/[pcre]/code/trunk/doc/pcrejit.3
ViewVC logotype

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

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 678 by ph10, Sun Aug 28 15:23:03 2011 UTC revision 761 by ph10, Tue Nov 22 12:24:26 2011 UTC
# Line 5  PCRE - Perl-compatible regular expressio Line 5  PCRE - Perl-compatible regular expressio
5  .rs  .rs
6  .sp  .sp
7  Just-in-time compiling is a heavyweight optimization that can greatly speed up  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  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  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  going to be matched many times. This does not necessarily mean many calls of
11  \fPpcre_exec()\fP; if the pattern is not anchored, matching attempts may take  \fPpcre_exec()\fP; 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 to  place many times at various positions in the subject, even for a single call to
13  \fBpcre_exec()\fP. If the subject string is very long, it may still pay to use  \fBpcre_exec()\fP. If the subject string is very long, it may still pay to use
14  JIT for one-off matches.  JIT for one-off matches.
15  .P  .P
16  JIT support applies only to the traditional matching function,  JIT support applies only to the traditional matching function,
17  \fBpcre_exec()\fP. It does not apply when \fBpcre_dfa_exec()\fP is being used.  \fBpcre_exec()\fP. It does not apply when \fBpcre_dfa_exec()\fP is being used.
18  The code for this support was written by Zoltan Herczeg.  The code for this support was written by Zoltan Herczeg.
19  .  .
# Line 26  JIT support is an optional feature of PC Line 26  JIT support is an optional feature of PC
26  JIT. The support is limited to the following hardware platforms:  JIT. The support is limited to the following hardware platforms:
27  .sp  .sp
28    ARM v5, v7, and Thumb2    ARM v5, v7, and Thumb2
   MIPS 32-bit  
   Power PC 32-bit and 64-bit  
29    Intel x86 32-bit and 64-bit    Intel x86 32-bit and 64-bit
30  .sp    MIPS 32-bit
31  If --enable-jit is set on an unsupported platform, compilation fails.    Power PC 32-bit and 64-bit (experimental)
32    .sp
33    The Power PC support is designated as experimental because it has not been
34    fully tested. If --enable-jit is set on an unsupported platform, compilation
35    fails.
36  .P  .P
37  A program can tell if JIT support is available by calling \fBpcre_config()\fP  A program can tell if JIT support is available by calling \fBpcre_config()\fP
38  with the PCRE_CONFIG_JIT option. The result is 1 when JIT is available, and 0  with the PCRE_CONFIG_JIT option. The result is 1 when JIT is available, and 0
39  otherwise. However, a simple program does not need to check this in order to  otherwise. However, a simple program does not need to check this in order to
40  use JIT. The API is implemented in a way that falls back to the ordinary PCRE  use JIT. The API is implemented in a way that falls back to the ordinary PCRE
41  code if JIT is not available.  code if JIT is not available.
# Line 47  You have to do two things to make use of Line 49  You have to do two things to make use of
49    (1) Call \fBpcre_study()\fP with the PCRE_STUDY_JIT_COMPILE option for    (1) Call \fBpcre_study()\fP with the PCRE_STUDY_JIT_COMPILE option for
50        each compiled pattern, and pass the resulting \fBpcre_extra\fP block to        each compiled pattern, and pass the resulting \fBpcre_extra\fP block to
51        \fBpcre_exec()\fP.        \fBpcre_exec()\fP.
52    .sp
53    (2) Use \fBpcre_free_study()\fP to free the \fBpcre_extra\fP block when it is    (2) Use \fBpcre_free_study()\fP to free the \fBpcre_extra\fP block when it is
54        no longer needed instead of just freeing it yourself. This ensures that        no longer needed instead of just freeing it yourself. This
55        any JIT data is also freed.        ensures that any JIT data is also freed.
56  .sp  .sp
57  In some circumstances you may need to call additional functions. These are  In some circumstances you may need to call additional functions. These are
58  described in the section entitled  described in the section entitled
59  .\" HTML <a href="#stackcontrol">  .\" HTML <a href="#stackcontrol">
60  .\" </a>  .\" </a>
# Line 60  described in the section entitled Line 62  described in the section entitled
62  .\"  .\"
63  below.  below.
