This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Omit frame pointer and fix %ebp by default on x86 (take 3)


On Mon, Aug 16, 2004 at 02:25:20PM -0600, Roger Sayle wrote:
> 
> On Mon, 16 Aug 2004, Jakub Jelinek wrote:
> > Why do you think it is an dramatic increase?
> > On x86-64 where -fasynchronous-unwind-tables is the default, .eh_frame
> > size is usually smaller than 5% of code size,
> 
> You forget so quickly :>  I posted the benchmarks for you here:
> http://gcc.gnu.org/ml/gcc-patches/2004-07/msg01202.html
> 
> On GCC's CSiBE benchmark, -fasynchronous-unwind-tables increases the
> object file sizes measured in that benchmark by over 30% on IA-32.

Well, the detailed numbers you posted suggest just 8.6% increase.
I have built now a bunch of programs on i386 with -Os -march=i386
-mtune=pentium4 -fomit-frame-pointer {,-funwind-tables,-fasynchronous-unwind-tables}
and the growth as reported by size(1) (which is questionable metrics,
see earlier mails) were never bigger than 9%.
Note that -funwind-tables and -fasynchronous-unwind-tables are
very similar in size with -fomit-frame-pointer.

Can you post what exact CSiBE tests were over 30% of codesize increase
that drag the total results that high?  I simply can't believe in 30%,
so would like to know what exactly grew up and why.

	Jakub


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]