paper.tex

DLX CPU VHDL CODE UNIVERSITY

% By Michael Schulte and James Stine in Jan, 1996. 

\documentstyle[12pt]{article}

% \textheight 25cm
% \topmargin -.5cm
% \textwidth 18cm
% \oddsidemargin -1cm
% \evensidemargin -1cm

\textheight 23.5cm
\textwidth 17cm
\topmargin 0cm
\oddsidemargin 0cm

\begin{document}
\include{boxedeps}
\SetRokickiEPSFSpecial
\HideDisplacementBoxes

\title{A Five Stage Multi-Cycle DLX RISC Microprocessor Design In VHDL}
\author{Louis P. Marquette\\
111 Sebring Rd\\
Dickson City, PA 18519, USA\\
[e-mail]\\
http://www.starrtechnologies.com/louis/}

\date{}

\maketitle

\vspace{-1cm}

\begin{abstract}

\end{abstract}

\section{Introduction}
	The DLX Microprocessor is a microprocessor in theory. To my
knowledge it has never been put onto silicon, at least for comercial
sale. It's main purpose is to study, teach, and practice with
microprocessor design. The Instruction set architecture is very
simmilar to MIPS, although they have some differences. It is a load/store
architecture and is RISC. Most instructions require 5 clock cycles to
complete; however, jumps need only 3. Floating point timing is not
accurate because of the fact that these instructions also take only 5
cycles to fully complete. 


\section{Accomplishments}
 The nature of my project confined most of my accomplishments to VHDL,
 although they did extend into other areas such as c++, since
 utilities were needed to help me with some problems that arose. Some
 of the more formidable accomplisments in this design would be: 
	The ability of the 'processor' to use a file on the hard drive as a 
 	 program ROM (this allowed for easy program modifications)
	Designing the processor to be multi-cycle ( this made going back and using
 	 Micro code easy )
 	Convering the processor to let it use Micro code that is
	 loaded from a data file ( This will hopefully make Piplining
 	 on the processor easy, should anyone ever attempt it. )
	Although I used a public domain floating point adder and 
         multiplyer, the concepts of floating point addition and
	 multiplication needed to be learned to make the necessary 
	 changes so that they could handle double precision numbers 
	 and rounding modes.
	The ploating point divider was a project in itself besause of
	 the fact that I built it from scratch. Unlike the floating 
	 point adder and multiplyer, which are procedurally designed,
         I created the divider as a structural entity. It, like the 
 	 other floating point units, is capable of both single and
 	 double precision as well as using the four rounding modes. 
	

\section{Experience}
	 During the time that I spent working on this project, I learned 
not only from my work, but also from the work of others and other
problems that arose in our lab that needed to be solved. These
include the basics of interval computations, some hard drive
considererations when installing linux, how to dissasemble a mammoth
piece of 40 year old test equipment, and too many computers in one
electrical outlet is a bad thing. The work that I did, of course,
brought msny more experiences such as learning more about unix,
VHDL and how to use it, IEEE 754 floating point numbers, floating 
point rounding modes, unions in c++, and about many more small 
pieces of information that, when viewed individually seem almost
inconsequential, but when viewed as a whole are a tramendous learning
experience.

\subsection{Unix experience}
	 Basicly, my knowledge of Unix was fairly limited before this 
project, but because of the consant use of it, I have slowly developed 
a firm working knowledge of it. Some of the more interesting learing
experiences from it included: not knowing that files could have a
'group' associated with it and only members of the group could access
it, realizing that there is no command to do a 'cp *.*  somewhereelse'
althought it would be nice, programs compile faster if your source 
code is on a drive physically in your machine and not somewhere
accross campus. I have also become familiar with utilities such as
ftp and sendmail. 

\subsection{Netscape experience}
	Although I have spent much time using this browser on my home
computer, there were a few diferences that I needed to become
familliar with to be able to use it fully. For example, forget about
a shockwave plugin for Unix, and saving files can be somewhat tricky.

\subsection{Emacs experience}
	Why do people use a few hundred dollar word processor made by a 
large corporation when there is a more functional editor out there for
free?  One of the mysteries of life I guess. Emacs seems to have more
utilities and functionality built in than any word processor I've
ever used before. Unfortunatly, it is not too user frienly unless you
know the methods by which to invoke all those nifty utilities.

\subsection{VHDL experience}
	Needless to say, VHDL is what I have gained to most experiencd
in. It was what the entire project revolved around. There were a few
books that I used as references, but I also used a CD-ROM tutorial
that came with some VHDL package used at my university. This CD-ROM
was probably the most informative of group, but you must consider that
I used it first when I knew the least.
   	VHDL seems to have many little quirks to it that have never been
improved on. This is not to say that it's designers did not think
ahead of themselfs, but there seems to be many context situations
where more code is needed then should be necessary. To be able to
compile my design I needed to become familiar with the Make utility
and therefor Makefile format. 
	

\subsection{C++ experience}
	While testing the functionality and accuracy of my design, I needed
and easy way to create Hexadecmal representations of floating point
numbers. to place in memory for use my the processor design. Rather
then take a minute or so each to do many of them, I wrote a c++
program to do this grunt-work for me. I also made a program that gave
me the sum, difference, product, and division value of two numbers in
each of the four rounding modes. This program had the help of
assembler routines that set the rounding mode, these were written by
a current graduate student at Lehigh University, James Stein.

\subsection{Debuging experience}
	Because of my being a novice at VHDL, I spent an extaordinary
amount of time debugging my code. There were times when I used the
wrong data types, or confused the names of variables, made infinite
loops, and otherwise did something that was a stupid mistake.

\section{Deviations from DLX Definition}
	There are some aspects of this design that are not consistant
with the DLX definition. The opcodes for loading and storing bytes and
half words was not seen as necessary and not implimented. The single
bit floating point status register was not sufficient for rounding
mode information and special case result codes, it was extended to 32
bits and uses the same format as in the MIPS architecture. Ther was an
opcode added to load the floating point status register with a
register. This was needed to set the rounding mode.

\section{Future Enhancements}
	There can probably be many improvements made on this
processor, since it is not very complex currently. One thing that I
had planned to do, but did not have the time for was to make it
pipelined. This would probably not be very hard since there are some
internmediate register buffers in the datapath already, and it is
multi-cycled and uses micro-code. I do not forsee and structural
hazards; althought, I am sure there would me plenty of othere hazards
to work around. 
	Pipelining the FP units or making their timing more accurate
might also be a improvement worth trying. 



\section{Conclusions} 
	VHDL seems to be a very versitile language, especially because it
allows you to describe an entity either structurally, behaviorly, or
any combination of two. 
	Microprocessor design is much less tedious and much easier when done
visually then when done in a text based language such as VHDL. I have
used software called Max Plus + to design a simpler 8 bit
microprocessor for a class and the work seemed to go much faster and
it was easier to see where everything was headed because I have an actual
picture of my design infront of me rather then text which I had to 
interprit. However, VHDL allows for not only the structural type of
designs, which would be described by a picture, but by procedural,
which are more powerful and complex but cannot be turned into a picture. 
	I quickly learned that sometimes there were errors that the
VHDL compiler did not recognise and somehow got around, but that the
simulator very nicly pointed out for me.



\bibliography{paper}

\bibliographystyle{ieeetr}

\end{document}