64  .P  .P
65  If JIT support is not available, PCRE_STUDY_JIT_COMPILE is ignored, and no JIT  If JIT support is not available, PCRE_STUDY_JIT_COMPILE is ignored, and no JIT
66  data is set up. Otherwise, the compiled pattern is passed to the JIT compiler,  data is set up. Otherwise, the compiled pattern is passed to the JIT compiler,
67  which turns it into machine code that executes much faster than the normal  which turns it into machine code that executes much faster than the normal
68  interpretive code. When \fBpcre_exec()\fP is passed a \fBpcre_extra\fP block  interpretive code. When \fBpcre_exec()\fP is passed a \fBpcre_extra\fP block
69  containing a pointer to JIT code, it obeys that instead of the normal code. The  containing a pointer to JIT code, it obeys that instead of the normal code. The
70  result is identical, but the code runs much faster.  result is identical, but the code runs much faster.
71  .P  .P
72  There are some \fBpcre_exec()\fP options that are not supported for JIT  There are some \fBpcre_exec()\fP options that are not supported for JIT
73  execution. There are also some pattern items that JIT cannot handle. Details  execution. There are also some pattern items that JIT cannot handle. Details
74  are given below. In both cases, execution automatically falls back to the  are given below. In both cases, execution automatically falls back to the
75  interpretive code.  interpretive code.
76  .P  .P
77  If the JIT compiler finds an unsupported item, no JIT data is generated. You  If the JIT compiler finds an unsupported item, no JIT data is generated. You
78  can find out if JIT execution is available after studying a pattern by calling  can find out if JIT execution is available after studying a pattern by calling
79  \fBpcre_fullinfo()\fP with the PCRE_INFO_JIT option. A result of 1 means that  \fBpcre_fullinfo()\fP with the PCRE_INFO_JIT option. A result of 1 means that
80  JIT compilationw was successful. A result of 0 means that JIT support is not  JIT compilation was successful. A result of 0 means that JIT support is not
81  available, or the pattern was not studied with PCRE_STUDY_JIT_COMPILE, or the  available, or the pattern was not studied with PCRE_STUDY_JIT_COMPILE, or the
82  JIT compiler was not able to handle the pattern.  JIT compiler was not able to handle the pattern.
83    .P
84    Once a pattern has been studied, with or without JIT, it can be used as many
85    times as you like for matching different subject strings.
86  .  .
87  .  .
88  .SH "UNSUPPORTED OPTIONS AND PATTERN ITEMS"  .SH "UNSUPPORTED OPTIONS AND PATTERN ITEMS"
89  .rs  .rs
90  .sp  .sp
91  The only \fBpcre_exec()\fP options that are supported for JIT execution are  The only \fBpcre_exec()\fP options that are supported for JIT execution are
92  PCRE_NO_UTF8_CHECK, PCRE_NOTBOL, PCRE_NOTEOL, PCRE_NOTEMPTY, and  PCRE_NO_UTF8_CHECK, PCRE_NOTBOL, PCRE_NOTEOL, PCRE_NOTEMPTY, and
93  PCRE_NOTEMPTY_ATSTART. Note in particular that partial matching is not  PCRE_NOTEMPTY_ATSTART. Note in particular that partial matching is not
94  supported.  supported.
95  .P  .P
96  The unsupported pattern items are:  The unsupported pattern items are:
97  .sp  .sp
98    \eC            match a single byte, even in UTF-8 mode    \eC            match a single byte; not supported in UTF-8 mode
99    (?Cn)          callouts    (?Cn)          callouts
   (?(<name>)...  conditional test on setting of a named subpattern  
   (?(R)...       conditional test on whole pattern recursion  
   (?(Rn)...      conditional test on recursion, by number  
   (?(R&name)...  conditional test on recursion, by name  
100    (*COMMIT)      )    (*COMMIT)      )
101    (*MARK)        )    (*MARK)        )
102    (*PRUNE)       ) the backtracking control verbs    (*PRUNE)       ) the backtracking control verbs
103    (*SKIP)        )    (*SKIP)        )
104    (*THEN)        )    (*THEN)        )
105  .sp  .sp
106  Support for some of these may be added in future.  Support for some of these may be added in future.
107  .  .
108  .  .
109  .SH "RETURN VALUES FROM JIT EXECUTION"  .SH "RETURN VALUES FROM JIT EXECUTION"
110  .rs  .rs
111  .sp  .sp
112  When a pattern is matched using JIT execution, the return values are the same  When a pattern is matched using JIT execution, the return values are the same
113  as those given by the interpretive \fBpcre_exec()\fP code, with the addition of  as those given by the interpretive \fBpcre_exec()\fP code, with the addition of
114  one new error code: PCRE_ERROR_JIT_STACKLIMIT. This means that the memory used  one new error code: PCRE_ERROR_JIT_STACKLIMIT. This means that the memory used
115  for the JIT stack was insufficient. See  for the JIT stack was insufficient. See
116  .\" HTML <a href="#stackcontrol">  .\" HTML <a href="#stackcontrol">
117  .\" </a>  .\" </a>
118  "Controlling the JIT stack"  "Controlling the JIT stack"
119  .\"  .\"
120  below for a discussion of JIT stack usage.  below for a discussion of JIT stack usage. For compatibility with the
121    interpretive \fBpcre_exec()\fP code, no more than two-thirds of the
122    \fIovector\fP argument is used for passing back captured substrings.
