Re: egcs-970910, powerpc-ibm-aix4.1.4.0, f771 Relocation overflow

[David Edelsohn <dje at watson dot ibm dot com>]

>>>>> dmitri  writes:

dmitri> At 10:46 PM 9/16/97 +0100, Dave Love wrote:
>> It's apparently an AIX linker bug.  I thought 970910 still had the

dmitri> this is definitely not a linker bug (the same thing will happen
dmitri> if you use gnu ld, rather than native AIX ld). TOC simply cannot
dmitri> be more than 64K. TOC entries accessed by using a TOC pointer
dmitri> (r2) and an offset, which is a 16 bit signed quantity.

	This definitely *IS* a linker bug.  A known linker bug.  Your
comments above about TOC overflow misses the point.  This bug originally
was uncovered by GNAT (which also requires -bbigtoc) last Spring.

	Kaveh's report shows that g77 uses the AIX4 linker -bbigtoc option
to handle TOC overflow (somewhat better than using GCC's -mminimal-toc
option).  The AIX linker, however, has a bug in the way that it handles
certain CSECTS and their proximity to the TOC anchor.  IBM's compilers do
not generate code which will elicit the problem, but GCC's somewhat stupid
handling of the TOC does generate such code which sometimes affects the
AIX linkers ability to handle programs requiring -bbigtoc, as in this
case.

	I do not know about the status of a fix.

David

	This message is not an official statement from IBM Corp and does
not imply any level, timetable, or commitment of support for this problem.
===============================================================================
David Edelsohn                                      T.J. Watson Research Center
dje@watson.ibm.com                                  P.O. Box 218
+1 914 945 4364 (TL 862)                            Yorktown Heights, NY 10598