Abstract:
Parallel and distributed programs often have hardware/problem specific optimizations for improving quality of the program such as efficiency and robustness. Those optimizations, unfortunately, degrade portability and re-usability as they are intertwined with the original algorithm description. Reflective languages, which provide the application programmer extensible and abstract implementation of the language, can describe such optimizations as extensions to the language. The separation of optimization descriptions gains portability and re-usability of both application programs and optimizations. However, the interpretive execution model of reflective languages imposes a large amount of performance overhead, which sometimes outweighs benefits of optimizations. Previous reflective languages prohibit some of operations being modified via reflection, so as to reduce the amount of interpretation overhead. The imperfection of this approach is that it still leaves a considerable amount of overhead, and it yields less flexible, unclear reflective architecture. This dissertation investigates design and compilation framework of meta-interpreters and meta-objects in an object-oriented concurrent language ABCL/R3. By using partial evaluation to compile reflective programs, ABCL/R3 achieves flexible and lucid reflective architecture and efficient execution at the same time. We design full-fledged meta-interpreters by examining several concurrent programming examples. A newly proposed delegation mechanism enables to define modular and scope controlled extensions to meta-interpreters. We design meta-objects by exploiting the notion of reader/writer methods in a concurrent object-oriented language Schematic, so that they can be effectively partially evaluated. The compilation frameworks of meta-interpreters and meta-objects basically translate concurrent object definitions into a sequential program, then apply partial evaluator for a sequential language, and generates a program in a (non-reflective) concurrent object-oriented language, in which base-level and meta-level objects are collapsed to single level objects. The effciency of generated programs is demonstrated by several benchmark programs, in which our compiler exhibits performance close to non-reflective languages.
Reference:
Architecture Design and Compilation Techniques Using Partial Evaluation in Reflective Concurrent Object-Oriented Languages (Hidehiko Masuhara), PhD thesis, Department of Information Science, Graduate School of Science, The University of Tokyo, 1999. (Supervisor: Akinori Yonezawa)
Bibtex Entry:
@phdthesis{masuhara99phd,
pdf = {phdthesis.pdf},
doi = {10.11501/3163477},
url = {https://www.is.s.u-tokyo.ac.jp/},
author = {Hidehiko Masuhara},
title = {Architecture Design and Compilation Techniques Using
Partial Evaluation in Reflective Concurrent
Object-Oriented Languages},
school = {Department of Information Science, Graduate School of
Science, The {University} of {Tokyo}},
year = 1999,
optaddress = {},
month = jan,
opttype = {},
optnote = {},
optaffil = {},
optkeywords = {},
optserialno = {},
optannote = {},
keywords = {ABCL/R},
note = {Supervisor: Akinori Yonezawa},
abstract = {Parallel and distributed programs often have hardware/problem specific optimizations for improving quality of the program such as efficiency and robustness. Those optimizations, unfortunately, degrade portability and re-usability as they are intertwined with the original algorithm description. Reflective languages, which provide the application programmer extensible and abstract implementation of the language, can describe such optimizations as extensions to the language. The separation of optimization descriptions gains portability and re-usability of both application programs and optimizations. However, the interpretive execution model of reflective languages imposes a large amount of performance overhead, which sometimes outweighs benefits of optimizations. Previous reflective languages prohibit some of operations being modified via reflection, so as to reduce the amount of interpretation overhead. The imperfection of this approach is that it still leaves a considerable amount of overhead, and it yields less flexible, unclear reflective architecture.
This dissertation investigates design and compilation framework of meta-interpreters and meta-objects in an object-oriented concurrent language ABCL/R3. By using partial evaluation to compile reflective programs, ABCL/R3 achieves flexible and lucid reflective architecture and efficient execution at the same time. We design full-fledged meta-interpreters by examining several concurrent programming examples. A newly proposed delegation mechanism enables to define modular and scope controlled extensions to meta-interpreters. We design meta-objects by exploiting the notion of reader/writer methods in a concurrent object-oriented language Schematic, so that they can be effectively partially evaluated. The compilation frameworks of meta-interpreters and meta-objects basically translate concurrent object definitions into a sequential program, then apply partial evaluator for a sequential language, and generates a program in a (non-reflective) concurrent object-oriented language, in which base-level and meta-level objects are collapsed to single level objects. The effciency of generated programs is demonstrated by several benchmark programs, in which our compiler exhibits performance close to non-reflective languages.}
}