ViewVC logotype

Contents of /code/trunk/doc/html/pcrejit.html

Parent Directory Parent Directory | Revision Log Revision Log

Revision 1298 - (show annotations)
Fri Mar 22 16:13:13 2013 UTC (6 years, 5 months ago) by ph10
File MIME type: text/html
File size: 20050 byte(s)
Fix COMMIT in recursion; document backtracking verbs in assertions and 
1 <html>
2 <head>
3 <title>pcrejit specification</title>
4 </head>
5 <body bgcolor="#FFFFFF" text="#00005A" link="#0066FF" alink="#3399FF" vlink="#2222BB">
6 <h1>pcrejit man page</h1>
7 <p>
8 Return to the <a href="index.html">PCRE index page</a>.
9 </p>
10 <p>
11 This page is part of the PCRE HTML documentation. It was generated automatically
12 from the original man page. If there is any nonsense in it, please consult the
13 man page, in case the conversion went wrong.
14 <br>
15 <ul>
16 <li><a name="TOC1" href="#SEC1">PCRE JUST-IN-TIME COMPILER SUPPORT</a>
17 <li><a name="TOC2" href="#SEC2">8-BIT, 16-BIT AND 32-BIT SUPPORT</a>
18 <li><a name="TOC3" href="#SEC3">AVAILABILITY OF JIT SUPPORT</a>
19 <li><a name="TOC4" href="#SEC4">SIMPLE USE OF JIT</a>
21 <li><a name="TOC6" href="#SEC6">RETURN VALUES FROM JIT EXECUTION</a>
23 <li><a name="TOC8" href="#SEC8">CONTROLLING THE JIT STACK</a>
24 <li><a name="TOC9" href="#SEC9">JIT STACK FAQ</a>
25 <li><a name="TOC10" href="#SEC10">EXAMPLE CODE</a>
26 <li><a name="TOC11" href="#SEC11">JIT FAST PATH API</a>
27 <li><a name="TOC12" href="#SEC12">SEE ALSO</a>
28 <li><a name="TOC13" href="#SEC13">AUTHOR</a>
29 <li><a name="TOC14" href="#SEC14">REVISION</a>
30 </ul>
31 <br><a name="SEC1" href="#TOC1">PCRE JUST-IN-TIME COMPILER SUPPORT</a><br>
32 <P>
33 Just-in-time compiling is a heavyweight optimization that can greatly speed up
34 pattern matching. However, it comes at the cost of extra processing before the
35 match is performed. Therefore, it is of most benefit when the same pattern is
36 going to be matched many times. This does not necessarily mean many calls of a
37 matching function; if the pattern is not anchored, matching attempts may take
38 place many times at various positions in the subject, even for a single call.
39 Therefore, if the subject string is very long, it may still pay to use JIT for
40 one-off matches.
41 </P>
42 <P>
43 JIT support applies only to the traditional Perl-compatible matching function.
44 It does not apply when the DFA matching function is being used. The code for
45 this support was written by Zoltan Herczeg.
46 </P>
47 <br><a name="SEC2" href="#TOC1">8-BIT, 16-BIT AND 32-BIT SUPPORT</a><br>
48 <P>
49 JIT support is available for all of the 8-bit, 16-bit and 32-bit PCRE
50 libraries. To keep this documentation simple, only the 8-bit interface is
51 described in what follows. If you are using the 16-bit library, substitute the
52 16-bit functions and 16-bit structures (for example, <i>pcre16_jit_stack</i>
53 instead of <i>pcre_jit_stack</i>). If you are using the 32-bit library,
54 substitute the 32-bit functions and 32-bit structures (for example,
55 <i>pcre32_jit_stack</i> instead of <i>pcre_jit_stack</i>).
56 </P>
57 <br><a name="SEC3" href="#TOC1">AVAILABILITY OF JIT SUPPORT</a><br>
58 <P>
59 JIT support is an optional feature of PCRE. The "configure" option --enable-jit
60 (or equivalent CMake option) must be set when PCRE is built if you want to use