123  .P  .P
124  The error code PCRE_ERROR_MATCHLIMIT is returned by the JIT code if searching a  The error code PCRE_ERROR_MATCHLIMIT is returned by the JIT code if searching a
125  very large pattern tree goes on for too long, as it is in the same circumstance  very large pattern tree goes on for too long, as it is in the same circumstance
# Line 128  execution. Line 131  execution.
131  .SH "SAVING AND RESTORING COMPILED PATTERNS"  .SH "SAVING AND RESTORING COMPILED PATTERNS"
132  .rs  .rs
133  .sp  .sp
134  The code that is generated by the JIT compiler is architecture-specific, and is  The code that is generated by the JIT compiler is architecture-specific, and is
135  also position dependent. For those reasons it cannot be saved and restored like  also position dependent. For those reasons it cannot be saved (in a file or
136  the bytecode and other data of a compiled pattern. You should be able run  database) and restored later like the bytecode and other data of a compiled
137  \fBpcre_study()\fP on a saved and restored pattern, and thereby recreate the  pattern. Saving and restoring compiled patterns is not something many people
138  JIT data, but because JIT compilation uses significant resources, it is  do. More detail about this facility is given in the
139  probably not worth doing.  .\" HREF
140    \fBpcreprecompile\fP
141    .\"
142    documentation. It should be possible to run \fBpcre_study()\fP on a saved and
143    restored pattern, and thereby recreate the JIT data, but because JIT
144    compilation uses significant resources, it is probably not worth doing this;
145    you might as well recompile the original pattern.
146  .  .
147  .  .
148  .\" HTML <a name="stackcontrol"></a>  .\" HTML <a name="stackcontrol"></a>
149  .SH "CONTROLLING THE JIT STACK"  .SH "CONTROLLING THE JIT STACK"
150  .rs  .rs
151  .sp  .sp
152  When the compiled JIT code runs, it needs a block of memory to use as a stack.  When the compiled JIT code runs, it needs a block of memory to use as a stack.
153  By default, it uses 32K on the machine stack. However, some large or  By default, it uses 32K on the machine stack. However, some large or
154  complicated patterns need more than this. The error PCRE_ERROR_JIT_STACKLIMIT  complicated patterns need more than this. The error PCRE_ERROR_JIT_STACKLIMIT
155  is given when there is not enough stack. Three functions are provided for  is given when there is not enough stack. Three functions are provided for
156  setting up alternative blocks of memory for use as JIT stacks.  managing blocks of memory for use as JIT stacks. There is further discussion
157  .P  about the use of JIT stacks in the section entitled
158  The \fBpcre_jit_stack_alloc()\fP function creates a JIT stack. Its arguments  .\" HTML <a href="#stackcontrol">
159  are a starting size and a maximum size, and it returns an opaque value  .\" </a>
160  of type \fBpcre_jit_stack\fP that represents a JIT stack, or NULL if there is  "JIT stack FAQ"
161  an error. The \fBpcre_jit_stack_free()\fP function can be used to free a stack  .\"
162  that is no longer needed.  below.
163    .P
164    The \fBpcre_jit_stack_alloc()\fP function creates a JIT stack. Its arguments
165    are a starting size and a maximum size, and it returns a pointer to an opaque
166    structure of type \fBpcre_jit_stack\fP, or NULL if there is an error. The
167    \fBpcre_jit_stack_free()\fP function can be used to free a stack that is no
168    longer needed. (For the technically minded: the address space is allocated by
169    mmap or VirtualAlloc.)
170    .P
171    JIT uses far less memory for recursion than the interpretive code,
172    and a maximum stack size of 512K to 1M should be more than enough for any
173    pattern.
