[committed] fix frame base for non-dwarf2 unwind info targets

[James E Wilson <wilson at specifix dot com>]

This is a partial fix for PR 24444.  When Richard Henderson rewrote the
frame base support in dwarf2out.c, he broke it for targets that don't
use dwarf2 unwind info.  This includes IA-64.  The fix is to readd some
of the code Richard removed, surrounded by ifdefs, so it is used for
only targets that don't use dwarf2 unwind info.

This isn't a complete fix, because we no longer have frame base info for
the prologue and epilogue.  The old way required some code in
var-tracking.c that was removed, and it doesn't seem to make sense to
readd it.  Instead, the plan here is to add another hook, or maybe make
use of DW_OP_call_frame_cfa.  This fix is good enough to make the gdb
testsuite happy though.

This was tested with an IA-64 linux bootstrap and make check for all
default languages.  There were no regressions.  This was also tested
with a gdb make check.  I tried running the gdb testsuite without this
patch, but so many tests were failing and timing out that I gave up
after 4 hours and killed it.  With this patch, I get reasonable gdb
testsuite results that are slightly better than the results I get with
gcc-3.3.5.  With the patched gcc-4.1, I got:
# of expected passes            10577
# of unexpected failures        176
# of unexpected successes       2
# of expected failures          42
# of known failures             40
# of unresolved testcases       5
# of untested testcases         3
# of unsupported tests          4
and with gcc-3.3.5 I got:
# of expected passes            10517
# of unexpected failures        186
# of unexpected successes       2
# of expected failures          41
# of known failures             44
# of unresolved testcases       5
# of untested testcases         7
# of unsupported tests          7
-- 
Jim Wilson, GNU Tools Support, http://www.specifix.com

Attachment: patch.frame.base
Description: Text document