[Beowulf] Stroustrup regarding multicore

Perry E. Metzger perry at piermont.com
Tue Aug 26 16:35:30 PDT 2008


"Robert G. Brown" <rgb at phy.duke.edu> writes:
> On Tue, 26 Aug 2008, Perry E. Metzger wrote:
>> "Robert G. Brown" <rgb at phy.duke.edu> writes:
>>> I think that matrix allocation per se is contiguous because of how and
>>> where it occurs at the compiler level.  Immediate matrices allocated
>>> like a[10][10] are just translated into displacements relative to the
>>> stack pointer, are they not?
>>
>> It depends. You're not guaranteed per se that they'll be allocated on
>> the stack, though in practice they are.
>
> Unless they are really large, I imagine.  I'm not sure what the stack
> limits are (I don't code to ever stress them) but I'm sure there are
> some.

In the kernel, stack limits are very tight for a variety of
reasons. In userland, there isn't in theory much of a limit other than
available RAM (and of course address space), though in practice the
ulimits are set pretty low. That's easily fixed, naturally.

Still, the right decision on what to use usually depends on what you
are doing, as always.

>>> Global/external matrices at the a[10][10]
>>> are prespecified in the data segment of the program,
>>
>> Depending on the particular OS. Certainly that's the usual model.
>
> What OS's do it some other way?  I'm just curious.  Is this a kernel
> feature or a compiler feature?

Kernel. Compilers have to live with what the OS model for a process
looks like. (Actually, more to the point, the linker and back end
tools have to live with it, though in practice that also means the
compiler has to know a little.)

There are a load of weird things out there that don't look like
this. Have a gander at Palm's operating system for their handhelds
sometime -- or the original Mac OS.

Perry
-- 
Perry E. Metzger		perry at piermont.com



More information about the Beowulf mailing list