https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122213
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Target Milestone|--- |15.3
See Also| |https://gcc.gnu.org/bugzill
| |a/show_bug.cgi?id=118637
Summary|{CEIL,ROUND}_DIV_EXPR |[15/16 Regression]
|produces wrong results when |{CEIL,ROUND}_DIV_EXPR
|constant operand is 2 or |produces wrong results when
|power of 2 with O2 flag |constant operand is 2 or
| |power of 2 with O2 flag
Keywords| |wrong-code
--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
In the inline case we get:
Matching expression match.pd:2944, gimple-match-8.cc:90
Matching expression match.pd:2948, gimple-match-4.cc:77
Matching expression match.pd:2958, gimple-match-6.cc:91
Matching expression match.pd:2964, gimple-match-10.cc:84
Matching expression match.pd:2960, gimple-match-6.cc:105
Matching expression match.pd:2964, gimple-match-10.cc:84
Matching expression match.pd:2940, gimple-match-8.cc:65
Matching expression match.pd:2948, gimple-match-4.cc:77
Matching expression match.pd:197, gimple-match-6.cc:22
gimple_simplified to D_6 = 2;
Oh:
Applying pattern match.pd:991, gimple-match-4.cc:5994
gimple_simplified to D_2 = a_1(D) >> 3;
That is:
```
(for div (trunc_div ceil_div floor_div round_div exact_div)
#if GIMPLE
/* Canonicalize unsigned t / 4 to t >> 2. */
(simplify
(div @0 integer_pow2p@1)
(if (INTEGRAL_TYPE_P (type)
&& (TYPE_UNSIGNED (type) || tree_expr_nonnegative_p (@0)))
(rshift @0 { build_int_cst (integer_type_node,
wi::exact_log2 (wi::to_wide (@1))); })))
```
Which was added in r15-7223-g92a5c5100c2519.
I think that is only valid for trunc_div, floor_div, and exact_div.
