[svn:parrot] r48634 - in trunk: compilers/imcc src/pmc

[Andy Dougherty]

On Wed, 25 Aug 2010, chromatic wrote:

> On Tuesday 24 August 2010 at 10:14, plobsing  wrote:
> 
> > Author: plobsing
> > Date: Tue Aug 24 17:14:54 2010
> > New Revision: 48634
> > URL: https://trac.parrot.org/parrot/changeset/48634
> > 
> > Log:
> > inline undocumented, used-once static functions
> 
> Shouldn't a decent optimizing compiler inline used-once static functions?  (I 
> know there's often a size limit for this, but before we manually inline such 
> functions I want to make sure we do so for good reasons.)

Yes, absolutely.  Moreso, if directed by the proper optimization flags, 
the compiler is usually able to make judgments about whether or not 
inlining is wise.  In looking at inline functions for perl5, I did some 
tests on compilers ranging from an ancient 1996-era Sun Workshop 4.2 
compiler up through Sun's 2010 compiler, as well as lots of versions of 
gcc.  Even without the inline keyword, all the compilers were already 
inling some functions anyway.  My conclusion:  The compiler is usually 
smarter than me.  Strive for clarity of code instead.  If the once-used 
static function is a nice logical unit, then keep it as a nice logical 
unit.  Forcing it inline might actually make things worse on some systems.

-- 
    Andy Dougherty		doughera at lafayette.edu