[patch] fix pr22550

Diego Novillo dnovillo@redhat.com
Fri Jul 29 16:37:00 GMT 2005


On Thu, Jul 28, 2005 at 11:30:04AM -0600, Jeffrey A Law wrote:
> On Thu, 2005-07-28 at 11:56 -0400, Diego Novillo wrote: 
>
> > I experimented with both approaches and, contrary to what I had
> > expected, iterating inside CFG cleanup is the winner.  Both
> > patches are attached.
> Wow.  I never would have expected this...
> 
Yeah.  So much for gut feelings, eh? ;)

> Hmmm, I can hazard a guess it's more a matter of preventing DOM
> iteration rather than fewer opportunities for DOM.
> 
Ah, yes.  I had forgotten about DOM iterations.  And indeed that
explains the difference:

Number of DOM iterations	Before	After	% diff

cc1-i-files			59000	52542	-10.95%
DLV				11478	10095	-12.05%
MICO				33032	30912	- 6.42%
SPEC2000			25382	22704	-10.55%
TRAMP3D				 4234	 3360	-20.64%
pr12850				 5069	 3790	-25.23%

This is what I committed.



	PR 22550
	* tree-cfgcleanup.c (cleanup_tree_cfg_1): Extract from ...
	(cleanup_tree_cfg): ... here.
	Call cleanup_tree_cfg_1 until there are no more cleanups to
	do.

testsuite/ChangeLog

	PR 22550
	* g++.dg/tree-ssa/pr22550.C: New test.

Index: tree-cfgcleanup.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/tree-cfgcleanup.c,v
retrieving revision 2.3
diff -d -u -p -r2.3 tree-cfgcleanup.c
--- tree-cfgcleanup.c	18 Jul 2005 23:20:09 -0000	2.3
+++ tree-cfgcleanup.c	29 Jul 2005 16:26:26 -0000
@@ -489,14 +489,12 @@ cleanup_forwarder_blocks (void)
   return changed;
 }
 
-/* Remove unreachable blocks and other miscellaneous clean up work.  */
+/* Do one round of CFG cleanup.  */
 
-bool
-cleanup_tree_cfg (void)
+static bool
+cleanup_tree_cfg_1 (void)
 {
-  bool retval = false;
-
-  timevar_push (TV_TREE_CLEANUP_CFG);
+  bool retval;
 
   retval = cleanup_control_flow ();
   retval |= delete_unreachable_blocks ();
@@ -516,6 +514,28 @@ cleanup_tree_cfg (void)
       end_recording_case_labels ();
     }
 
+  /* Merging the blocks may create new opportunities for folding
+     conditional branches (due to the elimination of single-valued PHI
+     nodes).  */
+  retval |= merge_seq_blocks ();
+
+  return retval;
+}
+
+
+/* Remove unreachable blocks and other miscellaneous clean up work.  */
+
+bool
+cleanup_tree_cfg (void)
+{
+  bool retval;
+  int i;
+
+  timevar_push (TV_TREE_CLEANUP_CFG);
+
+  for (retval = true, i = 0; i < 5 && retval; i++)
+    retval = cleanup_tree_cfg_1 ();
+
 #ifdef ENABLE_CHECKING
   if (retval)
     {
@@ -526,16 +546,14 @@ cleanup_tree_cfg (void)
     }
 #endif
 
-  /* Merging the blocks creates no new opportunities for the other
-     optimizations, so do it here.  */
-  retval |= merge_seq_blocks ();
-
   compact_blocks ();
 
 #ifdef ENABLE_CHECKING
   verify_flow_info ();
 #endif
+
   timevar_pop (TV_TREE_CLEANUP_CFG);
+
   return retval;
 }
 
Index: testsuite/g++.dg/tree-ssa/pr22550.C
===================================================================
RCS file: testsuite/g++.dg/tree-ssa/pr22550.C
diff -N testsuite/g++.dg/tree-ssa/pr22550.C
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ testsuite/g++.dg/tree-ssa/pr22550.C	29 Jul 2005 16:26:33 -0000
@@ -0,0 +1,28 @@
+/* { dg-do compile } */
+/* { dg-options "-O2" } */
+
+class X {
+public:
+  int mfunc1 () {
+    return 1;
+  }
+  int mfunc2 () {
+    return 2;
+  }
+  X (int a, int b) { }
+};
+
+typedef int (X::*memfunc_p_t) ();
+
+memfunc_p_t mf_arr[2] = { &X::mfunc1, &X::mfunc2 };
+
+int
+main ()
+{
+  // Get pntr to the array of pointers to member-funcs
+  memfunc_p_t (*mf_arr_p)[2] = &mf_arr;
+  // Compare indirect against direct access to an array element
+  if ((*mf_arr_p)[0] != mf_arr[0])
+    return 1;
+  return 0;
+}



More information about the Gcc-patches mailing list