1PCRE2JIT(3) Library Functions Manual PCRE2JIT(3)
2
6 PCRE2 - Perl-compatible regular expressions (revised API)
7
9 Just-in-time compiling is a heavyweight optimization that can greatly
11 speed up pattern matching. However, it comes at the cost of extra pro‐
12 cessing before the match is performed, so it is of most benefit when
13 the same pattern is going to be matched many times. This does not nec‐
14 essarily mean many calls of a matching function; if the pattern is not
15 anchored, matching attempts may take place many times at various posi‐
16 tions in the subject, even for a single call. Therefore, if the subject
17 string is very long, it may still pay to use JIT even for one-off
18 matches. JIT support is available for all of the 8-bit, 16-bit and
19 32-bit PCRE2 libraries.
20 JIT support applies only to the traditional Perl-compatible matching
22 function. It does not apply when the DFA matching function is being
23 used. The code for this support was written by Zoltan Herczeg.
24
26 JIT support is an optional feature of PCRE2. The "configure" option
28 --enable-jit (or equivalent CMake option) must be set when PCRE2 is
29 built if you want to use JIT. The support is limited to the following
30 hardware platforms:
31 ARM 32-bit (v5, v7, and Thumb2)
33 ARM 64-bit
34 Intel x86 32-bit and 64-bit
35 MIPS 32-bit and 64-bit
36 Power PC 32-bit and 64-bit
37 SPARC 32-bit
38 If --enable-jit is set on an unsupported platform, compilation fails.
40 A program can tell if JIT support is available by calling pcre2_con‐
42 fig() with the PCRE2_CONFIG_JIT option. The result is 1 when JIT is
43 available, and 0 otherwise. However, a simple program does not need to
44 check this in order to use JIT. The API is implemented in a way that
45 falls back to the interpretive code if JIT is not available. For pro‐
46 grams that need the best possible performance, there is also a "fast
47 path" API that is JIT-specific.
48
50 To make use of the JIT support in the simplest way, all you have to do
52 is to call pcre2_jit_compile() after successfully compiling a pattern
53 with pcre2_compile(). This function has two arguments: the first is the
54 compiled pattern pointer that was returned by pcre2_compile(), and the
55 second is zero or more of the following option bits: PCRE2_JIT_COM‐
56 PLETE, PCRE2_JIT_PARTIAL_HARD, or PCRE2_JIT_PARTIAL_SOFT.
57 If JIT support is not available, a call to pcre2_jit_compile() does
59 nothing and returns PCRE2_ERROR_JIT_BADOPTION. Otherwise, the compiled
60 pattern is passed to the JIT compiler, which turns it into machine code
61 that executes much faster than the normal interpretive code, but yields
62 exactly the same results. The returned value from pcre2_jit_compile()
63 is zero on success, or a negative error code.
64 There is a limit to the size of pattern that JIT supports, imposed by
66 the size of machine stack that it uses. The exact rules are not docu‐
67 mented because they may change at any time, in particular, when new
68 optimizations are introduced. If a pattern is too big, a call to
69 pcre2_jit_compile() returns PCRE2_ERROR_NOMEMORY.
70 PCRE2_JIT_COMPLETE requests the JIT compiler to generate code for com‐
72 plete matches. If you want to run partial matches using the PCRE2_PAR‐
73 TIAL_HARD or PCRE2_PARTIAL_SOFT options of pcre2_match(), you should
74 set one or both of the other options as well as, or instead of
75 PCRE2_JIT_COMPLETE. The JIT compiler generates different optimized code
76 for each of the three modes (normal, soft partial, hard partial). When
77 pcre2_match() is called, the appropriate code is run if it is avail‐
78 able. Otherwise, the pattern is matched using interpretive code.
79 You can call pcre2_jit_compile() multiple times for the same compiled
81 pattern. It does nothing if it has previously compiled code for any of
82 the option bits. For example, you can call it once with PCRE2_JIT_COM‐
83 PLETE and (perhaps later, when you find you need partial matching)
84 again with PCRE2_JIT_COMPLETE and PCRE2_JIT_PARTIAL_HARD. This time it
85 will ignore PCRE2_JIT_COMPLETE and just compile code for partial match‐
86 ing. If pcre2_jit_compile() is called with no option bits set, it imme‐
87 diately returns zero. This is an alternative way of testing whether JIT
88 is available.
