source: docs/Balisage09/Bal2009came0601.xml @ 1804

Last change on this file since 1804 was 294, checked in by ksherdy, 10 years ago

Update Table 6 to correct for HTML 4 whitespace normalization.

File size: 87.8 KB
1<?xml version="1.0" encoding="UTF-8"?>
3<!DOCTYPE article SYSTEM "balisage-1-1.dtd">
4<article xmlns="" version="5.0-subset Balisage-1.1"
5   xml:id="HR-23632987-8973">
6   <title>Parallel Bit Stream Technology as a Foundation for XML Parsing Performance</title>
7   <info>
8      <confgroup>
9         <conftitle>International Symposium on Processing XML Efficiently: Overcoming Limits on
10            Space, Time, or Bandwidth</conftitle>
11         <confdates>August 10 2009</confdates>
12      </confgroup>
13      <abstract>
14         <para>By first transforming the octets (bytes) of XML texts into eight parallel bit
15            streams, the SIMD features of commodity processors can be exploited for parallel
16            processing of blocks of 128 input bytes at a time. Established transcoding and parsing
17            techniques are reviewed followed by new techniques including parsing with bitstream
18            addition. Further opportunities are discussed in light of expected advances in CPU
19            architecture and compiler technology. Implications for various APIs and information
20            models are presented as well opportunities for collaborative open-source
21         development.</para>
22      </abstract>
23      <author>
24         <personname>
25            <firstname>Rob</firstname>
26            <surname>Cameron</surname>
27         </personname>
28         <personblurb>
29            <para>Dr. Rob Cameron is Professor and Director of Computing Science at Simon Fraser
30               University. With a broad spectrum of research interests related to programming
31               languages, software engineering and sociotechnical design of public computing
32               infrastructure, he has recently been focusing on high performance text processing
33               using parallel bit stream technology and its applications to XML. He is also a
34               patentleft evangelist, advocating university-based technology transfer models
35               dedicated to free use in open source. </para>
37         </personblurb>
38         <affiliation>
39            <jobtitle>Professor of Computing Science</jobtitle>
40            <orgname>Simon Fraser University</orgname>
41         </affiliation>
42         <email></email>
43      </author>
44      <author>
45         <personname>
46            <firstname>Ken</firstname>
47            <surname>Herdy</surname>
48         </personname>
49         <personblurb>
50            <para> Ken Herdy completed an Advanced Diploma of Technology in Geographical Information
51               Systems at the British Columbia Institute of Technology in 2003 and earned a Bachelor
52               of Science in Computing Science with a Certificate in Spatial Information Systems at
53               Simon Fraser University in 2005. </para>
54            <para> Ken is currently pursuing graduate studies in Computing Science at Simon Fraser
55               University with industrial scholarship support from the Natural Sciences and
56               Engineering Research Council of Canada, the Mathematics of Information Technology and
57               Complex Systems NCE, and the BC Innovation Council. His research focus is an analysis
58               of the principal techniques that may be used to improve XML processing performance in
59               the context of the Geography Markup Language (GML). </para>
61         </personblurb>
62         <affiliation>
63            <jobtitle>Graduate Student, School of Computing Science</jobtitle>
64            <orgname>Simon Fraser University </orgname>
65         </affiliation>
66         <email></email>
67      </author>
68      <author>
69         <personname>
70            <firstname>Ehsan</firstname>
71            <surname>Amiri</surname>
72         </personname>
73         <personblurb>
74            <para>Ehsan Amiri is a PhD student of Computer Science at Simon Fraser University.
75               Before that he studied at Sharif University of Technology, Tehran, Iran. While his
76               graduate research has been focused on theoretical problems like fingerprinting, Ehsan
77               has worked on some software projects like development of a multi-node firewall as
78               well. More recently he has been developing compiler technology for automatic
79               generation of bit stream processing code. </para>
81         </personblurb>
82         <affiliation>
83            <jobtitle>Graduate Student, School of Computing Science</jobtitle>
84            <orgname>Simon Fraser University</orgname>
85         </affiliation>
86         <email></email>
87      </author>
88      <legalnotice>
89         <para>Copyright &#x000A9; 2009 Robert D. Cameron, Kenneth S. Herdy and Ehsan Amiri.
90            This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative
91            Works 2.5 Canada License.</para>
92      </legalnotice>
93      <keywordset role="author">
94         <keyword/>
95         <keyword/>
96         <keyword/>
97      </keywordset>
98   </info>
99   <section>
100      <title>Introduction</title>
101      <para> While particular XML applications may benefit from special-purpose hardware such as XML
102         chips [<xref linkend="XMLChip09"/>] or appliances [<xref linkend="Datapower09"/>], the bulk
103         of the world's XML processing workload will continue to be handled by XML software stacks
104         on commodity processors. Exploiting the SIMD capabilities of such processors such as the
105         SSE instructions of x86 chips, parallel bit stream technology offers the potential of
106         dramatic improvement over byte-at-a-time processing for a variety of XML processing tasks.
107         Character set issues such as Unicode validation and transcoding [<xref linkend="PPoPP08"
108         />], normalization of line breaks and white space and XML character validation can be
109         handled fully in parallel using this representation. Lexical item streams, such as the bit
110         stream marking the positions of opening angle brackets, can also be formed in parallel.
111         Bit-scan instructions of commodity processors may then be used on lexical item streams to
112         implement rapid single-instruction scanning across variable-length multi-byte text blocks
113         as in the Parabix XML parser [<xref linkend="CASCON08"/>]. Overall, these techniques may be
114         combined to yield end-to-end performance that may be 1.5X to 15X faster than alternatives
115            [<xref linkend="SVGOpen08"/>].</para>
116      <para>Continued research in parallel bit stream techniques as well as more conventional
117         application of SIMD techniques in XML processing offers further prospects for improvement
118         of core XML components as well as for tackling performance-critical tasks further up the
119         stack. A newly prototyped technique for parallel tag parsing using bitstream addition is
120         expected to improve parsing performance even beyond that achieved using sequential bit
121         scans. Several techniques for improved symbol table performance are being investigated,
122         including parallel hash value calculation and length-based sorting using the cheap length
123         determination afforded by bit scans. To deliver the benefits of parallel bit stream
124         technology to the Java world, we are developing Array Set Model (ASM) representations of
125         XML Infoset and other XML information models for efficient transmission across the JNI
126         boundary.</para>
128      <para>Amplifying these software advances, continuing hardware advances in commodity processors
129         increase the relative advantage of parallel bit stream techniques over traditional
130         byte-at-a-time processors. For example, the Intel Core architecture improved SSE processing
131         to give superscalar execution of bitwise logic operations (3 instructions per cycle vs. 1
132         in Pentium 4). Upcoming 256-bit AVX technology extends the register set and replaces
133         destructive two-operand instructions with a nondestructive three-operand form. General
134         purpose programming on graphic processing units (GPGPU) such as the upcoming 512-bit
135         Larrabee processor may also be useful for XML applications using parallel bit streams. New
136         instruction set architectures may also offer dramatic improvements in core algorithms.
137         Using the relatively simple extensions to support the principle of inductive doubling, a 3X
138         improvement in several core parallel bit stream algorithms may be achieved [<xref
139            linkend="ASPLOS09"/>]. Other possibilities include direct implementation of parallel
140         extract and parallel deposit (pex/pdep) instructions [<xref linkend="Pex06"/>], and
141         bit-level interleave operations as in Larrabee, each of which would have important
142         application to parallel bit stream processing.</para>
144      <para>Further prospects for XML performance improvement arise from leveraging the
145         intraregister parallelism of parallel bit stream technology to exploit the interchip
146         parallelism of multicore computing. Parallel bit stream techniques can support multicore
147         parallelism in both data partitioning and task partitioning models. For example, the
148         datasection partitioning approach of Wu, Zhang, Yu and Li may be used to partition blocks
149         for speculative parallel parsing on separate cores followed by a postprocessing step to
150         join partial S-trees [<xref linkend="Wu08"/>].</para>
152      <para>In our view, the established and expected performance advantages of parallel bit stream
153         technology over traditional byte-at-a-time processing are so compelling that parallel bit
154         stream technology should ultimately form the foundation of every high-performance XML
155         software stack. We envision a common high-performance XML kernel that may be customized to
156         a variety of processor architectures and that supports a wide range of existing and new XML
157         APIs. Widespread deployment of this technology should greatly benefit the XML community in
158         addressing both the deserved and undeserved criticism of XML on performance grounds. A
159         further benefit of improved performance is a substantial greening of XML technologies.</para>
161      <para>To complement our research program investigating fundamental algorithms and issues in
162         high-performance XML processing, our work also involves development of open source software
163         implementing these algorithms, with a goal of full conformance to relevant specifications.
