source: docs/Working/re/ppopp-re.tex @ 3490

Last change on this file since 3490 was 3490, checked in by cameron, 6 years ago

Scripts and updates

File size: 34.5 KB
Line 
1%-----------------------------------------------------------------------------
2%
3%               Template for sigplanconf LaTeX Class
4%
5% Name:         sigplanconf-template.tex
6%
7% Purpose:      A template for sigplanconf.cls, which is a LaTeX 2e class
8%               file for SIGPLAN conference proceedings.
9%
10% Guide:        Refer to "Author's Guide to the ACM SIGPLAN Class,"
11%               sigplanconf-guide.pdf
12%
13% Author:       Paul C. Anagnostopoulos
14%               Windfall Software
15%               978 371-2316
16%               paul@windfall.com
17%
18% Created:      15 February 2005
19%
20%-----------------------------------------------------------------------------
21
22
23\documentclass[preprint]{sigplanconf}
24
25% The following \documentclass options may be useful:
26
27% preprint      Remove this option only once the paper is in final form.
28% 10pt          To set in 10-point type instead of 9-point.
29% 11pt          To set in 11-point type instead of 9-point.
30% authoryear    To obtain author/year citation style instead of numeric.
31
32\usepackage{amsmath}
33\usepackage{pgfplots}
34
35\begin{document}
36
37\special{papersize=8.5in,11in}
38\setlength{\pdfpageheight}{\paperheight}
39\setlength{\pdfpagewidth}{\paperwidth}
40
41\conferenceinfo{PPoPP 2014}{February 15-19, 2014, Orlando, Florida, United States} 
42\copyrightyear{2013} 
43\copyrightdata{978-1-nnnn-nnnn-n/yy/mm} 
44\doi{nnnnnnn.nnnnnnn}
45
46% Uncomment one of the following two, if you are not going for the
47% traditional copyright transfer agreement.
48
49%\exclusivelicense                % ACM gets exclusive license to publish,
50                                  % you retain copyright
51
52%\permissiontopublish             % ACM gets nonexclusive license to publish
53                                  % (paid open-access papers,
54                                  % short abstracts)
55
56\titlebanner{}        % These are ignored unless
57\preprintfooter{Bitwise Data Parallel Grep}   % 'preprint' option specified.
58
59\title{Bitwise Data Parallelism in Regular Expression Matching}
60%\subtitle{Subtitle Text, if any}
61
62\authorinfo{Anonymous Authors}{Institutions}{emails}
63%\authorinfo{Robert D. Cameron \and Kenneth S. Herdy \and Dan Lin \and Meng Lin \and Ben Hull \and Thomas S. Shermer \and Arrvindh Shriraman}
64%          {Simon Fraser University}
65%           {\{cameron,ksherdy,lindanl,linmengl,bhull,shermer,ashriram\}@cs.sfu.ca}
66
67\maketitle
68
69\begin{abstract}
70\input{abstract}
71\end{abstract}
72\category{Theory of computation}{Formal languages and automata theory}{Regular languages}
73\category{Computer systems organization}{Parallel architectures}{Single instruction, multiple data}
74
75% general terms are not compulsory anymore,
76% you may leave them out
77%\terms
78%term1, term2
79
80\keywords
81regular expression matching, grep, parallel bit stream technology
82
83\section{Introduction}
84
85The use of regular expressions to search texts for occurrences
86of string patterns has a long history and
87remains a pervasive technique throughout computing applications today.
88% {\em a brief history}
89The origins of regular expression matching date back to automata theory
90developed by Kleene in the 1950s \cite{kleene1951}.
91Thompson \cite{thompson1968} is credited with the first construction to convert regular expressions
92to nondeterministic finite automata (NFA).
93Following Thompson's approach, a regular expression of length $m$ is first converted
94to an NFA with $O(m)$ nodes. It is then possible to search a text of length $n$ using the
95NFA in worst case $O(mn)$ time. Often, a more efficient choice
96is to convert the NFA into a DFA. A DFA has only a single active state at any time
97in the matching process and
98hence it is possible to search a text at of length $n$ in $O(n)$ time.
99However, it is well known that the conversion of an NFA to an equivalent DFA may result
100in state explosion. That is, the number of resultant DFA states may increase exponentially.
101In \cite{Baeza-yates_anew} a new approach to text searching was proposed based on bit-parallelism \cite{baeza1992new}.
102This technique takes advantage of the intrinsic parallelism of bitwise operations
103within a computer word. Given a $w$-bit word, the Shift-Or algorithm \cite{Baeza-yates_anew} algorithm uses the
104bit-parallel approach to
105simulate an NFA in $O(nm/w)$ worst-case time.
106
107A disadvantage of the bit-parallel Shift-Or pattern matching approach
108in comparison to simple string matching algorithms is an inability to skip input characters.
109For example, the Boyer-Moore family of algorithms \cite{boyer1977fast} skip input characters
110to achieve sublinear times in the average case. Backward Dawg Matching
111(BDM) string matching algorithms \cite{crochemore1994text} based on suffix automata are able to skip characters.
