1.1 Binary file documents/Pygar-slides.odp has changed

     2.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     2.2 +++ b/documents/pygar-slides-handout.lyx	Wed May 12 02:25:34 2010 -0400
     2.3 @@ -0,0 +1,217 @@
     2.4 +#LyX 1.6.4 created this file. For more info see http://www.lyx.org/
     2.5 +\lyxformat 345
     2.6 +\begin_document
     2.7 +\begin_header
     2.8 +\textclass article
     2.9 +\use_default_options true
    2.10 +\language english
    2.11 +\inputencoding auto
    2.12 +\font_roman default
    2.13 +\font_sans default
    2.14 +\font_typewriter default
    2.15 +\font_default_family default
    2.16 +\font_sc false
    2.17 +\font_osf false
    2.18 +\font_sf_scale 100
    2.19 +\font_tt_scale 100
    2.20 +
    2.21 +\graphics default
    2.22 +\paperfontsize default
    2.23 +\use_hyperref false
    2.24 +\papersize default
    2.25 +\use_geometry false
    2.26 +\use_amsmath 1
    2.27 +\use_esint 1
    2.28 +\cite_engine basic
    2.29 +\use_bibtopic false
    2.30 +\paperorientation portrait
    2.31 +\secnumdepth 3
    2.32 +\tocdepth 3
    2.33 +\paragraph_separation indent
    2.34 +\defskip medskip
    2.35 +\quotes_language english
    2.36 +\papercolumns 1
    2.37 +\papersides 1
    2.38 +\paperpagestyle default
    2.39 +\tracking_changes false
    2.40 +\output_changes false
    2.41 +\author "" 
    2.42 +\author "" 
    2.43 +\end_header
    2.44 +
    2.45 +\begin_body
    2.46 +
    2.47 +\begin_layout Title
    2.48 +Pygar: Parallel Audio Processing
    2.49 +\end_layout
    2.50 +
    2.51 +\begin_layout Author
    2.52 +Laurel Pardue, Robert McIntyre
    2.53 +\end_layout
    2.54 +
    2.55 +\begin_layout Subsection*
    2.56 +Problem
    2.57 +\end_layout
    2.58 +
    2.59 +\begin_layout Standard
    2.60 +Music naturally comes in parallel sequences of samples called 
    2.61 +\emph on
    2.62 +voices
    2.63 +\emph default
    2.64 + (ex.
    2.65 + from multiple instruments).
    2.66 + Pure-software mixers are forced to pass these voices through the Von Neuman
    2.67 + bottleneck of a single processor, operating on these streams in series
    2.68 + and switching between each one quickly.
    2.69 + They are therefore naturally limited in the number of voices they can handle.
    2.70 + Worse, since the processing of each voice has to share the same processor,
    2.71 + too many voices at once can fully max out the processor and crash the system.
    2.72 + On typical laptop hardware and a high end software tool like ProTools,
    2.73 + this number is around 5.
    2.74 + Embedded devices have an even tougher time at meeting any sort of reasonable
    2.75 + timing requirements.
    2.76 +\end_layout
    2.77 +
    2.78 +\begin_layout Standard
    2.79 +[screenie Just 6 voices are enough to bring this session of ProTools to
    2.80 + it's knees.]
    2.81 +\end_layout
    2.82 +
    2.83 +\begin_layout Standard
    2.84 +We want the power of writing transforms for voices in a high level language
    2.85 + combined with a framework that applies these transforms to the voices in
    2.86 + parallel.
    2.87 +\end_layout
    2.88 +
    2.89 +\begin_layout Subsection*
    2.90 +Vision --- Pygar
    2.91 +\end_layout
    2.92 +
    2.93 +\begin_layout Standard
    2.94 +Our system addresses the limitations of pure software mixers.
    2.95 + It is a grid of SMIPS processors capped by a mixer.
    2.96 + The voices flow through the processors in parallel and are combined at
    2.97 + the final mixer into a single stream.
    2.98 + Each processor can be loaded with any arbitrary C program.
    2.99 +\end_layout
   2.100 +
   2.101 +\begin_layout Standard
   2.102 +\begin_inset Float figure
   2.103 +placement H
   2.104 +wide false
   2.105 +sideways false
   2.106 +status collapsed
   2.107 +
   2.108 +\begin_layout Plain Layout
   2.109 +\begin_inset Graphics
   2.110 +	filename ../../../../Pygar/documents/000402.png
   2.111 +	width 5in
   2.112 +
   2.113 +\end_inset
   2.114 +
   2.115 +
   2.116 +\begin_inset Caption
   2.117 +
   2.118 +\begin_layout Plain Layout
   2.119 +The audio data (“samples”) start in the memory, but are soon pulled into
   2.120 + action by the DMA (direct memory access).
