\documentclass{sig-alternate-05-2015}
\begin{document}

% ACM templates include ISBN and DOI...
\isbn{N.A.}
\doi{N.A.}

\title{EE480 Assignment 1: KySMet Encoding And Assembler}
\subtitle{Implementor's Notes}

\numberofauthors{1}
\author{
Hank Dietz\\
       \affaddr{Department of Electrical and Computer Engineering}\\
       \affaddr{University of Kentucky, Lexington, KY USA}\\
       \email{\texttt{hankd@engr.uky.edu}}
}

\maketitle
\begin{abstract}
This project involved the design of an encoding scheme for the
KySMet instruction set. The encoding was then to be embodied in
an AIK specification.
\end{abstract}


\section{General Approach}

The first issue is how to encode 25 instruction types in a
16-bit instruction words where 12 bits are needed for the
operands of some instructions. The remaining 4 bits can only
distinguish 16 instruction types. Complicating matters a bit, I
also wanted the encoding of similar operations to have more
instruction bits in common.  I further imposed the constraint
that I wanted the \texttt{trap} instruction to encode as 0, so
that any instruction sequence wandering into empty memory will
cause a trap to the operating system.

I decomposed the instruction set into groups. The first group
contains instructions that have no arguments. The second group
is control flow instructions that take a 16-bit address. The
next group take two register arguments. The three-register
instructions are the group that uses most of the 16 possible
opcode field values. The 8-bit load immediates form the last
group.  It's all very straightforward.

Although this instruction set is intended for a SIMD-parallel
implementation, that really doesn't impact the instruction
encoding. The only extra concern was that KySMet is a Harvard
architecture, so there are separate \texttt{.text} and
\texttt{.data} segments, both appearing to start at location 0.

% insert a break to roughly level columns...
\vfill\pagebreak

\section{The li Macro}

As required, there is an \texttt{li} macro operation
implemented to use a \texttt{li8} alone when possible
and to follow it with a \texttt{lu8} only when the
immediate value made it necessary. This is implemented
by having two patterns for \texttt{li \$.d,.i16}. The
first includes a condition that will fail if \texttt{.i16}
doesn't match the sign-extension of the bottom 8 bit of
\texttt{.i16} (i.e., \texttt{.i16 \& 255}). The second pattern
simply uses the pair \texttt{li8}, \texttt{lu8}.

\section{constants}

Not much to report here; all constants were given the values
that correspond to the order they were listed in the assignment.
However, the HTML document specifying KySMet was actually
self-inconsistent in that it swapped \texttt{IPROC} and
\texttt{NPROC} in the table describing the values. My design assumes
\texttt{IPROC} is 1 and \texttt{NPROC} is 2.

\section{Issues}

Really nothing terrible. Note that I was careful to avoid
potential name conflicts by using names that started with "."
in my specification.

The hardest thing to do is to test this. It really isn't
testable in any obvious way... so this was tested only by manual
inspection with a variety of simple test cases, the largest of
which is included with my assignment.  There were no errors
flagged by AIK.

\end{document}