[cpp-threads] Review comments on N2176 WRT dependency ordering

[Paul E. McKenney]

On Tue, Apr 17, 2007 at 02:59:11PM +0300, Peter Dimov wrote:
> Peter Dimov wrote:
> 
> >Are they important enough? We can add a minimalistic index-based
> >primitive along the lines of:
> >
> >template<class T, class U> inline U* atomic_load_index( T const * pi,
> >U * base );
> 
> Or maybe you had in mind the other kind of indexing that we can support by 
> adding an offset argument to atomic_load_address (defaulted to 0)? 

That might work in some cases.  A common expandable-hash-table case
looks something like this:

	struct foo {
		int size;
		struct bar *array[0]; /* or [1] if no zero-length arrays */
	};

	struct foo *head = NULL;
	int current_index;

-	Creation of table

	spin_lock(&mylock);
	p = (struct foo *)malloc(sizeof(*p) + N * sizeof(struct bar *));
	if (p == NULL) {
		die();
	}
	p->size = N;
	store_release_fence();
	head = p;
	spin_unlock(&mylock);

-	Addition of "current" element i

	spin_lock(&mylock);
	q = new struct bar;
	/* initialize q */
	if (head->array[i] != NULL) {
		die();
	}
	head->array[i] = q;
	store_release_fence();
	current_index = i;
	spin_unlock(&mylock);


-	Accessing "current" element

	p = atomic_load_address(head);
	q = p->array[atomic_load_address(current_index)];

Unless I am misunderstanding your proposal, the two separate
atomic_load_address()es are needed in this case.

							Thanx, Paul