[PATCH][vect] PR 92351: When peeling for alignment make alignment of epilogues unknown

["Andre Vieira (lists)" <andre dot simoesdiasvieira at arm dot com>]

Hi,

PR92351 reports a bug in which a wrongly aligned load is generated for 
an epilogue of a main loop for which we peeled for alignment.  There is 
no way to guarantee that epilogue data accesses are aligned when the 
main loop is peeling for alignment.

I also had to split vect-peel-2.c as there were scans there for the 
number of unaligned accesses that were vectorized, thanks to this change 
that now depends on whether we are vectorizing the epilogue, which will 
also contain unaligned accesses.  Since not all targets need to be able 
to vectorize the epilogue I decided to disable epilogue vectorization 
for the version in which we scan the dumps and add a version that 
attempts epilogue vectorization but does not scan the dumps.

Bootstrapped and regression tested on x86_64 and aarch64.

Is this OK for trunk?

In the future I would like to look at allowing for misalignment analysis 
for cases in which both the number of iterations and iterations to peel 
are known at compile time, as in that case we shouldn't ever be skipping 
the main loop as we shouldn't be generating it.

gcc/ChangeLog:
2019-11-07  Andre Vieira  <andre.simoesdiasvieira@arm.com>

        * tree-vect-data-refs.c (vect_compute_data_ref_alignment): When 
we are
        peeling the main loop for alignment, make sure to set the 
misalignment
        of the epilogue's data references to DR_MISALIGNMENT_UNKNOWN.

gcc/testsuite/ChangeLog:
2019-11-07  Andre Vieira  <andre.simoesdiasvieira@arm.com>

        * gcc.dg/vect/vect-peel-2.c: Disable epilogue vectorization and
        split the source of this test to...
        * gcc.dg/vect/vect-peel-2-src.c: ... This.
        * gcc.dg/vect/vect-peel-2-epilogues.c: New test.

diff --git a/gcc/testsuite/gcc.dg/vect/vect-peel-2-epilogues.c b/gcc/testsuite/gcc.dg/vect/vect-peel-2-epilogues.c
new file mode 100644
index 0000000000000000000000000000000000000000..c06fa442fafa36855d285d2336e0d69ee9bffe03
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/vect-peel-2-epilogues.c
@@ -0,0 +1,3 @@
+/* { dg-require-effective-target vect_int } */
+
+#include "vect-peel-2-src.c"
diff --git a/gcc/testsuite/gcc.dg/vect/vect-peel-2-src.c b/gcc/testsuite/gcc.dg/vect/vect-peel-2-src.c
new file mode 100644
index 0000000000000000000000000000000000000000..f6fc134c8705567a628dcd62c053ad6f2ca2904d
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/vect-peel-2-src.c
@@ -0,0 +1,48 @@
+#include <stdarg.h>
+#include "tree-vect.h"
+
+#define N 128
+
+/* unaligned store.  */
+
+int ib[N+7];
+
+__attribute__ ((noinline))
+int main1 ()
+{
+  int i;
+  int ia[N+1];
+
+  /* The store is aligned and the loads are misaligned with the same 
+     misalignment. Cost model is disabled. If misaligned stores are supported,
+     we peel according to the loads to align them.  */
+  for (i = 0; i <= N; i++)
+    {
+      ia[i] = ib[i+2] + ib[i+6];
+    }
+
+  /* check results:  */
+  for (i = 1; i <= N; i++)
+    {
+      if (ia[i] != ib[i+2] + ib[i+6])
+        abort ();
+    }
+
+  return 0;
+}
+
+int main (void)
+{ 
+  int i;
+
+  check_vect ();
+
+  for (i = 0; i <= N+6; i++)
+    {
+      asm volatile ("" : "+r" (i));
+      ib[i] = i;
+    }
+
+  return main1 ();
+}
+
diff --git a/gcc/testsuite/gcc.dg/vect/vect-peel-2.c b/gcc/testsuite/gcc.dg/vect/vect-peel-2.c
index b6061c3b8553b67ecdf56367b2f4128d7c0bd342..65e70bd44170c63ce3bc25c6a7ecf426ddcd39b1 100644
--- a/gcc/testsuite/gcc.dg/vect/vect-peel-2.c
+++ b/gcc/testsuite/gcc.dg/vect/vect-peel-2.c
@@ -1,52 +1,8 @@
 /* { dg-require-effective-target vect_int } */
+/* Disabling epilogues until we find a better way to deal with scans.  */
+/* { dg-additional-options "--param vect-epilogues-nomask=0" } */
 
-#include <stdarg.h>
-#include "tree-vect.h"
-
-#define N 128
-
-/* unaligned store.  */
-
-int ib[N+7];
-
-__attribute__ ((noinline))
-int main1 ()
-{
-  int i;
-  int ia[N+1];
-
-  /* The store is aligned and the loads are misaligned with the same 
-     misalignment. Cost model is disabled. If misaligned stores are supported,
-     we peel according to the loads to align them.  */
-  for (i = 0; i <= N; i++)
-    {
-      ia[i] = ib[i+2] + ib[i+6];
-    }
-
-  /* check results:  */
-  for (i = 1; i <= N; i++)
-    {
-      if (ia[i] != ib[i+2] + ib[i+6])
-        abort ();
-    }
-
-  return 0;
-}
-
-int main (void)
-{ 
-  int i;
-
-  check_vect ();
-
-  for (i = 0; i <= N+6; i++)
-    {
-      asm volatile ("" : "+r" (i));
-      ib[i] = i;
-    }
-
-  return main1 ();
-}
+#include "vect-peel-2-src.c"
 
 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
 /* { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 1 "vect" { target { { vect_element_align } && { vect_aligned_arrays } } } } } */
diff --git a/gcc/tree-vect-data-refs.c b/gcc/tree-vect-data-refs.c
index 9dd18d265361ba5635de685f9d898e355999bf4c..fc75ac4f112486934a007c90cc6b646b6115857b 100644
--- a/gcc/tree-vect-data-refs.c
+++ b/gcc/tree-vect-data-refs.c
@@ -938,6 +938,18 @@ vect_compute_data_ref_alignment (dr_vec_info *dr_info)
     = exact_div (vect_calculate_target_alignment (dr_info), BITS_PER_UNIT);
   DR_TARGET_ALIGNMENT (dr_info) = vector_alignment;
 
+  /* If the main loop has peeled for alignment we have no way of knowing
+     whether the data accesses in the epilogues are aligned.  We can't at
+     compile time answer the question whether we have entered the main loop or
+     not.  Fixes PR 92351.  */
+  if (loop_vinfo)
+    {
+      loop_vec_info orig_loop_vinfo = LOOP_VINFO_ORIG_LOOP_INFO (loop_vinfo);
+      if (orig_loop_vinfo
+	  && LOOP_VINFO_PEELING_FOR_ALIGNMENT (orig_loop_vinfo) != 0)
+	return;
+    }
+
   unsigned HOST_WIDE_INT vect_align_c;
   if (!vector_alignment.is_constant (&vect_align_c))
     return;