conferences/ngisearch2023/ngisearch2023.tex

   1 \documentclass[slidestop]{beamer}
   2 \usepackage{beamerthemesplit}
   3 \usepackage{graphics}
   4 \usepackage{pstricks}
   5
   6 \graphicspath{{./}}
   7
   8 \title{NGI SEARCH}
   9 \author{Luke Kenneth Casson Leighton}
  10
  11
  12 \begin{document}
  13
  14 \frame{
  15    \begin{center}
  16     \huge{NGI SEARCH}\\
  17     \vspace{32pt}
  18     \Large{SVP64 Power ISA}\\
  19     \Large{Vector Optimisation for Search}\\
  20     \vspace{24pt}
  21     \large{RED Semiconductor Ltd}\\
  22     \large{VectorCamp}\\
  23     \large{Vantosh}\\
  24     \vspace{6pt}
  25     \large{Assisted by Libre-SOC}\\
  26   \end{center}
  27 }
  28
  29
  30 \frame{\frametitle{SVP64 Power ISA Vector Optimisation for Search}
  31
  32 \vspace{6pt}
  33
  34  \begin{itemize}
  35     \item Take existing search algorithms and optimise a Vector ISA
  36         (at the hardware level) to
increase energy efficiency
  37          \vspace{6pt}
  38     \item Whilst normally a software developer has to make the best
  39         out of the situation of being provided with a "fait-accomplit"
  40         ISA (ARM, x86, MIPS) and attempt to optimise Search
  41     
   algorithms for it, we aim to turn it around: work out what
  42     
   features of an ISA, if added,
 would make the same algorithms
  43     
   much more power-efficient.
  44                 \vspace{6pt}
  45    \item RED:    Hardware/ISA concepts and project management\\
  46    \item VectorCamp:    Algorithm analysis and concepts\\
  47    \item VanTosh:    Validation, evolution and demonstration\\
  48   \end{itemize}
  49 }
  50
  51
  52 \frame{\frametitle{Low-level Libraries: used by all software}
  53 \vspace{6pt}
  54  \begin{itemize}
  55    \item Regardless of the high-level algorithm portable low-level
  56          libraries are needed. JSON parsing, UTF-8, strncpy, memcpy.
  57    \item Optimising libc6 standard routines (strncpy, memcpy, isascii)
  58                  would represent high value-for-money if made much more efficient
  59    \item strncpy already done under NLnet-funded Libre-SOC Grants: 11 Vector
  60                  instructions, representing a huge power-saving and high bang-per-buck
  61    \item We want to see how far this technical approach can be taken.
  62    \item All source code, all instructions, all HDL, all documentation:
  63          entirely public and FOSS Licensed or to be part of OpenPOWER
  64          Foundation Standards. Full transparency!
  65   \end{itemize}
  66 }
  67
  68 \frame{\frametitle{HyperScan}
  69 \vspace{6pt}
  70  \begin{itemize}
  71    \item Hyperscan is a high-performance multiple regex match library
  72                 \vspace{10pt}
  73   \end{itemize}
  74 }
  75
  76 \frame{
  77   \begin{center}
  78     {\Huge The end\vspace{15pt}\\
  79                    Thank you\vspace{15pt}\\
  80                    Questions?\vspace{10pt}
  81         }
  82   \end{center}
  83
  84   \begin{itemize}
  85         \item Discussion: http://lists.libre-soc.org
  86         \item Libera IRC \#libre-soc
  87         \item https://libre-soc.org/
  88         \item https://www.ngi.eu/ngi-projects/ngi-search/
  89   \end{itemize}
  90 }
  91
  92
  93 \end{document}