89 At present, it is not possible to free JIT compiled code except when
91 the entire compiled pattern is freed by calling pcre2_code_free().
92 In some circumstances you may need to call additional functions. These
94 are described in the section entitled "Controlling the JIT stack"
95 below.
96 There are some pcre2_match() options that are not supported by JIT, and
98 there are also some pattern items that JIT cannot handle. Details are
99 given below. In both cases, matching automatically falls back to the
100 interpretive code. If you want to know whether JIT was actually used
101 for a particular match, you should arrange for a JIT callback function
102 to be set up as described in the section entitled "Controlling the JIT
103 stack" below, even if you do not need to supply a non-default JIT
104 stack. Such a callback function is called whenever JIT code is about to
105 be obeyed. If the match-time options are not right for JIT execution,
106 the callback function is not obeyed.
107 If the JIT compiler finds an unsupported item, no JIT data is gener‐
109 ated. You can find out if JIT matching is available after compiling a
110 pattern by calling pcre2_pattern_info() with the PCRE2_INFO_JITSIZE
111 option. A non-zero result means that JIT compilation was successful. A
112 result of 0 means that JIT support is not available, or the pattern was
113 not processed by pcre2_jit_compile(), or the JIT compiler was not able
114 to handle the pattern.
115
117 The pcre2_match() options that are supported for JIT matching are
119 PCRE2_NOTBOL, PCRE2_NOTEOL, PCRE2_NOTEMPTY, PCRE2_NOTEMPTY_ATSTART,
120 PCRE2_NO_UTF_CHECK, PCRE2_PARTIAL_HARD, and PCRE2_PARTIAL_SOFT. The
121 PCRE2_ANCHORED option is not supported at match time.
122 If the PCRE2_NO_JIT option is passed to pcre2_match() it disables the
124 use of JIT, forcing matching by the interpreter code.
125 The only unsupported pattern items are \C (match a single data unit)
127 when running in a UTF mode, and a callout immediately before an asser‐
128 tion condition in a conditional group.
129
131 When a pattern is matched using JIT matching, the return values are the
133 same as those given by the interpretive pcre2_match() code, with the
134 addition of one new error code: PCRE2_ERROR_JIT_STACKLIMIT. This means
135 that the memory used for the JIT stack was insufficient. See "Control‐
136 ling the JIT stack" below for a discussion of JIT stack usage.
137 The error code PCRE2_ERROR_MATCHLIMIT is returned by the JIT code if
139 searching a very large pattern tree goes on for too long, as it is in
140 the same circumstance when JIT is not used, but the details of exactly
141 what is counted are not the same. The PCRE2_ERROR_RECURSIONLIMIT error
142 code is never returned when JIT matching is used.
143
145 When the compiled JIT code runs, it needs a block of memory to use as a
147 stack. By default, it uses 32K on the machine stack. However, some
148 large or complicated patterns need more than this. The error
149 PCRE2_ERROR_JIT_STACKLIMIT is given when there is not enough stack.
150 Three functions are provided for managing blocks of memory for use as
151 JIT stacks. There is further discussion about the use of JIT stacks in
152 the section entitled "JIT stack FAQ" below.
153 The pcre2_jit_stack_create() function creates a JIT stack. Its argu‐
155 ments are a starting size, a maximum size, and a general context (for
156 memory allocation functions, or NULL for standard memory allocation).
157 It returns a pointer to an opaque structure of type pcre2_jit_stack, or
158 NULL if there is an error. The pcre2_jit_stack_free() function is used
159 to free a stack that is no longer needed. (For the technically minded:
160 the address space is allocated by mmap or VirtualAlloc.)
161 JIT uses far less memory for recursion than the interpretive code, and
163 a maximum stack size of 512K to 1M should be more than enough for any
164 pattern.
165 The pcre2_jit_stack_assign() function specifies which stack JIT code
167 should use. Its arguments are as follows:
168 pcre2_match_context *mcontext
170 pcre2_jit_callback callback
171 void *data
172 The first argument is a pointer to a match context. When this is subse‐
174 quently passed to a matching function, its information determines which
175 JIT stack is used. There are three cases for the values of the other
176 two options:
177 (1) If callback is NULL and data is NULL, an internal 32K block
179 on the machine stack is used. This is the default when a match
180 context is created.
