re PR tree-optimization/68306 (ICE: in vectorizable_store, at tree-vect-stmts.c:5651)

2015-11-13  Richard Biener  <rguenther@suse.de>

	PR tree-optimization/68306
	* tree-vect-data-refs.c (verify_data_ref_alignment): Move
	loop related checks ...
	(vect_verify_datarefs_alignment): ... here.
	(vect_slp_analyze_and_verify_node_alignment): Compute and
	verify alignment of the single DR that it matters.
	* tree-vect-stmts.c (vectorizable_store): Add an assert.
	(vectorizable_load): Add a comment.
	* tree-vect-slp.c (vect_analyze_slp_cost_1): Fix DR used
	for determining load cost.

	* gcc.dg/pr68306.c: Adjust.
	* gcc.dg/pr68306-2.c: New testcase.
	* gcc.dg/pr68306-3.c: Likewise.

From-SVN: r230310

8 files changed, 94 insertions, 40 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index a7517f9..1b16e12 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,16 @@
+2015-11-13  Richard Biener  <rguenther@suse.de>
+
+	PR tree-optimization/68306
+	* tree-vect-data-refs.c (verify_data_ref_alignment): Move
+	loop related checks ...
+	(vect_verify_datarefs_alignment): ... here.
+	(vect_slp_analyze_and_verify_node_alignment): Compute and
+	verify alignment of the single DR that it matters.
+	* tree-vect-stmts.c (vectorizable_store): Add an assert.
+	(vectorizable_load): Add a comment.
+	* tree-vect-slp.c (vect_analyze_slp_cost_1): Fix DR used
+	for determining load cost.
+
 2015-11-13  Ilya Enkovich  <enkovich.gnu@gmail.com>
 
 	* tree-vect-loop.c (vect_determine_vectorization_factor): Check
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index ca62684..d0bce7f 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,10 @@
+2015-11-13  Richard Biener  <rguenther@suse.de>
+
+	PR tree-optimization/68306
+	* gcc.dg/pr68306.c: Adjust.
+	* gcc.dg/pr68306-2.c: New testcase.
+	* gcc.dg/pr68306-3.c: Likewise.
+
 2015-11-13  Ilya Enkovich  <enkovich.gnu@gmail.com>
 
 	* g++.dg/vect/simd-bool-comparison-1.cc: New test.
diff --git a/gcc/testsuite/gcc.dg/pr68306-2.c b/gcc/testsuite/gcc.dg/pr68306-2.c
new file mode 100644
index 0000000..1d84e96
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/pr68306-2.c
@@ -0,0 +1,12 @@
+/* { dg-do compile } */
+/* { dg-options "-O3" } */
+/* { dg-additional-options "-mno-sse -mno-mmx" { target x86_64-*-* } } */
+
+struct {
+    int tz_minuteswest;
+    int tz_dsttime;
+} a, b;
+void fn1() {
+    b.tz_minuteswest = a.tz_minuteswest;
+    b.tz_dsttime = a.tz_dsttime;
+}
diff --git a/gcc/testsuite/gcc.dg/pr68306-3.c b/gcc/testsuite/gcc.dg/pr68306-3.c
new file mode 100644
index 0000000..d03ce53
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/pr68306-3.c
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-options "-O3" } */
+/* { dg-additional-options "-mno-sse -mno-mmx" { target x86_64-*-* } } */
+/* { dg-additional-options "-mno-altivec -mno-vsx" { target powerpc*-*-* } } */
+
+extern void fn2();
+struct {
+    unsigned qp_num;
+    unsigned starting_psn;
+    void *private_data;
+} a;
+struct {
+    unsigned id;
+    unsigned qpn;
+    unsigned psn;
+} b;
+void fn1() {
+    a.qp_num = b.qpn;
+    a.starting_psn = b.psn;
+    fn2(b.id);
+}
diff --git a/gcc/testsuite/gcc.dg/pr68306.c b/gcc/testsuite/gcc.dg/pr68306.c
index b36fb34..e1805a7 100644
--- a/gcc/testsuite/gcc.dg/pr68306.c
+++ b/gcc/testsuite/gcc.dg/pr68306.c
@@ -1,5 +1,6 @@
 /* { dg-do compile } */
 /* { dg-options "-O3" } */
+/* { dg-additional-options "-mno-sse -mno-mmx" { target x86_64-*-* } } */
 
 enum powerpc_pmc_type { PPC_PMC_IBM };
 struct {
diff --git a/gcc/tree-vect-data-refs.c b/gcc/tree-vect-data-refs.c
index 590d363..f9327d7 100644
--- a/gcc/tree-vect-data-refs.c
+++ b/gcc/tree-vect-data-refs.c
@@ -916,22 +916,8 @@ vect_update_misalignment_for_peel (struct data_reference *dr,
 static bool
 verify_data_ref_alignment (data_reference_p dr)
 {
-  enum dr_alignment_support supportable_dr_alignment;
-  gimple *stmt = DR_STMT (dr);
-  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
-
-  /* For interleaving, only the alignment of the first access matters.   */
-  if (STMT_VINFO_GROUPED_ACCESS (stmt_info)
-      && GROUP_FIRST_ELEMENT (stmt_info) != stmt)
-    return true;
-
-  /* Strided accesses perform only component accesses, alignment is
-     irrelevant for them.  */
-  if (STMT_VINFO_STRIDED_P (stmt_info)
-      && !STMT_VINFO_GROUPED_ACCESS (stmt_info))
-    return true;
-
-  supportable_dr_alignment = vect_supportable_dr_alignment (dr, false);
+  enum dr_alignment_support supportable_dr_alignment
+    = vect_supportable_dr_alignment (dr, false);
   if (!supportable_dr_alignment)
     {
       if (dump_enabled_p ())
@@ -977,6 +963,18 @@ vect_verify_datarefs_alignment (loop_vec_info vinfo)
 