112The Backward Nondeterministic Dawg Matching (BNDM) pattern matching algorithm \cite{wu1992fast} 
113combines the bit-parallel advantages of Shift-Or and with the character skipping advantages of the BDM algorithm.
114The nrgrep pattern matching tool is built over the BNDM algorithm,
115and hence the name nrgrep \cite{navarro2000}.
116
117There has been considerable interest in using parallelization techniques
118to improve the performance of regular expression matching on parallel hardware
119such as multi-core processors (CPUs), graphics processing units (GPUs),
120field-programmable gate arrays (FPGAs), and specialized architectures such as
121the Cell Broadband Engine (Cell BE). % FPGA results (synthesis of patterns into logic circuits) vs. memory based approaches (STTs in memory)
122%CPU
123Scarpazza and Braudaway \cite{scarpazza2008fast} demonstrated that
124text processing algorithms that exhibit irregular memory access patterns
125can be efficiently executed on multicore hardware.
126In related work, Pasetto et al. presented a flexible tool that
127performs small-ruleset regular expression matching at a rate of
1282.88 Gbps per chip on Intel Xeon E5472 hardware \cite{pasetto2010}.
129Naghmouchi et al. demonstrated that the Aho-Corasick (AC)
130string matching algorithm \cite{aho1975} is well suited for parallel
131implementation on multi-core CPUs, GPUs and the Cell BE \cite{scarpazza2011top, naghmouchi2010}.
132On each hardware, both thread-level parallelism (additional cores) and data-level parallelism
133(wide SIMD units) are leveraged for performance.
134Salapura et. al., advocated the use of vector-style processing for regular expressions
135in business analytics applications and leveraged the SIMD hardware available
136on multi-core processors to acheive a speedup of better than 1.8 over a
137range of data sizes of interest \cite{salapura2012accelerating}.
138%Cell
139In \cite{scarpazza2008}, Scarpazza and Russell presented a SIMD tokenizer
140that delivered 1.00–1.78 Gbps on a single
141Cell BE chip and extended this approach for emulation on the Intel Larrabee
142instruction set \cite{scarpazza2009larrabee}.
143On the Cell BE, Scarpazza \cite{scarpazza2009cell} described a pattern matching
144implementation that delivered a throughput of 40
145Gbps for a small dictionary of approximately 100 patterns, and a throughput of 3.3-3.4
146Gbps for a larger dictionary of thousands of patterns. Iorio and van Lunteren \cite{iorio2008} 
147presented a string matching implementation for automata that achieves
1484 Gbps on the Cell BE.
149% GPU
150In more recent work, Tumeo et al. \cite{tumeo2010efficient} presented a chunk-based
151implementation of the AC algorithm for
152accelerating string matching on GPUs. Lin et al., proposed
153the Parallel Failureless Aho-Corasick (PFAC)
154algorithm to accelerate pattern matching on GPU hardware and
155achieved 143 Gbps throughput, 14.74 times faster
156than the AC algorithm performed on a four core
157multi-core processor using OpenMP \cite{lin2013accelerating}.
158
159Whereas the existing approaches to parallelization have been
160based on adapting traditional sequential algorithms to emergent
161parallel architectures, we introduce both a new algorithmic
162approach and its implementation on SIMD and GPU architectures.
163This approach relies on a bitwise data parallel view of text
164streams as well as a surprising use of addition to match
165runs of characters in a single step.  The closest previous
166work is that underlying bit-parallel XML parsing using 128-bit SSE2 SIMD
167technology together with a parallel scanning primitive also
168based on addition \cite{cameron2011parallel}.   
169However, in contrast to the deterministic, longest-match
170scanning associated with the ScanThru primitive of that
171work, we introduce here a new primitive MatchStar
172that can be used in full generality for nondeterministic
173regular expression matching.   We also introduce a long-stream
174addition technique involving a further application of MatchStar
175that enables us to scale the technique to $n$-bit addition
176in $\lceil\log_{64}{n}\rceil$ steps.   We ultimately apply this technique,
177for example, to perform
178synchronized 4096-bit addition on GPU warps of 64 threads.
179
180There is also a strong keyword match between the bit-parallel
181data streams used in our approach and the bit-parallelism
182used for NFA state transitions in the classical algorithms of
183Wu and Manber \cite{wu1992agrep}, Baez-Yates and Gonnet \cite{baeza1992new}
184and Navarro and Raffinot \cite{navarro1998bit}.
185However those algorithms use bit-parallelism in a fundamentally
186different way: representing all possible current NFA states
187as a bit vector and performing parallel transitions to a new
188set of states using table lookups and bitwise logic.    Whereas
189our approach can match multiple characters per step, bit-parallel
190NFA algorithms proceed through the input one byte at a time.
191Nevertheless, the agrep \cite{wu1992agrep} and
192nrgrep \cite{navarro2000} programs implemented using these techniques remain
193among the strongest competitors in regular expression matching
194performance, so we include them in our comparative evaluation.
