https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118669
Richard Sandiford <rsandifo at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords|wrong-code |internal-improvement
Assignee|unassigned at gcc dot gnu.org |rsandifo at gcc dot
gnu.org
Status|NEW |ASSIGNED
--- Comment #7 from Richard Sandiford <rsandifo at gcc dot gnu.org> ---
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index c0550acf6b2..06cd6e42355 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -9908,7 +9908,8 @@ vectorizable_store (vec_info *vinfo,
= fold_build2 (MEM_REF, vectype, dataref_ptr,
dataref_offset ? dataref_offset
: build_int_cst (ref_type, 0));
- if (alignment_support_scheme == dr_aligned)
+ if (alignment_support_scheme == dr_aligned
+ && align >= TYPE_ALIGN_UNIT (vectype))
;
else
TREE_TYPE (data_ref)
seems to fix it. (Loads will need similar treatment.)
The justification is that preferred_vector_alignment requests element alignment
for SVE, even though the vector mode itself has 128-bit alignment. In that
context, dr_aligned means that element alignment has been reached.
It looks like a long-standing issue, so I suppose it should wait for GCC 16.
