A note on hash linking

Abstract

In current machine designs, a machine address gives the user direct access to a single piece of information, namely the contents of that machine word. This note is based on the observation that it is often useful to associate additional information, with some (relatively few) address locations determined at run time, without the necessity of preallocating the storage at all possible such addresses. That is, it can be useful to have an effective extra bit, field, or address in some words without every word having to contain a bit (or bits) to mark this as a special case. The key idea is that this extra associated information can be found by a table search. Although it could be found by any search technique (e.g. linear, binary sorted, etc.), we suggest that an appropriate low overhead mechanism is to use hash search on a table in which the key is the address of the cell to be augmented.

An efficient representation for lazy constructors using 64-bit pointers

“…when data may only need a part of addressable memory [9,30], or when different addresses share a common prefix [49]. The technique has been applied successfully in Java virtual machines [1,39,46,47].…”

High-Level Language Implications of the Proposed IEEE Floating-Point Standard

“…It is feasible for the NaN to be a pointer into a storage area which would contain more information than could be stored immediately in the NaN. A run-time system using "hash links" [1] could be used if pointers are longer than the available space remaining in the datum (fraction part of the floating-point number).…”

“…For convenience, a set of names has been initialized so that the user need not know the particlar small numbers (1,2,4,8,16) This approach presents a problem in languages which try to prevent the manipulation of entry points as data. In the case of FORTRAN and other languages we have dealt with on the VAX UNIX system, it is solved by an abuse of the language.…”

Compact Encodings of List Structure

“…The key in the hash table is the address of the cell that should have contained the pointer. This idea is called "hash linking" by Bobrow [4]. (Since car and cdr could both have the escape value, one more bit needs to be hashed along with the address of the cell.…”

“…In Section 2 we described a use of Bobrow's [4] hash links as an escape mechanism. In this simple form, if the compact cdr (or car) field of cell x contains a particular escape codeword, the true value of cdr(x) (or car(x)) is found in a hash table, where x is the key.…”