61 JIT. The support is limited to the following hardware platforms:
62 <pre>
63 ARM v5, v7, and Thumb2
64 Intel x86 32-bit and 64-bit
65 MIPS 32-bit
66 Power PC 32-bit and 64-bit
67 SPARC 32-bit (experimental)
68 </pre>
69 If --enable-jit is set on an unsupported platform, compilation fails.
70 </P>
71 <P>
72 A program that is linked with PCRE 8.20 or later can tell if JIT support is
73 available by calling <b>pcre_config()</b> with the PCRE_CONFIG_JIT option. The
74 result is 1 when JIT is available, and 0 otherwise. However, a simple program
75 does not need to check this in order to use JIT. The normal API is implemented
76 in a way that falls back to the interpretive code if JIT is not available. For
77 programs that need the best possible performance, there is also a "fast path"
78 API that is JIT-specific.
79 </P>
80 <P>
81 If your program may sometimes be linked with versions of PCRE that are older
82 than 8.20, but you want to use JIT when it is available, you can test
83 the values of PCRE_MAJOR and PCRE_MINOR, or the existence of a JIT macro such
84 as PCRE_CONFIG_JIT, for compile-time control of your code.
85 </P>
86 <br><a name="SEC4" href="#TOC1">SIMPLE USE OF JIT</a><br>
87 <P>
88 You have to do two things to make use of the JIT support in the simplest way:
89 <pre>
90 (1) Call <b>pcre_study()</b> with the PCRE_STUDY_JIT_COMPILE option for
91 each compiled pattern, and pass the resulting <b>pcre_extra</b> block to
92 <b>pcre_exec()</b>.
94 (2) Use <b>pcre_free_study()</b> to free the <b>pcre_extra</b> block when it is
95 no longer needed, instead of just freeing it yourself. This ensures that
96 any JIT data is also freed.
97 </pre>
98 For a program that may be linked with pre-8.20 versions of PCRE, you can insert
99 <pre>
102 #endif
103 </pre>
104 so that no option is passed to <b>pcre_study()</b>, and then use something like
105 this to free the study data:
106 <pre>
107 #ifdef PCRE_CONFIG_JIT
108 pcre_free_study(study_ptr);
109 #else
110 pcre_free(study_ptr);
111 #endif
112 </pre>
113 PCRE_STUDY_JIT_COMPILE requests the JIT compiler to generate code for complete
114 matches. If you want to run partial matches using the PCRE_PARTIAL_HARD or
115 PCRE_PARTIAL_SOFT options of <b>pcre_exec()</b>, you should set one or both of
116 the following options in addition to, or instead of, PCRE_STUDY_JIT_COMPILE
117 when you call <b>pcre_study()</b>:
118 <pre>
121 </pre>
122 The JIT compiler generates different optimized code for each of the three
123 modes (normal, soft partial, hard partial). When <b>pcre_exec()</b> is called,
124 the appropriate code is run if it is available. Otherwise, the pattern is
125 matched using interpretive code.
126 </P>
127 <P>
128 In some circumstances you may need to call additional functions. These are
129 described in the section entitled
130 <a href="#stackcontrol">"Controlling the JIT stack"</a>
131 below.
132 </P>
133 <P>
134 If JIT support is not available, PCRE_STUDY_JIT_COMPILE etc. are ignored, and
135 no JIT data is created. Otherwise, the compiled pattern is passed to the JIT
136 compiler, which turns it into machine code that executes much faster than the
137 normal interpretive code. When <b>pcre_exec()</b> is passed a <b>pcre_extra</b>
138 block containing a pointer to JIT code of the appropriate mode (normal or
139 hard/soft partial), it obeys that code instead of running the interpreter. The
140 result is identical, but the compiled JIT code runs much faster.
141 </P>
142 <P>
143 There are some <b>pcre_exec()</b> options that are not supported for JIT
144 execution. There are also some pattern items that JIT cannot handle. Details
145 are given below. In both cases, execution automatically falls back to the
146 interpretive code. If you want to know whether JIT was actually used for a
147 particular match, you should arrange for a JIT callback function to be set up
148 as described in the section entitled
149 <a href="#stackcontrol">"Controlling the JIT stack"</a>
150 below, even if you do not need to supply a non-default JIT stack. Such a
151 callback function is called whenever JIT code is about to be obeyed. If the
152 execution options are not right for JIT execution, the callback function is not
153 obeyed.
154 </P>
155 <P>
156 If the JIT compiler finds an unsupported item, no JIT data is generated. You
157 can find out if JIT execution is available after studying a pattern by calling
158 <b>pcre_fullinfo()</b> with the PCRE_INFO_JIT option. A result of 1 means that
159 JIT compilation was successful. A result of 0 means that JIT support is not
160 available, or the pattern was not studied with PCRE_STUDY_JIT_COMPILE etc., or
161 the JIT compiler was not able to handle the pattern.
162 </P>
163 <P>
164 Once a pattern has been studied, with or without JIT, it can be used as many
165 times as you like for matching different subject strings.