164         From the research perspective, this approach is valuable in ensuring that the full
165         complexity of required XML processing is addressed in reporting and assessing processing
166         results. However, our goal is also to use this open source software as a basis of
167         technology transfer. A Simon Fraser University spin-off company, called International
168         Characters, Inc., has been created to commercialize the results of this work using a
169         patent-based open source model.</para>
171      <para>To date, we have not yet been successful in establishing a broader community of
172         participation with our open source code base. Within open-source communities, there is
173         often a general antipathy towards software patents; this may limit engagement with our
174         technology, even though it has been dedicated for free use in open source. </para>
176      <para>A further complication is the inherent difficulty of SIMD programming in general, and
177         parallel bit stream programming in particular. Considerable work is required with each new
178         algorithmic technique being investigated as well as in retargetting our techniques for each
179         new development in SIMD and multicore processor technologies. To address these concerns, we
180         have increasingly shifted the emphasis of our research program towards compiler technology
181         capable of generating parallel bit stream code from higher-level specifications.</para>
182   </section>
184   <section>
185      <title>A Catalog of Parallel Bit Streams for XML</title>
186      <section>
187         <title>Introduction</title>
188         <para>In this section, we introduce the fundamental concepts of parallel bit stream
189            technology and present a comprehensive catalog of parallel bit streams for use in XML
190            processing. In presenting this catalog, the focus is on the specification of the bit
191            streams as data streams in one-to-one correspondence with the character code units of an
192            input XML stream. The goal is to define these bit streams in the abstract without
193            initially considering memory layouts, register widths or other issues related to
194            particular target architectures. In cataloging these techniques, we also hope to convey
195            a sense of the breadth of applications of parallel bit stream technology to XML
196            processing tasks. </para>
197      </section>
199      <section>
200         <title>Basis Bit Streams</title>
201         <para>Given a byte-oriented text stream represented in UTF-8, for example, we define a
202            transform representation of this text consisting of a set of eight parallel bit streams
203            for the individual bits of each byte. Thus, the <code>Bit0</code> stream is the stream
204            of bits consisting of bit 0 of each byte in the input byte stream, <code>Bit1</code> is
205            the bit stream consisting of bit 1 of each byte in the input stream and so on. The set
206            of streams <code>Bit0</code> through <code>Bit7</code> are known as the <emphasis>basis
207               streams</emphasis> of the parallel bit stream representation. The following table
208            shows an example XML character stream together with its representation as a set of 8
209            basis streams. <table>
210               <caption>
211                  <para>XML Character Stream Transposition.</para>
212               </caption>
213               <colgroup>
214                  <col align="left" valign="top"/>
215                  <col align="left" valign="top"/>
216                  <col align="left" valign="top"/>
217                  <col align="left" valign="top"/>
218                  <col align="left" valign="top"/>
219                  <col align="left" valign="top"/>
220               </colgroup>
221               <tbody>
222                  <tr valign="top">
223                     <td>Input Data</td>
224                     <td>
225                        <code>&lt;</code>
226                     </td>
227                     <td>
228                        <code>t</code>
229                     </td>
230                     <td>
231                        <code>a</code>
232                     </td>
233                     <td>
234                        <code>g</code>
235                     </td>
236                     <td>
237                        <code>/</code>
238                     </td>
239                     <td>
240                        <code>&gt;</code>
241                     </td>
242                  </tr>
243                  <tr valign="top">
244                     <td>ASCII</td>
245                     <td>
246                        <code>00111100</code>
247                     </td>
248                     <td>
249                        <code>01110100</code>
250                     </td>
251                     <td>
252                        <code>01100001</code>
253                     </td>
254                     <td>
255                        <code>01100111</code>
256                     </td>
257                     <td>
258                        <code>00101111</code>
259                     </td>
260                     <td>
261                        <code>00111110</code>
262                     </td>
263                  </tr>
264                  <tr valign="top">
265                     <td>Bit0</td>
266                     <td>
267                        <code>0</code>
268                     </td>
269                     <td>
270                        <code>0</code>
271                     </td>
272                     <td>
273                        <code>0</code>
274                     </td>
275                     <td>
276                        <code>0</code>
277                     </td>
278                     <td>
279                        <code>0</code>
280                     </td>
281                     <td>
282                        <code>0</code>
283                     </td>
284                  </tr>
285                  <tr valign="top">
286                     <td>Bit1</td>
287                     <td>
288                        <code>0</code>
289                     </td>
290                     <td>
291                        <code>1</code>
292                     </td>
293                     <td>
294                        <code>1</code>
295                     </td>
296                     <td>
297                        <code>1</code>
298                     </td>
299                     <td>
300                        <code>0</code>
301                     </td>
302                     <td>
303                        <code>0</code>
304                     </td>
305                  </tr>
306                  <tr valign="top">
307                     <td>Bit2</td>
308                     <td>
309                        <code>1</code>
310                     </td>
311                     <td>
312                        <code>1</code>
313                     </td>
314                     <td>
315                        <code>1</code>
316                     </td>
317                     <td>
318                        <code>1</code>
319                     </td>
320                     <td>
321                        <code>1</code>
322                     </td>
323                     <td>
324                        <code>1</code>
325                     </td>
326                  </tr>
327                  <tr valign="top">
328                     <td>Bit3</td>
329                     <td>
330                        <code>1</code>
331                     </td>
332                     <td>
333                        <code>1</code>
334                     </td>
335                     <td>
336                        <code>0</code>
337                     </td>
338                     <td>
339                        <code>0</code>
340                     </td>
341                     <td>
342                        <code>0</code>
343                     </td>
344                     <td>
345                        <code>1</code>
346                     </td>
347                  </tr>
348                  <tr valign="top">
349                     <td>Bit4</td>
350                     <td>
351                        <code>1</code>
352                     </td>
353                     <td>
354                        <code>0</code>
355                     </td>
356                     <td>
357                        <code>0</code>
358                     </td>
359                     <td>
360                        <code>0</code>
361                     </td>
362                     <td>
363                        <code>1</code>
364                     </td>
365                     <td>
366                        <code>1</code>
367                     </td>
368                  </tr>
369                  <tr valign="top">
370                     <td>Bit5</td>
371                     <td>
372                        <code>1</code>
373                     </td>
374                     <td>
375                        <code>1</code>
376                     </td>
377                     <td>
378                        <code>0</code>
379                     </td>
380                     <td>
381                        <code>1</code>
382                     </td>
383                     <td>
384                        <code>1</code>
385                     </td>
386                     <td>
387                        <code>1</code>
388                     </td>
389                  </tr>
390                  <tr valign="top">
391                     <td>Bit6</td>
392                     <td>
393                        <code>0</code>
394                     </td>
395                     <td>
396                        <code>0</code>
397                     </td>
398                     <td>
399                        <code>0</code>
400                     </td>
401                     <td>
402                        <code>1</code>
403                     </td>
404                     <td>
405                        <code>1</code>
406                     </td>
407                     <td>
408                        <code>1</code>
409                     </td>
410                  </tr>
411                  <tr valign="top">
412                     <td>Bit7</td>
413                     <td>
414                        <code>0</code>
415                     </td>
416                     <td>
417                        <code>0</code>
418                     </td>
419                     <td>
420                        <code>1</code>
421                     </td>
422                     <td>
423                        <code>1</code>
424                     </td>
425                     <td>
426                        <code>1</code>
427                     </td>
428                     <td>
429                        <code>0</code>
430                     </td>
431                  </tr>
432               </tbody>
433            </table>
434         </para>
435         <para> Depending on the features of a particular processor architecture, there are a number
436            of algorithms for transposition to parallel bit stream form. Several of these algorithms
437            employ a three-stage structure. In the first stage, the input byte stream is divided
438            into a pair of half-length streams consisting of four bits for each byte, for example,
439            one stream for the high nybble of each byte and another for the low nybble of each byte.
440            In the second stage, these streams of four bits per byte are each divided into streams
441            consisting of two bits per original byte, for example streams for the
442            <code>Bit0/Bit1</code>, <code>Bit2/Bit3</code>, <code>Bit4/Bit5</code>, and
443               <code>Bit6/Bit7</code> pairs. In the final stage, the streams are further subdivided
444            in the individual bit streams. </para>
445         <para> Using SIMD capabilities, this process is quite efficient, with an amortized cost of
446            1.1 CPU cycles per input byte on Intel Core 2 with SSE, or 0.6 CPU cycles per input byte
447            on Power PC G4 with Altivec. With future advances in processor technology, this
448            transposition overhead is expected to reduce, possibly taking advantage of upcoming
449            parallel extract (pex) instructions on Intel technology. In the ideal, only 24
450            instructions are needed to transform a block of 128 input bytes using 128-bit SSE
451            registers using the inductive doubling instruction set architecture, representing an
452            overhead of less than 0.2 instructions per input byte. </para>
453      </section>
455      <section>
456         <title>General Streams</title>
457         <para>This section describes bit streams which support basic processing operations.</para>
459         <section>
460            <title>Deletion Mask Streams</title>
461            <para>DelMask (deletion mask) streams marks character code unit positions for deletion.
