Looking for a Reg. Alloc. research platform.

crowley@gradine.cis.upenn.edu (Albert T Crowley)
9 Dec 1995 19:43:53 -0500

          From comp.compilers

Related articles
Looking for a Reg. Alloc. research platform. crowley@gradine.cis.upenn.edu (1995-12-09)
| List of all articles for this month |

From: crowley@gradine.cis.upenn.edu (Albert T Crowley)
Newsgroups: comp.compilers
Date: 9 Dec 1995 19:43:53 -0500
Organization: University of Pennsylvania
Keywords: registers, experiment, question


My advisor and I are planning to implement a new register allocation
scheme to get some experimental data to go along with our current
theoretical results.

What we would like is a C compiler that has been created with future
modifications in mind (i.e. very modular). A good system that we are
looking at right now is a Fortran compiler from Rice University. It
was structured extremely well with each (sub)pass of the compiler as a
self contained program, but since it is based in Fortran, it isn't
exactly what we want. Fortran is (I believe) missing some of the
constructions available in C that we are interested in.

Our algorithm is centered around looking at the -structure- of the
control flow graph, and uses special consideration for break and
continue statements (as defined in C). This is why we want a C
compiler (Pascal would be better than Fortran, but not the best).

I hope to get a system that already optimizes the register allocation
in a method that is considered 'good' so that we can compare our
results against that. I've looked at the C compiler available from
Stanford, and the focus there seems to be on parallelization, and I
fear that it might not have a good register allocator built in that I
could compare against. Also since I am not an expert in
implementation, I would like a well build register allocator to work

So...I have two basic questions:

1) is there another compiler that I haven't heard about that would be
well suited to the research I want to do?

2) Am I mistaken about the Stanford compiler, and that it would be a
good system to work on. (for register allocation)

Al Crowley


Post a followup to this message

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