ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.pod
Revision: 1.35
Committed: Fri Jun 17 21:38:31 2011 UTC (13 years ago) by root
Branch: MAIN
Changes since 1.34: +22 -1 lines
Log Message:
*** empty log message ***

File Contents

# Content
1 =head1 LIBECB - e-C-Builtins
2
3 =head2 ABOUT LIBECB
4
5 Libecb is currently a simple header file that doesn't require any
6 configuration to use or include in your project.
7
8 It's part of the e-suite of libraries, other members of which include
9 libev and libeio.
10
11 Its homepage can be found here:
12
13 http://software.schmorp.de/pkg/libecb
14
15 It mainly provides a number of wrappers around GCC built-ins, together
16 with replacement functions for other compilers. In addition to this,
17 it provides a number of other lowlevel C utilities, such as endianness
18 detection, byte swapping or bit rotations.
19
20 Or in other words, things that should be built into any standard C system,
21 but aren't, implemented as efficient as possible with GCC, and still
22 correct with other compilers.
23
24 More might come.
25
26 =head2 ABOUT THE HEADER
27
28 At the moment, all you have to do is copy F<ecb.h> somewhere where your
29 compiler can find it and include it:
30
31 #include <ecb.h>
32
33 The header should work fine for both C and C++ compilation, and gives you
34 all of F<inttypes.h> in addition to the ECB symbols.
35
36 There are currently no object files to link to - future versions might
37 come with an (optional) object code library to link against, to reduce
38 code size or gain access to additional features.
39
40 It also currently includes everything from F<inttypes.h>.
41
42 =head2 ABOUT THIS MANUAL / CONVENTIONS
43
44 This manual mainly describes each (public) function available after
45 including the F<ecb.h> header. The header might define other symbols than
46 these, but these are not part of the public API, and not supported in any
47 way.
48
49 When the manual mentions a "function" then this could be defined either as
50 as inline function, a macro, or an external symbol.
51
52 When functions use a concrete standard type, such as C<int> or
53 C<uint32_t>, then the corresponding function works only with that type. If
54 only a generic name is used (C<expr>, C<cond>, C<value> and so on), then
55 the corresponding function relies on C to implement the correct types, and
56 is usually implemented as a macro. Specifically, a "bool" in this manual
57 refers to any kind of boolean value, not a specific type.
58
59 =head2 GCC ATTRIBUTES
60
61 A major part of libecb deals with GCC attributes. These are additional
62 attributes that you can assign to functions, variables and sometimes even
63 types - much like C<const> or C<volatile> in C.
64
65 While GCC allows declarations to show up in many surprising places,
66 but not in many expected places, the safest way is to put attribute
67 declarations before the whole declaration:
68
69 ecb_const int mysqrt (int a);
70 ecb_unused int i;
71
72 For variables, it is often nicer to put the attribute after the name, and
73 avoid multiple declarations using commas:
74
75 int i ecb_unused;
76
77 =over 4
78
79 =item ecb_attribute ((attrs...))
80
81 A simple wrapper that expands to C<__attribute__((attrs))> on GCC, and to
82 nothing on other compilers, so the effect is that only GCC sees these.
83
84 Example: use the C<deprecated> attribute on a function.
85
86 ecb_attribute((__deprecated__)) void
87 do_not_use_me_anymore (void);
88
89 =item ecb_unused
90
91 Marks a function or a variable as "unused", which simply suppresses a
92 warning by GCC when it detects it as unused. This is useful when you e.g.
93 declare a variable but do not always use it:
94
95 {
96 int var ecb_unused;
97
98 #ifdef SOMECONDITION
99 var = ...;
100 return var;
101 #else
102 return 0;
103 #endif
104 }
105
106 =item ecb_inline
107
108 This is not actually an attribute, but you use it like one. It expands
109 either to C<static inline> or to just C<static>, if inline isn't
110 supported. It should be used to declare functions that should be inlined,
111 for code size or speed reasons.
112
113 Example: inline this function, it surely will reduce codesize.
114
115 ecb_inline int
116 negmul (int a, int b)
117 {
118 return - (a * b);
119 }
120
121 =item ecb_noinline
122
123 Prevent a function from being inlined - it might be optimised away, but
124 not inlined into other functions. This is useful if you know your function
125 is rarely called and large enough for inlining not to be helpful.
126
127 =item ecb_noreturn
128
129 Marks a function as "not returning, ever". Some typical functions that
130 don't return are C<exit> or C<abort> (which really works hard to not
131 return), and now you can make your own:
132
133 ecb_noreturn void
134 my_abort (const char *errline)
135 {
136 puts (errline);
137 abort ();
138 }
139
140 In this case, the compiler would probably be smart enough to deduce it on
141 its own, so this is mainly useful for declarations.
