Moved existing code to determine the partial vector
style for a load or store into a new function
that will be reused for BB SLP with predicated tails.
---
gcc/tree-vect-stmts.cc | 71 ++++++++++++++++++++++++++++++++++--------
1 file changed, 58 insertions(+), 13 deletions(-)
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index 5ec65b2b2de..f9905652295 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -1489,6 +1489,56 @@ vect_get_len (vec_info *vinfo, slp_tree slp_node,
gimple_stmt_iterator *gsi,
static tree permute_vec_elements (vec_info *, tree, tree, tree, stmt_vec_info,
gimple_stmt_iterator *);
+/* If target supports vector load/store (determined by IS_LOAD) with length for
+ VECTYPE, return VECT_PARTIAL_VECTORS_LEN; otherwise, if it supports vector
+ masked load/store for mode for VECTYPE, return
+ VECT_PARTIAL_VECTORS_WHILE_ULT; otherwise, return VECT_PARTIAL_VECTORS_NONE.
+ If load/store with length or mask is supported and ELSVALS is nonzero, store
+ the possible else values in the vector it points to. Unless FACTOR is null,
+ *FACTOR is set to the number of separate subelements that each element of
+ VECTYPE should be split into. (This is always 1 except in the case of
+ VECT_PARTIAL_VECTORS_LEN when the machine mode of VECTYPE can only be
+ supported by using a length in bytes instead of a length in lanes.)
+ */
+static enum vect_partial_vector_style
+vect_get_partial_vector_style (tree vectype, bool is_load,
+ unsigned *factor = nullptr,
+ vec<int> *elsvals = nullptr)
+{
+ machine_mode vecmode = TYPE_MODE (vectype);
+
+ /* TYPE_MODE tries to map vector modes that are unsupported by the target to
+ same-sized scalar modes, so reject scalar modes here. */
+ if (VECTOR_MODE_P (vecmode))
+ {
+ machine_mode mask_mode, vmode;
+ if (get_len_load_store_mode (vecmode, is_load, nullptr, elsvals)
+ .exists (&vmode))
+ {
+ if (factor)
+ *factor = (vecmode == vmode) ? 1 : GET_MODE_UNIT_SIZE (vecmode);
+ return vect_partial_vectors_len;
+ }
+
+ if (factor)
+ *factor = 1;
+
+ if (targetm.vectorize.get_mask_mode (vecmode).exists (&mask_mode)
+ && can_vec_mask_load_store_p (vecmode, mask_mode, is_load, NULL,
+ elsvals))
+ {
+ return vect_partial_vectors_while_ult;
+ }
+ }
+
+ if (dump_enabled_p ())
+ dump_printf_loc (MSG_NOTE, vect_location,
+ "the target doesn't have the appropriate partial"
+ " vectorization %s for vector data type %T.\n",
+ is_load ? "load" : "store", vectype);
+ return vect_partial_vectors_none;
+}
+
/* Check whether a load or store statement in the loop or SLP group described
by
VINFO is possible using partial vectors. This is
testing whether the vectorizer pass has the appropriate support,
@@ -1642,28 +1692,23 @@ check_load_store_for_partial_vectors (vec_info *vinfo,
tree vectype,
poly_uint64 nunits = TYPE_VECTOR_SUBPARTS (vectype);
poly_uint64 vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo);
- machine_mode mask_mode;
- machine_mode vmode;
- bool using_partial_vectors_p = false;
- if (get_len_load_store_mode
- (vecmode, is_load, nullptr, elsvals).exists (&vmode))
+ unsigned factor;
+ vect_partial_vector_style partial_vector_style
+ = vect_get_partial_vector_style (vectype, is_load, &factor, elsvals);
+
+ if (partial_vector_style == vect_partial_vectors_len)
{
nvectors = group_memory_nvectors (group_size * vf, nunits);
- unsigned factor = (vecmode == vmode) ? 1 : GET_MODE_UNIT_SIZE (vecmode);
vect_record_len (vinfo, slp_node, nvectors, vectype, factor);
- using_partial_vectors_p = true;
}
- else if (targetm.vectorize.get_mask_mode (vecmode).exists (&mask_mode)
- && can_vec_mask_load_store_p (vecmode, mask_mode, is_load, NULL,
- elsvals))
+ else if (partial_vector_style == vect_partial_vectors_while_ult)
{
nvectors = group_memory_nvectors (group_size * vf, nunits);
vect_record_mask (vinfo, slp_node, nvectors, vectype, scalar_mask);
- using_partial_vectors_p = true;
}
-
- if (!using_partial_vectors_p)
+ else
{
+ gcc_assert (partial_vector_style == vect_partial_vectors_none);
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"can't operate on partial vectors because the"
--
2.43.0
