tree-affine.h (aff_combination_expand): Declare.

2008-03-27  Zdenek Dvorak  <ook@ucw.cz>

	* tree-affine.h (aff_combination_expand): Declare.
	(get_inner_reference_aff): Likewise.
	* tree-affine.c	(aff_combination_expand): Split out from
	tree_to_aff_combination_expand.
	(get_inner_reference_aff): New function.
	* tree-parloops.c (loop_parallel_p): Free vectorizer info.
        * tree-ssa-loop-im.c: Include tree-affine.h and pointer-set.h.
        (struct lim_aux_data): sm_done field removed.
        (mem_ref_loc_p, mem_ref_locs_p): New types.
        (struct mem_ref): Added id, stored, accesses_in_loop,
        indep_loop, dep_loop, indep_ref, dep_ref fields.
        Removed is_stored, locs and next fields.
        (memory_accesses): New variable.
        (movement_possibility): Do not allow moving statements
        that store to memory.
        (outermost_indep_loop, simple_mem_ref_in_stmt, mem_ref_in_stmt):
        New functions.
        (determine_max_movement): For statements with memory references,
        find the outermost loop in that the reference is independent.
        (move_computations_stmt): Mark the virtual operands for
        renaming.
        (memref_free, mem_ref_alloc, mem_ref_locs_alloc, mark_ref_stored,
        gather_mem_refs_stmt, gather_mem_refs_in_loops, vtoe_hash, vtoe_eq,
        vtoe_free, record_vop_access, get_vop_accesses, get_vop_stores,
        add_vop_ref_mapping, create_vop_ref_mapping_loop,
        create_vop_ref_mapping, analyze_memory_references,
        cannot_overlap_p, mem_refs_may_alias_p, rewrite_mem_ref_loc,
        get_all_locs_in_loop, ref_always_accessed_p,
        refs_independent_p, record_indep_loop, ref_indep_loop_p_1,
        ref_indep_loop_p, can_sm_ref_p, find_refs_for_sm,
        store_motion_loop, store_motion): New functions.
        (struct vop_to_refs_elt): New type.
        (record_mem_ref_loc, free_mem_ref_locs, rewrite_mem_refs,
        memref_hash, memref_eq, hoist_memory_references): Rewritten.
        (schedule_sm): Replaced by...
        (execute_sm): ... this.
        (determine_lsm_ref, hoist_memory_references,
        loop_suitable_for_sm, gather_mem_refs_stmt, gather_mem_refs,
        find_more_ref_vops, free_mem_ref, free_mem_refs,
        determine_lsm_loop, determine_lsm): Removed.
        (tree_ssa_lim_finalize): Free data structures used by store
        motion.
        (tree_ssa_lim): Call analyze_memory_references.  Use
        store_motion instead of determine_lsm.

	* gcc.dg/tree-ssa/loop-32.c: New testcase.
	* gcc.dg/tree-ssa/loop-33.c: Likewise.

From-SVN: r133637

gcc/ChangeLog

@@ -1,3 +1,50 @@
+2008-03-27  Zdenek Dvorak  <ook@ucw.cz>
+
+	* tree-affine.h (aff_combination_expand): Declare.
+	(get_inner_reference_aff): Likewise.
+	* tree-affine.c	(aff_combination_expand): Split out from
+	tree_to_aff_combination_expand.
+	(get_inner_reference_aff): New function.
+	* tree-parloops.c (loop_parallel_p): Free vectorizer info.
+        * tree-ssa-loop-im.c: Include tree-affine.h and pointer-set.h.
+        (struct lim_aux_data): sm_done field removed.
+        (mem_ref_loc_p, mem_ref_locs_p): New types.
+        (struct mem_ref): Added id, stored, accesses_in_loop,
+        indep_loop, dep_loop, indep_ref, dep_ref fields.
+        Removed is_stored, locs and next fields.
+        (memory_accesses): New variable.
+        (movement_possibility): Do not allow moving statements
+        that store to memory.
+        (outermost_indep_loop, simple_mem_ref_in_stmt, mem_ref_in_stmt):
+        New functions.
+        (determine_max_movement): For statements with memory references,
+        find the outermost loop in that the reference is independent.
+        (move_computations_stmt): Mark the virtual operands for
+        renaming.
+        (memref_free, mem_ref_alloc, mem_ref_locs_alloc, mark_ref_stored,
+        gather_mem_refs_stmt, gather_mem_refs_in_loops, vtoe_hash, vtoe_eq,
+        vtoe_free, record_vop_access, get_vop_accesses, get_vop_stores,
+        add_vop_ref_mapping, create_vop_ref_mapping_loop,
+        create_vop_ref_mapping, analyze_memory_references,
+        cannot_overlap_p, mem_refs_may_alias_p, rewrite_mem_ref_loc,
+        get_all_locs_in_loop, ref_always_accessed_p,
+        refs_independent_p, record_indep_loop, ref_indep_loop_p_1,
+        ref_indep_loop_p, can_sm_ref_p, find_refs_for_sm,
+        store_motion_loop, store_motion): New functions.
+        (struct vop_to_refs_elt): New type.
+        (record_mem_ref_loc, free_mem_ref_locs, rewrite_mem_refs,
+        memref_hash, memref_eq, hoist_memory_references): Rewritten.
+        (schedule_sm): Replaced by...
+        (execute_sm): ... this.
+        (determine_lsm_ref, hoist_memory_references,
+        loop_suitable_for_sm, gather_mem_refs_stmt, gather_mem_refs,
+        find_more_ref_vops, free_mem_ref, free_mem_refs,
+        determine_lsm_loop, determine_lsm): Removed.
+        (tree_ssa_lim_finalize): Free data structures used by store
+        motion.
+        (tree_ssa_lim): Call analyze_memory_references.  Use
+        store_motion instead of determine_lsm.
+
 2008-03-27  Paolo Bonzini  <bonzini@gnu.org>
 
