Re: Proposal for reference annotations

[Richard Henderson <rth at redhat dot com>]

On Wed, Jun 23, 2004 at 04:11:37PM -0400, Richard Kenner wrote:
> So my proposal is that we add an annotation mechanism for ARRAY_REF,
> ARRAY_RANGE_REF, and COMPONENT_REF.  Because in most cases these fields will
> be constants and usually the same for many nodes, we set up a hash table
> mechanism like MEM_ATTRS and have the rule that you need to make a new one in
> order to modify one.

Where are you going to put this annotation?  If on the node, then
you're no different than the current state that Mark hates.  If in
a look-aside datastructure, then you're going to require *every*
examination of a reference within the optimizer to do a search?

Further, if you share these, then putting the lower bound, stride,
offset, whathaveyou in SSA form will be impossible[1].  You'll have
in effect EXACTLY reconstructed the situation with the DECL/TYPE_SIZE
data that caused us to want to add the extra ref arguments in the
first place.

I see no advantage to be gained from this proposal.


r~


[1] get_expr_operands finds addresses of operands, and optimizers assume
that they can replace operands just by storing to the address.  I suppose
you could share iff everything is constant.