Abstract. We analyse the validity of several common program transformations in multi-threaded Java, as defined by the Java Memory Model (JMM) section of Chapter 17 of the Java Language Specification. The main design goal of the JMM was to allow as many optimisations as possible. However, we find that commonly used optimisations, such as common subexpression elimination, can introduce new behaviours and so are invalid for Java. In this paper, we describe several kinds of transformations and explain the problems with a number of counterexamples. More positively, we also examine some valid transformations, and prove their validity. Our study contributes to the understanding of the JMM, and has the practical impact of revealing some cases where the Sun Hotspot JVM does not comply with the Java Memory Model. 1

A central problem in extending the von Neumann architecture to petaflop computers with millions of hardware threads and with a shared memory is defining the memory model [Lam79,AG95,APP99]. Such a model must specify the behavior of concurrent (conditional) reads and writes to the same memory locations. We present a simple, general framework for the specification of memory models based on an abstract machine that uses sets of (interdependent) order and value constraints to communicate between threads and main memory. The separation of order constraints allows a parametric treatment of different order consistency models such as sequential consistency, location consistency, happens-before consistency (HB-consistency), etc. The use of value constraints allows a simple formulation of the semantics of concurrent dependent reads and writes in the presence of look-ahead and speculative computation. Memory is permitted to specify exactly those linkings (mappings from read