libunwind-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Libunwind-devel] Problems Compiling for Power


From: Thiago Jung Bauermann
Subject: Re: [Libunwind-devel] Problems Compiling for Power
Date: Thu, 06 Sep 2007 12:55:30 -0300

On Thu, 2007-09-06 at 09:39 -0600, David Mosberger-Tang wrote:
> Remember that libunwind is not just for debuggers.  Local unwinding is
> often performance-sensitive (think exception handling, code profiling,
> etc.), so it is desirable to have libunwind optimized as much as
> possible.  One way to achieve that is to have each libunwind support
> one and only one ABI.  You don't see this very much on the
> DWARF-unwinder yet, but if you look at the ia64-unwinder, you see that
> local unwinding results in highly efficient code (e.g., accessing
> memory is just dereferencing a pointer, etc.).

Ok, I'm not yet familiar with libunwind structure to have a opinion in
this regard. It's nice to know the motivations behind this approach.

> Another question of course is whether 32 and 64-bit variants of an ABI
> should be considered as related ABIs or as completely different ABIs.
> That's a matter of judgment.  x86 and x86-64 are different in ways
> other than pointer-size and so it makes sense to me that they should
> be treated as separate ABIs.  OTOH, the HP-UX 32-bit and 64-bit ABI
> variants for ia64 are close enough that it probably would make sense
> to cover them with a single libunwind.  I'm not sure where PPC lies,
> so somebody who's more familiar with PPC will have to make that call.

The ppc32 and ppc64 ABIs are different enough to be separate as well.
For instance, ppc64 uses function descriptors, has a TOC instead of a
GOT, and some other differences.
-- 
[]'s
Thiago Jung Bauermann
Software Engineer
IBM Linux Technology Center

> 
>   --david
> 
> On 9/4/07, Thiago Jung Bauermann <address@hidden> wrote:
> > On Tue, 2007-09-04 at 11:48 -0700, Arun Sharma wrote:
> >
> > > But you want a single 64 bit debugger to link with two copies of
> > > libunwind to
> > > debug two different ABIs. Did I state the problem correctly?
> >
> > It seems more natural to me to have one libunwind which understands the
> > two different ABIs, instead of linking with two versions of the same
> > library. Isn't it possible to use this approach?
> >
> > --
> > []'s
> > Thiago Jung Bauermann
> > Software Engineer
> > IBM Linux Technology Center
> >
> >
> >
> > _______________________________________________
> > Libunwind-devel mailing list
> > address@hidden
> > http://lists.nongnu.org/mailman/listinfo/libunwind-devel
> >
> 
> 






reply via email to

[Prev in Thread] Current Thread [Next in Thread]