middle-end: ensure that VEC_PERM operands get lowered to the same SSA_NAME. [PR107717]

Message ID patch-16601-tamar@arm.com
State Committed
Headers
Series middle-end: ensure that VEC_PERM operands get lowered to the same SSA_NAME. [PR107717] |

Commit Message

Tamar Christina Nov. 17, 2022, 4:09 a.m. UTC
  Hi All,

At the moment when the VEC_PERMs generated by this match.pd rule is generated
it creates two different SSA_NAMEs for the folded operand.  Because of this it
the permute switches from a single operand permute to a two operand permute and
the target may no longer support a permute for this.

This fixes it by ensuring we generate the same SSA_NAME for both operands.

Bootstrapped Regtested on aarch64-none-linux-gnu and no issues.

Ok for master?

Thanks,
Tamar

gcc/ChangeLog:

	PR tree-optimization/107717
	* match.pd: Ensure same SSA_NAME.

gcc/testsuite/ChangeLog:

	PR tree-optimization/107717
	* gcc.target/aarch64/sve2/pr107717.c: New test.

--- inline copy of patch -- 
diff --git a/gcc/match.pd b/gcc/match.pd
index 82f05bbc912e4f80f3984d930c4a8dcb010136e1..cd510d80d409a7c52ba2a78e2fd7df0fc85cab2b 100644




--
diff --git a/gcc/match.pd b/gcc/match.pd
index 82f05bbc912e4f80f3984d930c4a8dcb010136e1..cd510d80d409a7c52ba2a78e2fd7df0fc85cab2b 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -8317,7 +8317,7 @@ and,
  (simplify
   (op (vec_perm @0 @0 @2) (vec_perm @1 @1 @2))
    (if (VECTOR_INTEGER_TYPE_P (type))
-    (vec_perm (op @0 @1) (op @0 @1) @2))))
+    (vec_perm (op@3 @0 @1) @3 @2))))
 
 /* Similar for float arithmetic when permutation constant covers
    all vector elements.  */
@@ -8356,4 +8356,4 @@ and,
 	 }
       }
       (if (full_perm_p)
-	(vec_perm (op @0 @1) (op @0 @1) @2))))))
+	(vec_perm (op@3 @0 @1) @3 @2))))))
  

Comments

Richard Biener Nov. 17, 2022, 6:33 a.m. UTC | #1
> Am 17.11.2022 um 05:10 schrieb Tamar Christina via Gcc-patches <gcc-patches@gcc.gnu.org>:
> 
> ´╗┐Hi All,
> 
> At the moment when the VEC_PERMs generated by this match.pd rule is generated
> it creates two different SSA_NAMEs for the folded operand.  Because of this it
> the permute switches from a single operand permute to a two operand permute and
> the target may no longer support a permute for this.
> 
> This fixes it by ensuring we generate the same SSA_NAME for both operands.
> 
> Bootstrapped Regtested on aarch64-none-linux-gnu and no issues.
> 
> Ok for master?
> 

Ok.

Richard 

> Thanks,
> Tamar
> 
> gcc/ChangeLog:
> 
>    PR tree-optimization/107717
>    * match.pd: Ensure same SSA_NAME.
> 
> gcc/testsuite/ChangeLog:
> 
>    PR tree-optimization/107717
>    * gcc.target/aarch64/sve2/pr107717.c: New test.
> 
> --- inline copy of patch -- 
> diff --git a/gcc/match.pd b/gcc/match.pd
> index 82f05bbc912e4f80f3984d930c4a8dcb010136e1..cd510d80d409a7c52ba2a78e2fd7df0fc85cab2b 100644
> --- a/gcc/match.pd
> +++ b/gcc/match.pd
> @@ -8317,7 +8317,7 @@ and,
>  (simplify
>   (op (vec_perm @0 @0 @2) (vec_perm @1 @1 @2))
>    (if (VECTOR_INTEGER_TYPE_P (type))
> -    (vec_perm (op @0 @1) (op @0 @1) @2))))
> +    (vec_perm (op@3 @0 @1) @3 @2))))
> 
> /* Similar for float arithmetic when permutation constant covers
>    all vector elements.  */
> @@ -8356,4 +8356,4 @@ and,
>     }
>       }
>       (if (full_perm_p)
> -    (vec_perm (op @0 @1) (op @0 @1) @2))))))
> +    (vec_perm (op@3 @0 @1) @3 @2))))))
> 
> 
> 
> 
> -- 
> <rb16601.patch>
  

Patch

--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -8317,7 +8317,7 @@  and,
  (simplify
   (op (vec_perm @0 @0 @2) (vec_perm @1 @1 @2))
    (if (VECTOR_INTEGER_TYPE_P (type))
-    (vec_perm (op @0 @1) (op @0 @1) @2))))
+    (vec_perm (op@3 @0 @1) @3 @2))))
 
 /* Similar for float arithmetic when permutation constant covers
    all vector elements.  */
@@ -8356,4 +8356,4 @@  and,
 	 }
       }
       (if (full_perm_p)
-	(vec_perm (op @0 @1) (op @0 @1) @2))))))
+	(vec_perm (op@3 @0 @1) @3 @2))))))