ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.pod
Revision: 1.104
Committed: Fri Mar 25 08:44:14 2022 UTC (2 years, 3 months ago) by root
Branch: MAIN
Changes since 1.103: +109 -17 lines
Log Message:
add hilbert2d

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 many compiler built-ins,
16 together with replacement functions for other compilers. In addition
17 to this, it provides a number of other low-level C utilities, such as
18 endianness detection, byte swapping or bit rotations.
19
20 Or in other words, things that should be built into any standard C
21 system, but aren't, implemented as efficient as possible with GCC (clang,
22 MSVC...), and still 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 TYPES / TYPE SUPPORT
60
61 F<ecb.h> makes sure that the following types are defined (in the expected way):
62
63 int8_t uint8_
64 int16_t uint16_t
65 int32_t uint32_
66 int64_t uint64_t
67 int_fast8_t uint_fast8_t
68 int_fast16_t uint_fast16_t
69 int_fast32_t uint_fast32_t
70 int_fast64_t uint_fast64_t
71 intptr_t uintptr_t
72
73 The macro C<ECB_PTRSIZE> is defined to the size of a pointer on this
74 platform (currently C<4> or C<8>) and can be used in preprocessor
75 expressions.
76
77 For C<ptrdiff_t> and C<size_t> use C<stddef.h>/C<cstddef>.
78
79 =head2 LANGUAGE/ENVIRONMENT/COMPILER VERSIONS
80
81 All the following symbols expand to an expression that can be tested in
82 preprocessor instructions as well as treated as a boolean (use C<!!> to
83 ensure it's either C<0> or C<1> if you need that).
84
85 =over
86
87 =item ECB_C
88
89 True if the implementation defines the C<__STDC__> macro to a true value,
90 while not claiming to be C++, i..e C, but not C++.
91
92 =item ECB_C99
93
94 True if the implementation claims to be compliant to C99 (ISO/IEC
95 9899:1999) or any later version, while not claiming to be C++.
96
97 Note that later versions (ECB_C11) remove core features again (for
98 example, variable length arrays).
99
100 =item ECB_C11, ECB_C17
101
102 True if the implementation claims to be compliant to C11/C17 (ISO/IEC
103 9899:2011, :20187) or any later version, while not claiming to be C++.
104
105 =item ECB_CPP
106
107 True if the implementation defines the C<__cplusplus__> macro to a true
108 value, which is typically true for C++ compilers.
109
110 =item ECB_CPP11, ECB_CPP14, ECB_CPP17
111
112 True if the implementation claims to be compliant to C++11/C++14/C++17
113 (ISO/IEC 14882:2011, :2014, :2017) or any later version.
114
115 Note that many C++20 features will likely have their own feature test
116 macros (see e.g. L<http://eel.is/c++draft/cpp.predefined#1.8>).
117
118 =item ECB_OPTIMIZE_SIZE
119
120 Is C<1> when the compiler optimizes for size, C<0> otherwise. This symbol
121 can also be defined before including F<ecb.h>, in which case it will be
122 unchanged.
123
124 =item ECB_GCC_VERSION (major, minor)
125
126 Expands to a true value (suitable for testing by the preprocessor) if the
127 compiler used is GNU C and the version is the given version, or higher.
128
129 This macro tries to return false on compilers that claim to be GCC
130 compatible but aren't.
131
132 =item ECB_EXTERN_C
133
134 Expands to C<extern "C"> in C++, and a simple C<extern> in C.
135
136 This can be used to declare a single external C function:
137
138 ECB_EXTERN_C int printf (const char *format, ...);
139
140 =item ECB_EXTERN_C_BEG / ECB_EXTERN_C_END
141
142 These two macros can be used to wrap multiple C<extern "C"> definitions -
143 they expand to nothing in C.
144
145 They are most useful in header files:
146
147 ECB_EXTERN_C_BEG
148
149 int mycfun1 (int x);
150 int mycfun2 (int x);
151
152 ECB_EXTERN_C_END
153
154 =item ECB_STDFP
155
156 If this evaluates to a true value (suitable for testing by the
157 preprocessor), then C<float> and C<double> use IEEE 754 single/binary32
158 and double/binary64 representations internally I<and> the endianness of
159 both types match the endianness of C<uint32_t> and C<uint64_t>.
160
161 This means you can just copy the bits of a C<float> (or C<double>) to an
162 C<uint32_t> (or C<uint64_t>) and get the raw IEEE 754 bit representation
163 without having to think about format or endianness.
164
165 This is true for basically all modern platforms, although F<ecb.h> might
166 not be able to deduce this correctly everywhere and might err on the safe
167 side.
