re PR tree-optimization/69907 (wrong code at -O3 on x86_64-linux-gnu)

2016-02-24  Richard Biener  <rguenther@suse.de>

	PR tree-optimization/69907
	* tree-vect-stmts.c (vectorizable_load): Check for gaps at the
	end of permutations for BB vectorization.

	* gcc.dg/vect/bb-slp-pr69907.c: New testcase.
	* gcc.dg/vect/bb-slp-34.c: XFAIL.
	* gcc.dg/vect/bb-slp-pr68892.c: Likewise.

From-SVN: r233655

gcc/ChangeLog

+2016-02-24  Richard Biener  <rguenther@suse.de>
+
+	PR tree-optimization/69907
+	* tree-vect-stmts.c (vectorizable_load): Check for gaps at the
+	end of permutations for BB vectorization.
+
 2016-02-24  Christian Bruel  <christian.bruel@st.com>
 
 	* config/arm/arm-c.c (arm_option_override): Initialize

gcc/testsuite/ChangeLog

+2016-02-24  Richard Biener  <rguenther@suse.de>
+
+	PR tree-optimization/69907
+	* gcc.dg/vect/bb-slp-pr69907.c: New testcase.
+	* gcc.dg/vect/bb-slp-34.c: XFAIL.
+	* gcc.dg/vect/bb-slp-pr68892.c: Likewise.
+
 2016-02-24  Christian Bruel  <christian.bruel@st.com>
 
 	* gcc.target/arm/pragma_cpp_fma.c: New test.

gcc/testsuite/gcc.dg/vect/bb-slp-34.c

@@ -32,4 +32,5 @@ int main()
   return 0;
 }
 
-/* { dg-final { scan-tree-dump "basic block vectorized" "slp2" { target vect_perm } } } */
+/* ??? XFAILed because we access "excess" elements with the permutation.  */
+/* { dg-final { scan-tree-dump "basic block vectorized" "slp2" { target vect_perm xfail *-*-* } } } */

gcc/testsuite/gcc.dg/vect/bb-slp-pr68892.c

@@ -13,5 +13,7 @@ void foo(void)
   b[3] = a[3][0];
 }
 
-/* { dg-final { scan-tree-dump "not profitable" "slp2" } } */
+/* ???  The profitability check is not reached because we give up on the
+   gaps we access earlier.  */
+/* { dg-final { scan-tree-dump "not profitable" "slp2" { xfail *-*-* } } } */
 /* { dg-final { scan-tree-dump-times "Basic block will be vectorized" 0 "slp2" } } */

gcc/testsuite/gcc.dg/vect/bb-slp-pr69907.c

+/* { dg-do compile } */
+/* { dg-additional-options "-O3" } */
+/* { dg-require-effective-target vect_unpack } */
+
+void foo(unsigned *p1, unsigned short *p2)
+{
+  int n;
+  for (n = 0; n < 320; n++)
+    p1[n] = p2[n * 2];
+}
+
+/* { dg-final { scan-tree-dump "BB vectorization with gaps at the end of a load is not supported" "slp1" } } */

gcc/tree-vect-stmts.c

@@ -6395,6 +6395,19 @@ vectorizable_load (gimple *stmt, gimple_stmt_iterator *gsi, gimple **vec_stmt,
 	slp_perm = true;
 
       group_size = GROUP_SIZE (vinfo_for_stmt (first_stmt));
+
+      /* ???  The following is overly pessimistic (as well as the loop
+         case above) in the case we can statically determine the excess
+	 elements loaded are within the bounds of a decl that is accessed.
+	 Likewise for BB vectorizations using masked loads is a possibility.  */
+      if (bb_vinfo && slp_perm && group_size % nunits != 0)
+	{
+	  dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
+			   "BB vectorization with gaps at the end of a load "
+			   "is not supported\n");
+	  return false;
+	}
+
       if (!slp
 	  && !PURE_SLP_STMT (stmt_info)
 	  && !STMT_VINFO_STRIDED_P (stmt_info))