462               Since the deletion operation is dependency free across many stages of XML processing,
463               it is possible to simply mark and record deletion positions as deletion mask streams for future processing. A single
464               invocation of a SIMD based parallel deletion algorithm can then perform the deletion of
465               positions accumulated across a number of stages through a bitwise ORing of deletion
466               masks. For example, deletion arises in the replacement of predefined entities with a
467               single character, such as in the replacement of the &amp;amp; entity, with the
468               &amp; character. Deletion also arises in XML
469               end-of-line handling, and CDATA section delimeter processing. Several algorithms to
470               delete bits at positions marked by DelMask are possible [<xref linkend="u8u16"/>]. </para>
471            <para>The following table provides an example of generating a DelMask in the context of
472               bit stream based parsing of well-formed character references and predefined entities.
473               The result is the generation of a DelMask stream. <table>
474                  <caption>
475                     <para>DelMask Stream Generation</para>
476                  </caption>
477                  <colgroup>
478                     <col align="left" valign="top"/>
479                     <col align="left" valign="top"/>
480                  </colgroup>
481                  <tbody>
482                     <tr valign="top">
483                        <td>Input Data</td>
484                        <td>
485                           <code>&amp;gt; &amp;#13; &amp;#x0a;</code>
486                        </td>
487                     </tr>
488                     <tr valign="top">
489                        <td>GenRefs</td>
490                        <td>
491                           <code>_11______________</code>
492                        </td>
493                     </tr>
495                     <tr valign="top">
496                        <td>DecRefs</td>
497                        <td>
498                           <code>_______11________</code>
499                        </td>
500                     </tr>
501                     <tr valign="top">
502                        <td>HexRefs</td>
503                        <td>
504                           <code>______________11_</code>
505                        </td>
506                     </tr>
507                     <tr valign="top">
508                        <td>DelMask</td>
509                        <td>
510                           <code>111__1111__11111_</code>
511                        </td>
512                     </tr>
513                     <tr valign="top">
514                        <td>ErrorFlag</td>
515                        <td>
516                           <code>_________________</code>
517                        </td>
518                     </tr>
519                  </tbody>
520               </table>
521            </para>
522         </section>
524         <section>
525            <title>Error Flag Streams </title>
526            <para>Error flag streams indicates the character code unit positions of syntactical
527               errors. XML processing examples which benefit from the marking of error positions
528               include UTF-8 character sequence validation and XML parsing [<xref linkend="u8u16"
529               />].</para>
530            <para>The following table provides an example of using bit streams to parse character
531               references and predefined entities which fail to meet the XML 1.0 well-formedness
532               constraints. The result is the generation of an error flag stream that marks the
533               positions of mal-formed decimal and hexical character references respectively. <table>
534                  <caption>
535                     <para>Error Flag Stream Generation</para>
536                  </caption>
537                  <colgroup>
538                     <col align="left" valign="top"/>
539                     <col align="left" valign="top"/>
540                  </colgroup>
541                  <tbody>
542                     <tr valign="top">
543                        <td>Input Data</td>
544                        <td>
545                           <code>&amp;gt; &amp;#, &amp;#x; </code>
546                        </td>
547                     </tr>
548                     <tr valign="top">
549                        <td>GenRefs</td>
550                        <td>
551                           <code>_11___________</code>
552                        </td>
553                     </tr>
554                     <tr valign="top">
555                        <td>DecRefs</td>
556                        <td>
557                           <code>______________</code>
558                        </td>
559                     </tr>
560                     <tr valign="top">
561                        <td>HexRefs</td>
562                        <td>
563                           <code>______________</code>
564                        </td>
565                     </tr>
566                     <tr valign="top">
567                        <td>DelMask</td>
568                        <td>
569                           <code>111__11__111__</code>
570                        </td>
571                     </tr>
572                     <tr valign="top">
573                        <td>ErrorFlag</td>
574                        <td>
575                           <code>_______1____1_</code>
576                        </td>
577                     </tr>
578                  </tbody>
579               </table>
580            </para>
581         </section>
583      </section>
585      <section>
586         <title>Lexical Item Streams</title>
587         <para>Lexical item streams differ from traditional streams of tokens in that they are bit
588            streams that mark the positions of tokens, whitespace or delimiters. Additional bit
589            streams, such as the reference streams and callout streams, are subsequently constructed
590            based on the information held within the set of lexical items streams. Differentiation
591            between the actual tokens that may occur at a particular point (e.g., the different XML
592            tokens that begin “&lt;”) may be performed using multicharacter recognizers on the
593            bytestream representation [<xref linkend="CASCON08"/>].</para>
594         <para>A key role of lexical item streams in XML parsing is to facilitate fast scanning
595            operations. For example, a left angle bracket lexical item stream may be formed to
596            identify those character code unit positions at which a “&lt;” character occurs.
597            Hardware register bit scan operations may then be used by the XML parser on the left
598            angle bracket stream to efficiently identify the position of the next “&lt;”. Based
599            on the capabilities of current commodity processors, a single register bit scan
600            operation may effectively scan up to 64 byte positions with a single instruction.</para>
601         <para>Overall, the construction of the full set of lexical item stream computations
602            requires approximately 1.0 CPU cycles per byte when implemented for 128 positions at a
603            time using 128-bit SSE registers on Intel Core2 processors [<xref linkend="CASCON08"/>].
604            The following table defines the core lexical item streams defined by the Parabix XML
605            parser.</para>
606         <para>
607            <table>
608               <caption>
609                  <para>Lexical item stream descriptions.</para>
610               </caption>
611               <tbody>
612                  <tr>
613                     <td align="left"> LAngle </td>
614                     <td align="left"> Marks the position of any left angle bracket character.</td>
615                  </tr>
616                  <tr>
617                     <td align="left"> RAngle </td>
618                     <td align="left"> Marks the position of any right angle bracket character.</td>
619                  </tr>
620                  <tr>
621                     <td align="left"> LBracket </td>
622                     <td align="left"> Marks the position of any left square bracker character.</td>
623                  </tr>
624                  <tr>
625                     <td align="left"> RBracket </td>
626                     <td align="left"> Marks the position of any right square bracket
627                     character.</td>
628                  </tr>
629                  <tr>
630                     <td align="left"> Exclam </td>
631                     <td align="left"> Marks the position of any exclamation mark character.</td>
632                  </tr>
633                  <tr>
634                     <td align="left"> QMark </td>
635                     <td align="left"> Marks the position of any question mark character.</td>
636                  </tr>
637                  <tr>
638                     <td align="left"> Hyphen </td>
639                     <td align="left"> Marks the position of any hyphen character.</td>
640                  </tr>
641                  <tr>
642                     <td align="left"> Equals </td>
643                     <td align="left"> Marks the position of any equal sign character.</td>
644                  </tr>
645                  <tr>
646                     <td align="left"> SQuote </td>
647                     <td align="left"> Marks the position of any single quote character.</td>
648                  </tr>
649                  <tr>
650                     <td align="left"> DQuote </td>
651                     <td align="left"> Marks the position of any double quote character.</td>
652                  </tr>
653                  <tr>
654                     <td align="left"> Slash </td>
655                     <td align="left"> Marks the position of any forward slash character</td>
656                  </tr>
657                  <tr>
658                     <td align="left"> NameScan </td>
659                     <td align="left"> Marks the position of any XML name character.</td>
660                  </tr>
661                  <tr>
662                     <td align="left"> WS </td>
663                     <td align="left"> Marks the position of any XML 1.0 whitespace character.</td>
664                  </tr>
665                  <tr>
666                     <td align="left"> PI_start </td>
667                     <td align="left"> Marks the position of the start of any processing instruction
668                        at the '?' character position.</td>
669                  </tr>
670                  <tr>
671                     <td align="left"> PI_end </td>
672                     <td align="left"> Marks the position of any end of any processing instruction
673                        at the '>' character position.</td>
674                  </tr>
675                  <tr>
676                     <td align="left"> CtCD_start </td>
677                     <td align="left"> Marks the position of the start of any comment or CDATA
678                        section at the '!' character position.</td>
679                  </tr>
680                  <tr>
681                     <td align="left"> EndTag_start </td>
682                     <td align="left"> Marks the position of any end tag at the '/' character
683                        position.</td>
684                  </tr>
685                  <tr>
686                     <td align="left"> CD_end </td>
687                     <td align="left"> Marks the position of the end of any CDATA section at the '>'
688                        character position. </td>
689                  </tr>
690                  <tr>
691                     <td align="left"> DoubleHyphen </td>
692                     <td align="left"> Marks the position of any double hyphen character.</td>
693                  </tr>
694                  <tr>
695                     <td align="left"> RefStart </td>
696                     <td align="left"> Marks the position of any ampersand character.</td>
697                  </tr>
698                  <tr>
699                     <td align="left"> Hash </td>
700                     <td align="left"> Marks the position of any hash character.</td>
701                  </tr>
702                  <tr>
703                     <td align="left"> x </td>
704                     <td align="left"> Marks the position of any 'x' character.</td>
705                  </tr>
706                  <tr>
707                     <td align="left"> Digit </td>
708                     <td align="left"> Marks the position of any digit.</td>
709                  </tr>
710                  <tr>
711                     <td align="left"> Hex </td>
712                     <td align="left"> Marks the position of any hexidecimal character.</td>
713                  </tr>
714                  <tr>
715                     <td align="left"> Semicolon </td>
716                     <td align="left"> Marks the position of any semicolon character.</td>
717                  </tr>
718               </tbody>
719            </table>
720         </para>
721         <para> The following illustrates a number of the lexical item streams. </para>
722         <para>
723            <table>
724               <caption>
725                  <para>Lexical Item Streams</para>
726               </caption>
727               <colgroup>
728                  <col align="left" valign="top"/>
729                  <col align="left" valign="top"/>
730               </colgroup>
731               <tbody>
732                  <tr valign="top">
733                     <td>Input Data</td>
734                     <td>
735                        <code>&lt;tag&gt;&lt;tag&gt; text &amp;lt;
736                           &amp;#x3e; &lt;/tag&gt;&lt;/tag&gt;</code>
737                     </td>
738                  </tr>
740                  <tr valign="top">
741                     <td>LAngle</td>
742                     <td>
743                        <code>1____1______________________1_____1_____</code>
744                     </td>
745                  </tr>
746                  <tr valign="top">
747                     <td>RAngle</td>
748                     <td>
749                        <code>____1____1_______________________1_____1</code>
750                     </td>
751                  </tr>
752                  <tr valign="top">
753                     <td>WS</td>
754                     <td>
755                        <code>__________1____1____1______1____________</code>
756                     </td>
757                  </tr>
758                  <tr valign="top">
759                     <td>RefStart</td>
760                     <td>
761                        <code>________________1____1__________________</code>
762                     </td>
763                  </tr>
764                  <tr valign="top">
765                     <td>Hex</td>
766                     <td>
767                        <code>__1____1____1___________11_____1_____1__</code>
768                     </td>
769                  </tr>
771                  <tr valign="top">
772                     <td>Semicolon</td>
773                     <td>
774                        <code>___________________1______1_____________</code>
775                     </td>
776                  </tr>
777                  <tr valign="top">
778                     <td>Slash</td>
779                     <td>
780                        <code>_____________________________1_____1____</code>
781                     </td>
782                  </tr>
783               </tbody>
784            </table>
785         </para>
787      </section>
789      <section>
790         <title>UTF-8 Byte Classification, Scope and Validation Streams</title>
791         <para> An XML parser must accept the UTF-8 encoding of Unicode [<xref linkend="XML10"/>].