168
169 =item ECB_64BIT_NATIVE
170
171 Evaluates to a true value (suitable for both preprocessor and C code
172 testing) if 64 bit integer types on this architecture are evaluated
173 "natively", that is, with similar speeds as 32 bit integers. While 64 bit
174 integer support is very common (and in fact required by libecb), 32 bit
175 CPUs have to emulate operations on them, so you might want to avoid them.
176
177 =item ECB_AMD64, ECB_AMD64_X32
178
179 These two macros are defined to C<1> on the x86_64/amd64 ABI and the X32
180 ABI, respectively, and undefined elsewhere.
181
182 The designers of the new X32 ABI for some inexplicable reason decided to
183 make it look exactly like amd64, even though it's completely incompatible
184 to that ABI, breaking about every piece of software that assumed that
185 C<__x86_64> stands for, well, the x86-64 ABI, making these macros
186 necessary.
187
188 =back
189
190 =head2 MACRO TRICKERY
191
192 =over
193
194 =item ECB_CONCAT (a, b)
195
196 Expands any macros in C<a> and C<b>, then concatenates the result to form
197 a single token. This is mainly useful to form identifiers from components,
198 e.g.:
199
200 #define S1 str
201 #define S2 cpy
202
203 ECB_CONCAT (S1, S2)(dst, src); // == strcpy (dst, src);
204
205 =item ECB_STRINGIFY (arg)
206
207 Expands any macros in C<arg> and returns the stringified version of
208 it. This is mainly useful to get the contents of a macro in string form,
209 e.g.:
210
211 #define SQL_LIMIT 100
212 sql_exec ("select * from table limit " ECB_STRINGIFY (SQL_LIMIT));
213
214 =item ECB_STRINGIFY_EXPR (expr)
215
216 Like C<ECB_STRINGIFY>, but additionally evaluates C<expr> to make sure it
217 is a valid expression. This is useful to catch typos or cases where the
218 macro isn't available:
219
220 #include <errno.h>
221
222 ECB_STRINGIFY (EDOM); // "33" (on my system at least)
223 ECB_STRINGIFY_EXPR (EDOM); // "33"
224
225 // now imagine we had a typo:
226
227 ECB_STRINGIFY (EDAM); // "EDAM"
228 ECB_STRINGIFY_EXPR (EDAM); // error: EDAM undefined
229
230 =back
231
232 =head2 ATTRIBUTES
233
234 A major part of libecb deals with additional attributes that can be
235 assigned to functions, variables and sometimes even types - much like
236 C<const> or C<volatile> in C. They are implemented using either GCC
237 attributes or other compiler/language specific features. Attributes
238 declarations must be put before the whole declaration:
239
240 ecb_const int mysqrt (int a);
241 ecb_unused int i;
242
243 =over
244
245 =item ecb_unused
246
247 Marks a function or a variable as "unused", which simply suppresses a
248 warning by the compiler when it detects it as unused. This is useful when
249 you e.g. declare a variable but do not always use it:
250
251 {
252 ecb_unused int var;
253
254 #ifdef SOMECONDITION
255 var = ...;
256 return var;
257 #else
258 return 0;
259 #endif
260 }
261
262 =item ecb_deprecated
263
264 Similar to C<ecb_unused>, but marks a function, variable or type as
265 deprecated. This makes some compilers warn when the type is used.
266
267 =item ecb_deprecated_message (message)
268
269 Same as C<ecb_deprecated>, but if possible, the specified diagnostic is
270 used instead of a generic depreciation message when the object is being
271 used.
272
273 =item ecb_inline
274
275 Expands either to (a compiler-specific equivalent of) C<static inline> or
276 to just C<static>, if inline isn't supported. It should be used to declare
277 functions that should be inlined, for code size or speed reasons.
278
279 Example: inline this function, it surely will reduce code size.
280
281 ecb_inline int
282 negmul (int a, int b)
283 {
284 return - (a * b);
285 }
286
287 =item ecb_noinline
288
289 Prevents a function from being inlined - it might be optimised away, but
290 not inlined into other functions. This is useful if you know your function
291 is rarely called and large enough for inlining not to be helpful.
292
293 =item ecb_noreturn
294
295 Marks a function as "not returning, ever". Some typical functions that
296 don't return are C<exit> or C<abort> (which really works hard to not
297 return), and now you can make your own:
298
299 ecb_noreturn void
300 my_abort (const char *errline)
301 {
302 puts (errline);
303 abort ();
304 }
305
306 In this case, the compiler would probably be smart enough to deduce it on
307 its own, so this is mainly useful for declarations.
308
309 =item ecb_restrict
310
311 Expands to the C<restrict> keyword or equivalent on compilers that support
312 them, and to nothing on others. Must be specified on a pointer type or
313 an array index to indicate that the memory doesn't alias with any other
314 restricted pointer in the same scope.
315
316 Example: multiply a vector, and allow the compiler to parallelise the
317 loop, because it knows it doesn't overwrite input values.
