This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
Re: combine.c:1670: warning: left shift count >= width of type
- To: Andreas Jaeger <aj at suse dot de>
- Subject: Re: combine.c:1670: warning: left shift count >= width of type
- From: Alexandre Oliva <aoliva at redhat dot com>
- Date: 13 Feb 2001 08:18:59 -0200
- Cc: gcc-bugs at gcc dot gnu dot org, gcc-patches at gcc dot gnu dot org
- Organization: GCC Team, Red Hat
- References: <hozofr6i85.fsf@gee.suse.de>
On Feb 13, 2001, Andreas Jaeger <aj@suse.de> wrote:
> is your change to combine really correct? On i686-linux I get now:
> /cvs/gcc/gcc/combine.c: In function `try_combine':
> /cvs/gcc/gcc/combine.c:1670: warning: left shift count >= width of type
It's definitely wrong. I hadn't considered the possibility of
HOST_BITS_PER_WIDE_INT == BITS_PER_WORD, and didn't notice the
warnings, and, out of (bad) luck, it would just work on x86. Here's a
patch that fixes this problem. Ok to install, if it completes
bootstrap? Head and branch?
Index: gcc/ChangeLog
from Alexandre Oliva <aoliva@redhat.com>
* combine.c (UWIDE_SHIFT_LEFT_BY_BITS_PER_WORD): New macro.
(try_combine): Use it.
Index: gcc/combine.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/combine.c,v
retrieving revision 1.183
diff -u -p -r1.183 combine.c
--- gcc/combine.c 2001/02/06 12:39:15 1.183
+++ gcc/combine.c 2001/02/13 10:09:24
@@ -146,6 +146,12 @@ static int max_uid_cuid;
#define INSN_CUID(INSN) \
(INSN_UID (INSN) > max_uid_cuid ? insn_cuid (INSN) : uid_cuid[INSN_UID (INSN)])
+/* In case BITS_PER_WORD == HOST_BITS_PER_WIDE_INT, shifting by
+ BITS_PER_WORD would invoke undefined behavior. Work around it. */
+
+#define UWIDE_SHIFT_LEFT_BY_BITS_PER_WORD(val) \
+ (((unsigned HOST_WIDE_INT)(val) << (BITS_PER_WORD - 1)) << 1)
+
/* Maximum register number, which is the size of the tables below. */
static unsigned int combine_max_regno;
@@ -1667,7 +1673,7 @@ try_combine (i3, i2, i1, new_direct_jump
if (HOST_BITS_PER_WIDE_INT < BITS_PER_WORD)
abort ();
- lo &= ~(((unsigned HOST_WIDE_INT)1 << BITS_PER_WORD) - 1);
+ lo &= ~(UWIDE_SHIFT_LEFT_BY_BITS_PER_WORD (1) - 1);
lo |= INTVAL (SET_SRC (PATTERN (i3)));
}
else if (HOST_BITS_PER_WIDE_INT == BITS_PER_WORD)
@@ -1677,9 +1683,10 @@ try_combine (i3, i2, i1, new_direct_jump
int sign = -(int) ((unsigned HOST_WIDE_INT) lo
>> (HOST_BITS_PER_WIDE_INT - 1));
- lo &= ~((((unsigned HOST_WIDE_INT)1 << BITS_PER_WORD) - 1)
- << BITS_PER_WORD);
- lo |= INTVAL (SET_SRC (PATTERN (i3))) << BITS_PER_WORD;
+ lo &= ~ (UWIDE_SHIFT_LEFT_BY_BITS_PER_WORD
+ (UWIDE_SHIFT_LEFT_BY_BITS_PER_WORD (1) - 1));
+ lo |= (UWIDE_SHIFT_LEFT_BY_BITS_PER_WORD
+ (INTVAL (SET_SRC (PATTERN (i3)))));
if (hi == sign)
hi = lo < 0 ? -1 : 0;
}
--
Alexandre Oliva Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist *Please* write to mailing lists, not to me