ViewVC logotype

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

Parent Directory Parent Directory | Revision Log Revision Log

Revision 975 - (hide annotations) (download)
Sat Jun 2 11:03:06 2012 UTC (2 years, 10 months ago) by ph10
File size: 16427 byte(s)
Document update for 8.31-RC1 test release.

1 ph10 965 .TH PCREJIT 3 "04 May 2012" "PCRE 8.31"
2 ph10 678 .SH NAME
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 ph10 683 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 ph10 858 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 ph10 678 .P
16 ph10 858 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 ph10 678 .
20     .
21 ph10 858 .SH "8-BIT and 16-BIT SUPPORT"
22     .rs
23     .sp
24 ph10 903 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 ph10 858 \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 ph10 683 Intel x86 32-bit and 64-bit
40 ph10 678 MIPS 32-bit
41 ph10 859 Power PC 32-bit and 64-bit
42 ph10 683 .sp
43 ph10 921 If --enable-jit is set on an unsupported platform, compilation fails.
44 ph10 678 .P
45 ph10 836 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 ph10 921 way that falls back to the interpretive code if JIT is not available.
50 ph10 836 .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 ph10 678 .
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 ph10 691 .sp
66 ph10 678 (2) Use \fBpcre_free_study()\fP to free the \fBpcre_extra\fP block when it is
67 ph10 921 no longer needed, instead of just freeing it yourself. This
68 ph10 683 ensures that any JIT data is also freed.
69 ph10 678 .sp
70 ph10 836 For a program that may be linked with pre-8.20 versions of PCRE, you can insert
71     .sp
73     #define PCRE_STUDY_JIT_COMPILE 0
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
79     #ifdef PCRE_CONFIG_JIT
80     pcre_free_study(study_ptr);
81     #else
82     pcre_free(study_ptr);
83     #endif
84     .sp
85 ph10 975 PCRE_STUDY_JIT_COMPILE requests the JIT compiler to generate code for complete
86 ph10 921 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 ph10 683 In some circumstances you may need to call additional functions. These are
100 ph10 678 described in the section entitled
101     .\" HTML <a href="#stackcontrol">
102     .\" </a>
103     "Controlling the JIT stack"
104     .\"
105     below.
106     .P
107 ph10 921 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 ph10 975 block containing a pointer to JIT code of the appropriate mode (normal or
112 ph10 921 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 ph10 678 .P
115     There are some \fBpcre_exec()\fP options that are not supported for JIT
116 ph10 683 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 ph10 926 interpretive code. If you want to know whether JIT was actually used for a
119     particular match, you should arrange for a JIT callback function to be set up
120     as described in the section entitled
121     .\" HTML <a href="#stackcontrol">
122     .\" </a>
123     "Controlling the JIT stack"
124     .\"
125     below, even if you do not need to supply a non-default JIT stack. Such a
126     callback function is called whenever JIT code is about to be obeyed. If the
127     execution options are not right for JIT execution, the callback function is not
128     obeyed.
129 ph10 678 .P
130     If the JIT compiler finds an unsupported item, no JIT data is generated. You
131     can find out if JIT execution is available after studying a pattern by calling
132     \fBpcre_fullinfo()\fP with the PCRE_INFO_JIT option. A result of 1 means that
133 ph10 693 JIT compilation was successful. A result of 0 means that JIT support is not
134 ph10 921 available, or the pattern was not studied with PCRE_STUDY_JIT_COMPILE etc., or
135     the JIT compiler was not able to handle the pattern.
136 ph10 707 .P
137 ph10 708 Once a pattern has been studied, with or without JIT, it can be used as many
138 ph10 707 times as you like for matching different subject strings.
139 ph10 678 .
140     .
142     .rs
143     .sp
144     The only \fBpcre_exec()\fP options that are supported for JIT execution are
147 ph10 678 .P
148     The unsupported pattern items are:
149     .sp
150 ph10 836 \eC match a single byte; not supported in UTF-8 mode
151 ph10 678 (?Cn) callouts
152 ph10 975 (*PRUNE) )
153 ph10 965 (*SKIP) ) backtracking control verbs
154 ph10 678 (*THEN) )
155 ph10 683 .sp
156 ph10 678 Support for some of these may be added in future.
