Re: better atomics - v0.5

On Thu, Jun 26, 2014 at 7:21 AM, Andres Freund <andres(at)2ndquadrant(dot)com> wrote:
> On 2014-06-25 20:22:31 -0400, Robert Haas wrote:
>> On Wed, Jun 25, 2014 at 5:42 PM, Heikki Linnakangas
>> <hlinnakangas(at)vmware(dot)com> wrote:
>> > I think having a separate file for each architecture is nice. I totally
>> > agree that they don't belong in src/include/storage, though. s_lock.h has
>> > always been misplaced there, but we've let it be for historical reasons, but
>> > now that we're adding a dozen new files, it's time to move them out.
>>
>> I find the current organization pretty confusing, but maybe that could
>> be solved by better documentation of what's supposed to go in each
>> architecture or compiler-dependent file.
>
> The idea is that first a architecture specific file (atomics-arch-*.h)
> is included. That file can provide a (partial) implementation for the
> specific architecture. Or it can do pretty much nothing.
>
> After that a compiler specific file is included
> (atomics-generic-*.h). If atomics aren't yet implemented that can
> provide an intrinsics based implementation if the compiler version has
> support for it. At the very least a compiler barrier should be provided.
>
> After that the spinlock based fallback implementation
> (atomics-fallback.h) provides atomics and barriers if not yet
> available. By here we're sure that nothing else will provide them.
>
> Then we can provide operations (atomics-generic.h) that build ontop of
> the provided functions. E.g. implement _sub, _and et al.
>
> I'll include some more of that explanation in the header.

I get the general principle, but I think it would be good to have one
place that says something like:

Each architecture must provide A and B, may provide both or neither of
C and D, and may also any or all of E, F, and G.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company