166 </P>
167 <br><a name="SEC5" href="#TOC1">UNSUPPORTED OPTIONS AND PATTERN ITEMS</a><br>
168 <P>
169 The only <b>pcre_exec()</b> options that are supported for JIT execution are
173 </P>
174 <P>
175 The only unsupported pattern items are \C (match a single data unit) when
176 running in a UTF mode, and a callout immediately before an assertion condition
177 in a conditional group.
178 </P>
179 <br><a name="SEC6" href="#TOC1">RETURN VALUES FROM JIT EXECUTION</a><br>
180 <P>
181 When a pattern is matched using JIT execution, the return values are the same
182 as those given by the interpretive <b>pcre_exec()</b> code, with the addition of
183 one new error code: PCRE_ERROR_JIT_STACKLIMIT. This means that the memory used
184 for the JIT stack was insufficient. See
185 <a href="#stackcontrol">"Controlling the JIT stack"</a>
186 below for a discussion of JIT stack usage. For compatibility with the
187 interpretive <b>pcre_exec()</b> code, no more than two-thirds of the
188 <i>ovector</i> argument is used for passing back captured substrings.
189 </P>
190 <P>
191 The error code PCRE_ERROR_MATCHLIMIT is returned by the JIT code if searching a
192 very large pattern tree goes on for too long, as it is in the same circumstance
193 when JIT is not used, but the details of exactly what is counted are not the
194 same. The PCRE_ERROR_RECURSIONLIMIT error code is never returned by JIT
195 execution.
196 </P>
197 <br><a name="SEC7" href="#TOC1">SAVING AND RESTORING COMPILED PATTERNS</a><br>
198 <P>
199 The code that is generated by the JIT compiler is architecture-specific, and is
200 also position dependent. For those reasons it cannot be saved (in a file or
201 database) and restored later like the bytecode and other data of a compiled
202 pattern. Saving and restoring compiled patterns is not something many people
203 do. More detail about this facility is given in the
204 <a href="pcreprecompile.html"><b>pcreprecompile</b></a>
205 documentation. It should be possible to run <b>pcre_study()</b> on a saved and
206 restored pattern, and thereby recreate the JIT data, but because JIT
207 compilation uses significant resources, it is probably not worth doing this;
208 you might as well recompile the original pattern.
209 <a name="stackcontrol"></a></P>
210 <br><a name="SEC8" href="#TOC1">CONTROLLING THE JIT STACK</a><br>
211 <P>
212 When the compiled JIT code runs, it needs a block of memory to use as a stack.
213 By default, it uses 32K on the machine stack. However, some large or
214 complicated patterns need more than this. The error PCRE_ERROR_JIT_STACKLIMIT
215 is given when there is not enough stack. Three functions are provided for
216 managing blocks of memory for use as JIT stacks. There is further discussion
217 about the use of JIT stacks in the section entitled
218 <a href="#stackcontrol">"JIT stack FAQ"</a>
219 below.
220 </P>
221 <P>
222 The <b>pcre_jit_stack_alloc()</b> function creates a JIT stack. Its arguments
223 are a starting size and a maximum size, and it returns a pointer to an opaque
224 structure of type <b>pcre_jit_stack</b>, or NULL if there is an error. The
225 <b>pcre_jit_stack_free()</b> function can be used to free a stack that is no
226 longer needed. (For the technically minded: the address space is allocated by
227 mmap or VirtualAlloc.)
228 </P>
229 <P>
230 JIT uses far less memory for recursion than the interpretive code,
231 and a maximum stack size of 512K to 1M should be more than enough for any
232 pattern.
233 </P>
234 <P>
235 The <b>pcre_assign_jit_stack()</b> function specifies which stack JIT code
236 should use. Its arguments are as follows:
237 <pre>
238 pcre_extra *extra
239 pcre_jit_callback callback
240 void *data
241 </pre>
242 The <i>extra</i> argument must be the result of studying a pattern with
243 PCRE_STUDY_JIT_COMPILE etc. There are three cases for the values of the other
244 two options:
245 <pre>
246 (1) If <i>callback</i> is NULL and <i>data</i> is NULL, an internal 32K block
247 on the machine stack is used.
249 (2) If <i>callback</i> is NULL and <i>data</i> is not NULL, <i>data</i> must be
250 a valid JIT stack, the result of calling <b>pcre_jit_stack_alloc()</b>.
