This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: Change to search_n
- From: Paolo Carlini <pcarlini at suse dot de>
- To: Dimitris Xochellis <jimxoch at yahoo dot gr>
- Cc: Chris Jefferson <caj at cs dot york dot ac dot uk>, libstdc++ at gcc dot gnu dot org
- Date: Sat, 29 Jan 2005 09:39:21 +0100
- Subject: Re: Change to search_n
- References: <20050129081706.77568.qmail@web52801.mail.yahoo.com>
Dimitris Xochellis wrote:
What are the current status and your plans regarding
the search_n implementation? Can I help in any way?
As I see it, we are making progress ;)
I mean, Chris already contributed some testcases
that we were missing and, AFAIK, is refining the
actual implementation of the algorithm: maybe he
wants to add something.
My only doubt is about performance on "exotic"
targets: as you know, libstdc++-v3 supports a
rather wide range of targets (small and big) and
we want to be sure that the performance improve
on all of them. In particular, I'm concerned that
the backwards moves may penalize old processors
missing backward prefetchers. Things like this.
Do you have any performance numbers for "exotic"
targets? (say, anything != x86, x86_64, ia64,
powerpc, would do ;)
Thanks,
Paolo.