195
196The remainder of this paper is organized as follows.
197Section \ref{sec:grep} briefly describes regular expression
198notation and the grep problem.
199Section \ref{sec:bitwise} presents our basic algorithm and MatchStar
200using a model of arbitrary-length bit-parallel data streams.
201Section \ref{sec:blockwise} discusses the block-by-block
202implementation of our techniques including the long stream
203addition techniques for 256-bit addition with AVX2 and
2044096-bit additions with GPGPU SIMT.
205Section \ref{sec:analysis} 
206Section \ref{sec:SSE2} 
207Section \ref{sec:AVX2} 
208Section \ref{sec:GPU} 
209Section \ref{sec:Concl} concludes the paper with a discussion of areas for future work.
210
211\section{Regular Expression Notation and Grep}\label{sec:grep}
212
213We follow common Posix notation for regular expressions.
214A regular expression specifies a set of strings through
215a pattern notation.   Individual characters normally
216stand for themselves, unless they are one of the
217special characters \verb:*+?[{\(|^$.: that serve as metacharacters
218of the notation system.  Thus the regular expression \verb:cat:
219is a pattern for the set consisting of the single 3-character
220string ``\verb:cat:''.   The special characters must be escaped
221with a backslash to prevent interpretation as metacharacter, thus
222\verb:\$: represents the dollar-sign and \verb:\\\\: represent
223the string consisting of two backslash characters.
224Character class bracket expressions are pattern elements
225that allow any character in a given class to be used in a particular
226context.  For example, \verb:[@#%]: is a regular expression
227that stands for any of the three given symbols.  Contiguous
228ranges of characters may be specified using hyphens;
229for example \verb:[0-9]: for digits and \verb:[A-Za-z0-9_]:
230for any alphanumeric character or underscore.  If the
231caret character immediately follows the opening bracket,
232the class is negated, thus \verb:[^0-9]: stands for
233any character except a digit.  The period metacharacter
234\verb:.: stands for the class of all characters.
235
236Consecutive pattern elements stand for strings formed by
237concatenation, thus \verb:[cd][ao][tg]: stands for the
238set of 8 three-letter strings ``\verb:cat:'' through ``\verb:dog:''.
239The alternation operator \verb:|: allows a pattern to be
240defined to have to alternative forms, thus \verb:cat|dog:
241matches either ``\verb:cat:'' or ``\verb:dog:''.  Concatenation
242takes precedence over alternation, but parenthesis may be
243used to change this, thus \verb:(ab|cd)[0-9]: stands for any
244digit following one of the two prefixes  ``\verb:ab:'' or ``\verb:cd:''.
245
246Repetition operators may be appended to a pattern to specify
247a variable number of occurrences of that pattern. 
248The Kleene \verb:*: specifies zero-or-more occurrences
249matching the previous pattern, while \verb:+: specifies one-or
250more occurrences.  Thus \verb:[a-z][a-z]*: and \verb:[a-z]+:
251both specify the same set: strings of at least one lower-case
252letter.  The postfix operator \verb:?: specifies an optional
253component, i.e., zero-or-one occurrence of strings matching
254the element.  Specific bounds may be given within braces:
255\verb:(ab){3}: specifies the string ``\verb:ababab:'',
256\verb:[0-9A-Fa-f]{2,4}: specifies strings of two, three
257or four hexadecimal digits, and \verb:[A-Z]{4,}: specifies
258strings of at least 4 consecutive capital letters.
259
260The grep program searches a file for lines containing matches
261to a regular expression using any of the above notations.
262In addition, the pattern elements \verb:^: and \verb:$:
263may be used to match respectively the beginning or the
264end of a line.  In line-based tools such as grep, \verb:.:
265matches any character except newlines; matches cannot extend
266over lines.
267Normally, grep prints all matching
268lines to its output.  However, grep programs typically
269allow a command line flag such as \verb:-c: to specify
270that only a count of matching lines be produced; we use
271this option in our experimental evaluation to focus
272our comparisons on the performance of the underlying matching
273algorithms.
274
275\section{Matching with Bit-Parallel Data Streams}\label{sec:bitwise}
276
277Whereas the traditional approaches to regular expression matching
278using NFAs, DFAs or backtracking all rely on a byte-at-a-time
279processing model, the approach  we introduce in this paper is based
280on quite a different concept:  a data-parallel approach to simultaneous
281processing of data stream elements.  Indeed, our most abstract model
282is that of unbounded data parallelism: processing all elements of
283the input data stream simultaneously.   In essence, data streams are viewed
284as (very large) integers.   The fundamental operations are bitwise
285logic, stream shifting and long-stream addition.
286
287Depending on the available parallel processing resources, an actual
288implementation may divide an input stream into blocks  and process
289the blocks sequentially.   Within each block  all elements of the
290input stream are processed together, relying the availability of
291bitwise logic and addition scaled to the block size.   On commodity
292Intel and AMD processors with 128-bit SIMD capabilities (SSE2),
293we typically process input streams 128 bytes at a time.   In this
294case, we rely on the Parabix tool chain \cite{lin2012parabix}
295to handle the details of compilation to block-by-block processing.