157     .
158     .
160     .rs
161     .sp
162 ph10 683 When a pattern is matched using JIT execution, the return values are the same
163     as those given by the interpretive \fBpcre_exec()\fP code, with the addition of
164     one new error code: PCRE_ERROR_JIT_STACKLIMIT. This means that the memory used
165 ph10 678 for the JIT stack was insufficient. See
166     .\" HTML <a href="#stackcontrol">
167     .\" </a>
168     "Controlling the JIT stack"
169     .\"
170 ph10 683 below for a discussion of JIT stack usage. For compatibility with the
171     interpretive \fBpcre_exec()\fP code, no more than two-thirds of the
172     \fIovector\fP argument is used for passing back captured substrings.
173 ph10 678 .P
174     The error code PCRE_ERROR_MATCHLIMIT is returned by the JIT code if searching a
175     very large pattern tree goes on for too long, as it is in the same circumstance
176     when JIT is not used, but the details of exactly what is counted are not the
177     same. The PCRE_ERROR_RECURSIONLIMIT error code is never returned by JIT
178     execution.
179     .
180     .
182     .rs
183     .sp
184 ph10 683 The code that is generated by the JIT compiler is architecture-specific, and is
185 ph10 708 also position dependent. For those reasons it cannot be saved (in a file or
186 ph10 707 database) and restored later like the bytecode and other data of a compiled
187 ph10 708 pattern. Saving and restoring compiled patterns is not something many people
188 ph10 707 do. More detail about this facility is given in the
189     .\" HREF
190     \fBpcreprecompile\fP
191     .\"
192     documentation. It should be possible to run \fBpcre_study()\fP on a saved and
193     restored pattern, and thereby recreate the JIT data, but because JIT
194 ph10 708 compilation uses significant resources, it is probably not worth doing this;
195 ph10 707 you might as well recompile the original pattern.
196 ph10 678 .
197     .
198     .\" HTML <a name="stackcontrol"></a>
200     .rs
201     .sp
202 ph10 683 When the compiled JIT code runs, it needs a block of memory to use as a stack.
203     By default, it uses 32K on the machine stack. However, some large or
204     complicated patterns need more than this. The error PCRE_ERROR_JIT_STACKLIMIT
205     is given when there is not enough stack. Three functions are provided for
206 ph10 836 managing blocks of memory for use as JIT stacks. There is further discussion
207     about the use of JIT stacks in the section entitled
208     .\" HTML <a href="#stackcontrol">
209     .\" </a>
210     "JIT stack FAQ"
211     .\"
212     below.
213 ph10 678 .P
214 ph10 683 The \fBpcre_jit_stack_alloc()\fP function creates a JIT stack. Its arguments
215 ph10 691 are a starting size and a maximum size, and it returns a pointer to an opaque
216 ph10 686 structure of type \fBpcre_jit_stack\fP, or NULL if there is an error. The
217     \fBpcre_jit_stack_free()\fP function can be used to free a stack that is no
218     longer needed. (For the technically minded: the address space is allocated by
219     mmap or VirtualAlloc.)
220 ph10 678 .P
221 ph10 691 JIT uses far less memory for recursion than the interpretive code,
222 ph10 683 and a maximum stack size of 512K to 1M should be more than enough for any
223     pattern.
224     .P
225     The \fBpcre_assign_jit_stack()\fP function specifies which stack JIT code
226 ph10 678 should use. Its arguments are as follows:
227     .sp
228     pcre_extra *extra
229     pcre_jit_callback callback
230     void *data
231 ph10 683 .sp
232     The \fIextra\fP argument must be the result of studying a pattern with
233 ph10 921 PCRE_STUDY_JIT_COMPILE etc. There are three cases for the values of the other
234     two options:
235 ph10 678 .sp
236     (1) If \fIcallback\fP is NULL and \fIdata\fP is NULL, an internal 32K block
237     on the machine stack is used.
