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.
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.
(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. */
}
}
(if (full_perm_p)
- (vec_perm (op @0 @1) (op @0 @1) @2))))))
+ (vec_perm (op@3 @0 @1) @3 @2))))))
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-O3 -march=armv8-a+sve2" } */
+
+void foo(int n, char *restrict out, char *restrict in) {
+ for (int i=n; i-->0; ) {
+ out[i] += in[i];
+ }
+}