remove STMT_VINFO_NUM_SLP_USES

This removes STMT_VINFO_NUM_SLP_USES by pushing the setting of
the shared stmt_vec_info vector type to where we actually need it
which is alignment analysis and vectorizable_* analysis (where
we could eventually elide it for non-load/store operations).

In particular "uses" in the cache and in disqualified SLP
subgraphs should no longer provide conflicting vector types
this way.

2020-09-16  Richard Biener  <rguenther@suse.de>

	* tree-vectorizer.h (_stmt_vec_info::num_slp_uses): Remove.
	(STMT_VINFO_NUM_SLP_USES): Likewise.
	(vect_free_slp_instance): Adjust.
	(vect_update_shared_vectype): Declare.
	* tree-vectorizer.c (vec_info::~vec_info): Adjust.
	* tree-vect-loop.c (vect_analyze_loop_2): Likewise.
	(vectorizable_live_operation): Use vector type from
	SLP_TREE_REPRESENTATIVE.
	(vect_transform_loop): Adjust.
	* tree-vect-data-refs.c (vect_slp_analyze_node_alignment):
	Set the shared vector type.
	* tree-vect-slp.c (vect_free_slp_tree): Remove final_p
	parameter, remove STMT_VINFO_NUM_SLP_USES updating.
	(vect_free_slp_instance): Adjust.
	(vect_create_new_slp_node): Remove STMT_VINFO_NUM_SLP_USES
	updating.
	(vect_update_shared_vectype): Always compare with the
	present vector type, update if NULL.
	(vect_build_slp_tree_1): Do not update the shared vector
	type here.
	(vect_build_slp_tree_2): Adjust.
	(slp_copy_subtree): Likewise.
	(vect_attempt_slp_rearrange_stmts): Likewise.
	(vect_analyze_slp_instance): Likewise.
	(vect_analyze_slp): Likewise.
	(vect_slp_analyze_node_operations_1): Update the shared
	vector type.
	(vect_slp_analyze_operations): Adjust.
	(vect_slp_analyze_bb_1): Likewise.

5 files changed, 41 insertions, 67 deletions

diff --git a/gcc/tree-vect-data-refs.c b/gcc/tree-vect-data-refs.c
index 88c5fae..5bf93e2 100644
--- a/gcc/tree-vect-data-refs.c
+++ b/gcc/tree-vect-data-refs.c
@@ -2380,6 +2380,11 @@ vect_slp_analyze_node_alignment (vec_info *vinfo, slp_tree node)
   if (SLP_TREE_LOAD_PERMUTATION (node).exists ())
     first_stmt_info = DR_GROUP_FIRST_ELEMENT (first_stmt_info);
 
