conferences/fosdem2024/fosdem2024_ddffirst/fosdem2024_ddffirst.tex

   1 \documentclass[slidestop]{beamer}
   2 \usepackage{beamerthemesplit}
   3 \usepackage{graphics}
   4 \usepackage{pstricks}
   5 \usepackage{pgffor}
   6 \usepackage{listings}
   7
   8 \graphicspath{{./}}
   9
  10 \title{Data-Dependent-Fail-First}
  11 \author{Luke Kenneth Casson Leighton and Shriya Sharma}
  12
  13
  14 \begin{document}
  15
  16 \frame{
  17    \begin{center}
  18     \huge{The Libre-SOC Hybrid 3D CPU}\\
  19     \vspace{32pt}
  20     \Large{Data-Dependent-Fail-First}\\
  21
  22     \vspace{24pt}
  23     \Large{FOSDEM2024}\\
  24     \vspace{16pt}
  25     \large{Sponsored by NLnet's PET Programme}\\
  26     \vspace{6pt}
  27     \large{\today}
  28   \end{center}
  29 }
  30
  31
  32
  33 \begin{frame}[fragile]
  34 \frametitle{Simple-V CMPI in a nutshell}
  35
  36 \begin{semiverbatim}
  37 function op\_cmpi(BA, RA, SI) # cmpi not vector-cmpi!
  38   (assuming you know power-isa)
  39   int i, id=0, ira=0;
  40   for (i = 0; i < VL; i++)
  41     CR[BA+id] <= compare(ireg[RA+ira], SI);
  42     if (reg\_is\_vectorised[BA] ) \{ id += 1; \}
  43     if (reg\_is\_vectorised[RA])  \{ ira += 1; \}
  44 \end{semiverbatim}
  45
  46   \begin{itemize}
  47    \item Above is oversimplified: predication etc. left out
  48    \item Scalar-scalar and scalar-vector and vector-vector now all in one
  49    \item OoO may choose to push CMPIs into instr. queue (v. busy!)
  50   \end{itemize}
  51 \end{frame}
  52
  53
  54 \frame{\frametitle{Load/Store Fault-First}
  55
  56     \begin{itemize}
  57         \item Problem: vector load and store can cause a page fault
  58         \item Solution: a protocol that allows optional load/store
  59         \item instruction \textit{requests} a number of elements
  60         \item instruction \textit{informs} the number actually loaded
  61         \item first element load/store is not optional (cannot fail)
  62         \item ARM SVE: https://arxiv.org/pdf/1803.06185.pdf
  63         \item more: wikipedia Vector processor page: Fault/Fail First
  64         \vspace{10pt}
  65         \item Load/Store is Memory to/from Register, what about
  66               Register to Register?
  67         \item Register-to-register: "Data-Dependent Fail-First."
  68         \item Z80 LDIR: Mem-Register, CPIR: Register-Register
  69     \end{itemize}
  70 }
  71
  72 \begin{frame}[fragile]
  73     \frametitle{Data-Dependent-Fail-First in a nutshell}
  74
  75     \begin{semiverbatim}
  76 function op\_cmpi(BA, RA, SI) # cmpi not vector-cmpi!
  77 int i, id=0, ira=0;
  78 for (i = 0; i < VL; i++)
  79     CR[BA+id] <= compare(ireg[RA+ira], SI);
  80     if (reg\_is\_vectorised[BA] ) \{ id += 1; \}
  81     if (reg\_is\_vectorised[RA])  \{ ira += 1; \}
  82     if test (CR[BA+id]) == FAIL: \{ VL = i + 1; break \}
  83     \end{semiverbatim}
  84
  85     \begin{itemize}
  86         \item Parallelism still perfectly possible
  87               ("hold" writing results until sequential post-analysis
  88                carried out. Best done with OoO)
  89         \item VL truncation can be inclusive or exclusive
  90               (include or exclude a NULL pointer or a
  91               string-end character, or overflow result)
  92         \item \textit{Truncation can be to zero Vector Length}
  93     \end{itemize}
  94 \end{frame}
  95
  96 \frame{\frametitle{Power ISA v3.1 vstribr}
  97
  98     \lstinputlisting[language={}]{vstribr.txt}
  99
 100     \begin{itemize}
 101         \item ironically this hard-coded instruction is
 102         identical to general-purpose Simple-V DD-FFirst...
 103     \end{itemize}
 104
 105 }
 106
 107 \frame{\frametitle{Pospopcount}
 108
 109   \begin{itemize}
 110     \item     Positional popcount adds up the totals of each bit set to 1 in each bit-position, of an array of input values.
 111     \item   Notoriously difficult to do in SIMD assembler: typically 550 lines
 112     \item https://github.com/clausecker/pospop
 113
 114    \end{itemize}
 115
 116     \lstinputlisting[language={}]{pospopcount.c}
 117
 118
 119 }
 120
 121 \frame{\frametitle{Pospopcount}
 122
 123     \begin{center}
 124         \includegraphics[width=0.5\textwidth]{pospopcount.png}
 125     \end{center}
 126       \begin{itemize}
 127         \item   The challenge is to perform an appropriate transpose of the data (the CPU can only work on registers, horizontally),
 128         in blocks that suit the processor and the ISA capacity.
 129
 130
 131     \end{itemize}
 132 }
 133
 134 \frame{\frametitle{Pospopcount}
 135
 136     \begin{center}
 137         \includegraphics[width=0.6\textwidth]{array_popcnt.png}
 138     \end{center}
 139
 140   \begin{itemize}
 141
 142         \item     The draft gbbd instruction implements the transpose (shown above),
 143                 preparing the data to use standard popcount.