296For our GPGPU implementation, we have developed a long-stream
297addition technique that allows us to perform 4096-bit additions
298using 64 threads working in lock-step SIMT fashion.  Using scripts
299to modify the output of the Parabix tools, we effectively divide
300the input into blocks of 4K bytes processed in a fully data-parallel
301manner.
302
303A key concept in this streaming approach is the derivation of bit streams
304that are parallel to the input data stream, i.e., in one-to-one
305correspondence with the data element positions of the input
306streams.   Typically, the input stream is a byte stream comprising
307the 8-bit character code units of a particular encoding such
308as extended ASCII, ISO-8859-1 or UTF-8.   However, the method may also
309easily be used with wider code units such as the 16-bit code units of
310UTF-16.   In the case of a byte stream, the first step is to transpose
311the byte stream into eight parallel bit streams, such that bit stream
312$i$ comprises the $i^\text{th}$ bit of each byte.   These streams form
313a set of basis bit streams from which many other parallel bit
314streams can be calculated, such as character class bit
315streams such that each bit $j$ of the stream specifies
316whether character $j$ of the input stream is in the class
317or not.  Figure \ref{fig:streams} shows an example of an
318input byte stream in ASCII, the eight basis bit streams of the
319transposed representation, and the character class bit streams
320\verb:[a]:,
321\verb:[z]:, and
322\verb:[0-9]:
323that may be computed from the basis bit streams using bitwise logic.
324Transposition and character class construction are straightforward
325using the Parabix tool chain \cite{lin2012parabix}.
326
327\begin{figure}[tbh]
328\begin{center}
329\begin{tabular}{cr}\\
330input data  & \verb`a4534q--b29z---az---a4q--bca22z--`\\
331$B_7$ & \verb`.................................`\\
332$B_6$ & \verb`1....1..1..1...11...1.1..111..1..`\\
333$B_5$ & \verb`111111111111111111111111111111111`\\
334$B_4$ & \verb`.11111...111....1....11.....111..`\\
335$B_3$ & \verb`......11..11111.1111...11.....111`\\
336$B_2$ & \verb`.11.1.11....111..111.1.11......11`\\
337$B_1$ & \verb`...1....11.1....1........11.111..`\\
338$B_0$ & \verb`1.11.111..1.1111.1111.111.11...11`\\
339\verb:[a]: & \verb`1..............1....1......1.....`\\
340\verb:[z]: & \verb`...........1....1.............1..`\\
341\verb:[0-9]: & \verb`.1111....11..........1......11...`
342\end{tabular}
343
344\end{center}
345\caption{Basis and Character Class Streams}
346\label{fig:streams}
347\end{figure}
348
349
350\paragraph*{Marker Streams.}  Now consider how bit-parallel data
351streams can be used in regular expression matching.   Consider
352the problem of searching the input stream of Figure \ref{fig:streams}
353to finding occurrence of strings matching
354the regular expression \verb:a[0-9]*z:.
355The matching process involves the concept of {\em marker streams}, that
356is streams that mark the positions of current matches during the
357overall process.  In this case there are three marker streams computed
358during the match process, namely,
359$M_1$ representing match positions after an initial \verb:a:
360character has been found, $M_2$ representing positions
361reachable from positions marked by $M_1$ by further matching zero or
362more digits (\verb:[0-9]*:) and finally $M_3$ the stream
363marking positions after a final \verb:z: has been found.
364Without describing the details of how these streams are computed
365for the time being, Figure \ref{fig:streams} shows what each
366of these streams should be for our example matching problem.
367Note our convention that a marker stream contains a 1 bit
368at the next character position to be matched, that is,
369immediately past the last position that was matched.
370
371
372\paragraph*{MatchStar.}
373MatchStar takes a marker bitstream and a character class bitstream as input.  It returns all positions that can be reached by advancing the marker bitstream zero or more times through the character class bitstream.
374
375Figure \ref{fig:matchstar} illustrates the MatchStar method.  The second and third rows are the input bitstreams: the initial marker position bitstream and the character class bitstream derived from the source data.
376
377In the first operation ($T_0$), marker positions that cannot be advanced are temporarily removed from consideration by masking off marker positions that aren't character class positions using bitwise logic.  Next, the temporary marker bitstream is added to the character class bitstream.  $T_1$ has 1s in three types of positions.  There will be a 1 immediately following a block of character class positions that spanned one or more marker positions, at any character class positions that weren't affected by the addition (and are not part of the desired output), and at any marker position that wasn't the first in its block of character class positions.  Any character class positions that have a 0 in $T_1$ were affected by the addition and are part of the desired output.  These positions are obtained and the undesired 1 bits are removed by XORing with the character class stream. $T_2$ is now only missing marker positions that were removed in the first step as well as marker positions that were 1s in $T_1$.  The
378output marker stream is obtained by ORing $T_2$ with the initial marker stream.
