Re: Robert Dewar's great article about the Strengths of Ada over other langauges in multiprocessing!

["(see below)" <yaldnif.w@blueyonder.co.uk>]

On 01/04/2008 10:02, in article wvbr8wzyymkw.fsf@astra06.norway.sun.com,
"Ole-Hjalmar Kristensen"
<ole-hjalmar.kristensen@substitute_employer_here.com> wrote:

>>>>>> "(b" == (see below) <yaldnif.w@blueyonder.co.uk> writes:
> 
> <snip>
> 
>     (b> I think the wording is trying to cover all the bases.
>     (b> One clue is the phrase "_full_ cache consistency".
>     (b> The caches certainly need to be consistent with respect to the
>     (b> protected data, even for protected functions and procedures,
>     (b> but only entries ensure global consistency and so provide
>     (b> synchronization of data that is not local to the protected object.
> 
> Yes, that seems reasonable.
> 
>>> Btw., I ran a simple test on a SPRAC multiprocessor with an entryless
>>> protected object containing a single integer versus an integer
>>> declared with pragma atomic, and as expected the pragma atomic
>>> solution was much (40x) faster.
> 
>     (b> Unfortunately, we can't usefully apply that pragma even
>     (b> to a pair of integers. (I don't mean the pair's components!)

Also, simply declaring a variable atomic does not in itself
ensure global consistency of view. For that you also need
to execute appropriate memory barrier operations for the
architecture. (I'm sure you know that.)
 
> Agreed, but you may able to cheat and pack a pair of integers into
> a 64-bit atomic, and a compare-and-swap is also much cheaper than a
> protected object it seems.

Yes, but this is completely implementation-dependent.
Not much above the semantic level of assembly.
I would take a lot of convincing that the performance
improvement was actually necessary.

-- 
Bill Findlay
<surname><forename> chez blueyonder.co.uk