792            It is a fatal error if an XML document determined to be in UTF-8 contains byte sequences
793            that are not legal in that encoding. UTF-8 byte classification, scope, XML character
794            validation and error flag bit streams are defined to validate UTF-8 byte sequences and
795            support transcoding to UTF-16.</para>
797         <section>
798            <title>UTF-8 Byte Classification Streams</title>
799            <para>UTF-8 byte classification bit streams classify UTF-8 bytes based on their role in
800               forming single and multibyte sequences. The u8Prefix and u8Suffix bit streams
801               identify bytes that represent, respectively, prefix or suffix bytes of multibyte
802               UTF-8 sequences. The u8UniByte bit stream identifies those bytes that may be
803               considered single-byte sequences. The u8Prefix2, u8Prefix3, and u8Prefix4 refine the
804               u8Prefix respectively indicating prefixes of two, three or four byte
805            sequences respectively.</para>
806         </section>
808         <section>
809            <title>UTF-8 Scope Streams</title>
810            <para> Scope streams represent expectations established by UTF-8 prefix bytes. For
811               example, the u8Scope22 bit stream represents the positions at which the second byte of a
812               two-byte sequence is expected based on the occurrence of a two-byte prefix in the
813               immediately preceding position. The u8scope32, u8Scope33, u8Scope42, u8scope43, and
814               u8Scope44 complete the set of UTF-8 scope streams.</para>
815            <para> The following example demonstrates the UTF-8 character encoding validation
816               process using parallel bit stream techniques. The result of this validation process
817               is an error flag stream identifying the positions at which errors occur.</para>
818            <para>
819               <table>
820                  <caption>
821                     <para>UTF-8 Scope Streams</para>
822                  </caption>
823                  <colgroup>
824                     <col align="left" valign="top"/>
825                     <col align="left" valign="top"/>
826                  </colgroup>
827                  <tbody> <tr valign="top"><td>Input Data</td><td><code>A Text in Farsi:&#160;ى&#160;ك&#160;&#160;م&#160;ت&#160;ن&#160;&#160;ف&#160;ا&#160;ر&#160;س&#160;ى</code></td></tr>
829                     <tr valign="top">
830                        <td>High Nybbles</td>
831                        <td>
832                           <code>42567726624677632D8DBDBDAD82D8DAD82D8D8</code>
833                        </td>
834                     </tr>
835                     <tr valign="top">
836                        <td>Low Nybbles</td>
838                        <td>
839                           <code>10458409E061239A099838187910968A9509399</code>
840                        </td>
841                     </tr>
842                     <tr valign="top">
843                        <td>u8Unibyte</td>
844                        <td>
845                           <code>11111111111111111__________1______1____</code>
847                        </td>
848                     </tr>
849                     <tr valign="top">
850                        <td>u8Prefix</td>
851                        <td>
852                           <code>_________________1_1_1_1_1__1_1_1__1_1_</code>
853                        </td>
854                     </tr>
856                     <tr valign="top">
857                        <td>u8Suffix</td>
858                        <td>
859                           <code>__________________1_1_1_1_1__1_1_1__1_1</code>
860                        </td>
861                     </tr>
862                     <tr valign="top">
863                        <td>u8Prefix2</td>
865                        <td>
866                           <code>_________________1_1_1_1_1__1_1_1__1_1_</code>
867                        </td>
868                     </tr>
869                     <tr valign="top">
870                        <td>u8Scope22</td>
871                        <td>
872                           <code>__________________1_1_1_1_1__1_1_1__1_1</code>
874                        </td>
875                     </tr>
876                     <tr valign="top">
877                        <td>ErrorFlag</td>
878                        <td>
879                           <code>_______________________________________</code>
880                        </td>
881                     </tr>
883                  </tbody>
884               </table>
887            </para>
888         </section>
890         <section>
891            <title>UTF-8 Validation Streams</title>
892            <para> Proper formation of UTF-8 byte sequences requires that the correct number of
893               suffix bytes always follow a UTF-8 prefix byte, and that certain illegal byte
894               combinations are ruled out. For example, sequences beginning with the prefix bytes
895               0xF5 through 0xFF are illegal as they would represent code point values above 10FFFF.
896               In addition, there are constraints on the first suffix byte following certain special
897               prefixes, namely that a suffix following the prefix 0xE0 must fall in the range
898               0xA0–0xBF, a suffix following the prefix 0xED must fall in the range 0x80–0x9F, a
899               suffix following the prefix 0xF0 must fall in the range 0x90–0xBF and a suffix
900               following the prefix 0xF4 must fall in the range 0x80–0x8F. The task of ensuring that
901               each of these constraints hold is known as UTF-8 validation. The bit streams xE0,
902               xED, xF0, xF4, xA0_xBF, x80_x9F, x90_xBF, and x80_x8F are constructed to flag the
903               aforementioned UTF-8 validation errors. The result of UTF-8 validation is a UTF-8
904               error flag bit stream contructed as the ORing of a series of UTF-8 validation tests.
905            </para>
906         </section>
908         <section>
909            <title>XML Character Validation Streams</title>
910            <para>The UTF-8 character sequences <emphasis>0xEF 0xBF 0xBF</emphasis> and
911                  <emphasis>0xEF 0xBF 0xBE</emphasis> correspond to the Unicode code points 0xFFFE
912               and 0xFFFF respectively. In XML 1.0, 0xFFFE and 0xFFFF represent characters outside
913               the legal XML character ranges. As such, bit streams which mark 0xEF, 0xBF, and 0xBE
914               character are constructed to flag illegal UTF-8 character sequences. </para>
915         </section>
917         <section>
918            <title>UTF-8 to UTF-16 Transcoding</title>
919            <para>UTF-8 is often preferred for storage and data exchange, it is suitable for
920               processing, but it is significantly more complex to process than UTF-16 [<xref
921                  linkend="Unicode"/>]. As such, XML documents are typically encoded in UTF-8 for
922               serialization and transport, and subsequently transcoded to UTF-16 for processing
923               with programming languages such as Java and C#. Following the parallel bit stream
924               methods developed for the u8u16 transcoder, a high-performance standalone UTF-8 to
925               UTF-16 transcoder [<xref linkend="u8u16"/>], transcoding to UTF-16 may be achieved by
926               computing a series of 16 bit streams. One stream for each of the individual bits of a
927               UTF-16 code unit. </para>
928            <para>The bit streams for UTF-16 are conveniently divided into groups: the eight streams
929               u16Hi0, u16Hi1, ..., u16Hi7 for the high byte of each UTF-16 code unit and the eight
930               streams u16Lo1, ..., u16Lo7 for the low byte. Upon conversion of the parallel bit
931               stream data back to byte streams, eight sequential byte streams U16h0, U16h1, ...,
932               U16Hi7 are used for the high byte of each UTF-16 code unit, while U16Lo0, U16Lo1,...,
933               U16Lo7 are used for the corresponding low byte. Interleaving these streams then
934               produces the full UTF-16 doublebyte stream.</para>
935         </section>
937         <section>
938            <title>UTF-8 Indexed UTF-16 Streams</title>
939            <para>UTF-16 bit streams are initially defined in UTF-8 indexed form. That is, with sets
940               of bits in one-to-one correspondence with UTF-8 bytes. However, only one set of
941               UTF-16 bits is required for encoding two or three-byte UTF-8 sequences and only two
942               sets are required for surrogate pairs corresponding to four-byte UTF-8 sequences. The
943               u8LastByte (u8UniByte , u8Scope22 , u8Scope33 , and u8Scope44 ) and u8Scope42 streams
944               mark the positions at which the correct UTF-16 bits are computed. The bit sets at
945               other positions must be deleted to compress the streams to the UTF-16 indexed form.