       if (!STMT_VINFO_RELEVANT_P (stmt_info))
 	continue;
+
+      /* For interleaving, only the alignment of the first access matters.   */
+      if (STMT_VINFO_GROUPED_ACCESS (stmt_info)
+	  && GROUP_FIRST_ELEMENT (stmt_info) != stmt)
+	return true;
+
+      /* Strided accesses perform only component accesses, alignment is
+	 irrelevant for them.  */
+      if (STMT_VINFO_STRIDED_P (stmt_info)
+	  && !STMT_VINFO_GROUPED_ACCESS (stmt_info))
+	return true;
+
       if (! verify_data_ref_alignment (dr))
 	return false;
     }
@@ -2100,28 +2098,22 @@ vect_analyze_data_refs_alignment (loop_vec_info vinfo)
 static bool
 vect_slp_analyze_and_verify_node_alignment (slp_tree node)
 {
-  unsigned i;
-  gimple *stmt;
-  FOR_EACH_VEC_ELT (SLP_TREE_SCALAR_STMTS (node), i, stmt)
+  /* We vectorize from the first scalar stmt in the node unless
+     the node is permuted in which case we start from the first
+     element in the group.  */
+  gimple *first_stmt = SLP_TREE_SCALAR_STMTS (node)[0];
+  if (SLP_TREE_LOAD_PERMUTATION (node).exists ())
+    first_stmt = GROUP_FIRST_ELEMENT (vinfo_for_stmt (first_stmt));
+
+  data_reference_p dr = STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt));
+  if (! vect_compute_data_ref_alignment (dr)
+      || ! verify_data_ref_alignment (dr))
     {
-      stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
-
-      /* Strided accesses perform only component accesses, misalignment
-	 information is irrelevant for them.  */
-      if (STMT_VINFO_STRIDED_P (stmt_info)
-	  && !STMT_VINFO_GROUPED_ACCESS (stmt_info))
-	continue;
-
-      data_reference_p dr = STMT_VINFO_DATA_REF (stmt_info);
-      if (! vect_compute_data_ref_alignment (dr)
-	  || ! verify_data_ref_alignment (dr))
-	{
-	  if (dump_enabled_p ())
-	    dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
-			     "not vectorized: bad data alignment in basic "
-			     "block.\n");
-	  return false;
-	}
+      if (dump_enabled_p ())
+	dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
+			 "not vectorized: bad data alignment in basic "
+			 "block.\n");
+      return false;
     }
 
   return true;
diff --git a/gcc/tree-vect-slp.c b/gcc/tree-vect-slp.c
index bf6d1d8..f65837d 100644
--- a/gcc/tree-vect-slp.c
+++ b/gcc/tree-vect-slp.c
@@ -1429,6 +1429,13 @@ vect_analyze_slp_cost_1 (slp_instance instance, slp_tree node,
 	{
 	  int i;
 	  gcc_checking_assert (DR_IS_READ (STMT_VINFO_DATA_REF (stmt_info)));
+	  /* If the load is permuted then the alignment is determined by
+	     the first group element not by the first scalar stmt DR.  */
+	  if (SLP_TREE_LOAD_PERMUTATION (node).exists ())
+	    {
+	      stmt = GROUP_FIRST_ELEMENT (stmt_info);
+	      stmt_info = vinfo_for_stmt (stmt);
+	    }
 	  vect_model_load_cost (stmt_info, ncopies_for_cost, false,
 				node, prologue_cost_vec, body_cost_vec);
 	  /* If the load is permuted record the cost for the permutation.
diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
index f7eee91..0f64aaf 100644
--- a/gcc/tree-vect-stmts.c
+++ b/gcc/tree-vect-stmts.c
@@ -5464,6 +5464,7 @@ vectorizable_store (gimple *stmt, gimple_stmt_iterator *gsi, gimple **vec_stmt,
              group.  */
           vec_num = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node);
           first_stmt = SLP_TREE_SCALAR_STMTS (slp_node)[0]; 
+	  gcc_assert (GROUP_FIRST_ELEMENT (vinfo_for_stmt (first_stmt)) == first_stmt);
           first_dr = STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt));
 	  op = gimple_assign_rhs1 (first_stmt);
         } 
@@ -6658,9 +6659,9 @@ vectorizable_load (gimple *stmt, gimple_stmt_iterator *gsi, gimple **vec_stmt,
   if (grouped_load)
     {
       first_stmt = GROUP_FIRST_ELEMENT (stmt_info);
-      if (slp
-          && !SLP_TREE_LOAD_PERMUTATION (slp_node).exists ()
-	  && first_stmt != SLP_TREE_SCALAR_STMTS (slp_node)[0])
+      /* For BB vectorization we directly vectorize a subchain
+         without permutation.  */
+      if (slp && ! SLP_TREE_LOAD_PERMUTATION (slp_node).exists ())
         first_stmt = SLP_TREE_SCALAR_STMTS (slp_node)[0];
 
       /* Check if the chain of loads is already vectorized.  */