This is the mail archive of the 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: GCC 4.1.0 Released

H. J. Lu wrote:

> Here are diffs of "-O2 -mtune=nocona -ffast-math" vs
> "-O2 -mtune=generic -ffast-math" on Nocona:

A 1.5% performance improvement, while certainly significant for some
users, is not worth taking any serious risks on a release branch.  The
purpose of bug-fix releases on the release branch is to correct
regressions.  The goal is that (a) any happy user of 4.1.0 can upgrade
to 4.1.1, and (b) that users presently stuck with older releases because
of regressions in 4.1.0 can upgrade to 4.1.1.  In other words, the goal
is to shift the userbase forward to 4.1.x -- not to improve the
experience of people already using 4.1.0.

> My IA32 numbers show that -mtune=generic gives the better SPEC CPU 2K
> -O2 numbers on Dothan, Yonah, Northwood and Yonah than the previous
> -O2 option.

Yes, I understand.  I hope that 4.2 will in fact have quite a few
improvements on quite a few architectures of at least this magnitude!
However, we're not going to backport them all to 4.1.

> The last thing I want to see is the instability on IA32. I can't
> guarantee it won't happen. But I will do my best to fix it if the
> backported patch is the cause.

The first step is to address regressions on the mainline.  I have not
myself verified the claim, but there has been a suggestion that there is
at least one open regression due to the patch.  If there are any known
regressions from the patch, it's certainly not eligible for a backport.

Mark Mitchell
(650) 331-3385 x713

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