Changeset 237 for docs/ASPLOS09


Ignore:
Timestamp:
Dec 15, 2008, 1:19:21 PM (10 years ago)
Author:
cameron
Message:

end-to-end performance significance for S2P

Location:
docs/ASPLOS09
Files:
3 edited

Legend:

Unmodified
Added
Removed
  • docs/ASPLOS09/asplos094-cameron.bib

    r225 r237  
    2323booktitle = "{CASCON '08: Proceedings of the 2008 conference of the Centre for Advanced Studies on Collaborative research}",
    2424mon=oct, year = 2008, address="{Toronto, Ontario , Canada}"
     25}
     26
     27@inproceedings{Herdy,
     28author = "{Herdy, Kenneth S., Burggraf, David S., and Cameron, Robert D.}",
     29title =  "{High Performance GML to SVG Transformation for the Visual Presentation of Geographic Data in Web-Based Mapping Systems}",
     30booktitle = "{Proceedings of SVG Open 2008}",
     31mon=aug, year = 2008, address="{Nuremburg, Germany}"
    2532}
    2633
  • docs/ASPLOS09/asplos094-cameron.tex

    r236 r237  
    769769need to be moved to a register with a correct 3-bit address,
    770770a minimum of 24 operations is required.
     771
     772\subsection{End-to-End Significance}
     773
     774In a study of several XML technologies applied to
     775the problem of GML to SVG transformation, the parabix
     776implementation (parallel bit streams for XML) was
     777found to the fastest with a cost of approximately
     77815 CPU cycles per input byte \cite{Herdy}.  Within parabix,
     779transposition to parallel bit stream form requires
     780approximately 1.1 cycles per byte \cite{CASCON08}.
     781All other things being equal, a 3X speed-up of transposition
     782alone would improve end-to-end performance in a
     783complete XML processing application by more than 4\%.
     784
    771785
    772786\section{Parallel to Serial Conversion}
     
    12331247Thus, IDISA esssentially offers systematic support
    12341248for horizontal operations entirely through the
    1235 use of \verb:<h, l>: half-operand modifier
     1249use of \verb:<h,l>: half-operand modifier
    12361250combinations.
    12371251
     
    12471261in the context of particular architectures is a potential
    12481262area for further work.
    1249 
    1250 
    1251 \begin{figure*}
    1252 \begin{center}
    1253 \begin{tabular}{|c||c|c|c|}
    1254 \hline
    1255 Kernel & Altivec ops & IDISA ops & Speed-up  \\ \hline
    1256 pop\_count<32> &  & 5n & 3X  \\ \hline
    1257 
    1258 bit\_reverse<32> & & & \\ \hline
    1259 Gray2binary<32> & 10n & 5n &  2X\\ \hline
    1260 \end{tabular}
    1261 \end{center}
    1262 \label{perftable}
    1263 \caption{Performance Results}
    1264 \end{figure*}
    1265 
    12661263
    12671264
Note: See TracChangeset for help on using the changeset viewer.