ViewVC logotype

Contents of /code/trunk/doc/pcrejit.3

Parent Directory Parent Directory | Revision Log Revision Log

Revision 921 - (show annotations)
Sat Feb 18 18:45:55 2012 UTC (7 years, 7 months ago) by ph10
File size: 15254 byte(s)
Error occurred while calculating annotation data.
Update documentation for partial matching support in JIT.
3 PCRE - Perl-compatible regular expressions
5 .rs
6 .sp
7 Just-in-time compiling is a heavyweight optimization that can greatly speed up
8 pattern matching. However, it comes at the cost of extra processing before the
9 match is performed. Therefore, it is of most benefit when the same pattern is
10 going to be matched many times. This does not necessarily mean many calls of a
11 matching function; if the pattern is not anchored, matching attempts may take
12 place many times at various positions in the subject, even for a single call.
13 Therefore, if the subject string is very long, it may still pay to use JIT for
14 one-off matches.
15 .P
16 JIT support applies only to the traditional Perl-compatible matching function.
17 It does not apply when the DFA matching function is being used. The code for
18 this support was written by Zoltan Herczeg.
19 .
20 .
21 .SH "8-BIT and 16-BIT SUPPORT"
22 .rs
23 .sp
24 JIT support is available for both the 8-bit and 16-bit PCRE libraries. To keep
25 this documentation simple, only the 8-bit interface is described in what
26 follows. If you are using the 16-bit library, substitute the 16-bit functions
27 and 16-bit structures (for example, \fIpcre16_jit_stack\fP instead of
28 \fIpcre_jit_stack\fP).
29 .
30 .
32 .rs
33 .sp
34 JIT support is an optional feature of PCRE. The "configure" option --enable-jit
35 (or equivalent CMake option) must be set when PCRE is built if you want to use
36 JIT. The support is limited to the following hardware platforms:
37 .sp
38 ARM v5, v7, and Thumb2
39 Intel x86 32-bit and 64-bit
40 MIPS 32-bit
41 Power PC 32-bit and 64-bit
42 .sp
43 If --enable-jit is set on an unsupported platform, compilation fails.
44 .P
45 A program that is linked with PCRE 8.20 or later can tell if JIT support is
46 available by calling \fBpcre_config()\fP with the PCRE_CONFIG_JIT option. The
47 result is 1 when JIT is available, and 0 otherwise. However, a simple program
48 does not need to check this in order to use JIT. The API is implemented in a
49 way that falls back to the interpretive code if JIT is not available.
50 .P
51 If your program may sometimes be linked with versions of PCRE that are older
52 than 8.20, but you want to use JIT when it is available, you can test
53 the values of PCRE_MAJOR and PCRE_MINOR, or the existence of a JIT macro such
54 as PCRE_CONFIG_JIT, for compile-time control of your code.
55 .
56 .
58 .rs
59 .sp
60 You have to do two things to make use of the JIT support in the simplest way:
61 .sp
62 (1) Call \fBpcre_study()\fP with the PCRE_STUDY_JIT_COMPILE option for
63 each compiled pattern, and pass the resulting \fBpcre_extra\fP block to
64 \fBpcre_exec()\fP.
65 .sp
66 (2) Use \fBpcre_free_study()\fP to free the \fBpcre_extra\fP block when it is
67 no longer needed, instead of just freeing it yourself. This
68 ensures that any JIT data is also freed.
69 .sp
70 For a program that may be linked with pre-8.20 versions of PCRE, you can insert
71 .sp
74 #endif
75 .sp
76 so that no option is passed to \fBpcre_study()\fP, and then use something like
77 this to free the study data:
78 .sp
80 pcre_free_study(study_ptr);
81 #else
82 pcre_free(study_ptr);
83 #endif
84 .sp
85 PCRE_STUDY_JIT_COMPILE requests the JIT compiler to generate code for complete
86 matches. If you want to run partial matches using the PCRE_PARTIAL_HARD or
87 PCRE_PARTIAL_SOFT options of \fBpcre_exec()\fP, you should set one or both of
88 the following options in addition to, or instead of, PCRE_STUDY_JIT_COMPILE
89 when you call \fBpcre_study()\fP:
90 .sp
93 .sp
94 The JIT compiler generates different optimized code for each of the three
95 modes (normal, soft partial, hard partial). When \fBpcre_exec()\fP is called,
96 the appropriate code is run if it is available. Otherwise, the pattern is
97 matched using interpretive code.
