robert@techfak.uni-bielefeld.de (Robert Giegerich) (02/09/91)
Announcement Workshop on Code Generation International Conference and Research Center for Computer Science Schloss Dagstuhl, May 20-24, 1991 Current issues in code generation Current tools for code generation are built around code selection based on pattern matching. We now have a good understanding and efficient implementations for pattern matching. It is still a problem of current research how to integrate other subtasks of code generation with the pattern based approach in a satisfactory way. Meanwhile, RISC architectures tend to make code selection (by itself) trivial, while creating new tasks of effectively using pipelined CPUs. Instruction scheduling and delayed branch optimization are problems reoccuring with each new RISC processor, for which generally accepted models are not available. Another source of CPU parallelism is the presence of multiple functional units. VLIW-architectures require non-local code rearrangements, and introduce new complexity by the interaction between scheduling and global and local register allocation. In the implementation of non-imperative languages, abstract machines are used for code generation. Although these machines are different for logical, functional or object-oriented machines, many problems reoccur, like register allocation or pattern matching. While some of these abstract machines are being implemented in hardware, in most cases, concrete code for conventional machines must be produced from the abstract machine code. Generally, the sensitivity to the question of reliable systems is increasing. In hardware synthesis as well as in compiler construction, first attempts have been made to use automated proof systems for verification. The idea of a fully verified hard- and software system is still a challenging goal. What is the impact of code generation work on compiler and hardware design? Techniques used in code generation may prove useful also in other compilation phases. Retargettability can only be achieved based on an appropriate, modular compiler structure. Last not least, research on code generation should support hardware design by a classification of cpu properties that make effective cpu utilization feasible at a reasonable price. Goals of the Workshop on code generation While new problems and solutions have evolved in recent years, the work in the areas outlined above has proceeded in a rather isolated way. Recent conferences have not provided a forum to bring together a significant number of researchers in the field. This workshop is intended to provide such a forum for the exchange of results and ideas. Sufficient time will be allocated for in-depth presentation and detailed discussion, both in large and in small groups. The outcome of the workshop should be an in-depth documentation of the state of the field, and an outline of the major themes of research for the next years. Workshop Organization The workshop is sponsored by IFBI Schloss Dagstuhl, a new international conference and research center for computer science. The site of the IFBI institute and site of the workshop is a beautiful castle in the forests of a low mountain range in the Saarland (Germany). A computer science library as well as a local network of high performance workstations are available at the institute. The workshop is limited to ca. 40 participants and is by invitation only. The institute provides full board and lodging at a nominal fee, and supports, if needed, the participants' travel within Germany. For participants from USA, the organizers will attempt to obtain a block travel grant from NSF. Currently, the workshop is planned to include sessions on - Tools - Formal Methods - Code Generation for Nonimperative Languages - Register Allocation - Parallel Architectures. All kinds of contributions are solicited - survey papers that reflect the state of the art, reports on completed projects, and, in particular, long or short presentations of ongoing research. However, submission of a paper is not mandatory. For the workshop, preliminary versions of papers are required. Final versions of selected contributions will be published in a proceedings volume, to appear in Springer LNCS. Researchers who think they could contribute to and benefit from the workshop should send e-mail to both the coordinators. Please indicate your major field of interest, and the topic of an eventual contribution. Dates and Deadlines Coordinators: Susan L. Graham Robert Giegerich EE/CS Division Universitaet Bielefeld University of California Technische Fakultaet 511 Evans Hill P.O.Box 86 40 Berkeley, CA 94720 4800 Bielefeld 1 USA Germany graham@renoir.berkeley.edu robert@techfak.uni-bielefeld.de Workshop date: May 20-24th, 1991 Location: International Conference and Research Center for Computer Science, Schloss Dagstuhl, Germany (near Saarbruecken; easily reachable from Frankfurt Airport) Workshop fee: DM 150,- (ca. US$ 100,-; includes room and full board) Apply for invitation: before Feb. 28th, 1991 Preliminary papers due: April 1st, 1991 Accomodation for accompanying persons may be arranged in the surrounding communities. Note that the overall ambience is very rural. -- Robert Giegerich * University of Bielefeld * D-4800 Bielefeld 1 * W-GERMANY Email: robert@techfak.uni-bielefeld.de * Tel.: +49 521 106 2908 Horst Hogenkamp * University of Bielefeld * D-4800 Bielefeld 1 * W-GERMANY Email: horst@techfak.uni-bielefeld.de * Tel.: +49 521 106 2908 -- Send compilers articles to compilers@iecc.cambridge.ma.us or {ima | spdcc | world}!iecc!compilers. Meta-mail to compilers-request.