http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53284
--- Comment #2 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot
Uni-Bielefeld.DE> 2012-05-09 11:45:10 UTC ---
> --- Comment #1 from Richard Henderson <rth at gcc dot gnu.org> 2012-05-08
> 22:39:35 UTC ---
> Can you investigate why configure decides that __atomic_compare_exchange_8
> is implemented inline? That it isn't inline is obvious from the recursion.
That was the critical hint, thanks. Comparing i386-pc-solaris2.10 and
i686-unknown-linux-gnu config.log, I see that in the Linux case the test
is run as a link test and includes -fno-sync-libcalls $XCFLAGS, while in
the Solaris case, it's an compile test and lacks all of them.
> Was the configure change to CFLAGS in fact wrong?
No, that's unrelated.
The following patch works for me and lets all tests in the above
configuration pass. I've chosen to add to CFLAGS rather than adding
$XCFLAGS since otherwise we loose -fno-sync-libcalls.
Ok for mainline?
Thanks.
Rainer
2012-05-09 Rainer Orth <[email protected]>
* acinclude.m4 (LIBAT_TEST_ATOMIC_BUILTIN): Add -O0 -S to CFLAGS
instead of overriding.
* configure: Regenerate.
diff --git a/libatomic/acinclude.m4 b/libatomic/acinclude.m4
--- a/libatomic/acinclude.m4
+++ b/libatomic/acinclude.m4
@@ -67,7 +67,7 @@ AC_DEFUN([LIBAT_TEST_ATOMIC_BUILTIN],[
else
old_CFLAGS="$CFLAGS"
# Compile unoptimized.
- CFLAGS='-O0 -S'
+ CFLAGS="$CFLAGS -O0 -S"
if AC_TRY_EVAL(ac_compile); then
if grep __atomic_ conftest.s >/dev/null 2>&1 ; then
eval $2=no
