <f08d98e8-e04d-4d4e-b518-44bba735dba6@glegroupsg2000goo.googlegroups.com> <9b2aglFuj7U1@mid.individual.net> <j2hhnj$4js$1@dont-email.me> <slrnj4okqm.6gl.avl@gamma.logic.tuwien.ac.at> <__-dnaxcSIRV-NHTnZ2dnUVZ_tOdnZ2d@earthlink.com> <j2jaok$vll$1@dont-email.me> <slrnj4qmv5.6gl.avl@gamma.logic.tuwien.ac.at>
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Injection-Date: Thu, 18 Aug 2011 18:50:22 +0000 (UTC)
Injection-Info: mx04.eternal-september.org; posting-host="XjIWM99mD7Ijfdu600oVPA"; logging-data="31768"; mail-complaints-to="abuse@eternal-september.org";	posting-account="U2FsdGVkX19851NrnDRmsOFy3vsc75rTd1wyX1e0TW0="
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:6.0) Gecko/20110812 Thunderbird/6.0
In-Reply-To: <slrnj4qmv5.6gl.avl@gamma.logic.tuwien.ac.at>
Cancel-Lock: sha1:sRhlx026YCk8+ayrlXYPHMnqC04=
Xref: x330-a1.tempe.blueboxinc.net comp.lang.java.programmer:7225

On 8/18/2011 11:37 AM, Andreas Leitgeb wrote:

> Apart from that, it is a dirty hack to avoid conditionals.


One thing to keep in mind here I think is the relative execution time 
for each operation:

1. % and / are division.  Some CPUs execute this as quickly addition and 
subtraction, some require a large number of CPU cycles.

2. Conditionals tend to be a subtraction, followed by a branch. For a 
simple branch like this:

  (n < 0) ? -n : n

Might be implemented like this

   CMPZ n      ;; these are not byte codes, just psudeo-x86
   JLT  +2
   NEG n
   ...         ;; result of JLT

In other words, there's little opportunity for pipeline stall here.  The 
instruction after NEG n is executed regardless, and should be in the 
pipe no matter what.  I'm pretty sure that most CPUs will deal with 
executing one instruction like NEG conditional easily (most will compute 
the result and then just not use it if the conditional JLT is taken).

So preferring modulus to ?: or Math.abs() might be a misoptimization.