Re: Compiler Construction Tools?

Allyn Dimock <dimock@mass.eas.harvard.edu>
28 Jul 1998 11:37:43 -0400

          From comp.compilers

Related articles
Compilier Construction Tools? shindle@toocool.com (1998-07-26)
Re: Compilier Construction Tools? vmakarov@cygnus.com (Vladimir Makarov) (1998-07-27)
Re: Compiler Construction Tools? dimock@mass.eas.harvard.edu (Allyn Dimock) (1998-07-28)
Re: Compiler Construction Tools? friwi@prosun.first.gmd.de (1998-07-28)
Re: Compiler Construction Tools? lex@natlab.research.philips.com (Augusteijn, dr.ir. A.) (1998-07-30)
Re: Compiler Construction Tools? dwight@pentasoft.com (1998-07-31)
Re: Compiler Construction Tools? pjfarley@banet.net (1998-08-02)
| List of all articles for this month |

From: Allyn Dimock <dimock@mass.eas.harvard.edu>
Newsgroups: comp.compilers
Date: 28 Jul 1998 11:37:43 -0400
Organization: Harvard University, Cambridge, Massachusetts
References: 98-07-183 98-07-200
Keywords: tools

Just a note on SUIF:


It has not been designed as a front end tool: there are plenty of
those. It is a very flexible middle end and back end generator. I
was surprised to see how many papers at PLDI referred to SUIF as their
backend generator.


SUIF 1(.2?) is the version that I am most familiar with.
Supplied front ends to SUIF 1 are limited: it uses lcc as a front end
for C and uses Fortran to C translation followed by lcc to front end
Fortran. Other front ends may be used but you have to write them
yourself.


SUIF 1 is written without regard to compile time: individual passes
typically read input from a file ad write output to a file.


SUIF 1 is an excellent product for teaching courses on optimizing
compilers, because of the ease of introducing new passes.


SUIF 2 is in development, but is (pre?) released.
SUIF 2 fixes the read / unpickle / transform / pickle / write overhead
of SUIF 1 by putting in generalized piping between passes which may be
implemented in memory with low overhead. I believe that there are C++
and Java fromt ends for SUIF 2 in development.


For the SUIF package, please see http://suif.stanford.edu/ the
Stanford product is mostly concerned with middle end issues such as
parallelization.


For an associated product: machSUIF, please see
http://www.eecs.harvard.edu/hube/ the Harvard product is code
generation, register allocation, scheduling passes for SUIF, as well
as optimization passes that can be used at the machine code level.


-- Allyn
--


Post a followup to this message

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