Re: PAR (Was: Embedded languages based on early Ada)

["Dr. Adrian Wrigley" <amtw@linuxchip.demon.co.uk.uk.uk>]

On Wed, 07 Mar 2007 18:26:08 +0000, Ray Blaak wrote:

> "Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> writes:
>> Hmm, but checking is really the key issue here. I fail to see it in bare
>> PAR. It is just absent there. What would be the semantics of:
>> 
>> declare
>>    I : Integer := 0;
>> begin in parallel
>>    I := I + 1;
>>    I := I - 1;
>> end;  -- What is the postcondition here?
> 
> That would be bad programming, just as if you expressed it equivalently with
> expicit task bodies.

I think the issue is whether this is permitted, and whether compilers
can generally give a warning in such cases.  I think it's like
uninitialised variables - compilers often spot these and warn.
There are two problems in this case - reading and writing the same
variable in different statements, and writing to the same variable
twice. 

VHDL concurrent assignments can handle this multiple assignment very
nicely using signal resolution functions.  It would be nice if
software parallel constructs could do parallel combination,
such as "and"ing many boolean variables together, or computing
integer sums efficiently.  This would probably best be done by
an optimiser spotting specific patterns of use in serial code.

Another problem not mentioned is memory allocation.  Once you
fire off hundreds of threads, memory use may escalate unexpectedly.
Holding back some memory intensive threads may help, provided you
don't create new threads using up the last available store.
--
Adrian