   2.121 + The DMA sends the samples to a chain of 0 or more soft-cores, where they
   2.122 + are transformed according to the soft-cores’ algorithms.
   2.123 + After running the gauntlet of soft-cores, the samples flow first to a buffering
   2.124 + FIFO, and finally to a mixer, which sends the samples off to be played
   2.125 + by speakers or stored in a file.
   2.126 + 
   2.127 +\end_layout
   2.128 +
   2.129 +\end_inset
   2.130 +
   2.131 +
   2.132 +\end_layout
   2.133 +
   2.134 +\begin_layout Plain Layout
   2.135 +
   2.136 +\end_layout
   2.137 +
   2.138 +\end_inset
   2.139 +
   2.140 +
   2.141 +\end_layout
   2.142 +
   2.143 +\begin_layout Subsection*
   2.144 +Steps
   2.145 +\end_layout
   2.146 +
   2.147 +\begin_layout Standard
   2.148 +The difficult part of this project is managing code reuse.
   2.149 + We need three things for success.
   2.150 + 
   2.151 +\end_layout
   2.152 +
   2.153 +\begin_layout Itemize
   2.154 +SMIPS processor -- Easy.
   2.155 + Just use the Lab 5 processors.
   2.156 +\end_layout
   2.157 +
   2.158 +\begin_layout Itemize
   2.159 +Some way to program the processors
   2.160 +\end_layout
   2.161 +
   2.162 +\begin_layout Itemize
   2.163 +DMA (Direct Memory Access) to load voices into the processors.
   2.164 +\end_layout
   2.165 +
   2.166 +\begin_layout Standard
   2.167 +We use ScratchPad to load code into the processors.
   2.168 + ScratchPad is an Intel module which implements a cache hierarchy.
   2.169 + The hierarchy reaches all the way from RAM created on the FPGA to on-chip
   2.170 + DRAM to RAM on the host computer all the way to the Hard Disk of the host
   2.171 + computer.
   2.172 + The first time a processor tries to access one of its instructions, the
   2.173 + cache goes all the way to the hard disk of the host computer to retrieve
   2.174 + the data.
   2.175 + Subsequent attempts to access this data only go as far as the on-chip DRAM.
   2.176 + Each processor has its own ScratchPad and thus can be programmed independently.
   2.177 + The ScratchPad abstraction allows each processor to run a program of any
   2.178 + size.
   2.179 +\end_layout
   2.180 +
   2.181 +\begin_layout Standard
   2.182 +Music access is achieved through RRR, another Intel abstraction which allows
   2.183 + us to treat the hard disk of the host computer as if it were a normal FIFO.
   2.184 +\end_layout
   2.185 +
   2.186 +\begin_layout Subsection*
   2.187 +News
   2.188 +\end_layout
   2.189 +
   2.190 +\begin_layout Standard
   2.191 +We have run our system with 12 sample voices and various combinations of
   2.192 + simple c voice processing programs and the results have been better than
   2.193 + software implementations.
   2.194 + Significantly, increasing the number of voices does not increase the processing
   2.195 + load since each voice is processed in parallel.
   2.196 +\end_layout
   2.197 +
   2.198 +\begin_layout Subsection*
   2.199 +Contributions
   2.200 +\end_layout
   2.201 +
   2.202 +\begin_layout Itemize
   2.203 +Implemented Pygar, a system for quick parallel processing of audio.
   2.204 + 
   2.205 +\end_layout
   2.206 +
   2.207 +\begin_layout Itemize
   2.208 +Implemented 4 basic algorithms which serve as components for this system
   2.209 + (identity, bit-shift, volume-change, and delay) 
   2.210 +\end_layout
   2.211 +
   2.212 +\begin_layout Itemize
   2.213 +Demonstrated Pygar out-performs software-only systems.
   2.214 + Pure-software systems have a limit of around 6 voices, while our system
   2.215 + achieves 12 voices in parallel with no architecturally imposed limit on
   2.216 + the number of voices.
   2.217 +\end_layout
   2.218 +
   2.219 +\end_body
   2.220 +\end_document

     3.1 Binary file documents/pygar-slides.pdf has changed