Re: ARM: allow factorization of constants into addressing insns when optimizing for space

[Richard Earnshaw <rearnsha at arm dot com>]

[counter-example snipped]

> For a real example, also consider a saving of 13000 bytes of text in a Linux
> kernel binary.
> 

What percentage reduction is this?

> > See the CSiBE measurements results at http://sed.inf.u-szeged.hu/CSiBE
> > There is a code size increase of about 0,4% from the 20th to the 21st,
> > which was caused by allow factorization of constants.
> > 
> > I suggest using CSiBE for measuring code size effects. There is a
> > downloadable version of CSiBE, you can check it out at
> > http://sed.inf.u-szeged.hu/CSiBE/download.php
> 
> I don't think 0.4% increase in a single program should be considered 
> significant for backing out that patch.  You spotted a patological 
> regression case, but I don't think it's statistically significant given a 
> wider range of test cases.

CSiBE isn't a single application, it's a benchmark that comprises files 
from sereral applications and totals about 1Mb of compiled ARM code.  So 
it's not a trivial counter example.

0.4% is small, but not insignificant.  To judge its importance we need to 
weigh it against the benefit in the example you were working on (the Linux 
kernel in this case).

Gábor, I think the benchmark might well benefit from some larger examples. 
 It's a good start, but it really only has about a dozen small 
applications that it draws code from.

R.