On Wed, May 19, 2021 at 11:09:19AM +0200, Jakub Jelinek via Gcc-patches wrote:
> On Wed, May 19, 2021 at 10:15:53AM +0200, Christophe Lyon via Gcc-patches
> wrote:
> > After this update, the test fails on arm and aarch64: according to the
> > logs, the optimization is still performed 14 times.
>
> Seems this is because
> if (change
> && !flag_syntax_only
> && (load_extend_op (TYPE_MODE (TREE_TYPE (and0)))
> == ZERO_EXTEND))
> {
> tree uns = unsigned_type_for (TREE_TYPE (and0));
> and0 = fold_convert_loc (loc, uns, and0);
> and1 = fold_convert_loc (loc, uns, and1);
> }
> in fold-const.c adds on these targets extra casts that prevent the
> optimizations.
This patch seems to fix it (but I don't have an easy way to test on aarch64
or arm on the trunk and 11 branch would need numerous backports).
2021-05-19 Jakub Jelinek <ja...@redhat.com>
PR tree-optimization/94589
* match.pd ((X & Y) == X -> (X & ~Y) == 0): Simplify even in presence
of integral conversions.
--- gcc/match.pd.jj 2021-05-15 10:10:28.000000000 +0200
+++ gcc/match.pd 2021-05-19 11:34:42.130624557 +0200
@@ -4769,6 +4769,16 @@ (define_operator_list COND_TERNARY
(simplify
(cmp:c (bit_and:c @0 @1) @0)
(cmp (bit_and @0 (bit_not! @1)) { build_zero_cst (TREE_TYPE (@0)); }))
+ (simplify
+ (cmp:c (convert@3 (bit_and (convert@2 @0) INTEGER_CST@1)) (convert @0))
+ (if (INTEGRAL_TYPE_P (TREE_TYPE (@0))
+ && INTEGRAL_TYPE_P (TREE_TYPE (@2))
+ && INTEGRAL_TYPE_P (TREE_TYPE (@3))
+ && TYPE_PRECISION (TREE_TYPE (@2)) == TYPE_PRECISION (TREE_TYPE (@0))
+ && TYPE_PRECISION (TREE_TYPE (@3)) > TYPE_PRECISION (TREE_TYPE (@2))
+ && !wi::neg_p (wi::to_wide (@1)))
+ (cmp (bit_and @0 (convert (bit_not @1)))
+ { build_zero_cst (TREE_TYPE (@0)); })))
/* (X | Y) == Y becomes (X & ~Y) == 0. */
(simplify
Jakub
