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

[Matthew Dobson <colpatch@us.ibm.com>]

On Thu, 2004-03-18 at 17:56, Paul Jackson wrote:
> > These (cpumask_t/nodemask_t) are nice because they are
> > optimized for edge cases (UP for cpumask_t and Non-NUMA for nodemask_t)
> > as well as for long mask cases (passing by structs reference). 
> 
> When I looked at the assembly code generated on my one lung i386 box for
> native gcc 3.3.2, it looked pretty good (to my untrained eye) using a
> struct of an array of unsigned longs, both for the single unsigned long
> (<= 32 bits) and multiple unsigned long cases.

The code you wrote, or my patch?

> Except for the sparc64 guys and their friends who disparage passing
> structs on the stack, I conjecture that the single implementation of a
> struct of an array of unsigned longs is nearly ideal for all
> architectures.
> 
> ... go ahead ... prove me wrong.  It probably won't be hard ;).

Sounds like a good idea.  We certainly shouldn't be passing huge masks
on the stack, but for small masks like, i dunno, <= 4 ULs (the same
optimization Bill's code makes) it's no problem.

The thing about code specific to node and cpu masks is that we *know*
what the masks we're manipulating are used for, and that lets us do
things like not letting callers set bits other than 0 on UP cpumasks, or
throwing a BUG when they do.  Or optimizing first_node() to be smarter
than just calling find_first_set() on the passed in mask by just
checking whether bit 0 is set.

-Matt