+  /* We need to commit to a vector type for the group now.  */
+  if (is_a <bb_vec_info> (vinfo)
+      && !vect_update_shared_vectype (first_stmt_info, SLP_TREE_VECTYPE (node)))
+    return false;
+
   dr_vec_info *dr_info = STMT_VINFO_DR_INFO (first_stmt_info);
   vect_compute_data_ref_alignment (vinfo, dr_info);
   /* In several places we need alignment of the first element anyway.  */
diff --git a/gcc/tree-vect-loop.c b/gcc/tree-vect-loop.c
index c95ec5a..3af4cf7 100644
--- a/gcc/tree-vect-loop.c
+++ b/gcc/tree-vect-loop.c
@@ -2467,7 +2467,7 @@ again:
   LOOP_VINFO_VECT_FACTOR (loop_vinfo) = saved_vectorization_factor;
   /* Free the SLP instances.  */
   FOR_EACH_VEC_ELT (LOOP_VINFO_SLP_INSTANCES (loop_vinfo), j, instance)
-    vect_free_slp_instance (instance, false);
+    vect_free_slp_instance (instance);
   LOOP_VINFO_SLP_INSTANCES (loop_vinfo).release ();
   /* Reset SLP type to loop_vect on all stmts.  */
   for (i = 0; i < LOOP_VINFO_LOOP (loop_vinfo)->num_nodes; ++i)
@@ -8031,7 +8031,9 @@ vectorizable_live_operation (vec_info *vinfo,
   loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo);
   imm_use_iterator imm_iter;
   tree lhs, lhs_type, bitsize, vec_bitsize;
-  tree vectype = STMT_VINFO_VECTYPE (stmt_info);
+  tree vectype = (slp_node
+		  ? SLP_TREE_VECTYPE (SLP_TREE_REPRESENTATIVE (slp_node))
+		  : STMT_VINFO_VECTYPE (stmt_info));
   poly_uint64 nunits = TYPE_VECTOR_SUBPARTS (vectype);
   int ncopies;
   gimple *use_stmt;
@@ -9260,7 +9262,7 @@ vect_transform_loop (loop_vec_info loop_vinfo, gimple *loop_vectorized_call)
      won't work.  */
   slp_instance instance;
   FOR_EACH_VEC_ELT (LOOP_VINFO_SLP_INSTANCES (loop_vinfo), i, instance)
-    vect_free_slp_instance (instance, true);
+    vect_free_slp_instance (instance);
   LOOP_VINFO_SLP_INSTANCES (loop_vinfo).release ();
   /* Clear-up safelen field since its value is invalid after vectorization
      since vectorized loop can have loop-carried dependencies.  */
diff --git a/gcc/tree-vect-slp.c b/gcc/tree-vect-slp.c
index f9f4d70..ecce7a9 100644
--- a/gcc/tree-vect-slp.c
+++ b/gcc/tree-vect-slp.c
@@ -84,12 +84,10 @@ _slp_tree::~_slp_tree ()
   SLP_TREE_LANE_PERMUTATION (this).release ();
 }
 
-/* Recursively free the memory allocated for the SLP tree rooted at NODE.
-   FINAL_P is true if we have vectorized the instance or if we have
-   made a final decision not to vectorize the statements in any way.  */
+/* Recursively free the memory allocated for the SLP tree rooted at NODE.  */
 
 static void
-vect_free_slp_tree (slp_tree node, bool final_p)
+vect_free_slp_tree (slp_tree node)
 {
   int i;
   slp_tree child;
@@ -98,21 +96,7 @@ vect_free_slp_tree (slp_tree node, bool final_p)
     return;
 
   FOR_EACH_VEC_ELT (SLP_TREE_CHILDREN (node), i, child)
-    vect_free_slp_tree (child, final_p);
-
-  /* Don't update STMT_VINFO_NUM_SLP_USES if it isn't relevant.
-     Some statements might no longer exist, after having been
-     removed by vect_transform_stmt.  Updating the remaining
-     statements would be redundant.  */
-  if (!final_p)
-    {
-      stmt_vec_info stmt_info;
-      FOR_EACH_VEC_ELT (SLP_TREE_SCALAR_STMTS (node), i, stmt_info)
-	{
-	  gcc_assert (STMT_VINFO_NUM_SLP_USES (stmt_info) > 0);
-	  STMT_VINFO_NUM_SLP_USES (stmt_info)--;
-	}
-    }
+    vect_free_slp_tree (child);
 
   delete node;
 }
@@ -129,14 +113,12 @@ _slp_instance::location () const
 }
 
 
-/* Free the memory allocated for the SLP instance.  FINAL_P is true if we
-   have vectorized the instance or if we have made a final decision not
-   to vectorize the statements in any way.  */
+/* Free the memory allocated for the SLP instance.  */
 
