https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69421
Ilya Enkovich <ienkovich at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |ASSIGNED
--- Comment #9 from Ilya Enkovich <ienkovich at gcc dot gnu.org> ---
Here we don't check boolean values get compatible vectypes when analyze
VEC_COND statement. Here statement gets V16QI mode due to boolean type size.
And one of its operand gets V2DI mode because it is a result of comparison of
doubles.
We have patterns to catch such cases and add additional mask conversion for
comparison and bit operations but don't have them for VEC_COND.
Suppose vectorization of such cases should be delayed until stage1. For now
this patch helps to compile both reduced and original tests:
diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
index 1d2246d..8cb1bfe 100644
--- a/gcc/tree-vect-stmts.c
+++ b/gcc/tree-vect-stmts.c
@@ -7528,6 +7528,7 @@ vectorizable_condition (gimple *stmt,
gimple_stmt_iterator *gsi,
tree vectype = STMT_VINFO_VECTYPE (stmt_info);
int nunits = TYPE_VECTOR_SUBPARTS (vectype);
+ tree vectype1 = NULL_TREE, vectype2 = NULL_TREE;
if (slp_node || PURE_SLP_STMT (stmt_info))
ncopies = 1;
@@ -7547,9 +7548,19 @@ vectorizable_condition (gimple *stmt,
gimple_stmt_iterator *gsi,
return false;
gimple *def_stmt;
- if (!vect_is_simple_use (then_clause, stmt_info->vinfo, &def_stmt, &dt))
+ if (!vect_is_simple_use (then_clause, stmt_info->vinfo, &def_stmt, &dt,
+ &vectype1))
+ return false;
+ if (!vect_is_simple_use (else_clause, stmt_info->vinfo, &def_stmt, &dt,
+ &vectype2))
return false;
- if (!vect_is_simple_use (else_clause, stmt_info->vinfo, &def_stmt, &dt))
+
+ if (vectype1
+ && TYPE_VECTOR_SUBPARTS (vectype1) != TYPE_VECTOR_SUBPARTS (vectype))
+ return false;
+
+ if (vectype2
+ && TYPE_VECTOR_SUBPARTS (vectype2) != TYPE_VECTOR_SUBPARTS (vectype))
return false;
masked = !COMPARISON_CLASS_P (cond_expr);
