Code: Select all
\documentclass{beamer}
\usepackage{beamerthemesplit}
\title{TRIPS CPU}
\author{James Halliday}
\date{\today}
\begin{document}
\frame{\titlepage}
\frame{
\frametitle{What is it?}
{
\begin{itemize}
\item{Tera-op, Reliable, Intelligently adaptive Processing System}
\item[]{
\begin{itemize}
\item{University of Texas, Austin}
\item{IBM}
\item{Intel}
\item{Sun}
\end{itemize}
}
\end{itemize}
}
}
\frame{
\frametitle{What is it?}
{
\begin{itemize}
\item{Goals}
\item[]{
\begin{itemize}
\item{
One TFLOP on a single processor by 2012.
}
\item{
Scale across hundreds or thousands of cores
}
\end{itemize}
}
\end{itemize}
}
}
\frame{
\frametitle{How does it work?}
{
\begin{itemize}
\item{Explicit Data Graph Execution Instruction Set Architecture}
\item{Hyperblocks}
\item{
\begin{itemize}
\item{only 128 instructions}
\item{only 32 memory accesses}
\item{only 64 register accesses (32 read, 32 write)}
\item{one branch op (end)}
\end{itemize}
}
\item{Compilers already know enough to build these structures}
\end{itemize}
}
}
\frame{
\frametitle{How does it work? (terms)}
{
\begin{itemize}
\item{Dynamic Issue *}
\item[]{
\begin{itemize}
\item{Order of instruction execution is not known ahead of time}
\end{itemize}
}
\item{Static Issue}
\item[]{
\begin{itemize}
\item{Order of instruction execution is known ahead of time}
\item{
It's really hard to know whether or not memory is in cache
}
\item{
Waiting for memory to appear in cache holds up everything else
}
\end{itemize}
}
\item{Dynamic Placement}
\item[]{
\begin{itemize}
\item{Dependency analysis performed at runtime}
\end{itemize}
}
\item{Static Placement *}
\item[]{
\begin{itemize}
\item{Dependency analysis performed at compilation}
\item{Programs are only compiled once, but run many times}
\end{itemize}
}
\end{itemize}
}
}
\frame{
\frametitle{How does it work? (others)}
{
\begin{itemize}
\item{Other Approaches to Parallelism}
\item[]{
\begin{itemize}
\item{Superscalar}
\item[]{
\begin{itemize}
\item{Dynamic Placement}
\item{Dynamic Issue}
\end{itemize}
}
\item{Itanium}
\item[]{
\begin{itemize}
\item{Very Long Instruction Words}
\item{Static Placement}
\item{Static Issue}
\end{itemize}
}
\item{SIMD}
\item[]{
\begin{itemize}
\item{Hard-Coded Vector Sizes}
\item{Extra Circuit Complexity}
\end{itemize}
}
\end{itemize}
}
\end{itemize}
}
}
\frame{
\frametitle{How does it work?}
{
\begin{itemize}
\item{Static Placement}
\item{Dynamic Issue}
\item{TRIPS is lazy!}
\item{Local dependencies are optimized by the compiler, not the CPU}
\end{itemize}
}
}
\frame{
\frametitle{Why should I care?}
{
\begin{itemize}
\item{Exploit massive parallelism in sequential programs!}
\item{
Hundreds or thousands of identical cores on the same chip
}
\item{
Compilers are better at figuring out mundane details than you are
}
\end{itemize}
}
}
\frame{
\frametitle{Why should I care?}
{
\begin{itemize}
\item{Speaker: James Halliday}
\item{Subject: TRIPS CPU}
\item{What it is}
\item[]{
\begin{itemize}
\item{experimental, lots of identical cores}
\end{itemize}
}
\item{How it works:}
\item[]{
\begin{itemize}
\item{hyperblocks, static placement, dynamic issue}
\end{itemize}
}
\item{Why you care:}
\item[]{
\begin{itemize}
\item{compiler figures out parallelism details on thousands of cores}
\end{itemize}
}
\end{itemize}
}
}
\end{document}