98 .P
99 In some circumstances you may need to call additional functions. These are
100 described in the section entitled
101 .\" HTML <a href="#stackcontrol">
102 .\" </a>
103 "Controlling the JIT stack"
104 .\"
105 below.
106 .P
107 If JIT support is not available, PCRE_STUDY_JIT_COMPILE etc. are ignored, and
108 no JIT data is created. Otherwise, the compiled pattern is passed to the JIT
109 compiler, which turns it into machine code that executes much faster than the
110 normal interpretive code. When \fBpcre_exec()\fP is passed a \fBpcre_extra\fP
111 block containing a pointer to JIT code of the appropriate mode (normal or
112 hard/soft partial), it obeys that code instead of running the interpreter. The
113 result is identical, but the compiled JIT code runs much faster.
114 .P
115 There are some \fBpcre_exec()\fP options that are not supported for JIT
116 execution. There are also some pattern items that JIT cannot handle. Details
117 are given below. In both cases, execution automatically falls back to the
118 interpretive code.
119 .P
120 If the JIT compiler finds an unsupported item, no JIT data is generated. You
121 can find out if JIT execution is available after studying a pattern by calling
122 \fBpcre_fullinfo()\fP with the PCRE_INFO_JIT option. A result of 1 means that
123 JIT compilation was successful. A result of 0 means that JIT support is not
124 available, or the pattern was not studied with PCRE_STUDY_JIT_COMPILE etc., or
125 the JIT compiler was not able to handle the pattern.
126 .P
127 Once a pattern has been studied, with or without JIT, it can be used as many
128 times as you like for matching different subject strings.
129 .
130 .
132 .rs
133 .sp
134 The only \fBpcre_exec()\fP options that are supported for JIT execution are
137 .P
138 The unsupported pattern items are:
139 .sp
140 \eC match a single byte; not supported in UTF-8 mode
141 (?Cn) callouts
142 (*COMMIT) )
143 (*MARK) )
144 (*PRUNE) ) the backtracking control verbs
145 (*SKIP) )
146 (*THEN) )
147 .sp
148 Support for some of these may be added in future.
149 .
150 .
152 .rs
153 .sp
154 When a pattern is matched using JIT execution, the return values are the same
155 as those given by the interpretive \fBpcre_exec()\fP code, with the addition of
156 one new error code: PCRE_ERROR_JIT_STACKLIMIT. This means that the memory used
157 for the JIT stack was insufficient. See
158 .\" HTML <a href="#stackcontrol">
159 .\" </a>
160 "Controlling the JIT stack"
161 .\"
162 below for a discussion of JIT stack usage. For compatibility with the
163 interpretive \fBpcre_exec()\fP code, no more than two-thirds of the
164 \fIovector\fP argument is used for passing back captured substrings.
165 .P
166 The error code PCRE_ERROR_MATCHLIMIT is returned by the JIT code if searching a
167 very large pattern tree goes on for too long, as it is in the same circumstance
168 when JIT is not used, but the details of exactly what is counted are not the
169 same. The PCRE_ERROR_RECURSIONLIMIT error code is never returned by JIT
170 execution.
171 .
172 .
174 .rs
175 .sp
176 The code that is generated by the JIT compiler is architecture-specific, and is
177 also position dependent. For those reasons it cannot be saved (in a file or
178 database) and restored later like the bytecode and other data of a compiled
179 pattern. Saving and restoring compiled patterns is not something many people
180 do. More detail about this facility is given in the
181 .\" HREF
182 \fBpcreprecompile\fP
183 .\"
184 documentation. It should be possible to run \fBpcre_study()\fP on a saved and
185 restored pattern, and thereby recreate the JIT data, but because JIT
186 compilation uses significant resources, it is probably not worth doing this;
187 you might as well recompile the original pattern.
188 .
189 .
