On Wed, Sep 16, 2026 at 4:43 AM Kael Andrew Alonzo Franco
<[email protected]> wrote:
>
> When X is INT_MIN and N is -1, then
> r14-3644 converts ((X {+,-} M*N) / N) to
> ((X / N) {+,-} M). But INT_MIN / -1 overflows where
> ((X {+,-} M*N) / N) does not.
> Add a check for ((X {+,-} M*N) / N) via X > INT_MIN
> and N != -1.
>
> Bootstrapped and regtested on x86_64-pc-linux-gnu.
>
> PR tree-optimization/127406
>
> gcc/ChangeLog:
>
> * match.pd: Restrict (X {+,-} M*N) / N optimization.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.dg/torture/pr127406.c: New test.
>
> Signed-off-by: Kael Andrew Franco <[email protected]>
> ---
> gcc/match.pd | 10 +++++++--
> gcc/testsuite/gcc.dg/torture/pr127406.c | 29 +++++++++++++++++++++++++
> 2 files changed, 37 insertions(+), 2 deletions(-)
> create mode 100644 gcc/testsuite/gcc.dg/torture/pr127406.c
>
> diff --git a/gcc/match.pd b/gcc/match.pd
> index 4059ed787f0..08fe309c445 100644
> --- a/gcc/match.pd
> +++ b/gcc/match.pd
> @@ -1177,7 +1177,10 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
> /* "X+N*M" is not with opposite sign as "X". */
> && (TYPE_UNSIGNED (type)
> || (vr0.nonnegative_p () && vr4.nonnegative_p ())
> - || (vr0.nonpositive_p () && vr4.nonpositive_p ())))
> + || (vr0.nonpositive_p () && vr4.nonpositive_p ()
> + /* X = INT_MIN and N = -1 causes overflow. */
> + && !(vr0.contains_p (wi::min_value (TYPE_PRECISION (type),
> SIGNED))
> + && vr2.contains_p (wi::minus_one (TYPE_PRECISION
> (type)))))))
I think it is better to write this as:
vr0.nonpositive_p () && vr4.nonpositive_p () && (!vr0.contains_p(...)
|| !vr2.contains_p(...))
> (plus (div @0 @2) @1))))
>
> /* Simplify (X - M*N) / N -> X / N - M. */
> @@ -1197,7 +1200,10 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
> /* "X-N*M" is not with opposite sign as "X". */
> && (TYPE_UNSIGNED (type)
> || (vr0.nonnegative_p () && vr4.nonnegative_p ())
> - || (vr0.nonpositive_p () && vr4.nonpositive_p ())))
> + || (vr0.nonpositive_p () && vr4.nonpositive_p ()
> + /* X = INT_MIN and N = -1 causes overflow. */
> + && !(vr0.contains_p (wi::min_value (TYPE_PRECISION (type),
> SIGNED))
> + && vr2.contains_p (wi::minus_one (TYPE_PRECISION
> (type)))))))
Likewise.
This is ok with those two style changes.
Also this is ok to backport for all open affected branches.
> (minus (div @0 @2) @1)))))
>
> /* Simplify
> diff --git a/gcc/testsuite/gcc.dg/torture/pr127406.c
> b/gcc/testsuite/gcc.dg/torture/pr127406.c
> new file mode 100644
> index 00000000000..d81a548be46
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/torture/pr127406.c
> @@ -0,0 +1,29 @@
> +/* { dg-do run } */
> +/* { dg-additional-options "-O2" } */
> +
> +__attribute__((noipa))
> +int
> +fplus (int x, int m, int n)
> +{
> + if (x > -6 || m < -3 || m > -2 || n < -2 || n > -1)
> + return 0;
> + return (x + m * n) / n;
> +}
> +
> +__attribute__((noipa))
> +int
> +fminus (int x, int m, int n)
> +{
> + if (x > -6 || m < 2 || m > 3 || n < -2 || n > -1)
> + return 0;
> + return (x - m * n) / n;
> +}
> +
> +int
> +main (void)
> +{
> + if (fplus (-2147483647 - 1, -2, -1) != 2147483646
> + || fminus (-2147483647 - 1, 2, -1) != 2147483646)
> + __builtin_abort ();
> + return 0;
> +}
> --
> 2.55.0
>
