This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[PATCH] Fix PR52881, more loop preserving fallout with RTL optimizers


RTL optimizers perform CFG adjustments themselves rather than relying
on cleanup_cfg (as the gimple level does).  This makes it necessary
to handle things all-over-the-place.  This case is if-conversion
performing forwarder block removal (well, it calls it "speculation"),
removing empty loop latches.  That's unwanted, thus the following
patch makes sure we do not do this.

I'm sure more RTL optimiziation fallout will pop up - and I wonder
if we should simply avoid modifying the CFG all over the place and
instead do that in cleanup_cfg.  Thus, in the if-conversion case,
simply do the speculation and leave the empty forwarder blocks
to be cleaned up by cleanup_cfg (or for CSE to leave unconditional
jumps conditional, as if (0/1), for example).

Bootstrapped on x86_64-unknown-linux-gnu, testing in progress.

Richard.

2012-04-10  Richard Guenther  <rguenther@suse.de>

	PR rtl-optimization/52881
	* ifcvt.c (find_if_case_2): Avoid speculating loop latches.

	* gcc.dg/torture/pr52881.c: New testcase.
	* gcc.dg/torture/pr52913.c: Likewise.

Index: gcc/ifcvt.c
===================================================================
*** gcc/ifcvt.c	(revision 186274)
--- gcc/ifcvt.c	(working copy)
*************** find_if_case_2 (basic_block test_bb, edg
*** 3927,3932 ****
--- 3927,3937 ----
    edge else_succ;
    int then_prob, else_prob;
  
+   /* We do not want to speculate (empty) loop latches.  */
+   if (current_loops
+       && else_bb->loop_father->latch == else_bb)
+     return FALSE;
+ 
    /* If we are partitioning hot/cold basic blocks, we don't want to
       mess up unconditional or indirect jumps that cross between hot
       and cold sections.
Index: gcc/testsuite/gcc.dg/torture/pr52881.c
===================================================================
*** gcc/testsuite/gcc.dg/torture/pr52881.c	(revision 0)
--- gcc/testsuite/gcc.dg/torture/pr52881.c	(revision 0)
***************
*** 0 ****
--- 1,35 ----
+ /* { dg-do compile } */
+ 
+ int a, b, c, d, e, f, h, i, j, k, l, m, n, o;
+ static int g;
+ int
+ fn1 () {
+     for (;; ++f)
+       if (e)
+ 	break;
+     return 0;
+ }
+ unsigned char fn2 ();
+ void
+ fn3 () {
+ lbl_220:
+     if (j) {
+ lbl_221:
+ 	l = (g || b) <= fn1 ();
+ 	for (;;) {
+ 	    g = 0;
+ 	    fn2 ();
+ 	    if (k)
+ 	      goto lbl_220;
+ 	    break;
+ 	}
+ 	if (l)
+ 	  goto lbl_221;
+     }
+ }
+ unsigned char
+ fn2 () {
+     o = d ? 0 : c;
+     h = m | a % o != n;
+     return i;
+ }
Index: gcc/testsuite/gcc.dg/torture/pr52913.c
===================================================================
--- gcc/testsuite/gcc.dg/torture/pr52913.c	(revision 0)
+++ gcc/testsuite/gcc.dg/torture/pr52913.c	(revision 0)
@@ -0,0 +1,18 @@
+/* { dg-do compile } */
+
+int a, b, c, d, e;
+void
+fn1 ()
+{
+lbl_101:
+  e = 0;
+lbl_274:
+  for (c = 0; c < 1; c = a)
+    if (d)
+      if (b)
+	goto lbl_101;
+      else
+	break;
+  d = 1;
+  goto lbl_274;
+}


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]