https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127182
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Component|target |testsuite
--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Actually, we already have vect_usad_char effective target, so I'd just go with
2026-09-02 Jakub Jelinek <[email protected]>
PR target/127182
* gcc.dg/vect/vect-reduc-chain-3.c: Guard SAD_EXPR scan-tree-dump with
vect_usad_char effective target rather than vect_udot_qi.
--- a/gcc/testsuite/gcc.dg/vect/vect-reduc-chain-3.c 2026-03-27
10:17:17.727272108 +0100
+++ b/gcc/testsuite/gcc.dg/vect/vect-reduc-chain-3.c 2026-09-02
12:24:12.989003017 +0200
@@ -64,5 +64,5 @@ main (void)
__builtin_abort ();
}
-/* { dg-final { scan-tree-dump "vectorizing SLP node starting from: \\S+ =
SAD_EXPR" "vect" { target vect_udot_qi } } } */
+/* { dg-final { scan-tree-dump "vectorizing SLP node starting from: \\S+ =
SAD_EXPR" "vect" { target vect_usad_char } } } */
/* { dg-final { scan-tree-dump "vectorizing SLP node starting from: \\S+ =
DOT_PROD_EXPR" "vect" { target vect_sdot_hi } } } */
I wonder why vect_usad_char uses check_p9vector_hw_available rather than
check_effective_target_has_arch_pwr9, and am also confused by
check_vect_support_and_set_flags adding -mcpu=power9 if power9 hw is available
but only if power10 hw is not available, but for power10 hw it doesn't
anything. Am not sure how that can work at all.
Michael?
