Re: Loops and parallel execution

[Georg Bauhaus <rm.dash-bauhaus@futureapps.de>]

On 25.01.11 17:37, Dmitry A. Kazakov wrote:
> On Tue, 25 Jan 2011 16:40:38 +0100, Georg Bauhaus wrote:

>> Can this be borrowed from HPF (IIUC)?
>> Is pragma Pure (P) sufficient to signal reentrance?
> 
> No, it is not sufficient because it is wrong. P cannot be pure because all
> instances of P.Op must be synchronized at the end of the "loop." 

I was thinking of relatively small things, starting from this
observation:

If a compiler detects two independent paths inside a loop,
it will generate sequences of instructions that, while written
sequentially, will be executed in parallel by the processor
because the processor can do the corresponding distribution
across the many register words. Generalize that.

The Paraffin library just posted by Brad Moore looks like
it will resolve all other issues nicely! :-)

On an AS-IF basis, I thought of a tasking profile much < Ravenscar:
Other than exiting with a value, each P.Op (K) has no need for
communication. Each P.Op (K) is associated with a result object,
like this

   compiler_type Result_Object is limited record
       Ready : Boolean_Slot;  -- a CAS bit?
       Result : Value_Type;
   end record;

The idea of employing Pure or something similar was to have
the compiler prevent each P.Op (K) from meddling in P.Op (J)'s
affairs, J /= K, as with protected objects, again on an
AS-IF basis, as much as possible.