Subject: Re: CVS commit: src
To: Ignatios Souvatzis <email@example.com>
From: Scott Reynolds <firstname.lastname@example.org>
Date: 02/09/1999 21:16:06
On Tue, 09 Feb 1999 Ignatios Souvatzis wrote:
> On Tue, Feb 09, 1999 at 10:32:42AM -0800, Jason Thorpe wrote:
>> On Tue, 09 Feb 1999 13:30:05 -0500
>> Bill Sommerfeld <email@example.com> wrote:
>> > Would it make sense to make this selectable on a per-port basis?
>> It's already selectable on a per-person-system basis with the LDSTATIC
>> build variable.
> Uhm, no... thats _global_ right? For people doing lots of builds, or even
> kernel builds or make builds, a LOT of cc1 (and sh) are executed... but
> sh is already static. I think Scott Reynolds (but I might confuse him with
> another person) claims a (small) few percent speed difference.
No, it was me. I consistently measured ~2% speed difference in `make build'
time when I tried this about the time we moved from a static to a shared
libgcc. I hesitate to say it was done under strictly controlled conditions,
but it was certainly reproducible. I would be somewhat surprised if it were
actually as high as 15%, though. I'm looking forward to seeing Todd's
numbers, especially with the current toolchain.
My objection during the last discussion was that since it was consistent and
measurable, we need to think long and hard before committing to this path.
It's easy to write off several 1% and 2% penalties before we realize that
we're `suddenly' 10% slower than we were a year ago.