379
380\begin{figure*}[tbh]
381\begin{center}
382\begin{tabular}{cr}\\
383source data & \verb`--142578---125-231-----127--5394---94761205-`\\
384$M_0$ & \verb`.......1......1..1..1...1.............1..1..`\\
385$D = $\verb:[0-9]: & \verb`..111111...111.111.....111..1111...11111111.`\\
386$T_0 = M_0 \wedge D$ & \verb`.......1.........1......1.............1..1..`\\
387$T_1 = T_0 + D$ & \verb`.1.........1111.......1..1..1111..1...1...1.`\\
388$T_2 = T_1 \oplus D$ & \verb`.1111111......1111....111.........1111.111..`\\
389$M_1 = T_2 \, | \, M_0$ & \verb`.1111111......1111..1.111.........11111111..`
390\end{tabular}
391
392\end{center}
393\caption{Match Star}
394\label{fig:matchstar1}
395\end{figure*}
396
397
398In general, given a marker stream $M$ and a character class stream $C$,
399the operation of MatchStar is defined by the following equation. 
400\[\text{MatchStar}(M, C) = (((M \wedge C) + C)  \oplus C) | M\]
401Given a set of initial marker positions, the result stream marks
402all possible positions that can be reached by 0 or more occurrences
403of characters in class $C$ from each position in $M$
404
405
406
407
408\section{Block-at-a-Time Processing}\label{sec:blockwise}
409
410The unbounded stream model of the previous section must of course
411be translated an implementation that proceeds block-at-a-time for
412realistic application.  In this, we primarily rely on the Pablo
413compiler of the Parabix toolchain \cite{lin2012parabix}.  Given input
414statements expressed as arbitrary-length bitstream equations, Pablo
415produces block-at-a-time C++ code that initializes and maintains all the necessary
416carry bits for each of the additions and shifts involved in the
417bitstream calculations.   
418
419In the present work, our principal contribution to the block-at-a-time
420model is the technique of long-stream addition described below.
421Otherwise, we were able to use Pablo directly in compiling our
422SSE2 and AVX2 implementations.   Our GPU implementation required
423some scripting to modify the output of the Pablo compiler for our
424purpose.
425
426\paragraph*{Long-Stream Addition.}  The maximum word size for
427addition on commodity processors is typically 64 bits.  In order
428to implement long-stream addition for block sizes of 256 or larger,
429a method for propagating carries through the individual stages of
43064-bit addition is required.  However, the normal technique of
431sequential addition using add-with-carry instructions, for example,
432is far from ideal.
433
434We have developed a technique using SIMD or SIMT methods for constant-time
435long-stream addition up to 4096 bits.   
436We assume the availability of the following SIMD/SIMT operations
437operating on vectors of $f$ 64-bit fields.
438\begin{itemize}
439\item \verb#simd<64>::add(X, Y)#: vertical SIMD addition of corresponding 64-bit fields
440in two vectors to produce a result vector of $f$ 64-bit fields.
441\item  \verb#simd<64>::eq(X, -1)# :  comparison of the 64-bit fields
442of \verb:x: each with the constant value -1 (all bits 1), producing
443an $f$-bit mask value,
444\item  \verb#hsimd<64>::mask(X)# : gathering the high bit of each 64-bit
445field into a single compressed $f$-bit mask value, and
446\item normal bitwise logic operations on $f$-bit masks, and
447\item  \verb#simd<64>::spread(x)# : distributing the bits of
448an $f$ bit mask, one bit each to the $f$ 64-bit fields of a vector.
449\end{itemize}
450
451Given these operations, our method for long stream addition of
452two $f \times 64$ bit values \verb:X: and \verb:Y: is the following.
453\begin{enumerate}
454\item Form the vector of 64-bit sums of \verb:x: and \verb:y:.
455\[\text{\tt R} = \text{\tt simd<64>::add(X, Y)} \]
456
457\item Extract the $f$-bit masks of \verb:X:, \verb:Y: and \verb:R:.
458\[\text{\tt x} = \text{\tt hsimd<64>::mask(X)} \]
459\[\text{\tt y} = \text{\tt hsimd<64>::mask(Y)} \]
460\[\text{\tt r} = \text{\tt hsimd<64>::mask(R)} \]
461
462\item Compute an $f$-bit mask of carries generated for each of the
46364-bit additions of \verb:X: and \verb:Y:.
464\[\text{\tt c} = (\text{\tt x} \wedge \text{\tt y}) \vee ((\text{\tt x} \vee \text{\tt y}) \wedge \neg \text{\tt r})\]
465
466\item Compute an $f$-bit mask of all fields of {\tt R} that will overflow with
467an incoming carry bit.  This is the {\em bubble mask}.
468\[\text{\tt b} = \text{\tt simd<64>::eq(R, -1)}\]
469
470\item Determine an $f$-bit mask identifying the fields of {\tt R} that need to be
471incremented to produce the final sum.  Here we find a new application of
472MatchStar!
