Re: [Bug-apl] Parallel progress

[Blake McBride]

Dear Juergen,

Very exciting!  I think being able to take advantage of multiple processors is what will make the difference for APL.  

Thanks!

Blake

On Fri, Sep 12, 2014 at 5:17 AM, Juergen Sauermann <address@hidden> wrote:

Hi,

there are some good news from the parallel front. I have replaced
the semaphores that control the fork() and join() off threads by a new,
busy-wait based, mechanism.

That has reduced the fork() and join() times considerably: from the 20,000 cycle ballpark to
less than 1000 cycles (the numbers still vary a lot due to caching). As a consequence, the
break-even point (where parallel is faster than sequential) for e.g. Z←A+B has dropped from
⍴,Z of some 1000s to some 50s.

I will now start to integrate the Parallel.cc prototype (see attachment) into GNU APL...

/// Jürgen