946            </para>
947         </section>
948      </section>
950      <section>
951         <title>Control Character Streams</title>
952         <para>The control character bit streams marks ASCII control characters in the range
953            0x00-0x1F. Additional control character bit streams mark the tab, carriage return, line
954            feed, and space character. In addition, a bit stream to mark carriage return line
955            combinations is also constructed. Presently, control character bit streams support the
956            operations of XML 1.0 character validation and XML end-of-line handling.</para>
958         <section>
959            <title>XML Character Validation</title>
960            <para>Legal characters in XML are the tab, carriage return, and line feed characters,
961               together with all Unicode characters and excluding the surrogate blocks, as well as hexadecimal OxFFFE and
962               OxFFFF [<xref linkend="XML10"/>]. The x00_x1F bit stream is constructed and used in
963               combination with the additional control character bit streams to flags the positions
964               of illegal control characters.</para>
965         </section>
967         <section>
968            <title>XML 1.0 End-of-line Handling</title>
969            <para>In XML 1.0 the two-character sequence CR LF (carriage return, line feed) as well as
970               any CR character not followed by a LF character must be converted to a single LF
971               character [<xref linkend="XML10"/>].</para>
972            <para>By defining carriage return, line feed, and carriage return line feed bit streams,
973               dentoted CR, LF and CRLF respectively, end-of-line normalization processing can be
974               performed in parallel using only a small number of logical and shift operations.</para>
975            <para/>
976            <para>The following example demonstrates the generation of the CRLF deletion mask. In
977               this example, the position of all CR characters followed by LF characters are marked
978               for deletion. Isolated carriage returns are then replaced with LF characters.
979               Completion of this process satisfies the XML 1.0 end-of-line handling requirements.
980               For clarity, this example encodes input data carriage returns as
981               <emphasis>C</emphasis> characters, whereas line feed characters are shown as
982                  <emphasis>L</emphasis> characters.</para>
983            <para>
984               <table>
985                  <caption>
986                     <para>XML 1.0 End-of-line Handling</para>
987                  </caption>
988                  <colgroup>
989                     <col align="left" valign="top"/>
990                  </colgroup>
991                  <tbody>
992                     <tr valign="top">
993                        <td>Input Data</td>
994                        <td>
995                           <code>first line C second line CL third line L one more C nothing
996                           left</code>
997                        </td>
998                     </tr>
999                     <tr valign="top">
1000                        <td>CR</td>
1001                        <td>
1002                           <code>-----------1-------------1------------------------1-------------</code>
1003                        </td>
1004                     </tr>
1005                     <tr valign="top">
1006                        <td>LF</td>
1007                        <td>
1008                           <code>--------------------------1------------1------------------------</code>
1009                        </td>
1010                     </tr>
1011                     <tr valign="top">
1012                        <td>DelMask</td>
1013                        <td>
1014                           <code>--------------------------1-------------------------------------</code>
1015                        </td>
1016                     </tr>
1017                  </tbody>
1018               </table>
1020            </para>
1021         </section>
1023      </section>
1025      <section>
1026         <title>Call Out Streams</title>
1027         <para> Call out bit streams mark the extents of XML markup structures such as comments,
1028            processing instruction and CDATA sections as well as physical structures such as character and
1029            entity references and general references.  Call out streams are also formed for logical markup structures such
1030            start tags, end tags and empty element tags. </para>
1031         <section>
1032            <title>Comment, Processing Instruction and CDATA Section Call Out Streams</title>
1033            <para>Comments, processing instructions and CDATA sections call out streams, Ct_Span,
1034               PI_Span and CD_Span respectively, define sections of an XML document which
1035               contain markup that is not interpreted by an XML processor. As such, the union of
1036               Ct_Span, PI_Span and CD_Span streams defines the regions of non-interpreteable markup.
1037               The stream formed by this union is termed the CtCDPI_Mask.</para>
1038            <para>The following tables provides an example of constructing the CtCDPI_Mask. </para>
1039            <table>
1040               <caption>
1041                  <para>CtCDPI Mask Generation</para>
1042               </caption>
1043               <colgroup><col align="left" valign="top" /><col align="left" valign="top" /></colgroup>
1044               <tbody> <tr valign="top"><td>Input Data</td><td><code>&lt;?php?&gt; &lt;!-- example --&gt; &lt;![CDATA[ shift: a&lt;&lt;1 ]]&gt;</code></td></tr>
1046                  <tr valign="top"><td>CD_Span</td><td><code>___________________________1111111111111111111111_</code></td></tr>
1047                  <tr valign="top"><td>Ct_Span</td><td><code>___________111111111111___________________________</code></td></tr>
1048                  <tr valign="top"><td>PI_Span</td><td><code>_11111____________________________________________</code></td></tr>
1049                  <tr valign="top"><td>CtCDPI_Mask</td><td><code>_111111__111111111111111__111111111111111111111111</code></td></tr>
1050                  <tr valign="top"><td>ErrorFlag</td><td><code>__________________________________________________</code></td></tr>
1052               </tbody>
1053            </table>
1058            <para> With the removal of all non-interpreteable markup, several phases of parallel bit
1059               stream based SIMD operations may follow operating on up to 128 byte positions on
1060               current commondity processors and assured of XML markup relevancy. For
1061               example, with the extents identification of comments, processing instructions and
1062               CDATA secions, XML names may be identified and length sorted for efficient symbol
1063               table construction. </para>
1064            <para> As an aside, comments and CDATA sections must first be validated to ensure
1065               that comments do not contain "--" sequences and that CDATA sections do not contain illegal
1066               "]]&gt;" sequences prior to ignorable markup stream generation.</para>
1067         </section>
1069         <section>
1070            <title>Reference Call Out Streams</title>
1071            <para>The reference call out streams are the GenRefs, DecRefs, and HexRefs streams. This
1072               subset of the call out streams marks the extents of all but the closing semicolon of
1073               general and character references.</para>
1074            <para>Predefined character
1075               (<![CDATA[&lt;,&gt;,&amp;,&apos;,&quot;]]>) and numeric character
1076               references (&amp;#nnnn;, &amp;#xhhhh;) must be replaced by a single character
1077                  [<xref linkend="XML10"/>]. As previously shown, this subset of call out streams enables the construction of a DelMask for
1078               references.</para>
1079         </section>
1081         <section>
1082            <title>Tag Call Out Streams</title>
1083            <para>Whereas sequential bit scans over lexical item streams form the basis of XML
1084               parsing, in the current Parabix parser a new method of parallel parsing has been
1085               developed and prototyped using the concept of bitstream addition. Fundamental to this
1086               method is the concept of a <emphasis>cursor</emphasis> stream, a bit stream marking
1087               the positions of multiple parallel parses currently in process. </para>
1088            <para>The results of parallel parsing using the bit stream addition technique produces a
1089               set of tag call out bit streams. These streams mark the extents of each start tag,
1090               end tag and empty element tag. Within tags, additional streams mark start
1091               and end positions for tag names, as well as attribute names and values. An error flag
1092               stream marks the positions of any syntactic errors encountered during parsing.</para>
1093            <para> The set of tag call out streams consists of the ElemNames, AttNames, AttVals, Tags,
1094               EmptyTagEnds and EndTags bit streams. The following example demonstrates the bit
1095               stream output produced which from parallel parsing using bit stream addition. </para>
1096            <table>
1097               <caption>
1098                  <para>Tag Call Out Streams</para>
1099               </caption>
1100               <colgroup>
1101                  <col align="left" valign="top"/>
1102                  <col align="left" valign="top"/>
1103               </colgroup>
1104               <tbody>
1105                  <tr valign="top">
1106                     <td>Input Data</td>
1107                     <td>
1108                        <code>&lt;root&gt;&lt;t1&gt;text&lt;/t1&gt;&lt;t2
1109                           a1=&apos;foo&apos; a2 =
1110                           &apos;fie&apos;&gt;more&lt;/t2&gt;&lt;tag3
1111                           att3=&apos;b&apos;/&gt;&lt;/root&gt;</code>
1112                     </td>
1113                  </tr>
1115                  <tr valign="top">
1116                     <td>ElemNames</td>
1117                     <td>
1118                        <code>_1111__11___________11_______________________________1111__________________</code>
1119                     </td>
1120                  </tr>
1121                  <tr valign="top">
1122                     <td>AttNames</td>
1123                     <td>
1124                        <code>_______________________11_______11________________________1111_____________</code>
1125                     </td>
1126                  </tr>
1127                  <tr valign="top">
1128                     <td>AttrVals</td>
1129                     <td>
1130                        <code>__________________________11111______11111_____________________111_________</code>
1131                     </td>
1132                  </tr>
1133                  <tr valign="top">
1134                     <td>EmptyTagEnds</td>
1135                     <td>
1136                        <code>___________________________________________________________________1_______</code>
1137                     </td>
1138                  </tr>
1139                  <tr valign="top">
1140                     <td>EndTags</td>
1141                     <td>
