This is the mail archive of the
binutils@sources.redhat.com
mailing list for the binutils project.
Re: PATCH: speed up ar(1)
- From: Steven Bosscher <stevenb at suse dot de>
- To: binutils at sources dot redhat dot com
- Cc: Nick Clifton <nickc at redhat dot com>, Ben Elliston <bje at au1 dot ibm dot com>
- Date: Mon, 14 Mar 2005 02:05:47 +0100
- Subject: Re: PATCH: speed up ar(1)
- Organization: SUSE Labs
- References: <422CE744.7070409@au.ibm.com> <422EF1B4.4040104@redhat.com>
On Wednesday 09 March 2005 13:53, Nick Clifton wrote:
> Hi Ben,
>
> > The following patch alters the ar cache representation from a singly
> > linked list to a hash table using the hashtab implementation from
> > libiberty. When archiving files with thousands of archive members (eg.
> > libjava), this patch results in a 15-25% performance improvement.
> >
> > Tested with an --enable-targets=all build, make check-binutils and a
> > byte-wise comparison of the libjava libraries built with and without a
> > patched ar.
> >
> > Okay to commit?
>
> [Tut tut - no ChangeLog entry...]
>
> Approved for the mainline, but not the 2.16 branch.
Ouch, that's a shame.
I was all cheeded up for the day when I saw this patch and its
effect on ar times for libjava (Ben was my hero - he does that
sometimes). As a GCC hacker, I have to build libjava quite
often, and ar time is actually a serious problem for this huge
PITA library.
So the sooner I have a "production" binutils with this patch,
the better ;-) Is there really no chance that the patch can
go into 2.16?
Gr.
Steven