[PATCH] [PR116831] match.pd: Check trunc_mod vector obtap before folding.

[Richard Biener]

On Wed, 9 Oct 2024, Jennifer Schmitz wrote:

> 
> > On 8 Oct 2024, at 10:31, Richard Biener <rguenther@suse.de> wrote:
> > 
> > External email: Use caution opening links or attachments
> > 
> > 
> > On Fri, 4 Oct 2024, Jennifer Schmitz wrote:
> > 
> >> As in https://gcc.gnu.org/pipermail/gcc-patches/2024-September/663185.html,
> >> this patch guards the simplification x / y * y == x -> x % y == 0 in
> >> match.pd for vector types by a check for:
> >> 1) Support of the mod optab for vectors OR
> >> 2) Application before vector lowering for non-VL vectors.
> >> 
> >> The patch was bootstrapped and tested with no regression on
> >> aarch64-linux-gnu and x86_64-linux-gnu.
> >> OK for mainline?
> > 
> > -  (if (TREE_CODE (TREE_TYPE (@0)) != COMPLEX_TYPE)
> > +  (if (TREE_CODE (TREE_TYPE (@0)) != COMPLEX_TYPE
> > +       || (VECTOR_INTEGER_TYPE_P (type)
> > +          && ((optimize_vectors_before_lowering_p ()
> > +               && TREE_CODE (TYPE_SIZE (type)) == INTEGER_CST)
> > +              || target_supports_op_p (type, TRUNC_MOD_EXPR,
> > +                                       optab_vector))))
> > 
> > this looks a bit odd, VECTOR_INTEGER_TYPE_P (type) checks the
> > result type of the comparison.  I think the whole condition is
> > better written as
> > 
> > (if (TREE_CODE (TREE_TYPE (@0)) != COMPLEX_TYPE
> >      && (!VECTOR_MODE_P (TYPE_MODE (TREE_TYPE (@0)))
> >          || !target_supports_op_p (TREE_TYPE (@0), TRUNC_DIV_EXPR,
> >                                    optab_vector)
> >          || target_supports_op_p (TREE_TYPE (@0), TRUNC_MOD_EXPR,
> >                                    optab_vector)))
> > 
> > when we have non-vector mode we're before lowering, likewise when
> > the target doesn't support the division.  Even before lowering
> > we shouldn't replace a supported division (and multiplication)
> > with an unsupported modulo.
> Dear Richard,
> thanks for the review. I updated the patch with your suggestion and re-validated on aarch64 and x86_64.
> Best,
> Jennifer
> 
> This patch guards the simplification x / y * y == x -> x % y == 0 in
> match.pd by a check for:
> 1) Non-vector mode of x OR
> 2) Lack of support for vector division OR
> 3) Support of vector modulo
> 
> The patch was bootstrapped and tested with no regression on
> aarch64-linux-gnu and x86_64-linux-gnu.
> OK for mainline?

OK.

Thanks,
Richard.

> Signed-off-by: Jennifer Schmitz <jschmitz@nvidia.com>
> 
> gcc/
> 	PR tree-optimization/116831
> 	* match.pd: Guard simplification to trunc_mod with check for
> 	mod optab support.
> 
> gcc/testsuite/
> 	PR tree-optimization/116831
> 	* gcc.dg/torture/pr116831.c: New test.
> ---
>  gcc/match.pd                            |  9 +++++++--
>  gcc/testsuite/gcc.dg/torture/pr116831.c | 10 ++++++++++
>  2 files changed, 17 insertions(+), 2 deletions(-)
>  create mode 100644 gcc/testsuite/gcc.dg/torture/pr116831.c
> 
> diff --git a/gcc/match.pd b/gcc/match.pd
> index ba83f0f29e6..9b59b5c12f1 100644
> --- a/gcc/match.pd
> +++ b/gcc/match.pd
> @@ -5380,8 +5380,13 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
>  /* x / y * y == x -> x % y == 0.  */
>  (simplify
>    (eq:c (mult:c (trunc_div:s @0 @1) @1) @0)
> -  (if (TREE_CODE (TREE_TYPE (@0)) != COMPLEX_TYPE)
> -    (eq (trunc_mod @0 @1) { build_zero_cst (TREE_TYPE (@0)); })))
> +  (if (TREE_CODE (TREE_TYPE (@0)) != COMPLEX_TYPE
> +       && (!VECTOR_MODE_P (TYPE_MODE (TREE_TYPE (@0)))
> +	   || !target_supports_op_p (TREE_TYPE (@0), TRUNC_DIV_EXPR,
> +				     optab_vector)
> +	   || target_supports_op_p (TREE_TYPE (@0), TRUNC_MOD_EXPR,
> +				    optab_vector)))
> +   (eq (trunc_mod @0 @1) { build_zero_cst (TREE_TYPE (@0)); })))
>  
>  /* ((X /[ex] A) +- B) * A  -->  X +- A * B.  */
>  (for op (plus minus)
> diff --git a/gcc/testsuite/gcc.dg/torture/pr116831.c b/gcc/testsuite/gcc.dg/torture/pr116831.c
> new file mode 100644
> index 00000000000..92b2a130e69
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/torture/pr116831.c
> @@ -0,0 +1,10 @@
> +/* { dg-additional-options "-mcpu=neoverse-v2" { target aarch64*-*-* } } */
> +
> +long a;
> +int b, c;
> +void d (int e[][5], short f[][5][5][5]) 
> +{
> +  for (short g; g; g += 4)
> +    a = c ?: e[6][0] % b ? 0 : f[0][0][0][g];
> +}
> +
> 

-- 
Richard Biener <rguenther@suse.de>
SUSE Software Solutions Germany GmbH,
Frankenstrasse 146, 90461 Nuernberg, Germany;
GF: Ivo Totev, Andrew McDonald, Werner Knoblich; (HRB 36809, AG Nuernberg)