473\[\text{\tt i} = \text{\tt MatchStar(c*2, b)}\]
474
475This is the key step.  The mask {\tt c} of outgoing carries must be
476shifted one position ({\tt c*2}) so that each outgoing carry bit becomes associated
477with the next digit.  At the incoming position, the carry will
478increment the 64-bit digit.   However, if this digit is all ones (as
479signalled by the corresponding bit of bubble mask {\tt b}, then the addition
480will generate another carry.  In fact, if there is a sequence of
481digits that are all ones, then the carry must bubble through
482each of them.   This is just MatchStar!
483
484\item Compute the final result {\tt Z}.
485\[\text{\tt Z} = \text{\tt simd<64>::add(R, simd<64>::spread(i))}\]
486
487\end{enumerate}
488\begin{figure}
489\begin{center}
490\begin{tabular}{c||r|r|r|r|r|r|r|r||}\cline{2-9}
491{\tt X} & {\tt 19} & {\tt 31} & {\tt BA} & {\tt 4C} & {\tt 3D} & {\tt 45} & {\tt 21} & {\tt F1} \\ \cline{2-9}
492{\tt Y} & {\tt 22} & {\tt 12} & {\tt 45} & {\tt B3} & {\tt E2} & {\tt 16} & {\tt 17} & {\tt 36} \\ \cline{2-9}
493{\tt R} & {\tt 3B} & {\tt 43} & {\tt FF} & {\tt FF} & {\tt 1F} & {\tt 5B} & {\tt 38} & {\tt 27} \\ \cline{2-9}
494{\tt x} & {\tt 0} & {\tt 0} & {\tt 1} & {\tt 0} & {\tt 0} & {\tt 0} & {\tt 0} & {\tt 1} \\ \cline{2-9}
495{\tt y} & {\tt 0} & {\tt 0} & {\tt 0} & {\tt 1} & {\tt 1} & {\tt 0} & {\tt 0} & {\tt 0} \\ \cline{2-9}
496{\tt r} & {\tt 0} & {\tt 0} & {\tt 1} & {\tt 1} & {\tt 0} & {\tt 0} & {\tt 0} & {\tt 0} \\ \cline{2-9}
497{\tt c} & {\tt 0} & {\tt 0} & {\tt 0} & {\tt 0} & {\tt 1} & {\tt 0} & {\tt 0} & {\tt 1} \\ \cline{2-9}
498{\tt c*2} & {\tt 0} & {\tt 0} & {\tt 0} & {\tt 1} & {\tt 0} & {\tt 0} & {\tt 1} & {\tt 0} \\ \cline{2-9}
499{\tt b} & {\tt 0} & {\tt 0} & {\tt 1} & {\tt 1} & {\tt 0} & {\tt 0} & {\tt 0} & {\tt 0} \\ \cline{2-9}
500{\tt i} & {\tt 0} & {\tt 1} & {\tt 1} & {\tt 1} & {\tt 0} & {\tt 0} & {\tt 1} & {\tt 0} \\ \cline{2-9}
501{\tt Z} & {\tt 3B} & {\tt 44} & {\tt 0} & {\tt 0} & {\tt 1F} & {\tt 5B} & {\tt 39} & {\tt 27} \\ \cline{2-9}
502\end{tabular}
503\end{center}
504\caption{Long Stream Addition}\label{fig:longadd}
505\end{figure}
506
507Figure \ref{fig:longadd} illustrates the process.  In the figure,
508we illustrate the process with 8-bit fields rather than 64-bit fields
509and show all field values in hexadecimal notation.  Note that
510two of the individual 8-bit additions produce carries, while two
511others produce {\tt FF} values that generate bubble bits.  The
512net result is that four of the original 8-bit sums must be
513incremented to produce the long stream result.
514
515A slight extension to the process produces a long-stream full adder
516that can be used in chained addition.  In this case, the
517adder must take an additional carry-in bit
518{\tt p} and produce a carry-out bit {\tt q}.
519This may be accomplished by incorporating {\tt p}
520in calculating the increment mask in the low bit position,
521and then extracting the carry-out {\tt q} from the high bit position.
522\[\text{\tt i} = \text{\tt MatchStar(c*2+p, b)}\]
523\[\text{\tt q} = \text{\tt i >> f}\]
524
525As described subsequently, we use a two-level long-stream addition technique
526in both our AVX2 and GPU implementations.  In principle, one can extend
527the technique to additional levels.  Using 64-bit adders throughout,
528$\lceil\log_{64}{n}\rceil$ steps are needed for $n$-bit addition.
529A three-level scheme could coordinate
53064 groups each performing 4096-bit long additions in a two-level structure.
531However, whether there are reasonable architectures that can support fine-grained
532SIMT style coordinate at this level is an open question.
533
534Using the methods outlined, it is quite conceivable that instruction
535set extensions to support long-stream addition could be added for
536future SIMD and GPU processors.   Given the fundamental nature
537of addition as a primitive and its novel application to regular
538expression matching as shown herein, it seems reasonable to expect
539such instructions to become available.
540\raggedbottom
541\section{Analytical Comparison with DFA and NFA Implementations}\label{sec:analysis}
542
543\begin{enumerate}
544\item Operations
545\item Memory behaviour per input byte: note tables of DFA/NFA.
