Fold MASK_LOAD/STORE with an all-true mask

This patch folds IFN_MASK_LOAD and IFN_MASK_STOREs to normal accesses
if the mask is all-true.  This can happen for fully-masked loops that
didn't actually need to be (which could be handled by the vectoriser
instead), or for unrolled fully-masked loops whose first iteration is
guaranteed to operate on a full vector.  It's also useful when the
accesses are generated directly by intrinsics (to follow for SVE).

2019-08-05  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
	* gimple-fold.c (gimple_fold_mask_load_store_mem_ref)
	(gimple_fold_mask_load, gimple_fold_mask_store): New functions.
	(gimple_fold_call): Use them to fold IFN_MASK_LOAD and
	IFN_MASK_STORE.

gcc/testsuite/
	* gcc.target/aarch64/sve/mask_load_1.c: New test.

From-SVN: r274118

gcc/ChangeLog

 2019-08-05  Richard Sandiford  <richard.sandiford@arm.com>
 
+	* gimple-fold.c (gimple_fold_mask_load_store_mem_ref)
+	(gimple_fold_mask_load, gimple_fold_mask_store): New functions.
+	(gimple_fold_call): Use them to fold IFN_MASK_LOAD and
+	IFN_MASK_STORE.
+
+2019-08-05  Richard Sandiford  <richard.sandiford@arm.com>
+
 	* gimple.h (gimple_move_vops): Declare.
 	* gimple.c (gimple_move_vops): New function
 	* gimple-fold.c (replace_call_with_call_and_fold)

gcc/gimple-fold.c

@@ -4180,6 +4180,63 @@ arith_overflowed_p (enum tree_code code, const_tree type,
   return wi::min_precision (wres, sign) > TYPE_PRECISION (type);
 }
 
+/* If IFN_MASK_LOAD/STORE call CALL is unconditional, return a MEM_REF
+   for the memory it references, otherwise return null.  VECTYPE is the
+   type of the memory vector.  */
+
+static tree
+gimple_fold_mask_load_store_mem_ref (gcall *call, tree vectype)
+{
+  tree ptr = gimple_call_arg (call, 0);
+  tree alias_align = gimple_call_arg (call, 1);
+  tree mask = gimple_call_arg (call, 2);
+  if (!tree_fits_uhwi_p (alias_align) || !integer_all_onesp (mask))
+    return NULL_TREE;
+
+  unsigned HOST_WIDE_INT align = tree_to_uhwi (alias_align) * BITS_PER_UNIT;
+  if (TYPE_ALIGN (vectype) != align)
+    vectype = build_aligned_type (vectype, align);
+  tree offset = build_zero_cst (TREE_TYPE (alias_align));
+  return fold_build2 (MEM_REF, vectype, ptr, offset);
+}
+
+/* Try to fold IFN_MASK_LOAD call CALL.  Return true on success.  */
+
+static bool
+gimple_fold_mask_load (gimple_stmt_iterator *gsi, gcall *call)
+{
+  tree lhs = gimple_call_lhs (call);
+  if (!lhs)
+    return false;
+
+  if (tree rhs = gimple_fold_mask_load_store_mem_ref (call, TREE_TYPE (lhs)))
+    {
+      gassign *new_stmt = gimple_build_assign (lhs, rhs);
+      gimple_set_location (new_stmt, gimple_location (call));
+      gimple_move_vops (new_stmt, call);
+      gsi_replace (gsi, new_stmt, false);
+      return true;
+    }
+  return false;
+}
+
+/* Try to fold IFN_MASK_STORE call CALL.  Return true on success.  */
+
+static bool
+gimple_fold_mask_store (gimple_stmt_iterator *gsi, gcall *call)
+{
+  tree rhs = gimple_call_arg (call, 3);
+  if (tree lhs = gimple_fold_mask_load_store_mem_ref (call, TREE_TYPE (rhs)))
+    {
+      gassign *new_stmt = gimple_build_assign (lhs, rhs);
+      gimple_set_location (new_stmt, gimple_location (call));
+      gimple_move_vops (new_stmt, call);
+      gsi_replace (gsi, new_stmt, false);
+      return true;
+    }
+  return false;
+}
+
 /* Attempt to fold a call statement referenced by the statement iterator GSI.
    The statement may be replaced by another statement, e.g., if the call
    simplifies to a constant value. Return true if any changes were made.
@@ -4409,6 +4466,12 @@ gimple_fold_call (gimple_stmt_iterator *gsi, bool inplace)
 	  subcode = MULT_EXPR;
 	  cplx_result = true;
 	  break;
+	case IFN_MASK_LOAD:
+	  changed |= gimple_fold_mask_load (gsi, stmt);
+	  break;
+	case IFN_MASK_STORE:
+	  changed |= gimple_fold_mask_store (gsi, stmt);
+	  break;
 	default:
 	  break;
 	}

gcc/testsuite/ChangeLog

+2019-08-05  Richard Sandiford  <richard.sandiford@arm.com>
+
+	* gcc.target/aarch64/sve/mask_load_1.c: New test.
+
 2019-08-05  Jozef Lawrynowicz  <jozef.l@mittosystems.com>
 
 	* gcc.target/msp430/pr80993.c: Add cleanup-saved-temps to final

gcc/testsuite/gcc.target/aarch64/sve/mask_load_1.c

+/* { dg-do compile } */
+/* { dg-options "-O2 -ftree-vectorize -msve-vector-bits=256 -fdump-tree-optimized" } */
+
+void
+f (int *x)
+{
+  for (int i = 0; i < 8; ++i)
+    x[i] += 1;
+}
+
+/* { dg-final { scan-tree-dump { = MEM <vector\(8\) int>} "optimized" } } */
+/* { dg-final { scan-tree-dump { MEM <vector\(8\) int> \[[^]]*\] = } "optimized" } } */