174  .P  .P
175  The \fBpcre_assign_jit_stack()\fP function specifies which stack JIT code  The \fBpcre_assign_jit_stack()\fP function specifies which stack JIT code
176  should use. Its arguments are as follows:  should use. Its arguments are as follows:
177  .sp  .sp
178    pcre_extra         *extra    pcre_extra         *extra
179    pcre_jit_callback  callback    pcre_jit_callback  callback
180    void               *data    void               *data
181  .sp  .sp
182  The \fIextra\fP argument must be the result of studying a pattern with  The \fIextra\fP argument must be the result of studying a pattern with
183  PCRE_STUDY_JIT_COMPILE. There are three cases for the values of the other two  PCRE_STUDY_JIT_COMPILE. There are three cases for the values of the other two
184  options:  options:
185  .sp  .sp
186    (1) If \fIcallback\fP is NULL and \fIdata\fP is NULL, an internal 32K block    (1) If \fIcallback\fP is NULL and \fIdata\fP is NULL, an internal 32K block
# Line 170  options: Line 190  options:
190        a valid JIT stack, the result of calling \fBpcre_jit_stack_alloc()\fP.        a valid JIT stack, the result of calling \fBpcre_jit_stack_alloc()\fP.
191  .sp  .sp
192    (3) If \fIcallback\fP not NULL, it must point to a function that is called    (3) If \fIcallback\fP not NULL, it must point to a function that is called
193        with \fIdata\fP as an argument at the start of matching, in order to        with \fIdata\fP as an argument at the start of matching, in order to
194        set up a JIT stack. If the result is NULL, the internal 32K stack        set up a JIT stack. If the result is NULL, the internal 32K stack
195        is used; otherwise the return value must be a valid JIT stack,        is used; otherwise the return value must be a valid JIT stack,
196        the result of calling \fBpcre_jit_stack_alloc()\fP.        the result of calling \fBpcre_jit_stack_alloc()\fP.
197  .sp  .sp
198  You may safely assign the same JIT stack to more than one pattern, as long as  You may safely assign the same JIT stack to more than one pattern, as long as
199  they are all matched sequentially in the same thread. In a multithread  they are all matched sequentially in the same thread. In a multithread
200  application, each thread must use its own JIT stack.  application, each thread must use its own JIT stack.
201  .P  .P
202    Strictly speaking, even more is allowed. You can assign the same stack to any
203    number of patterns as long as they are not used for matching by multiple
204    threads at the same time. For example, you can assign the same stack to all
205    compiled patterns, and use a global mutex in the callback to wait until the
206    stack is available for use. However, this is an inefficient solution, and
207    not recommended.
208    .P
209    This is a suggestion for how a typical multithreaded program might operate:
210    .sp
211      During thread initalization
212        thread_local_var = pcre_jit_stack_alloc(...)
213    .sp
214      During thread exit
215        pcre_jit_stack_free(thread_local_var)
216    .sp
217      Use a one-line callback function
218        return thread_local_var
219    .sp
220  All the functions described in this section do nothing if JIT is not available,  All the functions described in this section do nothing if JIT is not available,
221  and \fBpcre_assign_jit_stack()\fP does nothing unless the \fBextra\fP argument  and \fBpcre_assign_jit_stack()\fP does nothing unless the \fBextra\fP argument
222  is non-NULL and points to a \fBpcre_extra\fP block that is the result of a  is non-NULL and points to a \fBpcre_extra\fP block that is the result of a
223  successful study with PCRE_STUDY_JIT_COMPILE.  successful study with PCRE_STUDY_JIT_COMPILE.
224  .  .
225  .  .
226    .\" HTML <a name="stackfaq"></a>
227    .SH "JIT STACK FAQ"
228    .rs
229    .sp
230    (1) Why do we need JIT stacks?
231    .sp
232    PCRE (and JIT) is a recursive, depth-first engine, so it needs a stack where
233    the local data of the current node is pushed before checking its child nodes.
234    Allocating real machine stack on some platforms is difficult. For example, the
235    stack chain needs to be updated every time if we extend the stack on PowerPC.
236    Although it is possible, its updating time overhead decreases performance. So
237    we do the recursion in memory.
238    .P
239    (2) Why don't we simply allocate blocks of memory with \fBmalloc()\fP?
240    .sp
241    Modern operating systems have a nice feature: they can reserve an address space
242    instead of allocating memory. We can safely allocate memory pages inside this
243    address space, so the stack could grow without moving memory data (this is
244    important because of pointers). Thus we can allocate 1M address space, and use
245    only a single memory page (usually 4K) if that is enough. However, we can still
246    grow up to 1M anytime if needed.
247    .P
248    (3) Who "owns" a JIT stack?
249    .sp
250    The owner of the stack is the user program, not the JIT studied pattern or
251    anything else. The user program must ensure that if a stack is used by
252    \fBpcre_exec()\fP, (that is, it is assigned to the pattern currently running),
253    that stack must not be used by any other threads (to avoid overwriting the same
254    memory area). The best practice for multithreaded programs is to allocate a
255    stack for each thread, and return this stack through the JIT callback function.
