Robert Haas <robertmhaas@gmail.com> writes:
> On Mon, Jun 29, 2015 at 10:32 PM, Andres Freund <andres@anarazel.de> wrote:
>> You removed a volatile at the same time, and volatile on IA64 has
>> acquire/release semantics.
> Can you explain what you mean by volatile having acquire/release
> semantics? I don't see how volatile can create a CPU barrier, but I'm
> guessing that it somehow can and that you're about to enlighten me.
It's late and I'm tired, but: gcc (and, apparently, icc) treats accesses
to volatile-qualified objects as cues to emit .acq or .rel memory ordering
qualifiers on IA64 instructions, per the comments in s_lock.h. I have not
seen any documentation stating specifically that aCC does the same, but
the buildfarm evidence is pretty clear that the 9.4 IA64-non-gcc version
of S_UNLOCK worked and the up-to-now-9.5 version does not. So personally,
I would be inclined to put back the volatile qualifier, independently of
any fooling around with _Asm_double_magic_xyzzy calls. Or to put it
differently: where is the evidence that removing the volatile qual is a
good idea?
regards, tom lane
