Abstract

Programs written in a higher level language are often less.efficient than equivalent assembly language programs, because they cannot exploit known invariances and optimizations which would violate the strict semantics of the target language. Moving code from Lisp into the kernel has been a traditional way of improving the performance of Lisp systems. Substantial sections of the PDP-10 implementation of Interlisp, for example, are in machine code for this reason. When a large proportion of AltoLisp was moved from Bcpl into Lisp in order to improve memory utilization and aid modification, the speed of the system decreased by nearly a factor of three [Deutsch, 1978]. Thus, to improve DoradoLisp performance, we first looked for Lisp-coded sections of the system that could be incorporated into the Bcpl'kerne[,.However,'~,e soon discovered that the poor performance was due more to the design of the algorithms in the kernel than to the language in which they

Citations

117	An Efficient, Incremental Automatic Garbage Collector - Deutsch, Bobrow - 1976
12	Compact encodings of list structure - Bobrow, Clark - 1979
7	A Lisp machine with very compact programs - Deutsch - 1973
7	The Maxc systems - Fiala - 1978
6	Experience with a microprogrammed Interlisp system - Deutsch - 1978
2	Local Optimization in a Compiler for Stack-based Lisp Machines - Masinter, Deutsch - 1980
1	A Processor for a High-Peformance Personal Computer - Lampson, Pier - 1980
1	The Interlisp virtual machine specification. Xerox PARC report CSL-76-5 - Moore - 1976