re PR tree-optimization/19831 (Missing DSE/malloc/free optimization)

2009-07-01  Richard Guenther  <rguenther@suse.de>

	PR tree-optimization/19831
	* tree-ssa-dce.c (propagate_necessity): Calls to functions
	that only act as barriers do not make any previous stores
	necessary.
	* tree-ssa-structalias.c (handle_lhs_call): Delay making
	HEAP variables global, do not add a constraint from nonlocal.
	(find_func_aliases): Handle escapes through return statements.
	(compute_points_to_sets): Make escaped HEAP variables global.

	* gcc.dg/tree-ssa/20041122-1.c: Enable TBAA, scan FRE dump,
	make allocated memory escape.  Un-XFAIL.
	* gcc.dg/vect/pr21591.c: Make allocated memory escape.
	* gcc.dg/vect/pr31699.c: Likewise.
	* gcc.dg/tree-ssa/ssa-dce-7.c: New testcase.

	libmudflap/
	* testsuite/libmudflap.c/fail11-frag.c: Make allocated memory
	escape.
	* testsuite/libmudflap.c/fail12-frag.c: Likewise.
	* testsuite/libmudflap.c/fail16-frag.c: Likewise.
	* testsuite/libmudflap.c/fail31-frag.c: Likewise.

From-SVN: r149140

gcc/ChangeLog

@@ -1,3 +1,14 @@
+2009-07-01  Richard Guenther  <rguenther@suse.de>
+
+	PR tree-optimization/19831
+	* tree-ssa-dce.c (propagate_necessity): Calls to functions
+	that only act as barriers do not make any previous stores
+	necessary.
+	* tree-ssa-structalias.c (handle_lhs_call): Delay making
+	HEAP variables global, do not add a constraint from nonlocal.
+	(find_func_aliases): Handle escapes through return statements.
+	(compute_points_to_sets): Make escaped HEAP variables global.
+
 2009-07-01  Paolo Bonzini  <bonzini@gnu.org>
 
 	PR bootstrap/40597

gcc/testsuite/ChangeLog

@@ -1,3 +1,12 @@
+2009-07-01  Richard Guenther  <rguenther@suse.de>
+
+	PR tree-optimization/19831
+	* gcc.dg/tree-ssa/20041122-1.c: Enable TBAA, scan FRE dump,
+	make allocated memory escape.  Un-XFAIL.
+	* gcc.dg/vect/pr21591.c: Make allocated memory escape.
+	* gcc.dg/vect/pr31699.c: Likewise.
+	* gcc.dg/tree-ssa/ssa-dce-7.c: New testcase.
+
 2009-06-30  Jakub Jelinek  <jakub@redhat.com>
 
 	PR c++/40566

gcc/testsuite/gcc.dg/tree-ssa/20041122-1.c

@@ -1,6 +1,5 @@
 /* { dg-do compile } */
-/* { dg-options "-O1 -fdump-tree-dom2" } */
-
+/* { dg-options "-O1 -fstrict-aliasing -fdump-tree-fre" } */
 
 __extension__ typedef __SIZE_TYPE__ size_t;
 extern void *xmalloc (size_t) __attribute__ ((__malloc__));
@@ -17,10 +16,10 @@ struct basic_block_def
 typedef struct basic_block_def *basic_block;
 extern int n_basic_blocks;
 extern edge frob ();
-void
-find_unreachable_blocks (int frobit)
+basic_block *
+find_unreachable_blocks (void)
 {
-  basic_block *tos, *worklist, bb;
+  basic_block *tos, *worklist;
   tos = worklist = xmalloc (sizeof (basic_block) * n_basic_blocks);
   edge e = frob();
   if (!(e->dest->flags & 4))
@@ -28,11 +27,12 @@ find_unreachable_blocks (int frobit)
       e->dest->flags |= 4;
       *tos++ = e->dest;
     }
+  return worklist;
 }
 