142
143 =item ecb_const
144
145 Declares that the function only depends on the values of its arguments,
146 much like a mathematical function. It specifically does not read or write
147 any memory any arguments might point to, global variables, or call any
148 non-const functions. It also must not have any side effects.
149
150 Such a function can be optimised much more aggressively by the compiler -
151 for example, multiple calls with the same arguments can be optimised into
152 a single call, which wouldn't be possible if the compiler would have to
153 expect any side effects.
154
155 It is best suited for functions in the sense of mathematical functions,
156 such as a function returning the square root of its input argument.
157
158 Not suited would be a function that calculates the hash of some memory
159 area you pass in, prints some messages or looks at a global variable to
160 decide on rounding.
161
162 See C<ecb_pure> for a slightly less restrictive class of functions.
163
164 =item ecb_pure
165
166 Similar to C<ecb_const>, declares a function that has no side
167 effects. Unlike C<ecb_const>, the function is allowed to examine global
168 variables and any other memory areas (such as the ones passed to it via
169 pointers).
170
171 While these functions cannot be optimised as aggressively as C<ecb_const>
172 functions, they can still be optimised away in many occasions, and the
173 compiler has more freedom in moving calls to them around.
174
175 Typical examples for such functions would be C<strlen> or C<memcmp>. A
176 function that calculates the MD5 sum of some input and updates some MD5
177 state passed as argument would I<NOT> be pure, however, as it would modify
178 some memory area that is not the return value.
179
180 =item ecb_hot
181
182 This declares a function as "hot" with regards to the cache - the function
183 is used so often, that it is very beneficial to keep it in the cache if
184 possible.
185
186 The compiler reacts by trying to place hot functions near to each other in
187 memory.
188
189 Whether a function is hot or not often depends on the whole program,
190 and less on the function itself. C<ecb_cold> is likely more useful in
191 practise.
192
193 =item ecb_cold
194
195 The opposite of C<ecb_hot> - declares a function as "cold" with regards to
196 the cache, or in other words, this function is not called often, or not at
197 speed-critical times, and keeping it in the cache might be a waste of said
198 cache.
199
200 In addition to placing cold functions together (or at least away from hot
201 functions), this knowledge can be used in other ways, for example, the
202 function will be optimised for size, as opposed to speed, and codepaths
203 leading to calls to those functions can automatically be marked as if
204 C<ecb_expect_false> had been used to reach them.
205
206 Good examples for such functions would be error reporting functions, or
207 functions only called in exceptional or rare cases.
208
209 =item ecb_artificial
210
211 Declares the function as "artificial", in this case meaning that this
212 function is not really mean to be a function, but more like an accessor
213 - many methods in C++ classes are mere accessor functions, and having a
214 crash reported in such a method, or single-stepping through them, is not
215 usually so helpful, especially when it's inlined to just a few instructions.
216
217 Marking them as artificial will instruct the debugger about just this,
218 leading to happier debugging and thus happier lives.
219
220 Example: in some kind of smart-pointer class, mark the pointer accessor as
221 artificial, so that the whole class acts more like a pointer and less like
222 some C++ abstraction monster.
223
224 template<typename T>
225 struct my_smart_ptr
226 {
227 T *value;
228
229 ecb_artificial
230 operator T *()
231 {
232 return value;
233 }
234 };
235
236 =back
237
238 =head2 OPTIMISATION HINTS
239
240 =over 4
241
242 =item bool ecb_is_constant(expr)
243
244 Returns true iff the expression can be deduced to be a compile-time
245 constant, and false otherwise.
246
247 For example, when you have a C<rndm16> function that returns a 16 bit
248 random number, and you have a function that maps this to a range from
249 0..n-1, then you could use this inline function in a header file:
250
251 ecb_inline uint32_t
252 rndm (uint32_t n)
253 {
254 return (n * (uint32_t)rndm16 ()) >> 16;
255 }
256
257 However, for powers of two, you could use a normal mask, but that is only
258 worth it if, at compile time, you can detect this case. This is the case
259 when the passed number is a constant and also a power of two (C<n & (n -
260 1) == 0>):
261
262 ecb_inline uint32_t
263 rndm (uint32_t n)
264 {
265 return is_constant (n) && !(n & (n - 1))
266 ? rndm16 () & (num - 1)
267 : (n * (uint32_t)rndm16 ()) >> 16;
268 }
269
270 =item bool ecb_expect (expr, value)
271
272 Evaluates C<expr> and returns it. In addition, it tells the compiler that
273 the C<expr> evaluates to C<value> a lot, which can be used for static
274 branch optimisations.
