re PR tree-optimization/58626 (possible array wrong code bug)

2013-10-25  Richard Biener  <rguenther@suse.de>

	PR tree-optimization/58626
	* tree-loop-distribution.c (enum rdg_dep_type): Remove
	anti_dd, output_dd and input_dd.
	(struct rdg_edge): Remove level and relation members.
	(RDGE_LEVEL, RDGE_RELATION): Remove.
	(dot_rdg_1): Adjust.
	(create_rdg_edge_for_ddr): Remove.
	(create_rdg_edges_for_scalar): Adjust.
	(create_edge_for_control_dependence): Likewise.
	(create_rdg_edges): Split into ...
	(create_rdg_flow_edges): ... this
	(create_rdg_cd_edges): ... and this.
	(free_rdg): Adjust.
	(build_rdg): Likewise, do not compute data dependences or
	add edges for them.
	(pg_add_dependence_edges): New function.
	(pgcmp): Likewise.
	(distribute_loop): First apply all non-dependence based
	partition mergings.  Then compute dependences between partitions
	and merge and order partitions according to them.

	* gcc.dg/torture/pr58626.c: New testcase.

From-SVN: r204062

1 files changed, 239 insertions, 160 deletions

diff --git a/gcc/tree-loop-distribution.c b/gcc/tree-loop-distribution.c
index 15f3809..353ce24 100644
--- a/gcc/tree-loop-distribution.c
+++ b/gcc/tree-loop-distribution.c
@@ -96,15 +96,6 @@ enum rdg_dep_type
   /* Read After Write (RAW).  */
   flow_dd = 'f',
 
-  /* Write After Read (WAR).  */
-  anti_dd = 'a',
-
-  /* Write After Write (WAW).  */
-  output_dd = 'o',
-
-  /* Read After Read (RAR).  */
-  input_dd = 'i',
-
   /* Control dependence (execute conditional on).  */
   control_dd = 'c'
 };
@@ -115,19 +106,9 @@ typedef struct rdg_edge
 {
   /* Type of the dependence.  */
   enum rdg_dep_type type;
-
-  /* Levels of the dependence: the depth of the loops that carry the
-     dependence.  */
-  unsigned level;
-
-  /* Dependence relation between data dependences, NULL when one of
-     the vertices is a scalar.  */
-  ddr_p relation;
 } *rdg_edge_p;
 
 #define RDGE_TYPE(E)        ((struct rdg_edge *) ((E)->data))->type
-#define RDGE_LEVEL(E)       ((struct rdg_edge *) ((E)->data))->level
-#define RDGE_RELATION(E)    ((struct rdg_edge *) ((E)->data))->relation
 
 /* Dump vertex I in RDG to FILE.  */
 
@@ -215,23 +196,11 @@ dot_rdg_1 (FILE *file, struct graph *rdg)
        for (e = v->succ; e; e = e->succ_next)
          switch (RDGE_TYPE (e))
            {
-           case input_dd:
-             fprintf (file, "%d -> %d [label=input] \n", i, e->dest);
-             break;
-
-           case output_dd:
-             fprintf (file, "%d -> %d [label=output] \n", i, e->dest);
-             break;
-
            case flow_dd:
              /* These are the most common dependences: don't print these. */
              fprintf (file, "%d -> %d \n", i, e->dest);
              break;
 
-           case anti_dd:
-             fprintf (file, "%d -> %d [label=anti] \n", i, e->dest);
-             break;
-
 	   case control_dd:
              fprintf (file, "%d -> %d [label=control] \n", i, e->dest);
              break;
@@ -273,52 +242,6 @@ rdg_vertex_for_stmt (struct graph *rdg ATTRIBUTE_UNUSED, gimple stmt)
   return index;
 }
 