 /* If the aliasing code does its job properly, then we should be
    able to determine that modifying e->dest->flags does not
-   modify e or e->dest.  The net result is that we only need one
-   load of e->dest.  */
-/* { dg-final { scan-tree-dump-times "->dest" 1 "dom2" { xfail *-*-* } } } */
-/* { dg-final { cleanup-tree-dump "dom2" } } */
+   modify e or e->dest if we can assert strict-aliasing rules.
+   The net result is that we only need one load of e->dest.  */
+/* { dg-final { scan-tree-dump-times "->dest" 1 "fre" } } */
+/* { dg-final { cleanup-tree-dump "fre" } } */

gcc/testsuite/gcc.dg/tree-ssa/ssa-dce-7.c

@@ -0,0 +1,33 @@
+/* { dg-do link } */
+/* { dg-options "-O -fdump-tree-optimized" } */
+
+extern void link_error (void);
+void foo(int n)
+{
+  int * f = (int*) __builtin_malloc (n * sizeof (int));
+  int * ff = (int*) __builtin_malloc (n * sizeof (int));
+  int i;
+
+  for (i = 0; i < n; ++i)
+    {
+      f[i] = 1;
+      ff[i] = 2;
+      if (f[i] != 1)
+	link_error ();
+      if (ff[i] != 2)
+	link_error ();
+    }
+
+  __builtin_free (f);
+  __builtin_free (ff);
+}
+int main()
+{
+  return 0;
+}
+
+/* We should have removed the calls to link_error () and all stores
+   to the allocated memory.  */
+
+/* { dg-final { scan-tree-dump-times "\\\*D" 0 "optimized" } } */
+/* { dg-final { cleanup-tree-dump "optimized" } } */

gcc/testsuite/gcc.dg/vect/pr21591.c

@@ -10,6 +10,8 @@ struct a
 struct a *malloc1(__SIZE_TYPE__) __attribute__((malloc));
 void free(void*);
 
+struct a *p, *q, *r;
+
 void f(void)
 {
    struct a *a = malloc1(sizeof(struct a));
@@ -26,9 +28,9 @@ void f(void)
    {
       a->a1[i] = b->a1[i] + c->a1[i];
    }
-   free(a);
-   free(b);
-   free(c);
+   p = a;
+   q = b;
+   r = c;
 }
 
 /* { dg-final { scan-tree-dump-times "vectorized 2 loops" 1 "vect" } } */

gcc/testsuite/gcc.dg/vect/pr31699.c

@@ -7,13 +7,15 @@
 float x[256];
 
 __attribute__ ((noinline))
-void foo(void)
+double *foo(void)
 {
  double *z = malloc (sizeof(double) * 256);
 
  int i;
  for (i=0; i<256; ++i)
    z[i] = x[i] + 1.0f;
+
+ return z;
 }
 
 

gcc/tree-ssa-dce.c

@@ -676,8 +676,19 @@ propagate_necessity (struct edge_list *el)
 
 	  if (is_gimple_call (stmt))
 	    {
+	      tree callee = gimple_call_fndecl (stmt);
 	      unsigned i;
 
+	      /* Calls to functions that are merely acting as barriers
+		 or that only store to memory do not make any previous
+		 stores necessary.  */
+	      if (callee != NULL_TREE
+		  && DECL_BUILT_IN_CLASS (callee) == BUILT_IN_NORMAL
+		  && (DECL_FUNCTION_CODE (callee) == BUILT_IN_MEMSET
+		      || DECL_FUNCTION_CODE (callee) == BUILT_IN_MALLOC
+		      || DECL_FUNCTION_CODE (callee) == BUILT_IN_FREE))
+		continue;
+
 	      /* Calls implicitly load from memory, their arguments
 	         in addition may explicitly perform memory loads.  */
 	      mark_all_reaching_defs_necessary (stmt);

