Re: [PATCH] match.pd: Fold CPROJ when nested inside CPROJ or CABS [PR 126690].

[Andrea Pinski]

On Wed, Aug 26, 2026 at 8:57 AM Dipesh Sharma <dipesh.sha...@amd.com> wrote:
>
> Hi,
>
> This patch aims to address the issue 
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126690
>
> regtested and bootstrapped on x86_64-linux-gnu. Ok for trunk ?
>
> gcc/ChangeLog:
>         PR tree-optimization/126690.
>         * match.pd: Add patterns to fold CPROJ when nested inside CPROJ or 
> CABS.
>
> gcc/testsuite/ChangeLog:
>
>         * gcc.dg/tree-ssa/pr126690.c: New test.
> ---
>  gcc/match.pd                             | 10 +++++
>  gcc/testsuite/gcc.dg/tree-ssa/pr126690.c | 52 ++++++++++++++++++++++++
>  2 files changed, 62 insertions(+)
>  create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr126690.c
>
> diff --git a/gcc/match.pd b/gcc/match.pd
> index c7841cab459..a7bf8092d25 100644
> --- a/gcc/match.pd
> +++ b/gcc/match.pd
> @@ -10186,6 +10186,16 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
>   (if (real_isinf (TREE_REAL_CST_PTR (@1)))
>    { build_complex_inf (type, TREE_REAL_CST_PTR (@1)->sign); }))
>
> +/* cproj(cproj(x)) -> cproj(x).  */
> +(simplify
> + (CPROJ (CPROJ @0))
> + (CPROJ @0))
> +
> +/* cabs(cproj(x)) -> cabs(x).  */
> +(simplify
> + (CABS (CPROJ @0))
> + (CABS @0))
> +
>  (for pows (POW)
>       sqrts (SQRT)
>       cbrts (CBRT)
> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr126690.c 
> b/gcc/testsuite/gcc.dg/tree-ssa/pr126690.c
> new file mode 100644
> index 00000000000..f4125fef6bc
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr126690.c
> @@ -0,0 +1,52 @@
> +/* { dg-do compile } */
> +/* { dg-require-effective-target c99_runtime } */
> +/* { dg-options "-O2 -fdump-tree-optimized" } */
> +
> +__attribute__((noinline))
> +_Complex float
> +cproj_idempotentf (_Complex float a)
> +{
> +  return __builtin_cprojf (__builtin_cprojf (a));
> +}
> +
> +__attribute__((noinline))
> +_Complex double
> +cproj_idempotent (_Complex double a)
> +{
> +  return __builtin_cproj (__builtin_cproj (a));
> +}
> +
> +__attribute__((noinline))
> +_Complex long double
> +cproj_idempotentl (_Complex long double a)
> +{
> +  return __builtin_cprojl (__builtin_cprojl (a));
> +}
> +
> +__attribute__((noinline))
> +float
> +cabs_absorbs_cprojf (_Complex float a)
> +{
> +  return __builtin_cabsf (__builtin_cprojf (a));
> +}
> +
> +__attribute__((noinline))
> +double
> +cabs_absorbs_cproj (_Complex double a)
> +{
> +  return __builtin_cabs (__builtin_cproj (a));
> +}
> +
> +__attribute__((noinline))
> +long double
> +cabs_absorbs_cprojl (_Complex long double a)
> +{
> +  return __builtin_cabsl (__builtin_cprojl (a));
> +}
> +
> +/* { dg-final { scan-tree-dump-not {__builtin_cprojf \(__builtin_cprojf} 
> "optimized" } } */
> +/* { dg-final { scan-tree-dump-not {__builtin_cproj \(__builtin_cproj} 
> "optimized" } } */
> +/* { dg-final { scan-tree-dump-not {__builtin_cprojl \(__builtin_cprojl} 
> "optimized" } } */
> +/* { dg-final { scan-tree-dump-not {__builtin_cabsf \(__builtin_cprojf} 
> "optimized" } } */
> +/* { dg-final { scan-tree-dump-not {__builtin_cabs \(__builtin_cproj} 
> "optimized" } } */
> +/* { dg-final { scan-tree-dump-not {__builtin_cabsl \(__builtin_cprojl} 
> "optimized" } } */

This testcase will pass even before your match.pd patch.
Maybe the best way to write this testcase is count the number of
cabs/cproj used.

> --
> 2.34.1
>