 void
-vect_free_slp_instance (slp_instance instance, bool final_p)
+vect_free_slp_instance (slp_instance instance)
 {
-  vect_free_slp_tree (SLP_INSTANCE_TREE (instance), final_p);
+  vect_free_slp_tree (SLP_INSTANCE_TREE (instance));
   SLP_INSTANCE_LOADS (instance).release ();
   instance->subgraph_entries.release ();
   instance->cost_vec.release ();
@@ -155,12 +137,6 @@ vect_create_new_slp_node (vec<stmt_vec_info> scalar_stmts, unsigned nops)
   SLP_TREE_DEF_TYPE (node) = vect_internal_def;
   SLP_TREE_REPRESENTATIVE (node) = scalar_stmts[0];
   SLP_TREE_LANES (node) = scalar_stmts.length ();
-
-  unsigned i;
-  stmt_vec_info stmt_info;
-  FOR_EACH_VEC_ELT (scalar_stmts, i, stmt_info)
-    STMT_VINFO_NUM_SLP_USES (stmt_info)++;
-
   return node;
 }
 
@@ -614,23 +590,23 @@ again:
    Return true if we can, meaning that this choice doesn't conflict with
    existing SLP nodes that use STMT_INFO.  */
 
-static bool
+bool
 vect_update_shared_vectype (stmt_vec_info stmt_info, tree vectype)
 {
   tree old_vectype = STMT_VINFO_VECTYPE (stmt_info);
-  if (old_vectype && useless_type_conversion_p (vectype, old_vectype))
-    return true;
+  if (old_vectype)
+    return useless_type_conversion_p (vectype, old_vectype);
 
-  if (STMT_VINFO_GROUPED_ACCESS (stmt_info)
-      && DR_IS_READ (STMT_VINFO_DATA_REF (stmt_info)))
+  if (STMT_VINFO_GROUPED_ACCESS (stmt_info))
     {
       /* We maintain the invariant that if any statement in the group is
 	 used, all other members of the group have the same vector type.  */
       stmt_vec_info first_info = DR_GROUP_FIRST_ELEMENT (stmt_info);
       stmt_vec_info member_info = first_info;
       for (; member_info; member_info = DR_GROUP_NEXT_ELEMENT (member_info))
-	if (STMT_VINFO_NUM_SLP_USES (member_info) > 0
-	    || is_pattern_stmt_p (member_info))
+	if (is_pattern_stmt_p (member_info)
+	    && !useless_type_conversion_p (vectype,
+					   STMT_VINFO_VECTYPE (member_info)))
 	  break;
 
       if (!member_info)
@@ -641,8 +617,7 @@ vect_update_shared_vectype (stmt_vec_info stmt_info, tree vectype)
 	  return true;
 	}
     }
-  else if (STMT_VINFO_NUM_SLP_USES (stmt_info) == 0
-	   && !is_pattern_stmt_p (stmt_info))
+  else if (!is_pattern_stmt_p (stmt_info))
     {
       STMT_VINFO_VECTYPE (stmt_info) = vectype;
       return true;
@@ -824,10 +799,6 @@ vect_build_slp_tree_1 (vec_info *vinfo, unsigned char *swap,
 
       gcc_assert (vectype);
 
-      if (is_a <bb_vec_info> (vinfo)
-	  && !vect_update_shared_vectype (stmt_info, vectype))
-	continue;
-
       gcall *call_stmt = dyn_cast <gcall *> (stmt);
       if (call_stmt)
 	{
@@ -1561,7 +1532,7 @@ fail:
 
       gcc_assert (child == NULL);
       FOR_EACH_VEC_ELT (children, j, child)
-	vect_free_slp_tree (child, false);
+	vect_free_slp_tree (child);
       vect_free_oprnd_info (oprnds_info);
       return NULL;
     }
@@ -1589,7 +1560,7 @@ fail:
 	  /* Roll back.  */
 	  matches[0] = false;
 	  FOR_EACH_VEC_ELT (children, j, child)
-	    vect_free_slp_tree (child, false);
+	    vect_free_slp_tree (child);
 