252 (3) If <i>callback</i> is not NULL, it must point to a function that is
253 called with <i>data</i> as an argument at the start of matching, in
254 order to set up a JIT stack. If the return from the callback
255 function is NULL, the internal 32K stack is used; otherwise the
256 return value must be a valid JIT stack, the result of calling
257 <b>pcre_jit_stack_alloc()</b>.
258 </pre>
259 A callback function is obeyed whenever JIT code is about to be run; it is not
260 obeyed when <b>pcre_exec()</b> is called with options that are incompatible for
261 JIT execution. A callback function can therefore be used to determine whether a
262 match operation was executed by JIT or by the interpreter.
263 </P>
264 <P>
265 You may safely use the same JIT stack for more than one pattern (either by
266 assigning directly or by callback), as long as the patterns are all matched
267 sequentially in the same thread. In a multithread application, if you do not
268 specify a JIT stack, or if you assign or pass back NULL from a callback, that
269 is thread-safe, because each thread has its own machine stack. However, if you
270 assign or pass back a non-NULL JIT stack, this must be a different stack for
271 each thread so that the application is thread-safe.
272 </P>
273 <P>
274 Strictly speaking, even more is allowed. You can assign the same non-NULL stack
275 to any number of patterns as long as they are not used for matching by multiple
276 threads at the same time. For example, you can assign the same stack to all
277 compiled patterns, and use a global mutex in the callback to wait until the
278 stack is available for use. However, this is an inefficient solution, and not
279 recommended.
280 </P>
281 <P>
282 This is a suggestion for how a multithreaded program that needs to set up
283 non-default JIT stacks might operate:
284 <pre>
285 During thread initalization
286 thread_local_var = pcre_jit_stack_alloc(...)
288 During thread exit
289 pcre_jit_stack_free(thread_local_var)
291 Use a one-line callback function
292 return thread_local_var
293 </pre>
294 All the functions described in this section do nothing if JIT is not available,
295 and <b>pcre_assign_jit_stack()</b> does nothing unless the <b>extra</b> argument
296 is non-NULL and points to a <b>pcre_extra</b> block that is the result of a
297 successful study with PCRE_STUDY_JIT_COMPILE etc.
298 <a name="stackfaq"></a></P>
299 <br><a name="SEC9" href="#TOC1">JIT STACK FAQ</a><br>
300 <P>
301 (1) Why do we need JIT stacks?
302 <br>
303 <br>
304 PCRE (and JIT) is a recursive, depth-first engine, so it needs a stack where
305 the local data of the current node is pushed before checking its child nodes.
306 Allocating real machine stack on some platforms is difficult. For example, the
307 stack chain needs to be updated every time if we extend the stack on PowerPC.
308 Although it is possible, its updating time overhead decreases performance. So
309 we do the recursion in memory.
310 </P>
311 <P>
312 (2) Why don't we simply allocate blocks of memory with <b>malloc()</b>?
313 <br>
314 <br>
315 Modern operating systems have a nice feature: they can reserve an address space
316 instead of allocating memory. We can safely allocate memory pages inside this
317 address space, so the stack could grow without moving memory data (this is
318 important because of pointers). Thus we can allocate 1M address space, and use
319 only a single memory page (usually 4K) if that is enough. However, we can still
320 grow up to 1M anytime if needed.
321 </P>
322 <P>
323 (3) Who "owns" a JIT stack?
324 <br>
325 <br>
326 The owner of the stack is the user program, not the JIT studied pattern or
327 anything else. The user program must ensure that if a stack is used by
328 <b>pcre_exec()</b>, (that is, it is assigned to the pattern currently running),
329 that stack must not be used by any other threads (to avoid overwriting the same
330 memory area). The best practice for multithreaded programs is to allocate a
331 stack for each thread, and return this stack through the JIT callback function.
332 </P>
333 <P>
334 (4) When should a JIT stack be freed?
335 <br>
336 <br>
337 You can free a JIT stack at any time, as long as it will not be used by
338 <b>pcre_exec()</b> again. When you assign the stack to a pattern, only a pointer
339 is set. There is no reference counting or any other magic. You can free the
340 patterns and stacks in any order, anytime. Just <i>do not</i> call
341 <b>pcre_exec()</b> with a pattern pointing to an already freed stack, as that
342 will cause SEGFAULT. (Also, do not free a stack currently used by
343 <b>pcre_exec()</b> in another thread). You can also replace the stack for a
344 pattern at any time. You can even free the previous stack before assigning a
345 replacement.
