[Bug rtl-optimization/46598] [4.4/4.5/4.6 Regression] Miscompiles computed goto

["rguenth at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46598

--- Comment #10 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-11-23 14:38:00 UTC ---
(In reply to comment #9)
> Looks just like bogus inline asm.  For rdtsc which sets 32-bits in %eax and
> 32-bits in %edx, for 32-bit code you want "=A" (long_long_variable) and
> for 64-bit code you need "=a" (one_int_variable), "=d" (another_int_variable)
> Otherwise you are lying on what the insn does.
> "=A" with an lvalue which is smaller or equal to word size means allocate it
> either in ax or in dx.  Only with double word sized lvalue it means both ax and
> dx.

The split result for 64bit code is non-intuitive - I would have expected
that "=A" (int64_t variable) always works.

We should issue a warning here - I suspect the broken variants never make
sense.