Supporting object accesses in a Java processor

Abstract

Due to Java's object-based nature and support for garbage collection, efficient object manipulation and relocation are critical to the execution speed of Java code. Java Virtual Machine implementations that utilise handle representation such as Sun's Java Development Kit 1.1 enable efficient object relocation at the cost of an additional indirection for each object access. The direct address object representations such as those used in CACAO and NET compiler eliminate the indirection overhead, but update during object relocation is complex. A virtual address object cache that reduces the indirection overhead while maintaining the efficiency of object relocation is proposed. The objects in the virtual address cache are addressed directly using the object reference and field offset pair. This eliminates the indirection overhead and off-set addition overhead associated with the handle representation model. A hardware object table that maintains the handles is used to obtain the actual object location on a virtual address cache miss. The performance of the virtual address cache is analysed using various Java programs, and is found to reduce 1.5 cycles per object access on an average as compared to the handle representation model for the various benchmarks studied.