tree-optimization/111115 - SLP of masked stores

The following adds the capability to do SLP on .MASK_STORE, I do not
plan to add interleaving support.

	PR tree-optimization/111115
gcc/
	* tree-vectorizer.h (vect_slp_child_index_for_operand): New.
	* tree-vect-data-refs.cc (can_group_stmts_p): Also group
	.MASK_STORE.
	* tree-vect-slp.cc (arg3_arg2_map): New.
	(vect_get_operand_map): Handle IFN_MASK_STORE.
	(vect_slp_child_index_for_operand): New function.
	(vect_build_slp_tree_1): Handle statements with no LHS,
	masked store ifns.
	(vect_remove_slp_scalar_calls): Likewise.
	* tree-vect-stmts.cc (vect_check_store_rhs): Lookup the
	SLP child corresponding to the ifn value index.
	(vectorizable_store): Likewise for the mask index.  Support
	masked stores.
	(vectorizable_load): Lookup the SLP child corresponding to the
	ifn mask index.

gcc/testsuite/
	* lib/target-supports.exp (check_effective_target_vect_masked_store):
	Supported with check_avx_available.
	* gcc.dg/vect/slp-mask-store-1.c: New testcase.

6 files changed, 94 insertions, 21 deletions

diff --git a/gcc/testsuite/gcc.dg/vect/slp-mask-store-1.c b/gcc/testsuite/gcc.dg/vect/slp-mask-store-1.c
new file mode 100644
index 0000000..50b7066
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/slp-mask-store-1.c
@@ -0,0 +1,39 @@
+/* { dg-do run } */
+/* { dg-additional-options "-mavx2" { target avx2 } } */
+
+#include "tree-vect.h"
+
+void __attribute__((noipa))
+foo (unsigned * __restrict x, int * __restrict flag)
+{
+  for (int i = 0; i < 32; ++i)
+    {
+      if (flag[2*i+0])
+        x[2*i+0] = x[2*i+0] + 3;
+      if (flag[2*i+1])
+        x[2*i+1] = x[2*i+1] + 177;
+    }
+}
+
+unsigned x[16];
+int flag[32] = { 1, 1, 0, 0, 0, 1, 1, 0, 1, 1, 0, 0, 0, 1, 1, 0,
+                 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+unsigned res[16] = { 3, 177, 0, 0, 0, 177, 3, 0, 3, 177, 0, 0, 0, 177, 3, 0 };
+
+int
+main ()
+{
+  check_vect ();
+
+  foo (x, flag);
+
+  if (__builtin_memcmp (x, res, sizeof (x)) != 0)
+    abort ();
+  for (int i = 0; i < 32; ++i)
+    if (flag[i] != 0 && flag[i] != 1)
+      abort ();
+
+  return 0;
+}
+
+/* { dg-final { scan-tree-dump-times "LOOP VECTORIZED" 1 "vect" { target { vect_masked_store && vect_masked_load } } } } */
diff --git a/gcc/testsuite/lib/target-supports.exp b/gcc/testsuite/lib/target-supports.exp
index d4623ee..d353cc0 100644
--- a/gcc/testsuite/lib/target-supports.exp
+++ b/gcc/testsuite/lib/target-supports.exp
@@ -8400,7 +8400,8 @@ proc check_effective_target_vect_masked_load { } {
 # Return 1 if the target supports vector masked stores.
 
 proc check_effective_target_vect_masked_store { } {
-    return [expr { [check_effective_target_aarch64_sve]
+    return [expr { [check_avx_available]
+		   || [check_effective_target_aarch64_sve]
 		   || [istarget amdgcn*-*-*] }]
 }
 
diff --git a/gcc/tree-vect-data-refs.cc b/gcc/tree-vect-data-refs.cc
index 3e9a284..a2caf6c 100644
--- a/gcc/tree-vect-data-refs.cc
+++ b/gcc/tree-vect-data-refs.cc
@@ -3048,8 +3048,7 @@ can_group_stmts_p (stmt_vec_info stmt1_info, stmt_vec_info stmt2_info,
 	 like those created by build_mask_conversion.  */
       tree mask1 = gimple_call_arg (call1, 2);
       tree mask2 = gimple_call_arg (call2, 2);
-      if (!operand_equal_p (mask1, mask2, 0)
-          && (ifn == IFN_MASK_STORE || !allow_slp_p))
+      if (!operand_equal_p (mask1, mask2, 0) && !allow_slp_p)
 	{
 	  mask1 = strip_conversion (mask1);
 	  if (!mask1)
diff --git a/gcc/tree-vect-slp.cc b/gcc/tree-vect-slp.cc
index 0bcc530..ace0ff3 100644
--- a/gcc/tree-vect-slp.cc
+++ b/gcc/tree-vect-slp.cc
@@ -503,6 +503,7 @@ static const int cond_expr_maps[3][5] = {
 static const int arg1_map[] = { 1, 1 };
 static const int arg2_map[] = { 1, 2 };
 static const int arg1_arg4_map[] = { 2, 1, 4 };
+static const int arg3_arg2_map[] = { 2, 3, 2 };
 static const int op1_op0_map[] = { 2, 1, 0 };
 
