Re: [PATCH] Introduce nodemask_t ADT [0/7]

[William Lee Irwin III <wli@holomorphy.com>]

On Mon, Mar 22, 2004 at 06:39:18PM -0800, Paul Jackson wrote:
> From this I conjecture that I can provide a single call:
>     cpumask_and(cpumask_t d, cpumask_t s1, cpumask_t s2);
> that works on both normal (1 to 32 cpu) systems and on big iron systems,
> with traditional 'C' pass by value semantics, all derived from a single
> mask type that works for both node and cpu masks.
> The one sticky point evident to me so far would be if some generic code
> were passing a cpumask_t across a function call boundary, and needed to
> be optimum for both small and sparc64 - one would want to pass by value,
> the other would want to pass a pointer to the cpumask.
> This is not your fathers 'C'.  The compile time inlining and
> optimization provided by gcc enables it to do a lot more than Dennis
> Ritchie's original C compiler that I learned on.

gcc flat out miscompiled such inlines last I checked (Zwane shipped the
bugreport IIRC). Either this kind of good behavior is not universally
observable or a miracle occurred and gcc's codegen went from incorrect
to 1980's (fscking patents).

Anyhow, this was also an observation of the code effectively made in
isolation; uninlining and other catastrophes do happen.

If people really thinks this works and/or don't care when it doesn't,
go for it. Last time I heard they did; who knows, the answer may be
different this time.


-- wli