546
547Bille and Throup \em{Faster regular expression matching}\cite{bille2009faster}
548
549\end{enumerate}
550
551
552
553\section{Commodity SIMD Implementation and Experimental Evaluation}\label{sec:SSE2}
554
555
556\subsection{Implementation Notes}
557\subsection{Evaluation Methodology}
558\subsection{Comparison}
559\begin{figure}
560\begin{center}
561\begin{tikzpicture}
562\begin{axis}[
563xtick=data,
564ylabel=Cycles per Byte,
565xticklabels={@,Date,Email,URIorEmail,xquote},
566tick label style={font=\tiny},
567enlargelimits=0.15,
568legend style={at={(0.5,-0.15)},
569anchor=north,legend columns=-1},
570ymax=8,
571ybar,
572bar width=7pt,
573]
574\addplot
575file {data/cycles1.dat};
576\addplot
577file {data/cycles2.dat};
578\addplot
579file {data/cycles3.dat};
580 
581\legend{Bitstreams,NRGrep,Grep,Annot}
582\end{axis}
583\end{tikzpicture}
584\end{center}
585\caption{Cycles per Byte}
586\end{figure}
587 
588\begin{figure}
589\begin{center}
590\begin{tikzpicture}
591\begin{axis}[
592xtick=data,
593ylabel=Instructions per Byte,
594xticklabels={@,Date,Email,URIorEmail,xquote},
595tick label style={font=\tiny},
596enlargelimits=0.15,
597legend style={at={(0.5,-0.15)},
598anchor=north,legend columns=-1},
599ymax=16,
600ybar,
601bar width=7pt,
602]
603\addplot
604file {data/instructions1.dat};
605\addplot
606file {data/instructions2.dat};
607\addplot
608file {data/instructions3.dat};
609 
610\legend{Bitstreams,NRGrep,Grep,Annot}
611\end{axis}
612\end{tikzpicture}
613\end{center}
614\caption{Instructions per Byte}
615\end{figure}
616
617\begin{figure}
618\begin{center}
619\begin{tikzpicture}
620\begin{axis}[
621xtick=data,
622ylabel=Instructions per Cycle,
623xticklabels={@,Date,Email,URIorEmail,xquote},
624tick label style={font=\tiny},
625enlargelimits=0.15,
626legend style={at={(0.5,-0.15)},
627anchor=north,legend columns=-1},
628ybar,
629bar width=7pt,
630]
631\addplot
632file {data/ipc1.dat};
633\addplot
634file {data/ipc2.dat};
635\addplot
636file {data/ipc3.dat};
637
638\legend{Bitstreams,NRGrep,Grep,Annot}
639\end{axis}
640\end{tikzpicture}
641\end{center}
642\caption{Instructions per Cycle}
643\end{figure}
644
645\begin{figure}
646\begin{center}
647\begin{tikzpicture}
648\begin{axis}[
649xtick=data,
650ylabel=Branch Misses per Byte,
651xticklabels={@,Date,Email,URIorEmail,xquote},
652tick label style={font=\tiny},
653enlargelimits=0.15,
654legend style={at={(0.5,-0.15)},
655anchor=north,legend columns=-1},
656ymax=0.03,
657ybar,
658bar width=7pt,
659]
660\addplot
661file {data/branch-misses1.dat};
662\addplot
663file {data/branch-misses2.dat};
664\addplot
665file {data/branch-misses3.dat};
666
667\legend{Bitstreams,NRGrep,Grep,Annot}
668\end{axis}
669\end{tikzpicture}
670\end{center}
671\caption{Branch Misses per Byte}
672\end{figure}
673
674
675
676\section{SIMD Scalability}\label{sec:AVX2}
677
678
679
680
681\begin{figure}
682\begin{center}
683\begin{tikzpicture}
684\begin{axis}[
685xtick=data,
686ylabel=Cycles per Byte,
687xticklabels={@,Date,Email,URIorEmail,xquote},
688tick label style={font=\tiny},
689enlargelimits=0.15,
690legend style={at={(0.5,-0.15)},
691anchor=north,legend columns=-1},
692ybar,
693bar width=7pt,
694]
695\addplot
696file {data/ssecycles.dat};
697\addplot
698file {data/avxcycles.dat};
699
700\legend{SSE2,AVX2,Annot}
701\end{axis}
702\end{tikzpicture}
703\end{center}
704\caption{Cycles per Byte}
705\end{figure}
706
707\begin{figure}
708\begin{center}
709\begin{tikzpicture}
710\begin{axis}[
711xtick=data,
712ylabel=Instructions per Byte,
713xticklabels={@,Date,Email,URIorEmail,xquote},
714tick label style={font=\tiny},
715enlargelimits=0.15,
716legend style={at={(0.5,-0.15)},
717anchor=north,legend columns=-1},
718ybar,
719bar width=7pt,
720]
721\addplot
722file {data/sseinstructions.dat};
723\addplot
724file {data/avxinstructions.dat};
725
726\legend{SSE2,AVX2,Annot}
727\end{axis}
728\end{tikzpicture}