 	* config.cc (m68hc11, m6811, m68hc12, m6812): Add usegas.h,

gcc/testsuite/ChangeLog

@@ -1,3 +1,8 @@
+2008-03-27  Zdenek Dvorak  <ook@ucw.cz>
+
+	* gcc.dg/tree-ssa/loop-32.c: New testcase.
+	* gcc.dg/tree-ssa/loop-33.c: Likewise.
+
 2008-03-27  Richard Guenther  <rguenther@suse.de>
 
 	* gcc.dg/fold-addr-1.c: New testcase.

gcc/testsuite/gcc.dg/tree-ssa/loop-32.c

@@ -0,0 +1,46 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-tree-lim-details" } */
+
+int x;
+int a[100];
+
+struct a
+{
+  int X;
+  int Y;
+};
+
+void bla(void);
+
+void test1(void)
+{
+  unsigned i;
+
+  /* We should perform store motion here.  */
+  for (x = 0; x < 100; x++)
+    a[x] = x;
+}
+
+void test2(void)
+{
+  unsigned i;
+
+  /* But not here.  */
+  for (x = 0; x < 100; x++)
+    bla ();
+}
+
+void test3(struct a *A)
+{
+  unsigned i;
+
+  /* But we should here (using base + offset analysis).  */
+  for (i = 0; i < 100; i++)
+    {
+      A[5].X += i;
+      A[5].Y += i;
+    }
+}
+
+/* { dg-final { scan-tree-dump-times "Executing store motion of" 3 "lim" } } */
+/* { dg-final { cleanup-tree-dump "lim" } } */

gcc/testsuite/gcc.dg/tree-ssa/loop-33.c

@@ -0,0 +1,40 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-tree-lim-details" } */
+
+int x;
+int a[100];
+
+struct a
+{
+  int X;
+  int Y;
+};
+
+struct a arr[100];
+
+void test4(unsigned b)
+{
+  unsigned i;
+
+  /* And here.  */
+  for (i = 0; i < 100; i++)
+    {
+      arr[b+8].X += i;
+      arr[b+9].X += i;
+    }
+}
+
+void test5(struct a *A, unsigned b)
+{
+  unsigned i;
+
+  /* And here as well.  */
+  for (i = 0; i < 100; i++)
+    {
+      A[b].X += i;
+      A[b+1].Y += i;
+    }
+}
+
+/* { dg-final { scan-tree-dump-times "Executing store motion of" 4 "lim" { xfail lp64 } } } */
+/* { dg-final { cleanup-tree-dump "lim" } } */

gcc/tree-affine.c

@@ -562,19 +562,11 @@ struct name_expansion
   unsigned in_progress : 1;
 };
 
-/* Similar to tree_to_aff_combination, but follows SSA name definitions
+/* Expands SSA names in COMB recursively.  CACHE is used to cache the
-   and expands them recursively.  CACHE is used to cache the expansions
+   results.  */
-   of the ssa names, to avoid exponential time complexity for cases
-   like
- 
-   a1 = a0 + a0;
-   a2 = a1 + a1;
-   a3 = a2 + a2;
-   ...  */
 
 void
-tree_to_aff_combination_expand (tree expr, tree type, aff_tree *comb,
+aff_combination_expand (aff_tree *comb, struct pointer_map_t **cache)
-				struct pointer_map_t **cache)
 {
   unsigned i;
   aff_tree to_add, current, curre;
@@ -583,8 +575,7 @@ tree_to_aff_combination_expand (tree expr, tree type, aff_tree *comb,
   void **slot;
   struct name_expansion *exp;
 
