[Bug tree-optimization/71437] [7 regression] Performance regression after r235817

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71437

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
The issue with the optimization opportunity is that IPA reference does not
compute that GetIN does not modify pL because its analysis is restricted
to variables local to the compilation unit.  Using -fwhole-program fixes that.
I didn't check whether PRE performs the desired optimization then - the
code gets quite obfuscated with -fwhole-program.

The r235817 issue is likely spurious - the SSA dumps of GCC 6 and trunk
agree and thus differences are likely due to SSA numbering differences.
The usual suspect is out-of SSA coalescing (without having verified that).