This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH] Re: Fix nested functions
- From: Ulrich Weigand <weigand at i1 dot informatik dot uni-erlangen dot de>
- To: jh at suse dot cz (Jan Hubicka)
- Cc: weigand at i1 dot informatik dot uni-erlangen dot de (Ulrich Weigand), hubicka at ucw dot cz (Jan Hubicka), jh at suse dot cz (Jan Hubicka), gcc-patches at gcc dot gnu dot org
- Date: Tue, 28 Sep 2004 22:04:52 +0200 (CEST)
- Subject: [PATCH] Re: Fix nested functions
Jan Hubicka wrote:
> Hmm, this reminds me the i386 case - fixing one problem leads to other
> one.
> I guess adding the mark_jump_label after validate_change will lead to
> later crash in shorten_branches.
Yes, this is because mark_jump_label also ignores non-local labels.
If I change this, there is an ICE due to non-matching edge counts
somewhere, and then I've stopped debugging ...
> We really need to bail out earlier and give up idea of propagating
> nonlocal LABEL_REF into the jump I would say :(
I did this for cse, but then combine attempted the same transformation,
and after I've fixed combine, local-alloc tried the same.
With these three places fixed, the jump finally stays indirect,
and my testcases are fixed.
Bootstrapped/regtested on s390-ibm-linux and s390x-ibm-linux.
OK for mainline?
Bye,
Ulrich
ChangeLog:
* cse.c (cse_insn): Avoid creating direct non-local jumps.
* combine.c (can_combine_p): Likewise.
* local-alloc. (update_equiv_regs): Likewise.
Index: gcc/combine.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/combine.c,v
retrieving revision 1.456
diff -c -p -r1.456 combine.c
*** gcc/combine.c 26 Sep 2004 14:55:37 -0000 1.456
--- gcc/combine.c 28 Sep 2004 16:53:21 -0000
*************** can_combine_p (rtx insn, rtx i3, rtx pre
*** 1176,1181 ****
--- 1176,1183 ----
/* Don't substitute into an incremented register. */
|| FIND_REG_INC_NOTE (i3, dest)
|| (succ && FIND_REG_INC_NOTE (succ, dest))
+ /* Don't substitute into a non-local goto, this confuses CFG. */
+ || (JUMP_P (i3) && find_reg_note (i3, REG_NON_LOCAL_GOTO, NULL_RTX))
#if 0
/* Don't combine the end of a libcall into anything. */
/* ??? This gives worse code, and appears to be unnecessary, since no
Index: gcc/cse.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/cse.c,v
retrieving revision 1.314
diff -c -p -r1.314 cse.c
*** gcc/cse.c 20 Sep 2004 20:38:17 -0000 1.314
--- gcc/cse.c 28 Sep 2004 16:53:22 -0000
*************** cse_insn (rtx insn, rtx libcall_insn)
*** 5395,5400 ****
--- 5395,5405 ----
|| (GET_CODE (trial) == LABEL_REF
&& ! condjump_p (insn))))
{
+ /* Don't substitute non-local labels, this confuses CFG. */
+ if (GET_CODE (trial) == LABEL_REF
+ && LABEL_REF_NONLOCAL_P (trial))
+ continue;
+
SET_SRC (sets[i].rtl) = trial;
cse_jumps_altered = 1;
break;
Index: gcc/local-alloc.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/local-alloc.c,v
retrieving revision 1.138
diff -c -p -r1.138 local-alloc.c
*** gcc/local-alloc.c 10 Sep 2004 11:02:21 -0000 1.138
--- gcc/local-alloc.c 28 Sep 2004 16:53:23 -0000
*************** update_equiv_regs (void)
*** 1010,1015 ****
--- 1010,1020 ----
if (! INSN_P (insn))
continue;
+ /* Don't substitute into a non-local goto, this confuses CFG. */
+ if (JUMP_P (insn)
+ && find_reg_note (insn, REG_NON_LOCAL_GOTO, NULL_RTX))
+ continue;
+
for (link = REG_NOTES (insn); link; link = XEXP (link, 1))
{
if (REG_NOTE_KIND (link) == REG_DEAD
--
Dr. Ulrich Weigand
weigand@informatik.uni-erlangen.de