1142                        <code>_______________111______________________________111__________________11111_</code>
1143                     </td>
1144                  </tr>
1146                  <tr valign="top">
1147                     <td>Start/EmptyTags</td>
1148                     <td>
1149                        <code>_1111__11___________1111111111111111111111___________11111111111111________</code>
1150                     </td>
1151                  </tr>
1152                  <tr valign="top">
1153                     <td>ErrorFlag</td>
1154                     <td>
1155                        <code>___________________________________________________________________________</code>
1156                     </td>
1157                  </tr>
1158               </tbody>
1159            </table>
1161         </section>
1162      </section>
1163   </section>
1164   <section>
1165      <title>SIMD Beyond Bitstreams: Names and Numbers</title>
1167      <para>Whereas the fundamental innovation of our work is the use of SIMD technology in
1168         implementing parallel bit streams for XML, there are also important ways in which more
1169         traditional byte-oriented SIMD operations can be useful in accelerating other aspects of
1170         XML processing.</para>
1172      <section>
1173         <title>Name Lookup</title>
1174         <para>Efficient symbol table mechanisms for looking up element and attribute names is
1175            important for almost all XML processing applications. It is also an important technique
1176            merely for assessing well-formedness of an XML document; rather than validating the
1177            character-by-character composition of each occurrence of an XML name as it is
1178            encountered, it is more efficient to validate all but the first occurrence by first
1179            determining whether the name already exists in a table of prevalidated names.</para>
1181         <para>The first symbol table mechanism deployed in the Parabix parser simply used the
1182            hashmaps of the C++ standard template library, without deploying any SIMD technology.
1183            However, with the overhead of character validation, transcoding and parsing dramatically
1184            reduced by parallel bit stream technology, we found that symbol lookups then accounted
1185            for about half of the remaining execution time in a statistics gathering application
1186               [<xref linkend="CASCON08"/>]. Thus, symbol table processing was identified as a major
1187            target for further performance improvement. </para>
1188         <para> Our first effort to improve symbol table performance was to employ the splash tables
1189            with cuckoo hashing as described by Ross [<xref linkend="Ross06"/>], using SIMD
1190            technology for parallel bucket processing. Although this technique did turn out to have
1191            the advantage of virtually constant-time performance even for very large vocabularies,
1192            it was not particularly helpful for the relatively small vocabularies typically found in
1193            XML document processing. </para>
1194         <para> However, a second approach has been found to be quite useful, taking advantage of
1195            parallel bit streams for cheap determination of symbol length. In essence, the length of
1196            a name can be determined very cheaply using a single bit scan operation. This then makes
1197            it possible to use length-sorted symbol table processing, as follows. First, the
1198            occurrences of all names are stored in arrays indexed by length. Then the length-sorted
1199            arrays may each be inserted into the symbol table in turn. The advantage of this is that
1200            a separate loop may be written for each length. Length sorting makes for very efficient
1201            name processing. For example hash value computations and name comparisons can be made by
1202            loading multibyte values and performing appropriate shifting and masking operations,
1203            without the need for a byte-at-a-time loop. In initial experiments, this length-sorting
1204            approach was found to reduce symbol lookup cost by a factor of two. </para>
1205         <para> Current research includes the application of SIMD technology to further enhance the
1206            performance of length-sorted lookup. We have identified a promising technique for
1207            parallel processing of multiple name occurrences using a parallel trie lookup technique.
1208            Given an array of occurrences of names of a particular length, the first one, two or
1209            four bytes of each name are gathered and stored in a linear array. SIMD techniques are
1210            then used to compare these prefixes with the possible prefixes for the current position
1211            within the trie. In general, a very small number of possibilities exist for each trie
1212            node, allowing for fast linear search through all possibilities. Typically, the
1213            parallelism is expected to exceed the number of possibilities to search through at each
1214            node. With length-sorting to separate the top-level trie into many small subtries, we
1215            expect only a single step of symbol lookup to be needed in most practical instances. </para>
1217         <para>The gather step of this algorithm is actually a common technique in SIMD processing.
1218            Instruction set support for gather operations is a likely future direction for SIMD
1219            technology.</para>
1220      </section>
1222      <section>
1223         <title>Numeric Processing</title>
1224         <para> Many XML applications involve numeric data fields as attribute values or element
1225            content. Although most current XML APIs uniformly return information to applications in
1226            the form of character strings, it is reasonable to consider direct API support for
1227            numeric conversions within a high-performance XML engine. With string to numeric
1228            conversion such a common need, why leave it to application programmers? </para>
1229         <para> High-performance string to numeric conversion using SIMD operations also can
1230            considerably outperform the byte-at-a-time loops that most application programmers or
1231            libraries might employ. A first step is reduction of ASCII bytes to corresponding
1232            decimal nybbles using a SIMD packing operation. Then an inductive doubling algorithm
1233            using SIMD operations may be employed. First, 16 sets of adjacent nybble values in the
1234            range 0-9 can be combined in just a few SIMD operations to 16 byte values in the range
1235            0-99. Then 8 sets of byte values may similarly be combined with further SIMD processing
1236            to produce doublebyte values in the range 0-9999. Further combination of doublebyte
1237            values into 32-bit integers and so on can also be performed using SIMD operations. </para>
1238         <para> Using appropriate gather operations to bring numeric strings into appropriate array
1239            structures, an XML engine could offer high-performance numeric conversion services to
1240            XML application programmers. We expect this to be an important direction for our future
1241            work, particularly in support of APIs that focus on direct conversion of XML data into
1242            business objects. </para>
1244      </section>
1245   </section>
1247   <section>
1248      <title>APIs and Parallel Bit Streams</title>
1250      <section>
1251         <title>The ILAX Streaming API</title>
1252         <para>The In-Line API for XML (ILAX) is the base API provided with the Parabix parser. It
1253            is intended for low-level extensions compiled right into the engine, with minimum
1254            possible overhead. It is similar to streaming event-based APIs such as SAX, but
1255            implemented by inline substitution rather than using callbacks. In essence, an extension
1256            programmer provides method bodies for event-processing methods declared internal to the
1257            Parabix parsing engine, compiling the event processing code directly with the core code
1258            of the engine. </para>
1259         <para> Although ILAX can be used directly for application programming, its primary use is
1260            for implementing engine extensions that support higher-level APIs. For example, the
1261            implementation of C or C++ based streaming APIs based on the Expat [<xref
1262               linkend="Expat"/>] or general SAX models can be quite directly implemented. C/C++ DOM
1263            or other tree-based APIs can also be fairly directly implemented. However, delivering
1264            Parabix performance to Java-based XML applications is challenging due to the
1265            considerable overhead of crossing the Java Native Interface (JNI) boundary. This issue
1266            is addressed with the Array Set Model (ASM) concept discussed in the following section. </para>
1267         <para> With the recent development of parallel parsing using bitstream addition, it is
1268            likely that the underlying ILAX interface of Parabix will change. In essence, ILAX
1269            suffers the drawback of all event-based interfaces: they are fundamentally sequential in
1270            number. As research continues, we expect efficient parallel methods building on parallel
1271            bit stream foundations to move up the stack of XML processing requirements. Artificially
1272            imposing sequential processing is thus expected to constrain further advances in XML
1273            performance. </para>
1274      </section>
1276      <section>
1277         <title>Efficient XML in Java Using Array Set Models</title>
1278         <para> In our GML-to-SVG case study, we identified the lack of high-performance XML
1279            processing solutions for Java to be of particular interest. Java byte code does not
1280            provide access to the SIMD capabilities of the underlying machine architecture. Java
1281            just-in-time compilers might be capable of using some SIMD facilities, but there is no
1282            real prospect of conventional compiler technology translating byte-at-a-time algorithms
1283            into parallel bit stream code. So the primary vehicle for delivering high-performance
1284            XML processing is to call native parallel bit stream code written in C through JNI
1285            capabilities. </para>
1286         <para>However, each JNI call is expensive, so it is desirable to minimize the number of
1287            calls and get as much work done during each call as possible. This mitigates against
1288            direct implementation of streaming APIs in Java through one-to-one mappings to an
1289            underlying streaming API in C. Instead, we have concentrated on gathering information on
1290            the C side into data structures that can then be passed to the Java side. However, using
1291            either C pointer-based structures or C++ objects is problematic because these are
1292            difficult to interpret on the Java side and are not amenable to Java's automatic storage
1293            management system. Similarly, Java objects cannot be conveniently created on the C side.
