[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: kdb and optimizations
From: |
Samuel Thibault |
Subject: |
Re: kdb and optimizations |
Date: |
Mon, 8 Apr 2013 09:20:11 +0200 |
User-agent: |
Mutt/1.5.21+34 (58baf7c9f32f) (2010-12-30) |
Thomas Schwinge, le Mon 08 Apr 2013 08:26:06 +0200, a écrit :
> On Mon, 8 Apr 2013 00:13:16 +0200, Samuel Thibault <samuel.thibault@gnu.org>
> wrote:
> > Thomas Schwinge, le Mon 08 Apr 2013 00:06:23 +0200, a écrit :
> > > On Sun, 7 Apr 2013 21:10:09 +0200, Samuel Thibault
> > > <samuel.thibault@gnu.org> wrote:
> > > > In the Debian package, when building the kdb-enabled kernel,
> > > > we pass -fno-omit-frame-pointer in CFLAGS, and I'll push
> > > > -fno-optimize-sibling-calls too. Both of them permit to get useful call
> > > > traces in kdb. They should rather be included in the upstream
> > > > configfrag.ac, don't they?
> > >
> > > With KDB not handling DWARF information, what you propose is probably the
> > > "right" thing to do indeed, depite the performance loss (which I don't
> > > know how big it is).
> >
> > It's probably negligible: it's only about replacing call with jmp.
>
> Ah, I've primarily been thinking about -fomit-frame-pointer; here it
> depends on whether GCC is able to do something useful with the additional
> register.
Ah, right. Indeed in some cases it could make a difference in some rare
loop cases if the compiler has to spill, but with nowadays' L1 speed,
that probably doesn't matter so much.
Samuel