190 .\" HTML <a name="stackcontrol"></a>
192 .rs
193 .sp
194 When the compiled JIT code runs, it needs a block of memory to use as a stack.
195 By default, it uses 32K on the machine stack. However, some large or
196 complicated patterns need more than this. The error PCRE_ERROR_JIT_STACKLIMIT
197 is given when there is not enough stack. Three functions are provided for
198 managing blocks of memory for use as JIT stacks. There is further discussion
199 about the use of JIT stacks in the section entitled
200 .\" HTML <a href="#stackcontrol">
201 .\" </a>
202 "JIT stack FAQ"
203 .\"
204 below.
205 .P
206 The \fBpcre_jit_stack_alloc()\fP function creates a JIT stack. Its arguments
207 are a starting size and a maximum size, and it returns a pointer to an opaque
208 structure of type \fBpcre_jit_stack\fP, or NULL if there is an error. The
209 \fBpcre_jit_stack_free()\fP function can be used to free a stack that is no
210 longer needed. (For the technically minded: the address space is allocated by
211 mmap or VirtualAlloc.)
212 .P
213 JIT uses far less memory for recursion than the interpretive code,
214 and a maximum stack size of 512K to 1M should be more than enough for any
215 pattern.
216 .P
217 The \fBpcre_assign_jit_stack()\fP function specifies which stack JIT code
218 should use. Its arguments are as follows:
219 .sp
220 pcre_extra *extra
221 pcre_jit_callback callback
222 void *data
223 .sp
224 The \fIextra\fP argument must be the result of studying a pattern with
225 PCRE_STUDY_JIT_COMPILE etc. There are three cases for the values of the other
226 two options:
227 .sp
228 (1) If \fIcallback\fP is NULL and \fIdata\fP is NULL, an internal 32K block
229 on the machine stack is used.
230 .sp
231 (2) If \fIcallback\fP is NULL and \fIdata\fP is not NULL, \fIdata\fP must be
232 a valid JIT stack, the result of calling \fBpcre_jit_stack_alloc()\fP.
233 .sp
234 (3) If \fIcallback\fP not NULL, it must point to a function that is called
235 with \fIdata\fP as an argument at the start of matching, in order to
236 set up a JIT stack. If the result is NULL, the internal 32K stack
237 is used; otherwise the return value must be a valid JIT stack,
238 the result of calling \fBpcre_jit_stack_alloc()\fP.
239 .sp
240 You may safely assign the same JIT stack to more than one pattern, as long as
241 they are all matched sequentially in the same thread. In a multithread
242 application, each thread must use its own JIT stack.
243 .P
244 Strictly speaking, even more is allowed. You can assign the same stack to any
245 number of patterns as long as they are not used for matching by multiple
246 threads at the same time. For example, you can assign the same stack to all
247 compiled patterns, and use a global mutex in the callback to wait until the
248 stack is available for use. However, this is an inefficient solution, and
249 not recommended.
250 .P
251 This is a suggestion for how a typical multithreaded program might operate:
252 .sp
253 During thread initalization
254 thread_local_var = pcre_jit_stack_alloc(...)
255 .sp
256 During thread exit
257 pcre_jit_stack_free(thread_local_var)
258 .sp
259 Use a one-line callback function
260 return thread_local_var
261 .sp
262 All the functions described in this section do nothing if JIT is not available,
263 and \fBpcre_assign_jit_stack()\fP does nothing unless the \fBextra\fP argument
264 is non-NULL and points to a \fBpcre_extra\fP block that is the result of a
265 successful study with PCRE_STUDY_JIT_COMPILE etc.
266 .
267 .
268 .\" HTML <a name="stackfaq"></a>
270 .rs
271 .sp
272 (1) Why do we need JIT stacks?
273 .sp
274 PCRE (and JIT) is a recursive, depth-first engine, so it needs a stack where
275 the local data of the current node is pushed before checking its child nodes.
276 Allocating real machine stack on some platforms is difficult. For example, the
277 stack chain needs to be updated every time if we extend the stack on PowerPC.
278 Although it is possible, its updating time overhead decreases performance. So
279 we do the recursion in memory.
280 .P
281 (2) Why don't we simply allocate blocks of memory with \fBmalloc()\fP?
