Re: [Sbcl-help] Signaling mechanism for thread termination

[<me...@ho...>]

On Monday 16 January 2006 03:25, David L. Rager wrote:
> Hello All,
>
> This is my first post to this mailing list, so thank you in advance
> for your time and thoughts.
>
> I am working on a parallelization library for our program ACL2
> (http://www.cs.utexas.edu/users/moore/acl2/).  ACL2 builds on several
> LISPs, including SBCL and OpenMCL.  I am trying to determine which of
> our parallelization features we can port to SBCL.  The OpenMCL
> maintainers were kind enough to provide a feature to the
> "process-run-function" macro that signals a semaphore upon the death
> of a thread.  This allows us to terminate child threads that have had
> zero CPU time (and thus have not entered any of the ACL2 level code)
> and still have them report to their parents that they are "done."=20
> Even if a parent does not need a child's result, the parent is
> required to wait for the child to announce that it is done before the
> parent is allowed to return its result.  Amongst other things, this
> requirement helps us maintain stability in our system.

So we need JOIN-THREAD[S]. The OpenMCL PROCESS-RUN-FUNCTION macro should=20
not be hard to port. Post it here if you have trouble with it.

> Is there a "signal something upon death" feature like this in SBCL?

No.

> I understand that SBCL uses condition variables, not semaphores, but
> I have built a semaphore wrapper around the condition variables and
> mutexes so that my code is more uniform across distributions.

There is an unexported semaphore implementation in sb-thread that's=20
built the same way, too.

>
> Thank you,
> David L. Rager

Cheers, G=C3=A1bor