https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103771
--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
Forcing the pattern to not trigger produces the expected
t.c:8:6: missed: not vectorized: relevant stmt not supported: iftmp.0_21 =
x.1_14 > 255 ? iftmp.0_19 : iftmp.0_20;
since condition vectorization itself doesn't know how to handle this, we end
up at
if (vectype1 && !useless_type_conversion_p (vectype, vectype1))
return false;
with vectype V32QI and vectype1 V8SI.
Splitting out the compare from the COND_EXPR in the pattern but leaving out
the attempt to "widen" it reveals the same fact that vectorizable_condition
doesn't support packing of multiple vector defs for the mask operand.
I think that is what we need to add. We also don't have a good representation
for "packing" of masks.
diff --git a/gcc/tree-vect-patterns.c b/gcc/tree-vect-patterns.c
index 3ea905538e1..729a1d32612 100644
--- a/gcc/tree-vect-patterns.c
+++ b/gcc/tree-vect-patterns.c
@@ -4679,8 +4679,10 @@ vect_recog_mask_conversion_pattern (vec_info *vinfo,
rhs1_type);
}
- if (maybe_ne (TYPE_VECTOR_SUBPARTS (vectype1),
- TYPE_VECTOR_SUBPARTS (vectype2)))
+ /* AVX512 style masks cannot be packed/unpacked. */
+ if (TYPE_PRECISION (TREE_TYPE (vectype2)) != 1
+ && maybe_ne (TYPE_VECTOR_SUBPARTS (vectype1),
+ TYPE_VECTOR_SUBPARTS (vectype2)))
tmp = build_mask_conversion (vinfo, rhs1, vectype1, stmt_vinfo);
else
tmp = rhs1;