 	  if (dump_enabled_p ())
 	    dump_printf_loc (MSG_NOTE, vect_location,
@@ -1833,12 +1804,7 @@ slp_copy_subtree (slp_tree node, hash_map<slp_tree, slp_tree> &map)
   copy->max_nunits = node->max_nunits;
   copy->refcnt = 0;
   if (SLP_TREE_SCALAR_STMTS (node).exists ())
-    {
-      SLP_TREE_SCALAR_STMTS (copy) = SLP_TREE_SCALAR_STMTS (node).copy ();
-      stmt_vec_info stmt_info;
-      FOR_EACH_VEC_ELT (SLP_TREE_SCALAR_STMTS (node), i, stmt_info)
-	STMT_VINFO_NUM_SLP_USES (stmt_info)++;
-    }
+    SLP_TREE_SCALAR_STMTS (copy) = SLP_TREE_SCALAR_STMTS (node).copy ();
   if (SLP_TREE_SCALAR_OPS (node).exists ())
     SLP_TREE_SCALAR_OPS (copy) = SLP_TREE_SCALAR_OPS (node).copy ();
   if (SLP_TREE_LOAD_PERMUTATION (node).exists ())
@@ -1968,7 +1934,7 @@ vect_attempt_slp_rearrange_stmts (slp_instance slp_instn)
   /* We have to unshare the SLP tree we modify.  */
   hash_map<slp_tree, slp_tree> map;
   slp_tree unshared = slp_copy_subtree (SLP_INSTANCE_TREE (slp_instn), map);
-  vect_free_slp_tree (SLP_INSTANCE_TREE (slp_instn), false);
+  vect_free_slp_tree (SLP_INSTANCE_TREE (slp_instn));
   unshared->refcnt++;
   SLP_INSTANCE_TREE (slp_instn) = unshared;
   FOR_EACH_VEC_ELT (SLP_INSTANCE_LOADS (slp_instn), i, node)
@@ -2280,7 +2246,7 @@ vect_analyze_slp_instance (vec_info *vinfo,
 				 "Build SLP failed: store group "
 				 "size not a multiple of the vector size "
 				 "in basic block SLP\n");
-	      vect_free_slp_tree (node, false);
+	      vect_free_slp_tree (node);
 	      return false;
 	    }
 	  /* Fatal mismatch.  */
@@ -2290,7 +2256,7 @@ vect_analyze_slp_instance (vec_info *vinfo,
 			     "splitting\n");
 	  matches[0] = true;
 	  matches[group_size / const_max_nunits * const_max_nunits] = false;
-	  vect_free_slp_tree (node, false);
+	  vect_free_slp_tree (node);
 	}
       else
 	{
@@ -2351,7 +2317,7 @@ vect_analyze_slp_instance (vec_info *vinfo,
 		    dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
 				     "Built SLP cancelled: can use "
 				     "load/store-lanes\n");
-		  vect_free_slp_instance (new_instance, false);
+		  vect_free_slp_instance (new_instance);
 		  return false;
 		}
 	    }
@@ -2517,7 +2483,7 @@ vect_analyze_slp (vec_info *vinfo, unsigned max_tree_size)
   for (scalar_stmts_to_slp_tree_map_t::iterator it = bst_map->begin ();
        it != bst_map->end (); ++it)
     if ((*it).second)
-      vect_free_slp_tree ((*it).second, false);
+      vect_free_slp_tree ((*it).second);
   delete bst_map;
 
   /* Optimize permutations in SLP reductions.  */
@@ -2839,6 +2805,10 @@ vect_slp_analyze_node_operations_1 (vec_info *vinfo, slp_tree node,
   if (SLP_TREE_CODE (node) == VEC_PERM_EXPR)
     return vectorizable_slp_permutation (vinfo, NULL, node, cost_vec);
 
