Re: Toy computational "benchmark" in Ada (new blog post)

[Paul Rubin <no.email@nospam.invalid>]

David Trudgett <dktrudgett@gmail.com> writes:
>> Yes I'll see if I can rework the C++ code to not launch new threads at
>> each run.

> It doesn't matter if you don't time that part of the execution. But
> I'm guessing that would be hard in C++.

I used the C++ futures/async feature which in principle could use a
thread pool, but in GCC 6 I'm told that it launches a new thread for
each async.  Otherwise I used what I think is the same strategy as the
Ada code:

      for Worker in Worker_Range loop
         Tasks (Worker).Run
           (Start_Idx  => Idx_Ranges (Worker, Low),
            Finish_Idx => Idx_Ranges (Worker, High));
      end loop;

It looks like you have a static array of worker threads, and I'm
guessing that the .Run message some kind of event to a worker to tell it
to run the task body.  I didn't try to figure it all out though.  Does
Ada automatically create an event loop in each task to receive those
messages?  If yes, that's pretty cool.

Is there a request queue between the main program and each worker?  What
happens if it gets too large?  Is there a way to share a single queue
between a bunch of workers?

I read "Ada Distilled" some time back but iirc it didn't say much about
this.  I've been wanting to get a more detailed book.