275
276 Usually, you want to use the more intuitive C<ecb_expect_true> and
277 C<ecb_expect_false> functions instead.
278
279 =item bool ecb_expect_true (cond)
280
281 =item bool ecb_expect_false (cond)
282
283 These two functions expect a expression that is true or false and return
284 C<1> or C<0>, respectively, so when used in the condition of an C<if> or
285 other conditional statement, it will not change the program:
286
287 /* these two do the same thing */
288 if (some_condition) ...;
289 if (ecb_expect_true (some_condition)) ...;
290
291 However, by using C<ecb_expect_true>, you tell the compiler that the
292 condition is likely to be true (and for C<ecb_expect_false>, that it is
293 unlikely to be true).
294
295 For example, when you check for a null pointer and expect this to be a
296 rare, exceptional, case, then use C<ecb_expect_false>:
297
298 void my_free (void *ptr)
299 {
300 if (ecb_expect_false (ptr == 0))
301 return;
302 }
303
304 Consequent use of these functions to mark away exceptional cases or to
305 tell the compiler what the hot path through a function is can increase
306 performance considerably.
307
308 You might know these functions under the name C<likely> and C<unlikely>
309 - while these are common aliases, we find that the expect name is easier
310 to understand when quickly skimming code. If you wish, you can use
311 C<ecb_likely> instead of C<ecb_expect_true> and C<ecb_unlikely> instead of
312 C<ecb_expect_false> - these are simply aliases.
313
314 A very good example is in a function that reserves more space for some
315 memory block (for example, inside an implementation of a string stream) -
316 each time something is added, you have to check for a buffer overrun, but
317 you expect that most checks will turn out to be false:
318
319 /* make sure we have "size" extra room in our buffer */
320 ecb_inline void
321 reserve (int size)
322 {
323 if (ecb_expect_false (current + size > end))
324 real_reserve_method (size); /* presumably noinline */
325 }
326
327 =item bool ecb_assume (cond)
328
329 Try to tell the compiler that some condition is true, even if it's not
330 obvious.
331
332 This can be used to teach the compiler about invariants or other
333 conditions that might improve code generation, but which are impossible to
334 deduce form the code itself.
335
336 For example, the example reservation function from the C<ecb_expect_false>
337 description could be written thus (only C<ecb_assume> was added):
338
339 ecb_inline void
340 reserve (int size)
341 {
342 if (ecb_expect_false (current + size > end))
343 real_reserve_method (size); /* presumably noinline */
344
345 ecb_assume (current + size <= end);
346 }
347
348 If you then call this function twice, like this:
349
350 reserve (10);
351 reserve (1);
352
353 Then the compiler I<might> be able to optimise out the second call
354 completely, as it knows that C<< current + 1 > end >> is false and the
355 call will never be executed.
356
357 =item bool ecb_unreachable ()
358
359 This function does nothing itself, except tell the compiler that it will
360 never be executed. Apart from suppressing a warning in some cases, this
361 function can be used to implement C<ecb_assume> or similar functions.
362
363 =item bool ecb_prefetch (addr, rw, locality)
364
365 Tells the compiler to try to prefetch memory at the given C<addr>ess
366 for either reading (C<rw> = 0) or writing (C<rw> = 1). A C<locality> of
367 C<0> means that there will only be one access later, C<3> means that
368 the data will likely be accessed very often, and values in between mean
369 something... in between. The memory pointed to by the address does not
370 need to be accessible (it could be a null pointer for example), but C<rw>
371 and C<locality> must be compile-time constants.
372
373 An obvious way to use this is to prefetch some data far away, in a big
374 array you loop over. This prefetches memory some 128 array elements later,
375 in the hope that it will be ready when the CPU arrives at that location.
376
377 int sum = 0;
378
379 for (i = 0; i < N; ++i)
380 {
381 sum += arr [i]
382 ecb_prefetch (arr + i + 128, 0, 0);
383 }
384
385 It's hard to predict how far to prefetch, and most CPUs that can prefetch
386 are often good enough to predict this kind of behaviour themselves. It
387 gets more interesting with linked lists, especially when you do some fair
388 processing on each list element:
389
390 for (node *n = start; n; n = n->next)
391 {
392 ecb_prefetch (n->next, 0, 0);
393 ... do medium amount of work with *n
394 }
395
396 After processing the node, (part of) the next node might already be in
397 cache.
398
399 =back
400
401 =head2 BIT FIDDLING / BITSTUFFS
402
403 =over 4
404
405 =item bool ecb_big_endian ()
406
407 =item bool ecb_little_endian ()
408
409 These two functions return true if the byte order is big endian
410 (most-significant byte first) or little endian (least-significant byte
411 first) respectively.