-  tree_to_aff_combination (expr, type, comb);
+  aff_combination_zero (&to_add, comb->type);
-  aff_combination_zero (&to_add, type);
   for (i = 0; i < comb->n; i++)
     {
       e = comb->elts[i].val;
@@ -616,7 +607,7 @@ tree_to_aff_combination_expand (tree expr, tree type, aff_tree *comb,
 	  exp = XNEW (struct name_expansion);
 	  exp->in_progress = 1;
 	  *slot = exp;
-	  tree_to_aff_combination_expand (rhs, type, &current, cache);
+	  tree_to_aff_combination_expand (rhs, comb->type, &current, cache);
 	  exp->expansion = current;
 	  exp->in_progress = 0;
 	}
@@ -632,7 +623,7 @@ tree_to_aff_combination_expand (tree expr, tree type, aff_tree *comb,
 	 COMB while traversing it; include the term -coef * E, to remove
          it from COMB.  */
       scale = comb->elts[i].coef;
-      aff_combination_zero (&curre, type);
+      aff_combination_zero (&curre, comb->type);
       aff_combination_add_elt (&curre, e, double_int_neg (scale));
       aff_combination_scale (&current, scale);
       aff_combination_add (&to_add, &current);
@@ -641,6 +632,24 @@ tree_to_aff_combination_expand (tree expr, tree type, aff_tree *comb,
   aff_combination_add (comb, &to_add);
 }
 
+/* Similar to tree_to_aff_combination, but follows SSA name definitions
+   and expands them recursively.  CACHE is used to cache the expansions
+   of the ssa names, to avoid exponential time complexity for cases
+   like
+
+   a1 = a0 + a0;
+   a2 = a1 + a1;
+   a3 = a2 + a2;
+   ...  */
+
+void
+tree_to_aff_combination_expand (tree expr, tree type, aff_tree *comb,
+				struct pointer_map_t **cache)
+{
+  tree_to_aff_combination (expr, type, comb);
+  aff_combination_expand (comb, cache);
+}
+
 /* Frees memory occupied by struct name_expansion in *VALUE.  Callback for
    pointer_map_traverse.  */
 
@@ -783,3 +792,36 @@ debug_aff (aff_tree *val)
   print_aff (stderr, val);
   fprintf (stderr, "\n");
 }
+
+/* Returns address of the reference REF in ADDR.  The size of the accessed
+   location is stored to SIZE.  */
+
+void
+get_inner_reference_aff (tree ref, aff_tree *addr, double_int *size)
+{
+  HOST_WIDE_INT bitsize, bitpos;
+  tree toff;
+  enum machine_mode mode;
+  int uns, vol;
+  aff_tree tmp;
+  tree base = get_inner_reference (ref, &bitsize, &bitpos, &toff, &mode,
+				   &uns, &vol, false);
+  tree base_addr = build_fold_addr_expr (base);
+
+  /* ADDR = &BASE + TOFF + BITPOS / BITS_PER_UNIT.  */
+
+  tree_to_aff_combination (base_addr, sizetype, addr);
+
+  if (toff)
+    {
+      tree_to_aff_combination (toff, sizetype, &tmp);
+      aff_combination_add (addr, &tmp);
+    }
+
+  aff_combination_const (&tmp, sizetype,
+			 shwi_to_double_int (bitpos / BITS_PER_UNIT));
+  aff_combination_add (addr, &tmp);
+
+  *size = shwi_to_double_int ((bitsize + BITS_PER_UNIT - 1) / BITS_PER_UNIT);
+}
+

gcc/tree-affine.h

@@ -71,8 +71,10 @@ void tree_to_aff_combination (tree, tree, aff_tree *);
 tree aff_combination_to_tree (aff_tree *);
 void unshare_aff_combination (aff_tree *);
 bool aff_combination_constant_multiple_p (aff_tree *, aff_tree *, double_int *);
+void aff_combination_expand (aff_tree *, struct pointer_map_t **);
 void tree_to_aff_combination_expand (tree, tree, aff_tree *,
 				     struct pointer_map_t **);
+void get_inner_reference_aff (tree, aff_tree *, double_int *);
 void free_affine_expand_cache (struct pointer_map_t **);
 
 /* Debugging functions.  */

gcc/tree-parloops.c

@@ -317,6 +317,9 @@ loop_parallel_p (struct loop *loop, htab_t reduction_list, struct tree_niter_des
 	}
     }
 
+  /* Get rid of the information created by the vectorizer functions.  */
+  destroy_loop_vec_info (simple_loop_info, true);
+
   for (phi = phi_nodes (exit->dest); phi; phi = PHI_CHAIN (phi))
     {
       struct reduction_info *red;