[Bug middle-end/126814] [17 Regression] ICE in RTL expand in fill_ranger_cache at gimple-range-cache.cc:1658 since r17-3171

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126814

--- Comment #6 from Drea Pinski <pinskia at gcc dot gnu.org> ---
So gimple_stmt_nonnegative_p uses get_global_range_query () for fp while
tree_single_nonnegative_p uses get_range_query (cfun).

This is why this didn't show up for fp before.

So my original patch used get_global_range_query but Andrew requested to move
it to get_range_query. 

Anyways the real issue is the call to gimple_stmt_nonnegative_p; it should be
removed and reimplemented in gimple-range-op.cc's maybe_builtin_call. (a bug
which I am about to file shows why).