1294            However, it is possible to transfer arrays of simple data values (bytes or integers)
1295            between C and Java, so that makes a reasonable focus for bulk data communication between
1296            C and Java. </para>
1297         <para><emphasis>Array Set Models</emphasis> are array-based representations of information
1298            representing an XML document in accord with XML InfoSet [<xref linkend="InfoSet"/>] or
1299            other XML data models relevant to particular APIs. As well as providing a mechanism for
1300            efficient bulk data communication across the JNI boundary, ASMs potentially have a
1301            number of other benefits in high-performance XML processing. <itemizedlist>
1302               <listitem>
1303                  <para>Prefetching. Commodity processors commonly support hardware and/or software
1304                     prefetching to ensure that data is available in a processor cache when it is
1305                     needed. In general, prefetching is most effective in conjunction with the
1306                     continuous sequential memory access patterns associated with array
1307                  processing.</para>
1308               </listitem>
1309               <listitem>
1310                  <para>DMA. Some processing environments provide Direct Memory Access (DMA)
1311                     controllers for block data movement in parallel with computation. For example,
1312                     the Cell Broadband Engine uses DMA controllers to move the data to and from the
1313                     local stores of the synergistic processing units. Arrays of contiguous data
1314                     elements are well suited to bulk data movement using DMA.</para>
1315               </listitem>
1316               <listitem>
1317                  <para>SIMD. Single Instruction Multiple Data (SIMD) capabilities of modern
1318                     processor instruction sets allow simultaneous application of particular
1319                     instructions to sets of elements from parallel arrays. For effective use of
1320                     SIMD capabilities, an SoA (Structure of Arrays) model is preferrable to an AoS
1321                     (Array of Structures) model. </para>
1322               </listitem>
1323               <listitem>
1324                  <para>Multicore processors. Array-oriented processing can enable the effective
1325                     distribution of work to the individual cores of a multicore system in two
1326                     distinct ways. First, provided that sequential dependencies can be minimized or
1327                     eliminated, large arrays can be divided into separate segments to be processed
1328                     in parallel on each core. Second, pipeline parallelism can be used to implement
1329                     efficient multipass processing with each pass consisting of a processing kernel
1330                     with array-based input and array-based output. </para>
1331               </listitem>
1332               <listitem>
1333                  <para>Streaming buffers for large XML documents. In the event that an XML document
1334                     is larger than can be reasonably represented entirely within processor memory,
1335                     a buffer-based streaming model can be applied to work through a document using
1336                     sliding windows over arrays of elements stored in document order. </para>
1337               </listitem>
1339            </itemizedlist>
1340         </para>
1342         <section>
1343            <title>Saxon-B TinyTree Example</title>
1344            <para>As a first example of the ASM concept, current work includes a proof-of-concept to
1345               deliver a high-performance replacement for building the TinyTree data structure used
1346               in Saxon-B 6.5.5, an open-source XSLT 2.0 processor written in Java [<xref
1347                  linkend="Saxon"/>]. Although XSLT stylesheets may be cached for performance, the
1348               caching of source XML documents is typically not possible. A new TinyTree object to
1349               represent the XML source document is thus commonly constructed with each new query so
1350               that the overall performance of simple queries on large source XML documents is
1351               highly dependent on TinyTree build time. Indeed, in a study of Saxon-SA, the
1352               commercial version of Saxon, query time was shown to be dominated by TinyTree build
1353               time [<xref linkend="Kay08"/>]. Similar performance results are demonstrable for the
1354               Saxon-B XSLT processor as well. </para>
1355            <para> The Saxon-B processor studied is a pure Java solution, converting a SAX (Simple
1356               API for XML) event stream into the TinyTree Java object using the efficient Aelfred
1357               XML parser [<xref linkend="AElfred"/>]. The TinyTree structure is itself an
1358               array-based structure mapping well suited to the ASM concept. It consists of six
1359               parallel arrays of integers indexed on node number and containing one entry for each
1360               node in the source document, with the exception of attribute and namespace nodes
1361                  [<xref linkend="Saxon"/>]. Four of the arrays respectively provide node kind, name
1362               code, depth, and next sibling information for each node, while the two others are
1363               overloaded for different purposes based on node kind value. For example, in the
1364               context of a text node , one of the overloaded arrays holds the text buffer offset
1365               value whereas the other holds the text buffer length value. Attributes and namespaces
1366               are represented using similiar parallel array of values. The stored TinyTree values
1367               are primarily primitive Java types, however, object types such as Java Strings and
1368               Java StringBuffers are also used to hold attribute values and comment values
1369               respectively. </para>
1370            <para> In addition to the TinyTree object, Saxon-B maintains a NamePool object which
1371               represents a collection of XML name triplets. Each triplet is composed of a Namespace
1372               URI, a Namespace prefix and a local name and encoded as an integer value known as a
1373               namecode. Namecodes permit efficient name search and look-up using integer
1374               comparison. Namecodes may also be subsequently decoded to recover namespace and local
1375               name information. </para>
1376            <para> Using the Parabix ILAX interface, a high-performance reimplementation of TinyTree
1377               and NamePool data structures was built to compare with the Saxon-B implementation. In
1378               fact, two functionally equivalent versions of the ASM java class were constructed. An
1379               initial version was constructed based on a set of primitive Java arrays constructed
1380               and allocated in the Java heap space via JNI New&lt;PrimitiveType&gt;Array
1381               method call. In this version, the JVM garbage collector is aware of all memory
1382               allocated in the native code. However, in this approach, large array copy operations
1383               limited overall performance to approximately a 2X gain over the Saxon-B build time. </para>
1384            <para>To further address the performance penalty imposed by copying large array values,
1385               a second version of the ASM Java object was constructed based on natively backed
1386               Direct Memory Byte Buffers [<xref linkend="JNI"/>]. In this version the JVM garbage
1387               collector is unaware any native memory resources backing the Direct Memory Byte
1388               Buffers. Large JNI-based copy operations are avoided; however, system memory must be
1389               explicitly deallocated via a Java native method call. Using this approach, our
1390               preliminary results show an approximate total 2.5X gain over Saxon-B build time.
1391            </para>
1392         </section>
1393      </section>
1394   </section>
1397   <section>
1398      <title>Compiler Technology</title>
1400      <para> An important focus of our recent work is on the development of compiler technology to
1401         automatically generate the low-level SIMD code necessary to implement bit stream processing
1402         given suitable high-level specifications. This has several potential benefits. First, it
1403         can eliminate the tedious and error-prone programming of bit stream operations in terms of
1404         register-at-a-time SIMD operations. Second, compilation technology can automatically employ
1405         a variety of performance improvement techniques that are difficult to apply manually. These
1406         include algorithms for instruction scheduling and register allocation as well as
1407         optimization techniques for common subexpression expression elimination and register
1408         rematerialization among others. Third, compiler technology makes it easier to make changes
1409         to the low-level code for reasons of perfective or adaptive maintenance.</para>
1411      <para>Beyond these reasons, compiler technology also offers the opportunity for retargetting
1412         the generation of code to accommodate different processor architectures and API
1413         requirements. Strategies for efficient parallel bit stream code can vary considerably
1414         depending on processor resources such as the number of registers available, the particular
1415         instruction set architecture supported, the size of L1 and L2 data caches, the number of
1416         available cores and so on. Separate implementation of custom code for each processor
1417         architecture would thus be likely to be prohibitively expensive, prone to errors and
1418         inconsistencies and difficult to maintain. Using compilation technology, however, the idea
1419         would be to implement a variety of processor-specific back-ends all using a common front
1420         end based on parallel bit streams. </para>
1422      <section>
1423         <title>Character Class Compiler</title>
1425         <para>The first compiler component that we have implemented is a character class compiler,
1426            capable of generation all the bit stream logic necessary to produce a set of lexical
1427            item streams each corresponding to some particular set of characters to be recognized.
