Re: Reduced Machine Description

dgb@june.cs.washington.edu (David Bradlee)
2 Mar 89 19:53:23 GMT

          From comp.compilers

Related articles
Reduced Machine Description martins@rwthinf.uucp (1989-02-22)
Re: Reduced Machine Description jhallen@wpi.wpi.edu (1989-03-15)
Re: Reduced Machine Description dgb@june.cs.washington.edu (1989-03-02)
Re: Reduced Machine Description cordy@qucis.queensu.ca (1989-02-27)
| List of all articles for this month |
Return-Path: <dgb@june.cs.washington.edu>
From: dgb@june.cs.washington.edu (David Bradlee)
Newsgroups: comp.compilers
Keywords: Portable compilers, Back ends
Date: 2 Mar 89 19:53:23 GMT
References: <3365@ima.ima.isc.com>
Organization: U of Washington, Computer Science, Seattle



Your idea to write a simpler machine description is a good one, but it has its
problems. First, you need a mapping between the intermediate language (IL) and
the instructions given in the description. This can be done, but it's
non-trivial. Second, you need to specify where instructions expect their
operands and where they leave results. On the 68000, some instructions allow
some operands in memory, while others do not. Third, what about different
object types (short int, long int, float)?


The bottom line is that you can make simplifying assumptions, such as
"arithmetic operators are take register operands and yield register results,"
and this will simplify the machine description. But, there is a significant
performance price to pay. I don't have exact figures, but without a more
complicated specification for the 68000 or the VAX, you would not be able to
produce production quality code. Since GNU C is intended to produce high
quality code, their machine descriptions are complicated.
-- Dave Bradlee
[From dgb@june.cs.washington.edu (David Bradlee)]
--


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.