On Sat, Oct 26, 2013 at 12:19:33AM +0200, Jakub Jelinek wrote:
> And here is a patch that allows vectorization without peeling for alignment
> and scalar loop for bound even for fn2, fn3 and fn4 in the following
> testcase, though as with the range __builtin_unreachable () notes, it is
> quite fragile, because it only works if there are no immediate uses of the
> tested SSA_NAME before the assertion. Perhaps more reliable way would be to
> convert those assertions info __builtin_assume_aligned, but that has the
> disadvantage that it's first argument is a pointer and it returns a pointer,
> so we'd need to cast integers to pointers and back, or add ASSUME_ALIGNED
> internal function.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux.
And here is as a follow-up to the whole patchset of 4 patches a testcase:
2013-10-29 Jakub Jelinek <ja...@redhat.com>
* gcc.dg/vect/vect-align-3.c: New test.
--- gcc/testsuite/gcc.dg/vect/vect-align-3.c.jj 2013-10-29 08:18:08.636348586
+0100
+++ gcc/testsuite/gcc.dg/vect/vect-align-3.c 2013-10-29 08:21:14.931384138
+0100
@@ -0,0 +1,54 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target vect_int } */
+
+int a[2048];
+
+void
+f1 (int x, int y)
+{
+ int i;
+ x &= -256;
+ y &= -256;
+ for (i = x + 256; i < y; i++)
+ a[i]++;
+}
+
+void
+f2 (int x, int y)
+{
+ int i;
+ if (x & 31)
+ __builtin_unreachable ();
+ if (y & 31)
+ __builtin_unreachable ();
+ for (i = x + 256; i < x + y; i++)
+ a[i]++;
+}
+
+void
+f3 (int x, int y)
+{
+ int i;
+ if (x % 256)
+ __builtin_unreachable ();
+ if (y % 256)
+ __builtin_unreachable ();
+ for (i = x + 256; i < x + y; i++)
+ a[i]++;
+}
+
+void
+f4 (int x, int y)
+{
+ int i;
+ if ((x % 256) != 0)
+ __builtin_unreachable ();
+ if ((y % 256) != 0)
+ __builtin_unreachable ();
+ for (i = x + 256; i < x + y; i++)
+ a[i]++;
+}
+
+/* { dg-final { scan-tree-dump-not "vect_do_peeling_for_loop_bound" "vect" } }
*/
+/* { dg-final { scan-tree-dump-not "loop peeled for vectorization" "vect" } }
*/
+/* { dg-final { cleanup-tree-dump "vect" } } */
Jakub
