PowerPC LEGITIMIZE_ADDRESS missing "goto WIN"

[Franz Sirl]

Am Thu, 19 Nov 1998 schrieb David Edelsohn:
>>>>>> Jeffrey A Law writes:
>
>David> Tue Nov 10 17:04:11 1998  David Edelsohn  <edelsohn@mhpcc.edu>
>David> * rs6000.h (LEGITIMIZE_ADDRESS): Add missing "goto WIN".
>
>Jeff> Is this something I should install for egcs-1.1.1?
>
>	If we still have time, this could not hurt.  I did not want to
>continue to add more patches at the last minute and cause you to rebuild
>the tarball yet again.
>
>	I cannot imagine that the previous version was correct.  I think
>that there was a bug report that might relate to this, but I cannot find
>it and have not heard from the LinuxPPC community about tracking down the
>bug report.  Without a real bug report to test against this fix, I did not
>want to unilaterally include this change in the egcs-1.1.1 distribution,
>using the "bugfix only" policy. 
>
>	I'm torn about this.  I think it should go in and I think it fixes
>a bug, but I cannot verify that.

Hmm, I feel like you :-). But I think I'm in favour of putting it in,
especially as I can see no testsuite regression on the mainline with it. BTW,
Jeff, forget about the profile1.C regression, this was a glibc-2.1pre problem
that is fixed now.

Since the 1.1.1 release date seems to be quite near, I would like to have only
one thing definetely in, and that's the "Build ppc libs with -fPIC patch" from
rth.

If some more time will pass til 1.1.1 these could be considered:
1. fix of the qt-1.41 miscompilation (can be worked around for now with
-fno-schedule-insns2, still checking if this also the bug that lets mozilla
segfault)
2. backport of ASM_OUTPUT_MI_THUNK from the mainline
3. change of the profiling code to use r12 instead of r11 like already done in
the mainline.

But don't spend too much time on 1.1.1, with the mainline now testing better
than 1.1.1pre2 I'm eagerly awaiting 1.2 already :-).

Franz.