1428            By taking advantage of common patterns between characters within classes, and special
1429            optimization logic for recognizing character-class ranges, our existing compiler is able
1430            to generate well-optimized code for complex sets of character classes involving numbers
1431            of special characters as well as characters within specific sets of ranges. </para>
1433      </section>
1434      <section>
1435         <title>Regular Expression Compilation</title>
1437         <para>Based on the character class compiler, we are currently investigating the
1438            construction of a regular expression compiler that can implement bit-stream based
1439            parallel regular-expression matching similar to that describe previously for parallel
1440            parsing by bistream addition. This compiler works with the assumption that bitstream
1441            regular-expression definitions are deterministic; no backtracking is permitted with the
1442            parallel bit stream representation. In XML applications, this compiler is primarily
1443            intended to enforce regular-expression constraints on string datatype specifications
1444            found in XML schema. </para>
1446      </section>
1448      <section>
1449         <title>Unbounded Bit Stream Compilation</title>
1451         <para>The Catalog of XML Bit Streams presented earlier consist of a set of abstract,
1452            unbounded bit streams, each in one-to-one correspondence with input bytes of a text
1453            file. Determining how these bit streams are implemented using fixed-width SIMD
1454            registers, and possibly processed in fixed-length buffers that represent some multiple
1455            of the register width is a source of considerable programming complexity. The general
1456            goal of our compilation strategy in this case is to allow operations to be programmed in
1457            terms of unbounded bit streams and then automatically reduced to efficient low-level
1458            code with the application of a systematic code generation strategy for handling block
1459            and buffer boundary crossing. This is work currently in progress. </para>
1461      </section>
1462   </section>
1464   <section>
1465      <title>Conclusion</title>
1466      <para>Parallel bit stream technology offers the opportunity to dramatically speed up the core
1467         XML processing components used to implement virtually any XML API. Character validation and
1468         transcoding, whitespace processing, and parsing up to including the full validation of tag
1469         syntax can be handled fully in parallel using bit stream methods. Bit streams to mark the
1470         positions of all element names, attribute names and attribute values can also be produced,
1471         followed by fast bit scan operations to generate position and length values. Beyond bit
1472         streams, byte-oriented SIMD processing of names and numerals can also accelerate
1473         performance beyond sequential byte-at-a-time methods. </para>
1474      <para>Advances in processor architecture are likely to further amplify the performance of
1475         parallel bit stream technology over traditional byte-at-a-time processing over the next
1476         decade. Improvements to SIMD register width, register complement and operation format can
1477         all result in further gains. New SIMD instruction set features such as inductive doubling
1478         support, parallel extract and deposit instructions, bit interleaving and scatter/gather
1479         capabilities should also result in significant speed-ups. Leveraging the intraregister
1480         parallelism of parallel bit stream technology within SIMD registers to take of intrachip
1481         parallelism on multicore processors should accelerate processing further. </para>
1482      <para>Technology transfer using a patent-based open-source business model is a further goal of
1483         our work with a view to widespread deployment of parallel bit stream technology in XML
1484         processing stacks implementing a variety of APIs. The feasibility of substantial
1485         performance improvement in replacement of technology implementing existing APIs has been
1486         demonstrated even in complex software architectures involving delivery of performance
1487         benefits across the JNI boundary. We are seeking to accelerate these deployment efforts
1488         both through the development of compiler technology to reliably apply these methods to a
1489         variety of architectures as well as to identify interested collaborators using open-source
1490         or commercial models. </para>
1491   </section>
1493   <section>
1494      <title>Acknowledgments</title>
1495      <para>This work is supported in part by research grants and scholarships from the Natural
1496         Sciences and Engineering Research Council of Canada, the Mathematics of Information
1497         Technology and Complex Systems Network and the British Columbia Innovation Council. </para>
1498      <para>We thank our colleague Dan Lin (Linda) for her work in high-performance symbol table
1499         processing. </para>
1500   </section>
1502   <bibliography>
1503      <title>Bibliography</title>
1504      <bibliomixed xml:id="XMLChip09" xreflabel="Leventhal and Lemoine 2009">Leventhal, Michael and
1505         Eric Lemoine 2009. The XML chip at 6 years. Proceedings of International Symposium on
1506         Processing XML Efficiently 2009, Montréal.</bibliomixed>
1507      <bibliomixed xml:id="Datapower09" xreflabel="Salz, Achilles and Maze 2009">Salz, Richard,
1508         Heather Achilles, and David Maze. 2009. Hardware and software trade-offs in the IBM
1509         DataPower XML XG4 processor card. Proceedings of International Symposium on Processing XML
1510         Efficiently 2009, Montréal.</bibliomixed>
1511      <bibliomixed xml:id="PPoPP08" xreflabel="Cameron 2007">Cameron, Robert D. 2007. A Case Study
1512         in SIMD Text Processing with Parallel Bit Streams UTF-8 to UTF-16 Transcoding. Proceedings
1513         of 13th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming 2008, Salt
1514         Lake City, Utah. On the Web at <link></link>.</bibliomixed>
1515      <bibliomixed xml:id="CASCON08" xreflabel="Cameron, Herdy and Lin 2008">Cameron, Robert D.,
1516         Kenneth S Herdy, and Dan Lin. 2008. High Performance XML Parsing Using Parallel Bit Stream
1517         Technology. Proceedings of CASCON 2008. 13th ACM SIGPLAN Symposium on Principles and
1518         Practice of Parallel Programming 2008, Toronto.</bibliomixed>
1519      <bibliomixed xml:id="SVGOpen08" xreflabel="Herdy, Burggraf and Cameron 2008">Herdy, Kenneth
1520         S., Robert D. Cameron and David S. Burggraf. 2008. High Performance GML to SVG
1521         Transformation for the Visual Presentation of Geographic Data in Web-Based Mapping Systems.
1522         Proceedings of SVG Open 6th International Conference on Scalable Vector Graphics,
1523         Nuremburg. On the Web at
1524            <link></link>.</bibliomixed>
1525      <bibliomixed xml:id="Ross06" xreflabel="Ross 2006">Ross, Kenneth A. 2006. Efficient hash
1526         probes on modern processors. Proceedings of ICDE, 2006. ICDE 2006, Atlanta. On the Web at
1527            <link></link>.</bibliomixed>
1528      <bibliomixed xml:id="ASPLOS09" xreflabel="Cameron and Lin 2009">Cameron, Robert D. and Dan
1529         Lin. 2009. Architectural Support for SWAR Text Processing with Parallel Bit Streams: The
1530         Inductive Doubling Principle. Proceedings of ASPLOS 2009, Washington, DC.</bibliomixed>
1531      <bibliomixed xml:id="Wu08" xreflabel="Wu et al. 2008">Wu, Yu, Qi Zhang, Zhiqiang Yu and
1532         Jianhui Li. 2008. A Hybrid Parallel Processing for XML Parsing and Schema Validation.
1533         Proceedings of Balisage 2008, Montréal. On the Web at
1534            <link></link>.</bibliomixed>
1535      <bibliomixed xml:id="u8u16" xreflabel="Cameron 2008">u8u16 - A High-Speed UTF-8 to UTF-16
1536         Transcoder Using Parallel Bit Streams Technical Report 2007-18. 2007. School of Computing
1537         Science Simon Fraser University, June 21 2007.</bibliomixed>
1538      <bibliomixed xml:id="XML10" xreflabel="XML 1.0">Extensible Markup Language (XML) 1.0 (Fifth
1539         Edition) W3C Recommendation 26 November 2008. On the Web at
1540            <link></link>.</bibliomixed>
1541      <bibliomixed xml:id="Unicode" xreflabel="Unicode">The Unicode Consortium. 2009. On the Web at
1542            <link></link>.</bibliomixed>
1543      <bibliomixed xml:id="Pex06" xreflabel="Hilewitz and Lee 2006"> Hilewitz, Y. and Ruby B. Lee.
1544         2006. Fast Bit Compression and Expansion with Parallel Extract and Parallel Deposit
1545         Instructions. Proceedings of the IEEE 17th International Conference on Application-Specific
1546         Systems, Architectures and Processors (ASAP), pp. 65-72, September 11-13, 2006.</bibliomixed>
1547      <bibliomixed xml:id="InfoSet" xreflabel="XML Infoset">XML Information Set (Second Edition) W3C
1548         Recommendation 4 February 2004. On the Web at
1549         <link></link>.</bibliomixed>
1550      <bibliomixed xml:id="Saxon" xreflabel="Saxon">SAXON The XSLT and XQuery Processor. On the Web
1551         at <link></link>.</bibliomixed>
1552      <bibliomixed xml:id="Kay08" xreflabel="Kay 2008"> Kay, Michael Y. 2008. Ten Reasons Why Saxon
1553         XQuery is Fast, IEEE Data Engineering Bulletin, December 2008.</bibliomixed>
1554      <bibliomixed xml:id="AElfred" xreflabel="Ælfred"> The Ælfred XML Parser. On the Web at
1555            <link></link>.</bibliomixed>
1556      <bibliomixed xml:id="JNI" xreflabel="Hitchens 2002">Hitchens, Ron. Java NIO. O'Reilly, 2002.</bibliomixed>
1557      <bibliomixed xml:id="Expat" xreflabel="Expat">The Expat XML Parser.
1558            <link></link>.</bibliomixed>
1559   </bibliography>
Note: See TracBrowser for help on using the repository browser.