[Bug tree-optimization/58143] [4.8/4.9 regression] wrong code at -O3

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58143

--- Comment #20 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Bernd Edlinger from comment #18)
> (In reply to Richard Biener from comment #17)
> > *** Bug 58731 has been marked as a duplicate of this bug. ***
> 
> Richard,
> 
> should I commit my patch now?

No, I would have approved it if so.  I am not happy with disabling loop
invariant motion for these cases.  Instead let's first fix the sequence
of bugs exposed in all testcases sofar, LIM moving sth, SCEV plus
niter analysis figuring out a wrong number of iterations, unroll killing
the exit test.  This restricts the operations we have to handle and
ultimately simplifies the patch which makes it more suitable for backporting.