 144                (gbbd is based on Power ISA vgbbd, v3.1 p445)
 145
 146     \end{itemize}
 147
 148 }
 149
 150 \frame{\frametitle{pospopcount assembler}
 151
 152
 153 \lstinputlisting[language={}]{pospopcount.s}
 154
 155 }
 156
 157 \frame{\frametitle{strncpy}
 158
 159     \lstinputlisting[language={}]{strncpy.c}
 160   \begin{itemize}
 161     \item two simple-looking for-loops,
 162           data-dependent in the first.
 163     \item sv.cmpi stops at the first zero, /vli includes the zero
 164           in VL.
 165     \item note the post-increment Load/Store: saves
 166           pre-decrementing
 167     \item a Vector of CRs is produced which then get tested
 168            by the sv.bc/all instruction, counting down CTR
 169            per item tested.
 170      \item Power ISA added hard-coded data-dependent capacity
 171            into vstribr, where SVP64 it is generic (applies
 172            to any instruction)
 173      \item even the null-ing part is not straightforward as
 174           it could be mis-aligned compared to the VSX width.
 175      \item end-result: assembler-optimised  strncpy on Power
 176           ISA v3.0 is a whopping 240 instructions. SVP64 is 10
 177           and parallel in HW
 178  \end{itemize}
 179 }
 180
 181
 182
 183 \frame{\frametitle{strncpy assembler}
 184
 185 \lstinputlisting[language={}]{strncpy.s}
 186
 187 }
 188
 189 \frame{\frametitle{sv.lbz/ff=RC1/vli *16,1(10)}
 190     \begin{center}
 191         \includegraphics[width=0.6\textwidth]{lbz_ff_vli.png}
 192     \end{center}
 193
 194   \begin{itemize}
 195     \item r10 points to memory address 0x001007
 196     \item sv.lbz (Power ISA load byte immediate) multiplies immediate
 197           offset by element step index, to get Effective Address (EA)
 198     \item LD/ST has no Rc=1 so Data-Dependent Fail-First specified
 199           as "ff=RC1". Not LD/ST Fault First! vli: VL inclusive
 200     \item Test done after each load. Fails at Memory contents
 201           0x001009. Inclusive Mode: VL is truncated to 5 (FIVE) not 4
 202  \end{itemize}
 203 }
 204
 205 \frame{\frametitle{linked-list walking}
 206
 207   \begin{itemize}
 208     \item "TODO
 209  \end{itemize}
 210 }
 211
 212 \frame{\frametitle{sv.ld/ff=RC1/vli *17, 8(*16)}
 213
 214         \begin{center}
 215                 \includegraphics[width=1.0\textwidth]{linked_list_dd.png}
 216         \end{center}
 217 }
 218
 219 \frame{\frametitle{maxloc}
 220         \begin{itemize}
 221                 \item "TODO
 222         \end{itemize}
 223 }
 224
 225 \frame{\frametitle{Summary}
 226
 227  \begin{itemize}
 228    \item Goal is to create a mass-volume low-power embedded SoC suitable
 229          for use in netbooks, chromebooks, tablets, smartphones, IoT SBCs.
 230    \item No way we could implement a project of this magnitude without
 231          nmigen (being able to use python OO to  HDL)
 232    \item Collaboration with OpenPOWER Foundation and Members absolutely
 233          essential. No short-cuts.  Standards to be developed and ratified
 234          so that everyone benefits.
 235    \item Riding the wave of huge stability of OpenPOWER ecosystem
 236    \item Greatly simplified open 3D and Video drivers reduces product
 237          development costs for customers
 238    \item It also happens to be fascinating, deeply rewarding technically
 239          challenging, and funded by NLnet
 240
 241   \end{itemize}
 242 }
 243
 244 \frame{\frametitle{How can you help?}
 245
 246     \vspace{5pt}
 247
 248     \begin{itemize}
 249         \item Start here! https://libre-soc.org \\
 250         Mailing lists https://lists.libre-soc.org \\
 251         IRC Freenode libre-soc \\
 252         etc. etc. (it's a Libre project, go figure) \\
 253         \vspace{3pt}
 254         \item Can I get paid? Yes!  NLnet funded\\
 255         See https://libre-soc.org/nlnet/\#faq \\
 256         \vspace{3pt}
 257         \item Also profit-sharing in any commercial ventures \\
 258         \vspace{3pt}
 259         \item How many opportunities to develop Libre SoCs exist,\\
 260         and actually get paid for it?
 261         \vspace{3pt}
 262         \item I'm not a developer, how can I help?\\
 263         - Plenty of research needed, artwork, website \\
 264         - Help find customers and OEMs willing to commit (LOI)
 265     \end{itemize}
 266 }
 267
 268
 269
 270 \frame{
 271   \begin{center}
 272     {\Huge The end\vspace{12pt}\\
 273            Thank you\vspace{12pt}\\
 274            Questions?\vspace{12pt}
 275     }
 276   \end{center}
 277
 278   \begin{itemize}
 279     \item Discussion: http://lists.libre-soc.org
 280     \item Freenode IRC \#libre-soc
 281     \item http://libre-soc.org/
 282     \item http://nlnet.nl/PET
 283     \item https://libre-soc.org/nlnet/\#faq
 284   \end{itemize}
 285 }
 286
 287
 288 \end{document}