gcc/tree-ssa-structalias.c

@@ -3473,7 +3473,9 @@ handle_lhs_call (tree lhs, int flags, VEC(ce_s, heap) *rhsc)
     {
       varinfo_t vi;
       vi = make_constraint_from_heapvar (get_vi_for_tree (lhs), "HEAP");
-      make_copy_constraint (vi, nonlocal_id);
+      /* We delay marking allocated storage global until we know if
+         it escapes.  */
+      vi->is_global_var = 0;
     }
   else if (VEC_length (ce_s, rhsc) > 0)
     {
@@ -3910,6 +3912,13 @@ find_func_aliases (gimple origt)
     {
       make_escape_constraint (gimple_assign_rhs1 (t));
     }
+  /* Handle escapes through return.  */
+  else if (gimple_code (t) == GIMPLE_RETURN
+	   && gimple_return_retval (t) != NULL_TREE
+	   && could_have_pointers (gimple_return_retval (t)))
+    {
+      make_escape_constraint (gimple_return_retval (t));
+    }
   /* Handle asms conservatively by adding escape constraints to everything.  */
   else if (gimple_code (t) == GIMPLE_ASM)
     {
@@ -5350,6 +5359,7 @@ compute_points_to_sets (void)
   struct scc_info *si;
   basic_block bb;
   unsigned i;
+  varinfo_t vi;
 
   timevar_push (TV_TREE_PTA);
 
@@ -5447,6 +5457,13 @@ compute_points_to_sets (void)
      points-to solution queries.  */
   cfun->gimple_df->escaped.escaped = 0;
 
+  /* Mark escaped HEAP variables as global.  */
+  for (i = 0; VEC_iterate (varinfo_t, varmap, i, vi); ++i)
+    if (vi->is_heap_var
+	&& !vi->is_global_var)
+      vi->is_global_var = pt_solution_includes (&cfun->gimple_df->escaped,
+						vi->decl);
+
   /* Compute the points-to sets for pointer SSA_NAMEs.  */
   for (i = 0; i < num_ssa_names; ++i)
     {

libmudflap/ChangeLog

@@ -1,3 +1,12 @@
+2009-07-01  Richard Guenther  <rguenther@suse.de>
+
+	PR tree-optimization/19831
+	* testsuite/libmudflap.c/fail11-frag.c: Make allocated memory
+	escape.
+	* testsuite/libmudflap.c/fail12-frag.c: Likewise.
+	* testsuite/libmudflap.c/fail16-frag.c: Likewise.
+	* testsuite/libmudflap.c/fail31-frag.c: Likewise.
+
 2009-06-30  Richard Sandiford  <r.sandiford@uk.ibm.com>
 
 	* testsuite/lib/libmudflap.exp (libmudflap-init): Don't add "."

libmudflap/testsuite/libmudflap.c/fail11-frag.c

@@ -1,11 +1,12 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+char *y;
 int main ()
 {
 int i = 10;
 char *x = (char *) malloc (i * sizeof (char));
-
+y = x;
 while (i--)
 {
   ++x;

libmudflap/testsuite/libmudflap.c/fail12-frag.c

@@ -1,11 +1,12 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+int *y;
 int main ()
 {
 int i = 10;
 int *x = (int *) malloc (i * sizeof (int));
-
+y = x;
 while (i--)
 {
   ++x;

libmudflap/testsuite/libmudflap.c/fail16-frag.c

@@ -1,6 +1,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+void *p;
 int main ()
 {
 struct base {
@@ -15,7 +16,7 @@ struct derived {
 struct base *bp;
 
 bp = (struct base *) malloc (sizeof (struct base));;
-
+p = bp;
 bp->basic = 10;
 ((struct derived *)bp)->extra = 'x';
 return 0;

libmudflap/testsuite/libmudflap.c/fail31-frag.c

@@ -8,11 +8,12 @@ int main ()
   int z = h (4, 10);
   return 0;
 }
-
+int *p;
 int h (int i, int j)
 {
   int k[i];
   k[j] = i;
+  p = k;
   return j;
 }