SPEC CPU2006 tests 464.h264ref and 482.sphinx3 fail to build on powerpc64-linux with "-m64 -lto -fsigned-char".  For some reason we've added the option -fsigned-char to the SPEC config files for those two tests; without it both tests fail at runtime for -m64, but it doesn't affect the results for -m32.  The failure to build with -flto might be due to some problem with -fsigned-char on powerpc64, rather than a problem with LTO.

Here's a minimized testcase:

---------------------
extern char *a;
extern int f;

void
foo (void)
{
  f = (*a != '-');
} 
---------------------

And here's what happens:

elm3b149% /home/janis/tools/gcc-trunk-anonsvn/bin/gcc -m64 -flto -fsigned-char bug.c -c
elm3b149% /home/janis/tools/gcc-trunk-anonsvn/bin/gcc -m64 -flto -fsigned-char bug.o -shared -o x.so
In function ‘foo’:
lto1: error: type mismatch in indirect reference
<unnamed-signed:8>
<unnamed-unsigned:8>

D.1228_2 = *a.0_1;

lto1: internal compiler error: verify_stmts failed
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.
lto-wrapper: /home/janis/tools/gcc-trunk-anonsvn/bin/gcc returned 1 exit status
collect2: lto-wrapper returned 1 exit status

If the type of 'a' is signed char then everything works fine.