Direct execution of lisp on a list_directed architecture

We have defined a direct-execution model dedicated to non-numerical processing which is based upon an internal representation of source programs derived from LISP. This model provides good support for both sophisticated editing (syntactical parsing, tree manipulation, pretty-printing, ...) of conventional languages and artificial intelligence languages. A high level microprogramming language (LEM) was designed to write the interpreters and the editors. A hardware processor was built and a LISP interpreter, microprogrammed in LEM, has been operational since September 1980. First, the influence of LISP on the LEM language and the architecture is discussed. At the LEM level, we will see that LISP has prompted the control constructs and the access functions to the tree-structured internal form. As for the architecture, we present the hardware implementation of a special garbage collector based upon reference counters. In turn, the machine has influenced the implementation of LISP. We present here the structure of our LISP interpreter and we give evaluation measures dealing with size, development effort, speed; they prove that programming in LEM is easy, short to debug and very concise. Moreover, the speed of our LISP interpreter confirms that the architecture is very efficient for symbolic processing.