[Bug middle-end/48377] [4.6 regression] miscompilation at -O3

["matt at use dot net" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48377

--- Comment #5 from Matt Hargett <matt at use dot net> 2011-04-05 03:53:19 UTC ---
(In reply to comment #4)
> Anyway, please read http://gcc.gnu.org/bugs.html, there is nothing we can do
> from the information you've provided.  Trying to find from a vague description
> of where you see the crash a difference in assembly is very hard, plus it may
> very well be just a bug in the sources you are compiling.  If -O2 works and -O3
> doesn't, try to add __attribute__((noinline)) resp.
> __attribute__((optimize(2)))
> resp. __attribute__((optimize(3))) to various suspect routines in the file to
> try to narrow it down to a single problematic routine (perhaps with some
> functions inlined into it), which will work fine if the whole file is compiled
> with -O3 just that routine is -O2 (and perhaps functions inlined into it) and
> will misbehave if everything is compled with -O2 just that function is compiled
> with -O3.  Then try to make a self-contained executable testcase out of it
> (just call that routine from some other CU's main, perhaps with the minimal
> necessary setup for it and stub everything it calls.  For (suspected)
> miscompilation we really need to be able to reproduce it ourselves, preferrably
> with a minimal testcase, otherwise we can't do anything on it unless you debug
> the problem yourself down to pointing a bug somewhere exact in the assembly.

I've got a greatly reduced testcase that I'm continuing to pare down. I'll
attach the self-contained example tomorrow.