-/* Creates an edge in RDG for each distance vector from DDR.  The
-   order that we keep track of in the RDG is the order in which
-   statements have to be executed.  */
-
-static void
-create_rdg_edge_for_ddr (struct graph *rdg, ddr_p ddr)
-{
-  struct graph_edge *e;
-  int va, vb;
-  data_reference_p dra = DDR_A (ddr);
-  data_reference_p drb = DDR_B (ddr);
-  unsigned level = ddr_dependence_level (ddr);
-
-  /* For non scalar dependences, when the dependence is REVERSED,
-     statement B has to be executed before statement A.  */
-  if (level > 0
-      && !DDR_REVERSED_P (ddr))
-    {
-      data_reference_p tmp = dra;
-      dra = drb;
-      drb = tmp;
-    }
-
-  va = rdg_vertex_for_stmt (rdg, DR_STMT (dra));
-  vb = rdg_vertex_for_stmt (rdg, DR_STMT (drb));
-
-  if (va < 0 || vb < 0)
-    return;
-
-  e = add_edge (rdg, va, vb);
-  e->data = XNEW (struct rdg_edge);
-
-  RDGE_LEVEL (e) = level;
-  RDGE_RELATION (e) = ddr;
-
-  /* Determines the type of the data dependence.  */
-  if (DR_IS_READ (dra) && DR_IS_READ (drb))
-    RDGE_TYPE (e) = input_dd;
-  else if (DR_IS_WRITE (dra) && DR_IS_WRITE (drb))
-    RDGE_TYPE (e) = output_dd;
-  else if (DR_IS_WRITE (dra) && DR_IS_READ (drb))
-    RDGE_TYPE (e) = flow_dd;
-  else if (DR_IS_READ (dra) && DR_IS_WRITE (drb))
-    RDGE_TYPE (e) = anti_dd;
-}
-
 /* Creates dependence edges in RDG for all the uses of DEF.  IDEF is
    the index of DEF in RDG.  */
 
@@ -339,7 +262,6 @@ create_rdg_edges_for_scalar (struct graph *rdg, tree def, int idef)
       e = add_edge (rdg, idef, use);
       e->data = XNEW (struct rdg_edge);
       RDGE_TYPE (e) = flow_dd;
-      RDGE_RELATION (e) = NULL;
     }
 }
 
@@ -366,7 +288,6 @@ create_edge_for_control_dependence (struct graph *rdg, basic_block bb,
 	  e = add_edge (rdg, c, v);
 	  e->data = XNEW (struct rdg_edge);
 	  RDGE_TYPE (e) = control_dd;
-	  RDGE_RELATION (e) = NULL;
 	}
     }
 }
@@ -374,38 +295,38 @@ create_edge_for_control_dependence (struct graph *rdg, basic_block bb,
 /* Creates the edges of the reduced dependence graph RDG.  */
 
 static void
-create_rdg_edges (struct graph *rdg, vec<ddr_p> ddrs, control_dependences *cd)
+create_rdg_flow_edges (struct graph *rdg)
 {
   int i;
-  struct data_dependence_relation *ddr;
   def_operand_p def_p;
   ssa_op_iter iter;
 
-  FOR_EACH_VEC_ELT (ddrs, i, ddr)
-    if (DDR_ARE_DEPENDENT (ddr) == NULL_TREE)
-      create_rdg_edge_for_ddr (rdg, ddr);
-    else
-      free_dependence_relation (ddr);
-
   for (i = 0; i < rdg->n_vertices; i++)
     FOR_EACH_PHI_OR_STMT_DEF (def_p, RDG_STMT (rdg, i),
 			      iter, SSA_OP_DEF)
       create_rdg_edges_for_scalar (rdg, DEF_FROM_PTR (def_p), i);
+}
 
-  if (cd)
-    for (i = 0; i < rdg->n_vertices; i++)
-      {
-	gimple stmt = RDG_STMT (rdg, i);
-	if (gimple_code (stmt) == GIMPLE_PHI)
-	  {
-	    edge_iterator ei;
-	    edge e;
-	    FOR_EACH_EDGE (e, ei, gimple_bb (stmt)->preds)
+/* Creates the edges of the reduced dependence graph RDG.  */
+
+static void
+create_rdg_cd_edges (struct graph *rdg, control_dependences *cd)
+{
+  int i;
+
+  for (i = 0; i < rdg->n_vertices; i++)
+    {
+      gimple stmt = RDG_STMT (rdg, i);
+      if (gimple_code (stmt) == GIMPLE_PHI)
+	{
+	  edge_iterator ei;
+	  edge e;
+	  FOR_EACH_EDGE (e, ei, gimple_bb (stmt)->preds)
 	      create_edge_for_control_dependence (rdg, e->src, i, cd);
-	  }
-	else
-	  create_edge_for_control_dependence (rdg, gimple_bb (stmt), i, cd);
-      }
+	}
+      else
+	create_edge_for_control_dependence (rdg, gimple_bb (stmt), i, cd);
+    }
 }
 