181 (2) If callback is NULL and data is not NULL, data must be
183 a pointer to a valid JIT stack, the result of calling
184 pcre2_jit_stack_create().
185 (3) If callback is not NULL, it must point to a function that is
187 called with data as an argument at the start of matching, in
188 order to set up a JIT stack. If the return from the callback
189 function is NULL, the internal 32K stack is used; otherwise the
190 return value must be a valid JIT stack, the result of calling
191 pcre2_jit_stack_create().
192 A callback function is obeyed whenever JIT code is about to be run; it
194 is not obeyed when pcre2_match() is called with options that are incom‐
195 patible for JIT matching. A callback function can therefore be used to
196 determine whether a match operation was executed by JIT or by the
197 interpreter.
198 You may safely use the same JIT stack for more than one pattern (either
200 by assigning directly or by callback), as long as the patterns are
201 matched sequentially in the same thread. Currently, the only way to set
202 up non-sequential matches in one thread is to use callouts: if a call‐
203 out function starts another match, that match must use a different JIT
204 stack to the one used for currently suspended match(es).
205 In a multithread application, if you do not specify a JIT stack, or if
207 you assign or pass back NULL from a callback, that is thread-safe,
208 because each thread has its own machine stack. However, if you assign
209 or pass back a non-NULL JIT stack, this must be a different stack for
210 each thread so that the application is thread-safe.
211 Strictly speaking, even more is allowed. You can assign the same non-
213 NULL stack to a match context that is used by any number of patterns,
214 as long as they are not used for matching by multiple threads at the
215 same time. For example, you could use the same stack in all compiled
216 patterns, with a global mutex in the callback to wait until the stack
217 is available for use. However, this is an inefficient solution, and not
218 recommended.
219 This is a suggestion for how a multithreaded program that needs to set
221 up non-default JIT stacks might operate:
222 During thread initalization
224 thread_local_var = pcre2_jit_stack_create(...)
225 During thread exit
227 pcre2_jit_stack_free(thread_local_var)
228 Use a one-line callback function
230 return thread_local_var
231 All the functions described in this section do nothing if JIT is not
233 available.
234
236 (1) Why do we need JIT stacks?
238 PCRE2 (and JIT) is a recursive, depth-first engine, so it needs a stack
240 where the local data of the current node is pushed before checking its
241 child nodes. Allocating real machine stack on some platforms is diffi‐
242 cult. For example, the stack chain needs to be updated every time if we
243 extend the stack on PowerPC. Although it is possible, its updating
244 time overhead decreases performance. So we do the recursion in memory.
245 (2) Why don't we simply allocate blocks of memory with malloc()?
247 Modern operating systems have a nice feature: they can reserve an
249 address space instead of allocating memory. We can safely allocate mem‐
250 ory pages inside this address space, so the stack could grow without
251 moving memory data (this is important because of pointers). Thus we can
252 allocate 1M address space, and use only a single memory page (usually
253 4K) if that is enough. However, we can still grow up to 1M anytime if
254 needed.
255 (3) Who "owns" a JIT stack?
257 The owner of the stack is the user program, not the JIT studied pattern
259 or anything else. The user program must ensure that if a stack is being
260 used by pcre2_match(), (that is, it is assigned to a match context that
261 is passed to the pattern currently running), that stack must not be
262 used by any other threads (to avoid overwriting the same memory area).
263 The best practice for multithreaded programs is to allocate a stack for
264 each thread, and return this stack through the JIT callback function.
265 (4) When should a JIT stack be freed?
267 You can free a JIT stack at any time, as long as it will not be used by
269 pcre2_match() again. When you assign the stack to a match context, only
270 a pointer is set. There is no reference counting or any other magic.
271 You can free compiled patterns, contexts, and stacks in any order, any‐
272 time. Just do not call pcre2_match() with a match context pointing to
273 an already freed stack, as that will cause SEGFAULT. (Also, do not free
274 a stack currently used by pcre2_match() in another thread). You can
275 also replace the stack in a context at any time when it is not in use.
276 You should free the previous stack before assigning a replacement.
277 (5) Should I allocate/free a stack every time before/after calling
279 pcre2_match()?