282 .sp
283 Modern operating systems have a nice feature: they can reserve an address space
284 instead of allocating memory. We can safely allocate memory pages inside this
285 address space, so the stack could grow without moving memory data (this is
286 important because of pointers). Thus we can allocate 1M address space, and use
287 only a single memory page (usually 4K) if that is enough. However, we can still
288 grow up to 1M anytime if needed.
289 .P
290 (3) Who "owns" a JIT stack?
291 .sp
292 The owner of the stack is the user program, not the JIT studied pattern or
293 anything else. The user program must ensure that if a stack is used by
294 \fBpcre_exec()\fP, (that is, it is assigned to the pattern currently running),
295 that stack must not be used by any other threads (to avoid overwriting the same
296 memory area). The best practice for multithreaded programs is to allocate a
297 stack for each thread, and return this stack through the JIT callback function.
298 .P
299 (4) When should a JIT stack be freed?
300 .sp
301 You can free a JIT stack at any time, as long as it will not be used by
302 \fBpcre_exec()\fP again. When you assign the stack to a pattern, only a pointer
303 is set. There is no reference counting or any other magic. You can free the
304 patterns and stacks in any order, anytime. Just \fIdo not\fP call
305 \fBpcre_exec()\fP with a pattern pointing to an already freed stack, as that
306 will cause SEGFAULT. (Also, do not free a stack currently used by
307 \fBpcre_exec()\fP in another thread). You can also replace the stack for a
308 pattern at any time. You can even free the previous stack before assigning a
309 replacement.
310 .P
311 (5) Should I allocate/free a stack every time before/after calling
312 \fBpcre_exec()\fP?
313 .sp
314 No, because this is too costly in terms of resources. However, you could
315 implement some clever idea which release the stack if it is not used in let's
316 say two minutes. The JIT callback can help to achive this without keeping a
317 list of the currently JIT studied patterns.
318 .P
319 (6) OK, the stack is for long term memory allocation. But what happens if a
320 pattern causes stack overflow with a stack of 1M? Is that 1M kept until the
321 stack is freed?
322 .sp
323 Especially on embedded sytems, it might be a good idea to release
324 memory sometimes without freeing the stack. There is no API for this at the
325 moment. Probably a function call which returns with the currently allocated
326 memory for any stack and another which allows releasing memory (shrinking the
327 stack) would be a good idea if someone needs this.
328 .P
329 (7) This is too much of a headache. Isn't there any better solution for JIT
330 stack handling?
331 .sp
332 No, thanks to Windows. If POSIX threads were used everywhere, we could throw
333 out this complicated API.
334 .
335 .
337 .rs
338 .sp
339 This is a single-threaded example that specifies a JIT stack without using a
340 callback.
341 .sp
342 int rc;
343 int ovector[30];
344 pcre *re;
345 pcre_extra *extra;
346 pcre_jit_stack *jit_stack;
347 .sp
348 re = pcre_compile(pattern, 0, &error, &erroffset, NULL);
349 /* Check for errors */
350 extra = pcre_study(re, PCRE_STUDY_JIT_COMPILE, &error);
351 jit_stack = pcre_jit_stack_alloc(32*1024, 512*1024);
352 /* Check for error (NULL) */
353 pcre_assign_jit_stack(extra, NULL, jit_stack);
354 rc = pcre_exec(re, extra, subject, length, 0, 0, ovector, 30);
355 /* Check results */
356 pcre_free(re);
357 pcre_free_study(extra);
358 pcre_jit_stack_free(jit_stack);
359 .sp
360 .
361 .
362 .SH "SEE ALSO"
363 .rs
364 .sp
365 \fBpcreapi\fP(3)
366 .
367 .
369 .rs
370 .sp
371 .nf
372 Philip Hazel (FAQ by Zoltan Herczeg)
373 University Computing Service
374 Cambridge CB2 3QH, England.
375 .fi
376 .
377 .
379 .rs
380 .sp
381 .nf
382 Last updated: 18 February 2012
383 Copyright (c) 1997-2012 University of Cambridge.
384 .fi

  ViewVC Help
Powered by ViewVC 1.1.5