 /* Build the vertices of the reduced dependence graph RDG.  Return false
@@ -494,10 +415,7 @@ free_rdg (struct graph *rdg)
       struct graph_edge *e;
 
       for (e = v->succ; e; e = e->succ_next)
-	{
-	  free_dependence_relation (RDGE_RELATION (e));
-	  free (e->data);
-	}
+	free (e->data);
 
       if (v->data)
 	{
@@ -520,7 +438,6 @@ build_rdg (vec<loop_p> loop_nest, control_dependences *cd)
   struct graph *rdg;
   vec<gimple> stmts;
   vec<data_reference_p> datarefs;
-  vec<ddr_p> dependence_relations;
 
   /* Create the RDG vertices from the stmts of the loop nest.  */
   stmts.create (10);
@@ -536,19 +453,10 @@ build_rdg (vec<loop_p> loop_nest, control_dependences *cd)
     }
   stmts.release ();
 
-  /* Create the RDG edges from the data dependences in the loop nest.  */
-  dependence_relations.create (100);
-  if (!compute_all_dependences (datarefs, &dependence_relations, loop_nest,
-				false)
-      || !known_dependences_p (dependence_relations))
-    {
-      free_dependence_relations (dependence_relations);
-      datarefs.release ();
-      free_rdg (rdg);
-      return NULL;
-    }
-  create_rdg_edges (rdg, dependence_relations, cd);
-  dependence_relations.release ();
+  create_rdg_flow_edges (rdg);
+  if (cd)
+    create_rdg_cd_edges (rdg, cd);
+
   datarefs.release ();
 
   return rdg;
@@ -1405,6 +1313,70 @@ partition_contains_all_rw (struct graph *rdg,
   return false;
 }
 
+/* Compute partition dependence created by the data references in DRS1
+   and DRS2 and modify and return DIR according to that.  */
+
+static int
+pg_add_dependence_edges (struct graph *rdg, vec<loop_p> loops, int dir,
+			 vec<data_reference_p> drs1,
+			 vec<data_reference_p> drs2)
+{
+  data_reference_p dr1, dr2;
+
+  /* dependence direction - 0 is no dependence, -1 is back,
+     1 is forth, 2 is both (we can stop then, merging will occur).  */
+  for (int ii = 0; drs1.iterate (ii, &dr1); ++ii)
+    for (int jj = 0; drs2.iterate (jj, &dr2); ++jj)
+      {
+	int this_dir = 1;
+	ddr_p ddr;
+	/* Re-shuffle data-refs to be in dominator order.  */
+	if (rdg_vertex_for_stmt (rdg, DR_STMT (dr1))
+	    > rdg_vertex_for_stmt (rdg, DR_STMT (dr2)))
+	  {
+	    data_reference_p tem = dr1;
+	    dr1 = dr2;
+	    dr2 = tem;
+	    this_dir = -this_dir;
+	  }
+	ddr = initialize_data_dependence_relation (dr1, dr2, loops);
+	compute_affine_dependence (ddr, loops[0]);
+	if (DDR_ARE_DEPENDENT (ddr) == chrec_dont_know)
+	  this_dir = 2;
+	else if (DDR_ARE_DEPENDENT (ddr) == NULL_TREE)
+	  {
+	    if (DDR_REVERSED_P (ddr))
+	      {
+		data_reference_p tem = dr1;
+		dr1 = dr2;
+		dr2 = tem;
+		this_dir = -this_dir;
+	      }
+	    /* Known dependences can still be unordered througout the
+	       iteration space, see gcc.dg/tree-ssa/ldist-16.c.  */
+	    if (DDR_NUM_DIST_VECTS (ddr) == 0)
+	      this_dir = 2;
+	  }
+	else
+	  this_dir = 0;
+	free_dependence_relation (ddr);
+	if (dir == 0)
+	  dir = this_dir;
+	else if (dir != this_dir)
+	  return 2;
+      }
+  return dir;
+}
+
+/* Compare postorder number of the partition graph vertices V1 and V2.  */
+
+static int
+pgcmp (const void *v1_, const void *v2_)
+{
+  const vertex *v1 = (const vertex *)v1_;
+  const vertex *v2 = (const vertex *)v2_;
+  return v2->post - v1->post;
+}
 
 /* Distributes the code from LOOP in such a way that producer
    statements are placed before consumer statements.  Tries to separate
@@ -1421,6 +1393,8 @@ distribute_loop (struct loop *loop, vec<gimple> stmts,
   partition_t partition;
   bool any_builtin;
   int i, nbp;
+  graph *pg = NULL;
+  int num_sccs = 1;
 