346 </P>
347 <P>
348 (5) Should I allocate/free a stack every time before/after calling
349 <b>pcre_exec()</b>?
350 <br>
351 <br>
352 No, because this is too costly in terms of resources. However, you could
353 implement some clever idea which release the stack if it is not used in let's
354 say two minutes. The JIT callback can help to achieve this without keeping a
355 list of the currently JIT studied patterns.
356 </P>
357 <P>
358 (6) OK, the stack is for long term memory allocation. But what happens if a
359 pattern causes stack overflow with a stack of 1M? Is that 1M kept until the
360 stack is freed?
361 <br>
362 <br>
363 Especially on embedded sytems, it might be a good idea to release memory
364 sometimes without freeing the stack. There is no API for this at the moment.
365 Probably a function call which returns with the currently allocated memory for
366 any stack and another which allows releasing memory (shrinking the stack) would
367 be a good idea if someone needs this.
368 </P>
369 <P>
370 (7) This is too much of a headache. Isn't there any better solution for JIT
371 stack handling?
372 <br>
373 <br>
374 No, thanks to Windows. If POSIX threads were used everywhere, we could throw
375 out this complicated API.
376 </P>
377 <br><a name="SEC10" href="#TOC1">EXAMPLE CODE</a><br>
378 <P>
379 This is a single-threaded example that specifies a JIT stack without using a
380 callback.
381 <pre>
382 int rc;
383 int ovector[30];
384 pcre *re;
385 pcre_extra *extra;
386 pcre_jit_stack *jit_stack;
388 re = pcre_compile(pattern, 0, &error, &erroffset, NULL);
389 /* Check for errors */
390 extra = pcre_study(re, PCRE_STUDY_JIT_COMPILE, &error);
391 jit_stack = pcre_jit_stack_alloc(32*1024, 512*1024);
392 /* Check for error (NULL) */
393 pcre_assign_jit_stack(extra, NULL, jit_stack);
394 rc = pcre_exec(re, extra, subject, length, 0, 0, ovector, 30);
395 /* Check results */
396 pcre_free(re);
397 pcre_free_study(extra);
398 pcre_jit_stack_free(jit_stack);
400 </PRE>
401 </P>
402 <br><a name="SEC11" href="#TOC1">JIT FAST PATH API</a><br>
403 <P>
404 Because the API described above falls back to interpreted execution when JIT is
405 not available, it is convenient for programs that are written for general use
406 in many environments. However, calling JIT via <b>pcre_exec()</b> does have a
407 performance impact. Programs that are written for use where JIT is known to be
408 available, and which need the best possible performance, can instead use a
409 "fast path" API to call JIT execution directly instead of calling
410 <b>pcre_exec()</b> (obviously only for patterns that have been successfully
411 studied by JIT).
412 </P>
413 <P>
414 The fast path function is called <b>pcre_jit_exec()</b>, and it takes exactly
415 the same arguments as <b>pcre_exec()</b>, plus one additional argument that
416 must point to a JIT stack. The JIT stack arrangements described above do not
417 apply. The return values are the same as for <b>pcre_exec()</b>.
418 </P>
419 <P>
420 When you call <b>pcre_exec()</b>, as well as testing for invalid options, a
421 number of other sanity checks are performed on the arguments. For example, if
422 the subject pointer is NULL, or its length is negative, an immediate error is
423 given. Also, unless PCRE_NO_UTF[8|16|32] is set, a UTF subject string is tested
424 for validity. In the interests of speed, these checks do not happen on the JIT
425 fast path, and if invalid data is passed, the result is undefined.
426 </P>
427 <P>
428 Bypassing the sanity checks and the <b>pcre_exec()</b> wrapping can give
429 speedups of more than 10%.
430 </P>
431 <br><a name="SEC12" href="#TOC1">SEE ALSO</a><br>
432 <P>
433 <b>pcreapi</b>(3)
434 </P>
435 <br><a name="SEC13" href="#TOC1">AUTHOR</a><br>
436 <P>
437 Philip Hazel (FAQ by Zoltan Herczeg)
438 <br>
439 University Computing Service
440 <br>
441 Cambridge CB2 3QH, England.
442 <br>
443 </P>
444 <br><a name="SEC14" href="#TOC1">REVISION</a><br>
445 <P>
446 Last updated: 17 March 2013
447 <br>
448 Copyright &copy; 1997-2013 University of Cambridge.
449 <br>
450 <p>
451 Return to the <a href="index.html">PCRE index page</a>.
452 </p>

  ViewVC Help
Powered by ViewVC 1.1.5