 /* For most SLP statements, there is a one-to-one mapping between
@@ -543,6 +544,9 @@ vect_get_operand_map (const gimple *stmt, unsigned char swap = 0)
 	  case IFN_MASK_GATHER_LOAD:
 	    return arg1_arg4_map;
 
+	  case IFN_MASK_STORE:
+	    return arg3_arg2_map;
+
 	  default:
 	    break;
 	  }
@@ -550,6 +554,20 @@ vect_get_operand_map (const gimple *stmt, unsigned char swap = 0)
   return nullptr;
 }
 
+/* Return the SLP node child index for operand OP of STMT.  */
+
+int
+vect_slp_child_index_for_operand (const gimple *stmt, int op)
+{
+  const int *opmap = vect_get_operand_map (stmt);
+  if (!opmap)
+    return op;
+  for (int i = 1; i < 1 + opmap[0]; ++i)
+    if (opmap[i] == op)
+      return i - 1;
+  gcc_unreachable ();
+}
+
 /* Get the defs for the rhs of STMT (collect them in OPRNDS_INFO), check that
    they are of a valid type and that they match the defs of the first stmt of
    the SLP group (stored in OPRNDS_INFO).  This function tries to match stmts
@@ -1003,8 +1021,12 @@ vect_build_slp_tree_1 (vec_info *vinfo, unsigned char *swap,
           return false;
         }
 
+      gcall *call_stmt = dyn_cast <gcall *> (stmt);
       lhs = gimple_get_lhs (stmt);
-      if (lhs == NULL_TREE)
+      if (lhs == NULL_TREE
+	  && (!call_stmt
+	      || !gimple_call_internal_p (stmt)
+	      || !internal_store_fn_p (gimple_call_internal_fn (stmt))))
 	{
 	  if (dump_enabled_p ())
 	    dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
@@ -1041,7 +1063,6 @@ vect_build_slp_tree_1 (vec_info *vinfo, unsigned char *swap,
 
       gcc_assert (vectype);
 
-      gcall *call_stmt = dyn_cast <gcall *> (stmt);
       if (call_stmt)
 	{
 	  combined_fn cfn = gimple_call_combined_fn (call_stmt);
@@ -1054,6 +1075,8 @@ vect_build_slp_tree_1 (vec_info *vinfo, unsigned char *swap,
 	      || cfn == CFN_GATHER_LOAD
 	      || cfn == CFN_MASK_GATHER_LOAD)
 	    load_p = true;
+	  else if (cfn == CFN_MASK_STORE)
+	    rhs_code = CFN_MASK_STORE;
 	  else if ((internal_fn_p (cfn)
 		    && !vectorizable_internal_fn_p (as_internal_fn (cfn)))
 		   || gimple_call_tail_p (call_stmt)
@@ -1212,7 +1235,9 @@ vect_build_slp_tree_1 (vec_info *vinfo, unsigned char *swap,
 	      continue;
 	    }
 
-	  if (call_stmt && first_stmt_code != CFN_MASK_LOAD)
+	  if (call_stmt
+	      && first_stmt_code != CFN_MASK_LOAD
+	      && first_stmt_code != CFN_MASK_STORE)
 	    {
 	      if (!compatible_calls_p (as_a <gcall *> (stmts[0]->stmt),
 				       call_stmt))
@@ -1266,9 +1291,11 @@ vect_build_slp_tree_1 (vec_info *vinfo, unsigned char *swap,
       /* Grouped store or load.  */
       if (STMT_VINFO_GROUPED_ACCESS (stmt_info))
 	{
-	  if (REFERENCE_CLASS_P (lhs))
+	  if (!load_p)
 	    {
 	      /* Store.  */
+	      gcc_assert (rhs_code == CFN_MASK_STORE
+			  || REFERENCE_CLASS_P (lhs));
 	      ;
 	    }
 	  else
@@ -9095,10 +9122,17 @@ vect_remove_slp_scalar_calls (vec_info *vinfo,
 	  || !PURE_SLP_STMT (stmt_info))
 	continue;
       lhs = gimple_call_lhs (stmt);
-      new_stmt = gimple_build_assign (lhs, build_zero_cst (TREE_TYPE (lhs)));
+      if (lhs)
+	new_stmt = gimple_build_assign (lhs, build_zero_cst (TREE_TYPE (lhs)));
+      else
+	{
+	  new_stmt = gimple_build_nop ();
+	  unlink_stmt_vdef (stmt_info->stmt);
+	}
       gsi = gsi_for_stmt (stmt);
       vinfo->replace_stmt (&gsi, stmt_info, new_stmt);
-      SSA_NAME_DEF_STMT (gimple_assign_lhs (new_stmt)) = new_stmt;
+      if (lhs)
+	SSA_NAME_DEF_STMT (lhs) = new_stmt;
     }
 }
 
