re PR tree-optimization/56918 (incorrect auto-vectorization of array initialization)

	PR tree-optimization/56918
	PR tree-optimization/56920
	* fold-const.c (int_const_binop_1): Use op1.mul_with_sign (op2, ...)
	instead of op1 - op2.  Pass 2 * TYPE_PRECISION (type) as second
	argument to rshift method.  For 2 * HOST_BITS_PER_WIDE_INT precision
	use wide_mul_with_sign method.

	* gcc.dg/vect/pr56918.c: New test.
	* gcc.dg/vect/pr56920.c: New test.

From-SVN: r197846

5 files changed, 83 insertions, 5 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index afcdb10..aa33272 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,12 @@
+2013-04-12  Jakub Jelinek  <jakub@redhat.com>
+
+	PR tree-optimization/56918
+	PR tree-optimization/56920
+	* fold-const.c (int_const_binop_1): Use op1.mul_with_sign (op2, ...)
+	instead of op1 - op2.  Pass 2 * TYPE_PRECISION (type) as second
+	argument to rshift method.  For 2 * HOST_BITS_PER_WIDE_INT precision
+	use wide_mul_with_sign method.
+
 2013-04-12  Richard Biener  <rguenther@suse.de>
 
 	* gimple.c (is_gimple_constant): Vector CONSTRUCTORs should
diff --git a/gcc/fold-const.c b/gcc/fold-const.c
index e8187cb..59dbc03 100644
--- a/gcc/fold-const.c
+++ b/gcc/fold-const.c
@@ -984,12 +984,22 @@ int_const_binop_1 (enum tree_code code, const_tree arg1, const_tree arg2,
       break;
 
     case MULT_HIGHPART_EXPR:
-      /* ??? Need quad precision, or an additional shift operand
-	 to the multiply primitive, to handle very large highparts.  */
       if (TYPE_PRECISION (type) > HOST_BITS_PER_WIDE_INT)
-	return NULL_TREE;
-      tmp = op1 - op2;
-      res = tmp.rshift (TYPE_PRECISION (type), TYPE_PRECISION (type), !uns);
+	{
+	  bool dummy_overflow;
+	  if (TYPE_PRECISION (type) != 2 * HOST_BITS_PER_WIDE_INT)
+	    return NULL_TREE;
+	  op1.wide_mul_with_sign (op2, uns, &res, &dummy_overflow);
+	}
+      else
+	{
+	  bool dummy_overflow;
+	  /* MULT_HIGHPART_EXPR can't ever oveflow, as the multiplication
+	     is performed in twice the precision of arguments.  */
+	  tmp = op1.mul_with_sign (op2, false, &dummy_overflow);
+	  res = tmp.rshift (TYPE_PRECISION (type),
+			    2 * TYPE_PRECISION (type), !uns);
+	}
       break;
 
     case TRUNC_DIV_EXPR:
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index bbf27e6..2c8c878 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,10 @@
+2013-04-12  Jakub Jelinek  <jakub@redhat.com>
+
+	PR tree-optimization/56918
+	PR tree-optimization/56920
+	* gcc.dg/vect/pr56918.c: New test.
+	* gcc.dg/vect/pr56920.c: New test.
+
 2013-04-12  Tobias Burnus  <burnus@net-b.de>
 
 	PR fortran/56845
diff --git a/gcc/testsuite/gcc.dg/vect/pr56918.c b/gcc/testsuite/gcc.dg/vect/pr56918.c
new file mode 100644
index 0000000..581faa0
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/pr56918.c
@@ -0,0 +1,31 @@
+/* PR tree-optimization/56918 */
+/* { dg-additional-options "-O3" } */
+
+#include "tree-vect.h"
+
+extern void abort (void);
+double data[8];
+
+__attribute__((noinline, noclone)) void
+foo ()
+{
+  int i;
+  for (i = 0; i < 8; ++i)
+    data[i] = ((i + 2) % 3) + 1;
+}
+
+int
+main ()
+{
+  int i;
+  check_vect ();
+  foo ();
+  if (data[0] != 3 || data[7] != 1)
+    abort ();
+  for (i = 1; i < 4; ++i)
+    if (data[i] != i || data[i + 3] != i)
+      abort ();
+  return 0;
+}
+
+/* { dg-final { cleanup-tree-dump "vect" } } */
diff --git a/gcc/testsuite/gcc.dg/vect/pr56920.c b/gcc/testsuite/gcc.dg/vect/pr56920.c
new file mode 100644
index 0000000..c6c7cca
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/pr56920.c
@@ -0,0 +1,21 @@
+/* PR tree-optimization/56920 */
+/* { dg-additional-options "-O3" } */
+
+#include "tree-vect.h"
+
+extern void abort (void);
+
+int
+main ()
+{
+  unsigned int a[15], i;
+  check_vect ();
+  for (i = 0; i < 15; ++i)
+    a[i] = (i * 2) % 15;
+  for (i = 0; i < 15; ++i)
+    if (a[i] != (i * 2) % 15)
+      abort ();
+  return 0;
+}
+
+/* { dg-final { cleanup-tree-dump "vect" } } */