An architecture for interpreted dynamic object-oriented languages

of the thesis has been supplied on the condition that anyone who consults it is understood to recognise that its copyright rests with its author and that no quotation from the thesis and no information derived from it may be published without the prior written consent of the author. This thesis may be made available for consultation within the University Library and may be photocopied or lent to other libraries for the purposes of consultation. Summary This thesis is concerned with the implementation of object-oriented dynamic programming languages based on bytecode interpretation. A new interpretive implementation architecture is proposed that meets the requirements of code and system portability, execution performance , static and dynamic memory eeciency as well as language interoperability. The diierent quality of the architecture compared to other virtual machine approaches is related to the key techniques developed within this work: (i) C embedded virtual machine code, (ii) indexed code threading, (iii) optimal virtual instruction ordering and (iv) quasi-inline method caching. C embedded virtual machine code refers to the representation of bytecodes as constant C arrays that are located in sharable text segments after compilation. Interoperability, application start-up and dynamic memory usage beneet from this representation. Indexed code threading addresses the performance problem with virtual instruction mapping (i.e. loading, decoding and invoking) by using a fast threaded instruction transfer. Unlike with standard code threading, virtual machine code remains compact and executable also with a non-threaded virtual machine emulator. A further performance boost is achieved with optimal virtual instruction ordering. This technique helps to cluster the native code implementing virtual instructions so that native instruction cache performance is increased. Finally, the eeciency problem involved with dynamic method lookup is alleviated with an inline caching scheme that is applicable with constant bytecode vectors. The scheme exploits type locality similar to polymorphic inline caching. However, dynamic memory is saved by avoiding redundant method entries and by being adaptable to generic function invocation which typically comes in waves with hot-spots on particular methods. A realization of the architecture is presented in form of an implementation of the dynamic object-oriented language EuLisp. The implementation demonstrates the feasibility and eeec-tiveness of the proposed architecture. The average performance increase with indexed code threading is 14% (P5) and 17% (MIPS). The average increase with optimal instruction ordering in the indexed threaded interpreter is 21% (P5) and 15% (MIPS). Sharable read-only data is increased on average by …