https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88057
--- Comment #4 from David Binderman <dcb314 at hotmail dot com> ---
(In reply to Martin Liška from comment #3)
> I've briefly looked at the issue and I don't know a configuration that uses
> libdecnumber. So I'm leaving that..
I just checked my latest build of trunk gcc on x86_64 and I got:
$ fgrep -c libdecnumber mk.out
613
$
I think it would be sufficient to move local variable "buffer" out
to a wider scope.
Something like:
$ svn diff decCommon.c
Index: decCommon.c
===================================================================
--- decCommon.c (revision 266300)
+++ decCommon.c (working copy)
@@ -427,6 +427,8 @@
/* if exponent is >=emax may have to clamp, overflow, or fold-down */
if (num->exponent>DECEMAX-(DECPMAX-1)) { /* is edge case */
+ uByte buffer[ROUNDUP(DECPMAX+3, 4)]; /* [+3 allows uInt padding] */
+
/* printf("overflow checks...\n"); */
if (*ulsd==0 && ulsd==umsd) { /* have zero */
num->exponent=DECEMAX-(DECPMAX-1); /* clamp to max */
@@ -468,7 +470,6 @@
/* fold down needed; must copy to buffer in order to pad */
/* with zeros safely; fortunately this is not the worst case */
/* path because cannot have had a round */
- uByte buffer[ROUNDUP(DECPMAX+3, 4)]; /* [+3 allows uInt padding] */
uByte *s=umsd; /* source */
uByte *t=buffer; /* safe target */
uByte *tlsd=buffer+(ulsd-umsd)+shift; /* target LSD */
$
