Re: we should wait for 1sided implementations

David C. DiNucci (dinucci@nas.nasa.gov)
Mon, 3 Jun 1996 17:24:59 -0700

dennis@rats-b.nosc.mil (Dennis Cottel) writes:
> Eric says,
> > But in the specific case of data alignment I still don't see the
> > problem. Let's say just for the sake of argument that Hell freezes over
> > and we actually do vote to require all data chunks to be 64-bit aligned.
>
> I've been resisting commenting on this discussion since my experience is
> entirely with message passing rather than get/put style communication.
> But from a user's point of view, requiring all the put/get operations to
> be X-bit aligned (for any X) is a terrible idea. What about the new
> 128-bit chip architectures which will no doubt be announced soon? What
> about machines with 256-bit cache lines?

C programmers hardly ever need to worry about whether data is X-bit aligned,
but it usually must be. One of the reasons that they don't need to worry
is that C ensures that the memory returned by dynamic allocation routines
is "suitably aligned for any use" (according to the "malloc" man page on
my machine). If dynamic allocation and communication are combined (as they
are in CDS1 and the Reactive Kernel and Leslie Hart's proposal), then any
desired alignment for simple transfer can be had, essentially for free.

> > This is trivial for both users and implementations to check.
>
> Fooey. I shouldn't have to.
> --Dennis

I agree.

-Dave
===============================================================================
David C. DiNucci | MRJ, Inc., Rsrch Scntst |USMail: NASA Ames Rsrch Ctr
dinucci@nas.nasa.gov| NAS (Num. Aerospace Sim.)| M/S T27A-2
(415)604-4430 | Parallel Tools Group | Moffett Field, CA 94035