Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: Branch & bound for not everywhere defiend constraints



Dan

On 8 Sep 2009, at 04:53, Dan Zuras Intervals wrote:
	So long as we specify that a decorated interval consist
	of an 'undecorated' interval together with an associated
	decoration, we have said nothing about how that decoration
	appears or how it is associated with the interval.

	An adequately clever implementation might choose to store
	the interval part in one place & the decoration somewhere
	else & count on a similarly smart compiler to maintain
	the association.  So, for example, arrays of intervals
	parts could be packed up efficiently on power-of-two
	boundaries with arrays of their decorations stored
	elsewhere (perhaps ALSO on smaller power-of-two
	boundaries).

	Such an approach has its merits (it eliminates all your
	valid complaints) & its costs (the compiler has to keep
	track of things & the hardware assist needs an extra
	address path).

	But if our definition of decorated intervals is kept as
	an abstraction, this sort of thing is an implementation
	possibility.

Thank you for pointing this out. I had wondered if it was a reasonable possibility, and a way for composite data structures of funny widths to be stored and accessed efficiently. Other low-level experts: do you agree?

John