+  if (is_a <bb_vec_info> (vinfo)
+      && !vect_update_shared_vectype (stmt_info, SLP_TREE_VECTYPE (node)))
+    return false;
+
   bool dummy;
   return vect_analyze_stmt (vinfo, stmt_info, &dummy,
 			    node, node_instance, cost_vec);
@@ -3164,7 +3134,7 @@ vect_slp_analyze_operations (vec_info *vinfo)
 	    dump_printf_loc (MSG_NOTE, vect_location,
 			     "removing SLP instance operations starting from: %G",
 			     stmt_info->stmt);
-	  vect_free_slp_instance (instance, false);
+	  vect_free_slp_instance (instance);
           vinfo->slp_instances.ordered_remove (i);
 	  cost_vec.release ();
 	}
@@ -3590,7 +3560,7 @@ vect_slp_analyze_bb_1 (bb_vec_info bb_vinfo, int n_stmts, bool &fatal,
 	    dump_printf_loc (MSG_NOTE, vect_location,
 			     "removing SLP instance operations starting from: %G",
 			     stmt_info->stmt);
-	  vect_free_slp_instance (instance, false);
+	  vect_free_slp_instance (instance);
 	  BB_VINFO_SLP_INSTANCES (bb_vinfo).ordered_remove (i);
 	  continue;
 	}
diff --git a/gcc/tree-vectorizer.c b/gcc/tree-vectorizer.c
index bbe2de5..37e36f5 100644
--- a/gcc/tree-vectorizer.c
+++ b/gcc/tree-vectorizer.c
@@ -473,7 +473,7 @@ vec_info::~vec_info ()
   unsigned int i;
 
   FOR_EACH_VEC_ELT (slp_instances, i, instance)
-    vect_free_slp_instance (instance, true);
+    vect_free_slp_instance (instance);
 
   destroy_cost_data (target_cost_data);
   free_stmt_vec_infos ();
diff --git a/gcc/tree-vectorizer.h b/gcc/tree-vectorizer.h
index 6c29ee6..9dffc55 100644
--- a/gcc/tree-vectorizer.h
+++ b/gcc/tree-vectorizer.h
@@ -1223,9 +1223,6 @@ public:
   /* Whether on this stmt reduction meta is recorded.  */
   bool is_reduc_info;
 
-  /* The number of scalar stmt references from active SLP instances.  */
-  unsigned int num_slp_uses;
-
   /* If nonzero, the lhs of the statement could be truncated to this
      many bits without affecting any users of the result.  */
   unsigned int min_output_precision;
@@ -1340,7 +1337,6 @@ struct gather_scatter_info {
 #define STMT_VINFO_LOOP_PHI_EVOLUTION_BASE_UNCHANGED(S) (S)->loop_phi_evolution_base_unchanged
 #define STMT_VINFO_LOOP_PHI_EVOLUTION_PART(S) (S)->loop_phi_evolution_part
 #define STMT_VINFO_MIN_NEG_DIST(S)	(S)->min_neg_dist
-#define STMT_VINFO_NUM_SLP_USES(S)	(S)->num_slp_uses
 #define STMT_VINFO_REDUC_TYPE(S)	(S)->reduc_type
 #define STMT_VINFO_REDUC_CODE(S)	(S)->reduc_code
 #define STMT_VINFO_REDUC_FN(S)		(S)->reduc_fn
@@ -2024,7 +2020,7 @@ extern int vect_get_known_peeling_cost (loop_vec_info, int, int *,
 extern tree cse_and_gimplify_to_preheader (loop_vec_info, tree);
 
 /* In tree-vect-slp.c.  */
-extern void vect_free_slp_instance (slp_instance, bool);
+extern void vect_free_slp_instance (slp_instance);
 extern bool vect_transform_slp_perm_load (vec_info *, slp_tree, vec<tree>,
 					  gimple_stmt_iterator *, poly_uint64,
 					  bool, unsigned *);
@@ -2047,6 +2043,7 @@ extern bool can_duplicate_and_interleave_p (vec_info *, unsigned int, tree,
 extern void duplicate_and_interleave (vec_info *, gimple_seq *, tree,
 				      vec<tree>, unsigned int, vec<tree> &);
 extern int vect_get_place_in_interleaving_chain (stmt_vec_info, stmt_vec_info);
+extern bool vect_update_shared_vectype (stmt_vec_info, tree);
 
 /* In tree-vect-patterns.c.  */
 /* Pattern recognition functions.