Changeset 4858
- Timestamp:
- Nov 3, 2015, 4:06:30 PM (3 years ago)
- File:
-
- 1 edited
Legend:
- Unmodified
- Added
- Removed
-
docs/Working/icGrep/pres.tex
r4857 r4858 356 356 \end{tabular} 357 357 \end{center} 358 359 \end{frame} 360 361 \begin{frame}[fragile] 362 \frametitle{If Hierarchies} 363 364 \begin{itemize} 365 \item Unicode property classes may contain many thousands of codepoints. 366 \item Evaluating all required equations may be very expensive. 367 \item However, any 128-byte segment of input will involve only a few codepoint ranges. 368 \item Evaluation of property classes is embedded in if-hierarchies of successively 369 finer ranges. 370 \item This techique greatly reduces the number of equations evaluated for each property. 371 \end{itemize} 358 372 359 373 \end{frame} … … 488 502 \path (S2P.south east)+(+1.7,-0.1) node (b) {}; 489 503 \path[fill=green!20,rounded corners, draw=green, solid] (a) rectangle (b); 490 \node [draw=none,above=-0.04cm of S2P.north east] (t1) { };504 \node [draw=none,above=-0.04cm of S2P.north east] (t1) {\small Transposition Thread}; 491 505 492 506 \path (RequiredStreamsGenerator.north west)+(-.1,.38) node (a) {}; 493 507 \path (RequiredStreamsGenerator.south east)+(+2,-0.1) node (b) {}; 494 508 \path[fill=blue!20,rounded corners, draw=blue, solid] (a) rectangle (b); 495 \node [draw=none,above=-0.04cm of RequiredStreamsGenerator.north east] (t1) { };509 \node [draw=none,above=-0.04cm of RequiredStreamsGenerator.north east] (t1) {\small Stream Generator Thread}; 496 510 497 511 \path (JITFunction.north west)+(-.1,.38) node (a) {}; 498 512 \path (NamedPropertyLibrary.east |- MatchScanner.south)+(+.1,-0.1) node (b) {}; 499 513 \path[fill=red!20,rounded corners, draw=red, solid] (a) rectangle (b); 500 \node [draw=none,above=-0.04cm of JITFunction.north east] (t1) { };514 \node [draw=none,above=-0.04cm of JITFunction.north east] (t1) {\small Matcher Thread}; 501 515 502 516 \end{pgfonlayer}{threads} … … 504 518 505 519 \end{center} 506 \end{frame}507 508 \section{Performance Study: Unicode Set Operations}509 \begin{frame}[fragile]510 \frametitle{}511 \begin{center}512 \huge Performance Study: Unicode Set Operations513 \end{center}514 515 520 \end{frame} 516 521 … … 557 562 558 563 564 565 559 566 \end{tikzpicture} 560 567 \end{center} 561 568 %\caption{Matching Performance for Simple Property Expressions}\label{fig:property_test} 569 \end{frame} 570 571 \begin{frame} 572 \frametitle{Performance Comparison (Multithread)} 573 574 \begin{itemize} 575 \item Pipeline parallelism 576 \item Up to 40\% speedup by hiding transposition and required stream generation. 577 \item Combining the AVX2 ISA with multithreading gives and average overall 61\% 578 \item Future work 579 \begin{itemize} 580 \item Pipeline + data parallelism 581 \item AVX-512 ISA 582 \end{itemize} 583 \end{itemize} 562 584 \end{frame} 563 585
Note: See TracChangeset
for help on using the changeset viewer.