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index be9f3a2..cd7c109 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -2629,12 +2629,14 @@ vect_check_store_rhs (vec_info *vinfo, stmt_vec_info stmt_info,
       return false;
     }
 
-  unsigned op_no = 0;
+  int op_no = 0;
   if (gcall *call = dyn_cast <gcall *> (stmt_info->stmt))
     {
       if (gimple_call_internal_p (call)
 	  && internal_store_fn_p (gimple_call_internal_fn (call)))
 	op_no = internal_fn_stored_value_index (gimple_call_internal_fn (call));
+      if (slp_node)
+	op_no = vect_slp_child_index_for_operand (call, op_no);
     }
 
   enum vect_def_type rhs_dt;
@@ -8244,15 +8246,9 @@ vectorizable_store (vec_info *vinfo,
       if (!internal_store_fn_p (ifn))
 	return false;
 
-      if (slp_node != NULL)
-	{
-	  if (dump_enabled_p ())
-	    dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
-			     "SLP of masked stores not supported.\n");
-	  return false;
-	}
-
       int mask_index = internal_fn_mask_index (ifn);
+      if (mask_index >= 0 && slp_node)
+	mask_index = vect_slp_child_index_for_operand (call, mask_index);
       if (mask_index >= 0
 	  && !vect_check_scalar_mask (vinfo, stmt_info, slp_node, mask_index,
 				      &mask, NULL, &mask_dt, &mask_vectype))
@@ -9093,8 +9089,10 @@ vectorizable_store (vec_info *vinfo,
 	    {
 	      /* Get vectorized arguments for SLP_NODE.  */
 	      vect_get_vec_defs (vinfo, stmt_info, slp_node, 1, op,
-				 &vec_oprnds);
+				 &vec_oprnds, mask, &vec_masks);
 	      vec_oprnd = vec_oprnds[0];
+	      if (mask)
+		vec_mask = vec_masks[0];
 	    }
 	  else
 	    {
@@ -9191,6 +9189,8 @@ vectorizable_store (vec_info *vinfo,
 	    final_mask = vect_get_loop_mask (loop_vinfo, gsi, loop_masks,
 					     vec_num * ncopies, vectype,
 					     vec_num * j + i);
+	  if (slp && vec_mask)
+	    vec_mask = vec_masks[i];
 	  if (vec_mask)
 	    final_mask = prepare_vec_mask (loop_vinfo, mask_vectype, final_mask,
 					   vec_mask, gsi);
@@ -9575,9 +9575,8 @@ vectorizable_load (vec_info *vinfo,
 	return false;
 
       mask_index = internal_fn_mask_index (ifn);
-      /* ??? For SLP the mask operand is always last.  */
       if (mask_index >= 0 && slp_node)
-	mask_index = SLP_TREE_CHILDREN (slp_node).length () - 1;
+	mask_index = vect_slp_child_index_for_operand (call, mask_index);
       if (mask_index >= 0
 	  && !vect_check_scalar_mask (vinfo, stmt_info, slp_node, mask_index,
 				      &mask, NULL, &mask_dt, &mask_vectype))
diff --git a/gcc/tree-vectorizer.h b/gcc/tree-vectorizer.h
index 53a3d78..f1d0cd7 100644
--- a/gcc/tree-vectorizer.h
+++ b/gcc/tree-vectorizer.h
@@ -2429,6 +2429,7 @@ extern int vect_get_place_in_interleaving_chain (stmt_vec_info, stmt_vec_info);
 extern slp_tree vect_create_new_slp_node (unsigned, tree_code);
 extern void vect_free_slp_tree (slp_tree);
 extern bool compatible_calls_p (gcall *, gcall *);
+extern int vect_slp_child_index_for_operand (const gimple *, int op);
 
 /* In tree-vect-patterns.cc.  */
 extern void