Re: comparison of 2.95.2 and 3.2 optimizers on SPARC

[Brad Spencer <spencer at infointeractive dot com>]

On Thu, Jan 02, 2003 at 01:38:52PM -0600, Loren James Rittle wrote:
> Martin,
> 
> Oops and, likewise, sorry I didn't see that issue.  I can now confirm
> your report.  Thank you (again) for taking the time to report it and
> the follow up.
> 
> On FreeBSD/i386 with 2.95.X:
> 
> S rittle@latour; time a.out 1000000 stdio >/dev/null
>      2r     1.6u     0.0s       a.out 1000000 stdio
> S rittle@latour; time a.out 1000000 iostream >/dev/null
>      1r     0.9u     0.0s       a.out 1000000 iostream
> 
> On FreeBSD/i386 with mainline:
> 
> S rittle@latour; time a.out 1000000 stdio >/dev/null
>      1r     1.6u     0.0s       a.out 1000000 stdio
> S rittle@latour; time a.out 1000000 iostream >/dev/null
>     14r    14.4u     0.0s       a.out 1000000 iostream
> 

I posted a patch in October[1] that tried to avoid some unnecessary
work that is done every time a numeric is converted to a string.  It
was against the 3.2 release, so it's a bit out of date now, but the
idea is still valid.   I've attached a somewhat more modern version
for the gcc-3.3 series.  Perhaps this addresses part of the slowdown?

References:

[1] http://gcc.gnu.org/ml/libstdc++/2002-10/msg00056.html 

-- 
------------------------------------------------------------------
Brad Spencer - spencer@infointeractive.com - "It's quite nice..."
Systems Architect | InfoInterActive Corp. | A Canadian AOL Company

Attachment: gcc-3.3-libstdc++-v3-locale_facets-optimized.patch
Description: Text document