412
413 On systems that are neither, their return values are unspecified.
414
415 =item int ecb_ctz32 (uint32_t x)
416
417 =item int ecb_ctz64 (uint64_t x)
418
419 Returns the index of the least significant bit set in C<x> (or
420 equivalently the number of bits set to 0 before the least significant bit
421 set), starting from 0. If C<x> is 0 the result is undefined.
422
423 For example:
424
425 ecb_ctz32 (3) = 0
426 ecb_ctz32 (6) = 1
427
428 =item int ecb_ld32 (uint32_t x)
429
430 =item int ecb_ld64 (uint64_t x)
431
432 Returns the index of the most significant bit set in C<x>, or the number
433 of digits the number requires in binary (so that C<< 2**ld <= x <
434 2**(ld+1) >>). If C<x> is 0 the result is undefined. A common use case is
435 to compute the integer binary logarithm, i.e. C<floor (log2 (n))>, for
436 example to see how many bits a certain number requires to be encoded.
437
438 This function is similar to the "count leading zero bits" function, except
439 that that one returns how many zero bits are "in front" of the number (in
440 the given data type), while C<ecb_ld> returns how many bits the number
441 itself requires.
442
443 =item int ecb_popcount32 (uint32_t x)
444
445 =item int ecb_popcount64 (uint64_t x)
446
447 Returns the number of bits set to 1 in C<x>. For example:
448
449 ecb_popcount32 (7) = 3
450 ecb_popcount32 (255) = 8
451
452 =item uint32_t ecb_bswap16 (uint32_t x)
453
454 =item uint32_t ecb_bswap32 (uint32_t x)
455
456 =item uint64_t ecb_bswap64 (uint64_t x)
457
458 These functions return the value of the 16-bit (32-bit, 64-bit) value
459 C<x> after reversing the order of bytes (0x11223344 becomes 0x44332211 in
460 C<ecb_bswap32>).
461
462 =item uint8_t ecb_rotl8 (uint8_t x, unsigned int count)
463
464 =item uint16_t ecb_rotl16 (uint16_t x, unsigned int count)
465
466 =item uint32_t ecb_rotl32 (uint32_t x, unsigned int count)
467
468 =item uint64_t ecb_rotl64 (uint64_t x, unsigned int count)
469
470 =item uint8_t ecb_rotr8 (uint8_t x, unsigned int count)
471
472 =item uint16_t ecb_rotr16 (uint16_t x, unsigned int count)
473
474 =item uint32_t ecb_rotr32 (uint32_t x, unsigned int count)
475
476 =item uint64_t ecb_rotr64 (uint64_t x, unsigned int count)
477
478 These two families of functions return the value of C<x> after rotating
479 all the bits by C<count> positions to the right (C<ecb_rotr>) or left
480 (C<ecb_rotl>).
481
482 Current GCC versions understand these functions and usually compile them
483 to "optimal" code (e.g. a single C<rol> or a combination of C<shld> on
484 x86).
485
486 =back
487
488 =head2 ARITHMETIC
489
490 =over 4
491
492 =item x = ecb_mod (m, n)
493
494 Returns C<m> modulo C<n>, which is the same as the positive remainder
495 of the division operation between C<m> and C<n>, using floored
496 division. Unlike the C remainder operator C<%>, this function ensures that
497 the return value is always positive and that the two numbers I<m> and
498 I<m' = m + i * n> result in the same value modulo I<n> - in other words,
499 C<ecb_mod> implements the mathematical modulo operation, which is missing
500 in the language.
501
502 C<n> must be strictly positive (i.e. C<< >= 1 >>), while C<m> must be
503 negatable, that is, both C<m> and C<-m> must be representable in its
504 type (this typically excludes the minimum signed integer value, the same
505 limitation as for C</> and C<%> in C).
506
507 Current GCC versions compile this into an efficient branchless sequence on
508 almost all CPUs.
509
510 For example, when you want to rotate forward through the members of an
511 array for increasing C<m> (which might be negative), then you should use
512 C<ecb_mod>, as the C<%> operator might give either negative results, or
513 change direction for negative values:
514
515 for (m = -100; m <= 100; ++m)
516 int elem = myarray [ecb_mod (m, ecb_array_length (myarray))];
517
518 =back
519
520 =head2 UTILITY
521
522 =over 4
523
524 =item element_count = ecb_array_length (name)
525
526 Returns the number of elements in the array C<name>. For example:
527
528 int primes[] = { 2, 3, 5, 7, 11 };
529 int sum = 0;
530
531 for (i = 0; i < ecb_array_length (primes); i++)
532 sum += primes [i];
533
534 =back
535
536