   *nb_calls = 0;
   loop_nest.create (3);
@@ -1455,8 +1429,8 @@ distribute_loop (struct loop *loop, vec<gimple> stmts,
       any_builtin |= partition_builtin_p (partition);
     }
 
-  /* If we did not detect any builtin but are not asked to apply
-     regular loop distribution simply bail out.  */
+  /* If we are only distributing patterns but did not detect any,
+     simply bail out.  */
   if (!flag_tree_loop_distribution
       && !any_builtin)
     {
@@ -1464,9 +1438,56 @@ distribute_loop (struct loop *loop, vec<gimple> stmts,
       goto ldist_done;
     }
 
+  /* If we are only distributing patterns fuse all partitions that
+     were not classified as builtins.  This also avoids chopping
+     a loop into pieces, separated by builtin calls.  That is, we
+     only want no or a single loop body remaining.  */
+  partition_t into;
+  if (!flag_tree_loop_distribution)
+    {
+      for (i = 0; partitions.iterate (i, &into); ++i)
+	if (!partition_builtin_p (into))
+	  break;
+      for (++i; partitions.iterate (i, &partition); ++i)
+	if (!partition_builtin_p (partition))
+	  {
+	    if (dump_file && (dump_flags & TDF_DETAILS))
+	      {
+		fprintf (dump_file, "fusing non-builtin partitions\n");
+		dump_bitmap (dump_file, into->stmts);
+		dump_bitmap (dump_file, partition->stmts);
+	      }
+	    partition_merge_into (into, partition);
+	    partitions.unordered_remove (i);
+	    partition_free (partition);
+	    i--;
+	  }
+    }
+
+  /* Due to limitations in the transform phase we have to fuse all
+     reduction partitions into the last partition so the existing
+     loop will contain all loop-closed PHI nodes.  */
+  for (i = 0; partitions.iterate (i, &into); ++i)
+    if (partition_reduction_p (into))
+      break;
+  for (i = i + 1; partitions.iterate (i, &partition); ++i)
+    if (partition_reduction_p (partition))
+      {
+	if (dump_file && (dump_flags & TDF_DETAILS))
+	  {
+	    fprintf (dump_file, "fusing partitions\n");
+	    dump_bitmap (dump_file, into->stmts);
+	    dump_bitmap (dump_file, partition->stmts);
+	    fprintf (dump_file, "because they have reductions\n");
+	  }
+	partition_merge_into (into, partition);
+	partitions.unordered_remove (i);
+	partition_free (partition);
+	i--;
+      }
+
   /* Apply our simple cost model - fuse partitions with similar
      memory accesses.  */
-  partition_t into;
   for (i = 0; partitions.iterate (i, &into); ++i)
     {
       if (partition_builtin_p (into))
@@ -1486,61 +1507,119 @@ distribute_loop (struct loop *loop, vec<gimple> stmts,
 			   "memory accesses\n");
 		}
 	      partition_merge_into (into, partition);
-	      partitions.ordered_remove (j);
+	      partitions.unordered_remove (j);
 	      partition_free (partition);
 	      j--;
 	    }
 	}
     }
 