729\end{center}
730\caption{Instructions per Byte}
731\end{figure}
732
733\begin{figure}
734\begin{center}
735\begin{tikzpicture}
736\begin{axis}[
737xtick=data,
738ylabel=Instructions per Cycle,
739xticklabels={@,Date,Email,URIorEmail,xquote},
740tick label style={font=\tiny},
741enlargelimits=0.15,
742legend style={at={(0.5,-0.15)},
743anchor=north,legend columns=-1},
744ybar,
745bar width=7pt,
746]
747\addplot
748file {data/sseipc.dat};
749\addplot
750file {data/avxipc.dat};
751
752
753\legend{SSE2,AVX2,Annot}
754\end{axis}
755\end{tikzpicture}
756\end{center}
757\caption{Instructions per Cycle}
758\end{figure}
759
760\begin{figure}
761\begin{center}
762\begin{tikzpicture}
763\begin{axis}[
764xtick=data,
765ylabel=Branch Misses per Byte,
766xticklabels={@,Date,Email,URIorEmail,xquote},
767tick label style={font=\tiny},
768enlargelimits=0.15,
769legend style={at={(0.5,-0.15)},
770anchor=north,legend columns=-1},
771ybar,
772bar width=7pt,
773]
774\addplot
775file {data/ssebranch-misses.dat};
776\addplot
777file {data/avxbranch-misses.dat};
778
779\legend{SSE2,AVX2,Annot}
780\end{axis}
781\end{tikzpicture}
782\end{center}
783\caption{Branch Misses per Byte}
784\end{figure}
785
786
787
788
789\subsection{AVX Stream Addition}
790 \begin{figure*}[tbh]
791\begin{center}
792\begin{verbatim}
793void add_ci_co(bitblock256_t x, bitblock256_t y, carry_t carry_in, carry_t & carry_out, bitblock256_t & sum) {
794  bitblock256_t all_ones = simd256<1>::constant<1>();
795  bitblock256_t gen = simd_and(x, y);
796  bitblock256_t prop = simd_xor(x, y);
797  bitblock256_t partial_sum = simd256<64>::add(x, y);
798  bitblock256_t carry = simd_or(gen, simd_andc(prop, partial_sum));
799  bitblock256_t bubble = simd256<64>::eq(partial_sum, all_ones);
800  uint64_t carry_mask = hsimd256<64>::signmask(carry) * 2 + convert(carry_in);
801  uint64_t bubble_mask = hsimd256<64>::signmask(bubble);
802  uint64_t carry_scan_thru_bubbles = (carry_mask + bubble_mask) &~ bubble_mask;
803  uint64_t increments = carry_scan_thru_bubbles | (carry_scan_thru_bubbles - carry_mask);
804  carry_out = convert(increments >> 4);
805  uint64_t spread = 0x0000200040008001 * increments & 0x0001000100010001;
806  sum = simd256<64>::add(partial_sum, _mm256_cvtepu16_epi64(avx_select_lo128(convert(spread))));
807}
808
809\end{verbatim}
810
811\end{center}
812\caption{AVX2 256-bit Addition}
813\label{fig:AVX2add}
814\end{figure*}
815
816\section{GPU Implementation}\label{sec:GPU}
817\begin{figure}
818\begin{center}
819\begin{tikzpicture}
820\begin{axis}[
821xtick=data,
822ylabel=Running Time (ms per byte),
823xticklabels={@,Date,Email,URIorEmail,xquote},
824tick label style={font=\tiny},
825enlargelimits=0.15,
826legend style={at={(0.5,-0.15)},
827anchor=north,legend columns=-1},
828ybar,
829bar width=7pt,
830]
831\addplot
832file {data/ssetime.dat};
833\addplot
834file {data/avxtime.dat};
835\addplot
836file {data/gputime.dat};
837
838\legend{SSE2,AVX2,GPU,Annot}
839\end{axis}
840\end{tikzpicture}
841\end{center}
842\caption{Running Time}
843\end{figure}
844
845
846
847
848\section{Miscellaneous}
849\subsection{Skipping}
850\input{re-Unicode}
851
852\input{conclusion}
853
854
855
856%\appendix
857%\section{Appendix Title}
858
859%This is the text of the appendix, if you need one.
860
861\acks
862
863This research was supported by grants from the Natural Sciences and Engineering Research Council of Canada and
864MITACS, Inc.
865
866% We recommend abbrvnat bibliography style.
867
868\bibliographystyle{abbrvnat}
869
870% The bibliography should be embedded for final submission.
871 
872\bibliography{reference}
873
874%\begin{thebibliography}{}
875%\softraggedright
876
877%\bibitem[Smith et~al.(2009)Smith, Jones]{smith02}
878%P. Q. Smith, and X. Y. Jones. ...reference text...
879%
880%\end{thebibliography}
881
882
883\end{document}
884
885
Note: See TracBrowser for help on using the repository browser.