This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH] Fix SUBREG_BYTE related ICE on ia64
- To: rth at redhat dot com
- Subject: [PATCH] Fix SUBREG_BYTE related ICE on ia64
- From: Jakub Jelinek <jakub at redhat dot com>
- Date: Mon, 23 Apr 2001 09:36:15 +0200
- Cc: gcc-patches at gcc dot gnu dot org
- Reply-To: Jakub Jelinek <jakub at redhat dot com>
Hi!
The following testcase ICEs on ia64. The issue is that combiner throws
random crap into subregs, so the checks cannot be too tight if combining.
subreg_hard_regno already has a check_mode argument which is only cleared
when called from combine, so this looks like the right thing to do to me.
Ok to commit?
2001-04-23 Jakub Jelinek <jakub@redhat.com>
* emit-rtl.c (subreg_hard_regno): Only do HARD_REGNO_MODE_OK check
if check_mode is non-zero.
* gcc.dg/20010423-1.c: New test.
--- gcc/testsuite/gcc.dg/20010423-1.c.jj Mon Apr 23 10:37:45 2001
+++ gcc/testsuite/gcc.dg/20010423-1.c Mon Apr 23 10:38:57 2001
@@ -0,0 +1,17 @@
+/* { dg-do compile { target ia64-*-* } } */
+/* { dg-options "-O2" } */
+
+int __sync_fetch_and_add_si (int *, int);
+
+inline unsigned int
+bar (volatile unsigned int *mem, unsigned int val)
+{
+ return __sync_fetch_and_add_si((int *)mem, (int)val);
+}
+
+volatile unsigned int x;
+
+void foo (unsigned short *a)
+{
+ *a = bar (&x, 1) + 1;
+}
--- gcc/emit-rtl.c.jj Thu Apr 5 00:12:39 2001
+++ gcc/emit-rtl.c Mon Apr 23 10:35:29 2001
@@ -719,7 +719,7 @@ subreg_hard_regno (x, check_mode)
base_regno = REGNO (reg);
if (base_regno >= FIRST_PSEUDO_REGISTER)
abort ();
- if (! HARD_REGNO_MODE_OK (base_regno, GET_MODE (reg)))
+ if (check_mode && ! HARD_REGNO_MODE_OK (base_regno, GET_MODE (reg)))
abort ();
/* Catch non-congruent offsets too. */
Jakub