Re: 5% code size and compilation time regression at -Os in last 2 days

[Daniel Jacobowitz <drow at false dot org>]

On Sun, Dec 05, 2004 at 07:59:43PM -0500, Andrew Pinski wrote:
> 
> On Dec 2, 2004, at 5:28 PM, Richard Earnshaw wrote:
> 
> >
> >Come on guys, what are you doing?
> >
> >Code size at -Os has regressed by a whopping 5% over the last 2 days:
> >
> >http://www.inf.u-szeged.hu/csibe/l-sbs.php?branchid=mainline&flags=- 
> >Os&data
> >view=Timeline&finish_button=Finish
> 
> 
> But looking at todays results it looks like back to normal,  I think  
> this is not
> really a gcc bug but a linker one.  The only thing which went in  
> yesterday which
> could have even caused this was a patch which broke adding hidden  
> symbols into
> libgcc but I fixed that today.  So if tomorrow's results will show that  
> we
> regressed again but this would mean the linker is not doing its job.
> 
> So in conclusion, this is not a gcc regression at all.

CSiBE measures object file size, not linked size.

-- 
Daniel Jacobowitz