From 2b8775f335a64ac0c7a4d32b38ee50463862d6fb Mon Sep 17 00:00:00 2001
From: Ira Rosen <irar@il.ibm.com>
Date: Wed, 14 May 2008 12:28:53 +0000
Subject: [PATCH] re PR tree-optimization/36098 (435.gromacs miscompares at
 -O3)

	PR tree-optimization/36098
	* tree-vect-analyze.c (vect_analyze_group_access): Set the gap
	value for the first load in the group in case of a gap.
	(vect_build_slp_tree): Check that there are no gaps in loads.

From-SVN: r135290
---
 gcc/ChangeLog                          |  7 ++++++
 gcc/testsuite/ChangeLog                |  6 +++++
 gcc/testsuite/gcc.dg/vect/O3-pr36098.c | 22 ++++++++++++++++++
 gcc/testsuite/gcc.dg/vect/vect.exp     |  2 +-
 gcc/tree-vect-analyze.c                | 32 ++++++++++++++++++--------
 5 files changed, 58 insertions(+), 11 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/vect/O3-pr36098.c

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 223fe39ad03..9e151ea6984 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,10 @@
+2008-05-14  Ira Rosen  <irar@il.ibm.com>
+
+	PR tree-optimization/36098
+	* tree-vect-analyze.c (vect_analyze_group_access): Set the gap
+	value for the first load in the group in case of a gap.
+	(vect_build_slp_tree): Check that there are no gaps in loads.
+
 2008-05-14  Kenneth Zadeck <zadeck@naturalbridge.com>
 
 	* doc/rtl.texi: Removed reference to REG_NO_CONFLICT notes.
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 05a2d2569a0..bd147b45a52 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,9 @@
+2008-05-14  Ira Rosen  <irar@il.ibm.com>
+
+	PR tree-optimization/36098	
+	* gcc.dg/vect/vect.exp: Compile with -O3 all the tests named "O3-*".
+	* gcc.dg/vect/O3-pr36098.c: New test.
+
 2008-05-14  Samuel Tardieu  <sam@rfc1149.net>
 
 	* gnat.dg/old_errors.ads, gnat.dg/old_errors.adb: New.
diff --git a/gcc/testsuite/gcc.dg/vect/O3-pr36098.c b/gcc/testsuite/gcc.dg/vect/O3-pr36098.c
new file mode 100644
index 00000000000..1c445be9e20
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/O3-pr36098.c
@@ -0,0 +1,22 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target vect_int } */
+
+typedef struct {
+  int iatom[3];
+  int blocknr;
+} t_sortblock;
+
+#define DIM 3
+
+void foo (int ncons, t_sortblock *sb, int *iatom)
+{
+ int i, m;
+
+ for(i=0; (i<ncons); i++,iatom+=3)
+   for(m=0; (m<DIM); m++)
+     iatom[m]=sb[i].iatom[m];
+}
+
+/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" { xfail *-*-* } } } */
+/* { dg-final { cleanup-tree-dump "vect" } } */
+
diff --git a/gcc/testsuite/gcc.dg/vect/vect.exp b/gcc/testsuite/gcc.dg/vect/vect.exp
index 18877095fa9..2fb14b1b08e 100644
--- a/gcc/testsuite/gcc.dg/vect/vect.exp
+++ b/gcc/testsuite/gcc.dg/vect/vect.exp
@@ -245,7 +245,7 @@ dg-runtest [lsort [glob -nocomplain $srcdir/$subdir/Os-vect-*.\[cS\]]]  \
 # With -O3
 set DEFAULT_VECTCFLAGS $SAVED_DEFAULT_VECTCFLAGS
 lappend DEFAULT_VECTCFLAGS "-O3"
-dg-runtest [lsort [glob -nocomplain $srcdir/$subdir/O3-vect-*.\[cS\]]]  \
+dg-runtest [lsort [glob -nocomplain $srcdir/$subdir/O3-*.\[cS\]]]  \
         "" $DEFAULT_VECTCFLAGS
 
 # With -O1
diff --git a/gcc/tree-vect-analyze.c b/gcc/tree-vect-analyze.c
index 0ef7f9cc444..1e2012f85d1 100644
--- a/gcc/tree-vect-analyze.c
+++ b/gcc/tree-vect-analyze.c
@@ -2230,7 +2230,13 @@ vect_analyze_group_access (struct data_reference *dr)
       if (dr_step != count_in_bytes)
         {
           if (DR_IS_READ (dr))
-            slp_impossible = true;
+            {
+              slp_impossible = true;
+              /* There is a gap after the last load in the group. This gap is a
+                 difference between the stride and the number of elements. When 
+                 there is no gap, this difference should be 0.  */ 
+              DR_GROUP_GAP (vinfo_for_stmt (stmt)) = stride - count; 
+            }
           else
             {
               if (vect_print_dump_info (REPORT_DETAILS))
@@ -2644,7 +2650,7 @@ vect_build_slp_tree (loop_vec_info loop_vinfo, slp_tree *node,
   enum machine_mode vec_mode;
   tree first_stmt_const_oprnd = NULL_TREE;
   struct data_reference *first_dr;
- 
+
   /* For every stmt in NODE find its def stmt/s.  */
   for (i = 0; VEC_iterate (tree, stmts, i, stmt); i++)
     {
@@ -2776,15 +2782,17 @@ vect_build_slp_tree (loop_vec_info loop_vinfo, slp_tree *node,
 		if (i == 0)
 		  {
 		    /* First stmt of the SLP group should be the first load of 
-		       the interleaving loop if data permutation is not 
-		       allowed.  */
-		    if  (DR_GROUP_FIRST_DR (vinfo_for_stmt (stmt)) != stmt) 
+		       the interleaving loop if data permutation is not allowed.
+		       Check that there is no gap between the loads.  */
+		    if (DR_GROUP_FIRST_DR (vinfo_for_stmt (stmt)) != stmt
+                        || DR_GROUP_GAP (vinfo_for_stmt (stmt)) != 0) 
 		      {
-			/* FORNOW: data permutations are not supported.  */
+			/* FORNOW: data permutations and gaps in loads are not 
+                           supported.  */
 			if (vect_print_dump_info (REPORT_SLP)) 
 			  {
 			    fprintf (vect_dump, "Build SLP failed: strided "
-				     " loads need permutation ");
+				     " loads need permutation or have gaps ");
 			    print_generic_expr (vect_dump, stmt, TDF_SLIM);
 			  }
 
@@ -2811,13 +2819,17 @@ vect_build_slp_tree (loop_vec_info loop_vinfo, slp_tree *node,
 		  }
 		else
 		  {
-		    if (DR_GROUP_NEXT_DR (vinfo_for_stmt (prev_stmt)) != stmt)
+                    /* Check that we have consecutive loads from interleaving
+                       chain and that there is no gap between the loads.  */
+		    if (DR_GROUP_NEXT_DR (vinfo_for_stmt (prev_stmt)) != stmt
+                        || DR_GROUP_GAP (vinfo_for_stmt (stmt)) != 1)
 		      {
-			/* FORNOW: data permutations are not supported.  */
+			/* FORNOW: data permutations and gaps in loads are not
+                           supported.  */
 			if (vect_print_dump_info (REPORT_SLP)) 
 			  {
 			    fprintf (vect_dump, "Build SLP failed: strided "
-				     " loads need permutation ");
+				     " loads need permutation or have gaps ");
 			    print_generic_expr (vect_dump, stmt, TDF_SLIM);
 			  }
 			return false;
-- 
2.30.2