280 No, because this is too costly in terms of resources. However, you
282 could implement some clever idea which release the stack if it is not
283 used in let's say two minutes. The JIT callback can help to achieve
284 this without keeping a list of patterns.
285 (6) OK, the stack is for long term memory allocation. But what happens
287 if a pattern causes stack overflow with a stack of 1M? Is that 1M kept
288 until the stack is freed?
289 Especially on embedded sytems, it might be a good idea to release mem‐
291 ory sometimes without freeing the stack. There is no API for this at
292 the moment. Probably a function call which returns with the currently
293 allocated memory for any stack and another which allows releasing mem‐
294 ory (shrinking the stack) would be a good idea if someone needs this.
295 (7) This is too much of a headache. Isn't there any better solution for
297 JIT stack handling?
298 No, thanks to Windows. If POSIX threads were used everywhere, we could
300 throw out this complicated API.
301
303 void pcre2_jit_free_unused_memory(pcre2_general_context *gcontext);
305 The JIT executable allocator does not free all memory when it is possi‐
307 ble. It expects new allocations, and keeps some free memory around to
308 improve allocation speed. However, in low memory conditions, it might
309 be better to free all possible memory. You can cause this to happen by
310 calling pcre2_jit_free_unused_memory(). Its argument is a general con‐
311 text, for custom memory management, or NULL for standard memory manage‐
312 ment.
313
315 This is a single-threaded example that specifies a JIT stack without
317 using a callback. A real program should include error checking after
318 all the function calls.
319 int rc;
321 pcre2_code *re;
322 pcre2_match_data *match_data;
323 pcre2_match_context *mcontext;
324 pcre2_jit_stack *jit_stack;
325 re = pcre2_compile(pattern, PCRE2_ZERO_TERMINATED, 0,
327 &errornumber, &erroffset, NULL);
328 rc = pcre2_jit_compile(re, PCRE2_JIT_COMPLETE);
329 mcontext = pcre2_match_context_create(NULL);
330 jit_stack = pcre2_jit_stack_create(32*1024, 512*1024, NULL);
331 pcre2_jit_stack_assign(mcontext, NULL, jit_stack);
332 match_data = pcre2_match_data_create(re, 10);
333 rc = pcre2_match(re, subject, length, 0, 0, match_data, mcontext);
334 /* Process result */
335 pcre2_code_free(re);
337 pcre2_match_data_free(match_data);
338 pcre2_match_context_free(mcontext);
339 pcre2_jit_stack_free(jit_stack);
340
343 Because the API described above falls back to interpreted matching when
345 JIT is not available, it is convenient for programs that are written
346 for general use in many environments. However, calling JIT via
347 pcre2_match() does have a performance impact. Programs that are written
348 for use where JIT is known to be available, and which need the best
349 possible performance, can instead use a "fast path" API to call JIT
350 matching directly instead of calling pcre2_match() (obviously only for
351 patterns that have been successfully processed by pcre2_jit_compile()).
352 The fast path function is called pcre2_jit_match(), and it takes
354 exactly the same arguments as pcre2_match(). The return values are also
355 the same, plus PCRE2_ERROR_JIT_BADOPTION if a matching mode (partial or
356 complete) is requested that was not compiled. Unsupported option bits
357 (for example, PCRE2_ANCHORED) are ignored, as is the PCRE2_NO_JIT
358 option.
359 When you call pcre2_match(), as well as testing for invalid options, a
361 number of other sanity checks are performed on the arguments. For exam‐
362 ple, if the subject pointer is NULL, an immediate error is given. Also,
363 unless PCRE2_NO_UTF_CHECK is set, a UTF subject string is tested for
364 validity. In the interests of speed, these checks do not happen on the
365 JIT fast path, and if invalid data is passed, the result is undefined.
366 Bypassing the sanity checks and the pcre2_match() wrapping can give
368 speedups of more than 10%.
369
371 pcre2api(3)
373
375 Philip Hazel (FAQ by Zoltan Herczeg)
377 University Computing Service
378 Cambridge, England.
379
381 Last updated: 05 June 2016
383 Copyright (c) 1997-2016 University of Cambridge.
384PCRE2 10.22 05 June 2016 PCRE2JIT(3)