238     .sp
239     (2) If \fIcallback\fP is NULL and \fIdata\fP is not NULL, \fIdata\fP must be
240     a valid JIT stack, the result of calling \fBpcre_jit_stack_alloc()\fP.
241     .sp
242 ph10 975 (3) If \fIcallback\fP is not NULL, it must point to a function that is
243     called with \fIdata\fP as an argument at the start of matching, in
244     order to set up a JIT stack. If the return from the callback
245     function is NULL, the internal 32K stack is used; otherwise the
246     return value must be a valid JIT stack, the result of calling
247 ph10 926 \fBpcre_jit_stack_alloc()\fP.
248 ph10 678 .sp
249 ph10 975 A callback function is obeyed whenever JIT code is about to be run; it is not
250     obeyed when \fBpcre_exec()\fP is called with options that are incompatible for
251 ph10 926 JIT execution. A callback function can therefore be used to determine whether a
252     match operation was executed by JIT or by the interpreter.
253     .P
254 ph10 928 You may safely use the same JIT stack for more than one pattern (either by
255     assigning directly or by callback), as long as the patterns are all matched
256     sequentially in the same thread. In a multithread application, if you do not
257 ph10 975 specify a JIT stack, or if you assign or pass back NULL from a callback, that
258     is thread-safe, because each thread has its own machine stack. However, if you
259     assign or pass back a non-NULL JIT stack, this must be a different stack for
260 ph10 928 each thread so that the application is thread-safe.
261 ph10 678 .P
262 ph10 928 Strictly speaking, even more is allowed. You can assign the same non-NULL stack
263     to any number of patterns as long as they are not used for matching by multiple
264 ph10 683 threads at the same time. For example, you can assign the same stack to all
265     compiled patterns, and use a global mutex in the callback to wait until the
266 ph10 928 stack is available for use. However, this is an inefficient solution, and not
267     recommended.
268 ph10 683 .P
269 ph10 928 This is a suggestion for how a multithreaded program that needs to set up
270     non-default JIT stacks might operate:
271 ph10 683 .sp
272     During thread initalization
273     thread_local_var = pcre_jit_stack_alloc(...)
274 ph10 691 .sp
275 ph10 683 During thread exit
276     pcre_jit_stack_free(thread_local_var)
277 ph10 691 .sp
278 ph10 683 Use a one-line callback function
279     return thread_local_var
280     .sp
281 ph10 678 All the functions described in this section do nothing if JIT is not available,
282 ph10 683 and \fBpcre_assign_jit_stack()\fP does nothing unless the \fBextra\fP argument
283     is non-NULL and points to a \fBpcre_extra\fP block that is the result of a
284 ph10 921 successful study with PCRE_STUDY_JIT_COMPILE etc.
285 ph10 678 .
286     .
287 ph10 836 .\" HTML <a name="stackfaq"></a>
288     .SH "JIT STACK FAQ"
289     .rs
290     .sp
291     (1) Why do we need JIT stacks?
292     .sp
293     PCRE (and JIT) is a recursive, depth-first engine, so it needs a stack where
294     the local data of the current node is pushed before checking its child nodes.
295     Allocating real machine stack on some platforms is difficult. For example, the
296     stack chain needs to be updated every time if we extend the stack on PowerPC.
297     Although it is possible, its updating time overhead decreases performance. So
298     we do the recursion in memory.
299     .P
300     (2) Why don't we simply allocate blocks of memory with \fBmalloc()\fP?
301     .sp
302     Modern operating systems have a nice feature: they can reserve an address space
303     instead of allocating memory. We can safely allocate memory pages inside this
304     address space, so the stack could grow without moving memory data (this is
305     important because of pointers). Thus we can allocate 1M address space, and use
306     only a single memory page (usually 4K) if that is enough. However, we can still
307     grow up to 1M anytime if needed.
308     .P
309     (3) Who "owns" a JIT stack?