-  /* If we are only distributing patterns fuse all partitions that
-     were not properly classified as builtins.  */
-  if (!flag_tree_loop_distribution)
+  /* Build the partition dependency graph.  */
+  if (partitions.length () > 1)
     {
-      partition_t into;
-      /* Only fuse adjacent non-builtin partitions, see PR53616.
-         ???  Use dependence information to improve partition ordering.  */
-      i = 0;
-      do
+      pg = new_graph (partitions.length ());
+      struct pgdata {
+	  partition_t partition;
+	  vec<data_reference_p> writes;
+	  vec<data_reference_p> reads;
+      };
+#define PGDATA(i) ((pgdata *)(pg->vertices[i].data))
+      for (i = 0; partitions.iterate (i, &partition); ++i)
+	{
+	  vertex *v = &pg->vertices[i];
+	  pgdata *data = new pgdata;
+	  data_reference_p dr;
+	  /* FIXME - leaks.  */
+	  v->data = data;
+	  bitmap_iterator bi;
+	  unsigned j;
+	  data->partition = partition;
+	  data->reads = vNULL;
+	  data->writes = vNULL;
+	  EXECUTE_IF_SET_IN_BITMAP (partition->stmts, 0, j, bi)
+	    for (int k = 0; RDG_DATAREFS (rdg, j).iterate (k, &dr); ++k)
+	      if (DR_IS_READ (dr))
+		data->reads.safe_push (dr);
+	      else
+		data->writes.safe_push (dr);
+	}
+      partition_t partition1, partition2;
+      for (i = 0; partitions.iterate (i, &partition1); ++i)
+	for (int j = i + 1; partitions.iterate (j, &partition2); ++j)
+	  {
+	    /* dependence direction - 0 is no dependence, -1 is back,
+	       1 is forth, 2 is both (we can stop then, merging will occur).  */
+	    int dir = 0;
+	    dir = pg_add_dependence_edges (rdg, loop_nest, dir,
+					   PGDATA(i)->writes,
+					   PGDATA(j)->reads);
+	    if (dir != 2)
+	      dir = pg_add_dependence_edges (rdg, loop_nest, dir,
+					     PGDATA(i)->reads,
+					     PGDATA(j)->writes);
+	    if (dir != 2)
+	      dir = pg_add_dependence_edges (rdg, loop_nest, dir,
+					     PGDATA(i)->writes,
+					     PGDATA(j)->writes);
+	    if (dir == 1 || dir == 2)
+	      add_edge (pg, i, j);
+	    if (dir == -1 || dir == 2)
+	      add_edge (pg, j, i);
+	  }
+
+      /* Add edges to the reduction partition (if any) to force it last.  */
+      unsigned j;
+      for (j = 0; partitions.iterate (j, &partition); ++j)
+	if (partition_reduction_p (partition))
+	  break;
+      if (j < partitions.length ())
 	{
-	  for (; partitions.iterate (i, &into); ++i)
-	    if (!partition_builtin_p (into))
+	  for (unsigned i = 0; partitions.iterate (i, &partition); ++i)
+	    if (i != j)
+	      add_edge (pg, i, j);
+	}
+
+      /* Compute partitions we cannot separate and fuse them.  */
+      num_sccs = graphds_scc (pg, NULL);
+      for (i = 0; i < num_sccs; ++i)
+	{
+	  partition_t first;
+	  int j;
+	  for (j = 0; partitions.iterate (j, &first); ++j)
+	    if (pg->vertices[j].component == i)
 	      break;
-	  for (++i; partitions.iterate (i, &partition); ++i)
-	    if (!partition_builtin_p (partition))
+	  for (j = j + 1; partitions.iterate (j, &partition); ++j)
+	    if (pg->vertices[j].component == i)
 	      {
-		partition_merge_into (into, partition);
-		partitions.ordered_remove (i);
+		if (dump_file && (dump_flags & TDF_DETAILS))
+		  {
+		    fprintf (dump_file, "fusing partitions\n");
+		    dump_bitmap (dump_file, first->stmts);
+		    dump_bitmap (dump_file, partition->stmts);
+		    fprintf (dump_file, "because they are in the same "
+			     "dependence SCC\n");
+		  }
+		partition_merge_into (first, partition);
+		partitions[j] = NULL;
 		partition_free (partition);
-		i--;
+		PGDATA (j)->partition = NULL;
 	      }
-	    else
-	      break;
 	}
-      while ((unsigned) i < partitions.length ());
-    }
 
-  /* Fuse all reduction partitions into the last.  */
-  if (partitions.length () > 1)
-    {
-      partition_t into = partitions.last ();
-      for (i = partitions.length () - 2; i >= 0; --i)
+      /* Now order the remaining nodes in postorder.  */
+      qsort (pg->vertices, pg->n_vertices, sizeof (vertex), pgcmp);
+      partitions.truncate (0);
+      for (i = 0; i < pg->n_vertices; ++i)
 	{
-	  partition_t what = partitions[i];
-	  if (partition_reduction_p (what))
-	    {
-	      if (dump_file && (dump_flags & TDF_DETAILS))
-		{
-		  fprintf (dump_file, "fusing partitions\n");
-		  dump_bitmap (dump_file, into->stmts);
-		  dump_bitmap (dump_file, what->stmts);
-		  fprintf (dump_file, "because the latter has reductions\n");
-		}
-	      partition_merge_into (into, what);
-	      partitions.ordered_remove (i);
-	      partition_free (what);
-	    }
+	  pgdata *data = PGDATA (i);
+	  if (data->partition)
+	    partitions.safe_push (data->partition);
+	  data->reads.release ();
+	  data->writes.release ();
+	  delete data;
 	}
+      gcc_assert (partitions.length () == (unsigned)num_sccs);
+      free_graph (pg);
     }
 
   nbp = partitions.length ();