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]

bb-reorder crashes on potential-fall-through computed jumps


gcc.c-torture/compile/950613-1.c fails on SH with -m4 -O2.  The
problem is that the subsequent basic block is indeed one of the
potential edges, so the jump is considered as a potential
fall-through, so we end up crashing.

Here's a patch that arranges for us not to abort() on non-conditional
and conditional computed jumps (the latter was needed by another port
I have worked on, so I'm taking the opportunity to contribute the
code).

Ok to install?

Index: gcc/ChangeLog
from  Alexandre Oliva  <aoliva@redhat.com>
	* jump.c (any_cond_computed_jump_p): New.
	* bb-reorder.c (fixup_reorder_chain): Don't crash on computed
	jumps.

Index: gcc/jump.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/jump.c,v
retrieving revision 1.152
diff -u -p -r1.152 jump.c
--- gcc/jump.c 2001/01/30 19:34:40 1.152
+++ gcc/jump.c 2001/02/02 19:10:36
@@ -2304,6 +2304,32 @@ any_condjump_p (insn)
 	  || (a == PC && (b == LABEL_REF || b == RETURN)));
 }
 
+/* Return true when insn is a conditional jump to a register.  This
+   function works for instructions containing PC sets in PARALLELs.
+   The instruction may have various other effects so before removing
+   the jump you must verify onlyjump_p.
+
+   Note that unlike condjump_p it returns false for unconditional jumps.  */
+
+int
+any_cond_computed_jump_p (insn)
+     rtx insn;
+{
+  rtx x = pc_set (insn);
+  enum rtx_code a, b;
+
+  if (!x)
+    return 0;
+  if (GET_CODE (SET_SRC (x)) != IF_THEN_ELSE)
+    return 0;
+
+  a = GET_CODE (XEXP (SET_SRC (x), 1));
+  b = GET_CODE (XEXP (SET_SRC (x), 2));
+
+  return ((b == PC && a == REG)
+	  || (a == PC && b == REG));
+}
+
 /* Return the label of a conditional jump.  */
 
 rtx
Index: gcc/bb-reorder.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/bb-reorder.c,v
retrieving revision 1.25
diff -u -p -r1.25 bb-reorder.c
--- gcc/bb-reorder.c 2001/01/24 05:57:46 1.25
+++ gcc/bb-reorder.c 2001/02/02 19:10:37
@@ -638,6 +638,9 @@ fixup_reorder_chain ()
 		 99% case, there should not have been a fallthru edge.  */
 	      if (! e_fall)
 		continue;
+	      if (computed_jump_p (bb_end_insn)
+		  || any_cond_computed_jump_p (bb_end_insn))
+		continue;
 #ifdef CASE_DROPS_THROUGH
 	      /* Except for VAX.  Since we didn't have predication for the
 		 tablejump, the fallthru block should not have moved.  */

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