310     .sp
311     The owner of the stack is the user program, not the JIT studied pattern or
312     anything else. The user program must ensure that if a stack is used by
313     \fBpcre_exec()\fP, (that is, it is assigned to the pattern currently running),
314     that stack must not be used by any other threads (to avoid overwriting the same
315     memory area). The best practice for multithreaded programs is to allocate a
316     stack for each thread, and return this stack through the JIT callback function.
317     .P
318     (4) When should a JIT stack be freed?
319     .sp
320     You can free a JIT stack at any time, as long as it will not be used by
321     \fBpcre_exec()\fP again. When you assign the stack to a pattern, only a pointer
322     is set. There is no reference counting or any other magic. You can free the
323     patterns and stacks in any order, anytime. Just \fIdo not\fP call
324     \fBpcre_exec()\fP with a pattern pointing to an already freed stack, as that
325     will cause SEGFAULT. (Also, do not free a stack currently used by
326     \fBpcre_exec()\fP in another thread). You can also replace the stack for a
327     pattern at any time. You can even free the previous stack before assigning a
328     replacement.
329     .P
330     (5) Should I allocate/free a stack every time before/after calling
331     \fBpcre_exec()\fP?
332     .sp
333     No, because this is too costly in terms of resources. However, you could
334     implement some clever idea which release the stack if it is not used in let's
335     say two minutes. The JIT callback can help to achive this without keeping a
336     list of the currently JIT studied patterns.
337     .P
338     (6) OK, the stack is for long term memory allocation. But what happens if a
339     pattern causes stack overflow with a stack of 1M? Is that 1M kept until the
340     stack is freed?
341     .sp
342 ph10 928 Especially on embedded sytems, it might be a good idea to release memory
343     sometimes without freeing the stack. There is no API for this at the moment.
344     Probably a function call which returns with the currently allocated memory for
345     any stack and another which allows releasing memory (shrinking the stack) would
346     be a good idea if someone needs this.
347 ph10 836 .P
348     (7) This is too much of a headache. Isn't there any better solution for JIT
349     stack handling?
350     .sp
351     No, thanks to Windows. If POSIX threads were used everywhere, we could throw
352     out this complicated API.
353     .
354     .
355 ph10 678 .SH "EXAMPLE CODE"
356     .rs
357     .sp
358 ph10 683 This is a single-threaded example that specifies a JIT stack without using a
359 ph10 691 callback.
360 ph10 678 .sp
361     int rc;
362 ph10 683 int ovector[30];
363 ph10 678 pcre *re;
364 ph10 683 pcre_extra *extra;
365     pcre_jit_stack *jit_stack;
366     .sp
367 ph10 678 re = pcre_compile(pattern, 0, &error, &erroffset, NULL);
368     /* Check for errors */
369     extra = pcre_study(re, PCRE_STUDY_JIT_COMPILE, &error);
370 ph10 683 jit_stack = pcre_jit_stack_alloc(32*1024, 512*1024);
371 ph10 678 /* Check for error (NULL) */
372     pcre_assign_jit_stack(extra, NULL, jit_stack);
373 ph10 683 rc = pcre_exec(re, extra, subject, length, 0, 0, ovector, 30);
374 ph10 678 /* Check results */
375     pcre_free(re);
376 ph10 683 pcre_free_study(extra);
377 ph10 691 pcre_jit_stack_free(jit_stack);
378 ph10 678 .sp
379     .
380     .
381     .SH "SEE ALSO"
382     .rs
383     .sp
384     \fBpcreapi\fP(3)
385     .
386     .
387     .SH AUTHOR
388     .rs
389     .sp
390     .nf
391 ph10 836 Philip Hazel (FAQ by Zoltan Herczeg)
392 ph10 678 University Computing Service
393     Cambridge CB2 3QH, England.
394     .fi
395     .
396     .
397     .SH REVISION
398     .rs
399     .sp
400     .nf
401 ph10 965 Last updated: 04 May 2012
402 ph10 858 Copyright (c) 1997-2012 University of Cambridge.
403 ph10 678 .fi

ViewVC Help
Powered by ViewVC 1.1.12