Bug#233633: gcc-3.3: recent ICE regression in 3.3 while building xfree86
on Thu, Feb 19, 2004 at 05:38:08PM +0100, Falk Hueffner wrote:
> Probably related to gcc PR 14166. There's a patch for that here:
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14166
Thanks. That's kind of related, but actually a separate problem. 14166
is a Thumb-specific case, which Debian doesn't hit because Thumb isn't
enabled in our configurations.
I hope that a patch like this one will fix the problem, though I haven't
tested it yet.
p.
--- arm.h.old Thu Feb 19 13:48:41 2004
+++ arm.h Thu Feb 19 13:56:31 2004
@@ -2003,10 +2003,17 @@
&& INTVAL (op) <= 31) \
goto LABEL; \
} \
- /* NASTY: Since this limits the addressing of unsigned \
- byte loads. */ \
- range = ((MODE) == HImode || (MODE) == QImode) \
- ? (arm_arch4 ? 256 : 4095) : 4096; \
+ /* XXX For ARM v4 we may be doing a sign-extend operation \
+ during the load, but that has a restricted addressing \
+ range and we are unable to tell here whether that is the \
+ case. To be safe we restrict all loads to that \
+ range. */ \
+ if (arm_arch4) \
+ range = (mode == HImode || mode == QImode) ? 256 : 4096; \
+ else if (mode == HImode) \
+ range = 4095; \
+ else \
+ range = 4096; \
if (code == CONST_INT && INTVAL (INDEX) < range \
&& INTVAL (INDEX) > -range) \
goto LABEL; \
Reply to: