[Bug optimization/11655] New: [3.3 regression] long long/double -O0 produces incorrect code

[nick at ilm dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11655

           Summary: [3.3 regression] long long/double -O0 produces incorrect
                    code
           Product: gcc
           Version: 3.3.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: nick at ilm dot com
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu

This is a cut down version of some fp classification code.  The odd
thing is that it gets the correct answer with -O1 or with -mcpu=pentium4
-mfpmath=sse, but is incorrect at -O0.  Using 3.2.3 -O0 and -O1 both work.

the following code should print out 4094:

----
extern int printf(const char *,...);

typedef union { double d; unsigned long long ull; } doublebits;

inline int func(double x) {
    doublebits bits;
    bits.d = x;
    printf("%llu\n", bits.ull >> 51);
    return 0;
}

int main(int argc, const char **argv) {
  doublebits snan;
  snan.ull = 0x7ff0000000000001ULL;
  func(snan.d);
  return 0;
}
----

> /dept/rnd/vendor/gcc-3.3.1-20030720/bin/gcc -O0 longlong.c -o longlong
> ./longlong
4095
> /dept/rnd/vendor/gcc-3.3.1-20030720/bin/gcc -O1 longlong.c -o longlong
> ./longlong
4094
> /dept/rnd/vendor/gcc-3.3.1-20030720/bin/gcc -O0 -march=pentium4 -mfpmath=sse
longlong.c -o longlong
> ./longlong 
4094
> /dept/rnd/vendor/gcc-3.2.3/bin/gcc -O0 longlong.c -o longlong
> ./longlong
4094