[Bug tree-optimization/84478] [8 Regression] pdftex miscompilation on i386

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84478

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 43474
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=43474&action=edit
gcc8-pr84478.patch

Untested fix.  Sadly the warning code for some weird reason relies on the bogus
PHI handling (strangely COND_EXPR is handled differently), but obviously in
code generation we can't allow that, if there is any PHI argument or any
COND_EXPR we don't know anything about, we can't use it for range info.