and uses _BM(32) in several places. For some reason clang doesn't
realize that the ~(-(1UL << (x))) branch is dead and warns that we
shift a 32-bit integer by 32 bits. Arguably a bug in clang, but I
noticed that the i386 code simply uses:
#define _BM(x) (~(-(1ULL << (x))))
which makes the problem go away. I also changed the type of
reloc_target_bitmask from 'int' to 'long' to match the i386 code.