> Am 23.07.2026 um 18:15 schrieb Jeffrey Law <[email protected]>:
>
>
>
>> On 7/23/2026 2:59 AM, Richard Biener wrote:
>>> On Thu, Jul 23, 2026 at 9:12 AM Eikansh Gupta
>>> <[email protected]> wrote:
>>> Fold (signbit (x) cmp1 0) cmp (signbit (y) cmp2 0) ? y : -y
>>> to copysign (y, +-x). The result keeps the magnitude of Y and takes its
>>> sign from X (or -X). Emitted as IFN_COPYSIGN when the target supports it.
>>>
>>> PR tree-optimization/109843
>>>
>>> gcc/ChangeLog:
>>>
>>> * match.pd ((signbit (x) cmp 0) cmp (signbit (y) cmp 0) ? y : -y):
>>> New simplification to copysign (y, +-x).
>>>
>>> gcc/testsuite/ChangeLog:
>>>
>>> * gcc.dg/tree-ssa/pr109843.c: New test.
>>>
>>> Signed-off-by: Eikansh Gupta <[email protected]>
>>> ---
>>> gcc/match.pd | 20 +++++++
>>> gcc/testsuite/gcc.dg/tree-ssa/pr109843.c | 76 ++++++++++++++++++++++++
>>> 2 files changed, 96 insertions(+)
>>> create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr109843.c
>>>
>>> diff --git a/gcc/match.pd b/gcc/match.pd
>>> index d1a12c35ed3..59ea0e1cc0c 100644
>>> --- a/gcc/match.pd
>>> +++ b/gcc/match.pd
>>> @@ -9864,6 +9864,26 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
>>> (abs @0)
>>> (negate (abs @0))))))
>>>
>>> +/* (signbit (x) cmp1 0) cmp (signbit (y) cmp2 0) ? y : -y
>>> + -> copysign (y, +-x). */
>>> +#if GIMPLE
>>> +(for cmp (eq ne)
>>> + (for icmp1 (eq ne)
>>> + (for icmp2 (eq ne)
>>> + (for sign (SIGNBIT)
>> That's an awful a lot of combinations (2*2*2*4)
>>
>>> + (simplify
>>> + (cond (cmp:c (icmp1 (sign @0) integer_zerop)
>> Plus :c here.
>>
>> I wonder where the pattern request comes from ...
> Probably not spec given Andrea has been tagging things she finds with spec
> when relevant.
>
> Are you objecting to the number of patterns this generates, or just noting
> that it's a bunch?
Just noticed it’s a bunch but without a concrete idea how to reduce
(canonicalization) or doing it in a pass in some other ways (copy-of-sign
propagation?).
>
> jeff
>
>