256    .P
257    (4) When should a JIT stack be freed?
258    .sp
259    You can free a JIT stack at any time, as long as it will not be used by
260    \fBpcre_exec()\fP again. When you assign the stack to a pattern, only a pointer
261    is set. There is no reference counting or any other magic. You can free the
262    patterns and stacks in any order, anytime. Just \fIdo not\fP call
263    \fBpcre_exec()\fP with a pattern pointing to an already freed stack, as that
264    will cause SEGFAULT. (Also, do not free a stack currently used by
265    \fBpcre_exec()\fP in another thread). You can also replace the stack for a
266    pattern at any time. You can even free the previous stack before assigning a
267    replacement.
268    .P
269    (5) Should I allocate/free a stack every time before/after calling
270    \fBpcre_exec()\fP?
271    .sp
272    No, because this is too costly in terms of resources. However, you could
273    implement some clever idea which release the stack if it is not used in let's
274    say two minutes. The JIT callback can help to achive this without keeping a
275    list of the currently JIT studied patterns.
276    .P
277    (6) OK, the stack is for long term memory allocation. But what happens if a
278    pattern causes stack overflow with a stack of 1M? Is that 1M kept until the
279    stack is freed?
280    .sp
281    Especially on embedded sytems, it might be a good idea to release
282    memory sometimes without freeing the stack. There is no API for this at the
283    moment. Probably a function call which returns with the currently allocated
284    memory for any stack and another which allows releasing memory (shrinking the
285    stack) would be a good idea if someone needs this.
286    .P
287    (7) This is too much of a headache. Isn't there any better solution for JIT
288    stack handling?
289    .sp
290    No, thanks to Windows. If POSIX threads were used everywhere, we could throw
291    out this complicated API.
292    .
293    .
294  .SH "EXAMPLE CODE"  .SH "EXAMPLE CODE"
295  .rs  .rs
296  .sp  .sp
297  This is a single-threaded example that specifies a JIT stack without using a  This is a single-threaded example that specifies a JIT stack without using a
298  callback.  callback.
299  .sp  .sp
300    int rc;    int rc;
301      int ovector[30];
302    pcre *re;    pcre *re;
303    pcre_extra *extra;    pcre_extra *extra;
304    pcre_jit_stack *jit_stack;    pcre_jit_stack *jit_stack;
305  .sp  .sp
306    re = pcre_compile(pattern, 0, &error, &erroffset, NULL);    re = pcre_compile(pattern, 0, &error, &erroffset, NULL);
307    /* Check for errors */    /* Check for errors */
308    extra = pcre_study(re, PCRE_STUDY_JIT_COMPILE, &error);    extra = pcre_study(re, PCRE_STUDY_JIT_COMPILE, &error);
309    jit_stack = pcre_jit_stack_alloc(1, 512 * 1024);    jit_stack = pcre_jit_stack_alloc(32*1024, 512*1024);
310    /* Check for error (NULL) */    /* Check for error (NULL) */
311    pcre_assign_jit_stack(extra, NULL, jit_stack);    pcre_assign_jit_stack(extra, NULL, jit_stack);
312    rc = pcre_exec(re, extra, subject, length, 0, 0, ovector, ovecsize);    rc = pcre_exec(re, extra, subject, length, 0, 0, ovector, 30);
313    /* Check results */    /* Check results */
314    pcre_free(re);    pcre_free(re);
315    pcre_free_study(extra);    pcre_free_study(extra);
316      pcre_jit_stack_free(jit_stack);
317  .sp  .sp
318  .  .
319  .  .
# Line 219  callback. Line 327  callback.
327  .rs  .rs
328  .sp  .sp
329  .nf  .nf
330  Philip Hazel  Philip Hazel (FAQ by Zoltan Herczeg)
331  University Computing Service  University Computing Service
332  Cambridge CB2 3QH, England.  Cambridge CB2 3QH, England.
333  .fi  .fi
# Line 229  Cambridge CB2 3QH, England. Line 337  Cambridge CB2 3QH, England.
337  .rs  .rs
338  .sp  .sp
339  .nf  .nf
340  Last updated: 28 August 2011  Last updated: 22 November 2011
341  Copyright (c) 1997-2011 University of Cambridge.  Copyright (c) 1997-2011 University of Cambridge.
342  .fi  .fi

Legend:
Removed from v.678  
changed lines
  Added in v.761

  ViewVC Help
Powered by ViewVC 1.1.5