[Bug middle-end/90648] [8/9/10 Regression] ICE tree check: accessed operand 4 of call_expr with 3 operands in generic_simplify_MULT_EXPR, at generic-match.c:27222

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90648

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The fact that it is a call to builtin is determined just by the call calling a
fndecl which has builtin class different from none.
At GIMPLE perhaps we could cache whether it is a valid builtin call in some
GF_CALL_* bit, just would to trigger recomputation of that if we make some
significant changes (change types of arguments or similar).  As the usual case
is correct prototypes and correct arguments to builtins, I think caching this
call to builtin is ok is all we need.
But for GENERIC we don't even have a routine similar to
gimple_builtin_call_types_compatible_p that would check a CALL_EXPR, at least
I'm not aware of it.