Hi!
I really can't figure out why one would need to add extra casts.
type must be an integral type which has BIT_NOT_EXPR applied on it
which yields all ones and we need a type in which negating 0 or 1
range will yield 0 or all ones, I think all integral types satisfy
that.
This fixes PR111369, where one of the bitint*.c tests FAILs with
GCC_TEST_RUN_EXPENSIVE=1.
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
2023-09-30 Jakub Jelinek <ja...@redhat.com>
PR middle-end/111369
* match.pd (a?~t:t -> (-(a))^t): Always convert to type rather
than using build_nonstandard_integer_type.
--- gcc/match.pd.jj 2023-09-28 11:32:16.122434235 +0200
+++ gcc/match.pd 2023-09-29 18:05:50.554640268 +0200
@@ -6742,12 +6742,7 @@ (define_operator_list SYNC_FETCH_AND_AND
(if (INTEGRAL_TYPE_P (type)
&& bitwise_inverted_equal_p (@1, @2, wascmp)
&& (!wascmp || element_precision (type) == 1))
- (with {
- auto prec = TYPE_PRECISION (type);
- auto unsign = TYPE_UNSIGNED (type);
- tree inttype = build_nonstandard_integer_type (prec, unsign);
- }
- (convert (bit_xor (negate (convert:inttype @0)) (convert:inttype @2)))))))
+ (bit_xor (negate (convert:type @0)) (convert:type @2)))))
#endif
/* Simplify pointer equality compares using PTA. */
Jakub
