This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Bootstrap failure with thunks change + patch to fix it.
- From: Geoff Keating <geoffk at geoffk dot org>
- To: mark at codesourcery dot com
- Cc: ghazi at caip dot rutgers dot edu, rth at redhat dot com, davem at redhat dot com, gcc-patches at gcc dot gnu dot org
- Date: Tue, 22 Oct 2002 11:56:47 -0700
- Subject: Re: Bootstrap failure with thunks change + patch to fix it.
- References: <109090000.1035259243@warlock.codesourcery.com>
> Date: Mon, 21 Oct 2002 21:00:43 -0700
> From: Mark Mitchell <mark@codesourcery.com>
> > > /* 64-bit constants. If "int" is 32 bits, we'll never hit this
> > abort. */ > else if (TARGET_64BIT && (delta < -2147483647 - 1 ||
> > delta > 2147483647)) > abort ();
> > > /* Large constants that can be done by one addis instruction. */
> > > else if ((delta & 0xffff) == 0)
> > > asm_fprintf (file, "\t{cau|addis} %s,%s,%d\n", this_reg, this_reg,
> > > (int) (delta >> 16));
> >
> > What this looks like to me is that on a 64-bit machine, if delta is
> > outside the range of a 32-bit int then abort.
>
> Ah. Well, that problem was just as there before -- it was just that
> we silently generated bad code when the HOST_WIDE_INT was smushed into
> an int. So, the new behavior is actually better.
>
> We'll have to ask the RS6000 maintainers to figure out what they want
> to do to fix this. 4-gigabyte objects -- especially with virtual
> functions in subobjects beyond the 4-gigabyte boundary -- are probably
> pretty rare, but still...
IMO, that code is OK, except that it should use sorry() rather than
abort() since it's theoretically possible for the user to trip it.
I am not too worried about 4G objects with inheritance. If anyone
does want to implement support, I would ask that they supply a
testcase, since I expect there are problems before this code is reached.
--
- Geoffrey Keating <geoffk@geoffk.org>