Fast subtype checking in the HotSpot JVM

Abstract

We present the fast subtype checking implemented in Sun's HotSpot JVM. Subtype checks occur when a program wishes to know if class S implements class T, where S and T are not both known at compile-time. Large Java programs will make millions or even billions of such checks, hence a fast check is essential. In actual benchmark runs our technique performs complete subtype checks in 3 instructions (and only 1 memory reference) essentially all the time. In rare instances it reverts to a slower array scan. Memory usage is moderate (11 words per class) and can be traded off for time. Class loading does not require recomputing any data structures associated with subtype checking.

Empirical assessment of object-oriented implementations with multiple inheritance and static typing

“…For instance, with JAVA interfaces, the underlying technique could be PH or IC and the method table would cache the interface ID and the group offset of the last succeeding access [Alpern et al 2001a,b, Click andRose 2002]. Of course this cache might be used for any table-based subtyping technique and for all three mechanisms, at the expense of caching three data, namely class ID and method and attribute group offsets.…”

Perfect hashing as an almost perfect subtype test

“…Despite many works from the beginning of object-oriented programming [Schubert et al 1983;Agrawal et al 1989;Aït-Kaci et al 1989;Cohen 1991;Caseau 1993;Habib and Nourine 1994;Habib et al 1995;Habib et al 1997;Vitek et al 1997;Alpern et al 2001b;Raynaud and Thierry 2001;Gil and Zibin 2005;Click and Rose 2002;Palacz and Vitek 2003;Alavi et al 2008], general and efficient implementation of subtype checks remains an open issue.…”

“…This somewhat amounts to padding each method table, instead of padding only the single-or the very few-memory areas dedicated to method tables. Click and Rose [2002] attribute the technique to Pfister and Templ [1991]. However, the statistics in Section 5 show that, on the same aforementioned benchmarks, the maximum superclass number may be 5-fold greater than its average and, on most benchmarks, the maximum is more than twice the average (Table III).…”

“…Cohen's test is commonly used in the implementation of Java virtual machines for class subtyping tests [Alpern et al 2001b;Click and Rose 2002]. However, it is often used in a non-optimal way, with uniform-size separate tables (Variant 2, page 11).…”

“…For instance, according to [Palacz and Vitek 2003], the Sun Microsystems Research VM uses a simple linear search improved with a cache in the source type memorizing the last target type for which the test succeeded. The technique of Click and Rose [2002], which is considered to be very efficient, improves this by combining it with Cohen's test. Of course this cache improvement might serve for any table-based subtyping technique.…”