318
319 void
320 multiply (ecb_restrict float *src,
321 ecb_restrict float *dst,
322 int len, float factor)
323 {
324 int i;
325
326 for (i = 0; i < len; ++i)
327 dst [i] = src [i] * factor;
328 }
329
330 =item ecb_const
331
332 Declares that the function only depends on the values of its arguments,
333 much like a mathematical function. It specifically does not read or write
334 any memory any arguments might point to, global variables, or call any
335 non-const functions. It also must not have any side effects.
336
337 Such a function can be optimised much more aggressively by the compiler -
338 for example, multiple calls with the same arguments can be optimised into
339 a single call, which wouldn't be possible if the compiler would have to
340 expect any side effects.
341
342 It is best suited for functions in the sense of mathematical functions,
343 such as a function returning the square root of its input argument.
344
345 Not suited would be a function that calculates the hash of some memory
346 area you pass in, prints some messages or looks at a global variable to
347 decide on rounding.
348
349 See C<ecb_pure> for a slightly less restrictive class of functions.
350
351 =item ecb_pure
352
353 Similar to C<ecb_const>, declares a function that has no side
354 effects. Unlike C<ecb_const>, the function is allowed to examine global
355 variables and any other memory areas (such as the ones passed to it via
356 pointers).
357
358 While these functions cannot be optimised as aggressively as C<ecb_const>
359 functions, they can still be optimised away in many occasions, and the
360 compiler has more freedom in moving calls to them around.
361
362 Typical examples for such functions would be C<strlen> or C<memcmp>. A
363 function that calculates the MD5 sum of some input and updates some MD5
364 state passed as argument would I<NOT> be pure, however, as it would modify
365 some memory area that is not the return value.
366
367 =item ecb_hot
368
369 This declares a function as "hot" with regards to the cache - the function
370 is used so often, that it is very beneficial to keep it in the cache if
371 possible.
372
373 The compiler reacts by trying to place hot functions near to each other in
374 memory.
375
376 Whether a function is hot or not often depends on the whole program,
377 and less on the function itself. C<ecb_cold> is likely more useful in
378 practise.
379
380 =item ecb_cold
381
382 The opposite of C<ecb_hot> - declares a function as "cold" with regards to
383 the cache, or in other words, this function is not called often, or not at
384 speed-critical times, and keeping it in the cache might be a waste of said
385 cache.
386
387 In addition to placing cold functions together (or at least away from hot
388 functions), this knowledge can be used in other ways, for example, the
389 function will be optimised for size, as opposed to speed, and code paths
390 leading to calls to those functions can automatically be marked as if
391 C<ecb_expect_false> had been used to reach them.
392
393 Good examples for such functions would be error reporting functions, or
394 functions only called in exceptional or rare cases.
395
396 =item ecb_artificial
397
398 Declares the function as "artificial", in this case meaning that this
399 function is not really meant to be a function, but more like an accessor
400 - many methods in C++ classes are mere accessor functions, and having a
401 crash reported in such a method, or single-stepping through them, is not
402 usually so helpful, especially when it's inlined to just a few instructions.
403
404 Marking them as artificial will instruct the debugger about just this,
405 leading to happier debugging and thus happier lives.
406
407 Example: in some kind of smart-pointer class, mark the pointer accessor as
408 artificial, so that the whole class acts more like a pointer and less like
409 some C++ abstraction monster.
410
411 template<typename T>
412 struct my_smart_ptr
413 {
414 T *value;
415
416 ecb_artificial
417 operator T *()
418 {
419 return value;
420 }
421 };
422
423 =back
424
425 =head2 OPTIMISATION HINTS
426
427 =over
428
429 =item bool ecb_is_constant (expr)
430
431 Returns true iff the expression can be deduced to be a compile-time
432 constant, and false otherwise.
433
434 For example, when you have a C<rndm16> function that returns a 16 bit
435 random number, and you have a function that maps this to a range from
436 0..n-1, then you could use this inline function in a header file:
437
438 ecb_inline uint32_t
439 rndm (uint32_t n)
440 {
441 return (n * (uint32_t)rndm16 ()) >> 16;
442 }
443
444 However, for powers of two, you could use a normal mask, but that is only
445 worth it if, at compile time, you can detect this case. This is the case
446 when the passed number is a constant and also a power of two (C<n & (n -
447 1) == 0>):
448
449 ecb_inline uint32_t
450 rndm (uint32_t n)
451 {
452 return is_constant (n) && !(n & (n - 1))
453 ? rndm16 () & (num - 1)
454 : (n * (uint32_t)rndm16 ()) >> 16;
455 }
456
457 =item ecb_expect (expr, value)
458
459 Evaluates C<expr> and returns it. In addition, it tells the compiler that
460 the C<expr> evaluates to C<value> a lot, which can be used for static
461 branch optimisations.
462
463 Usually, you want to use the more intuitive C<ecb_expect_true> and
464 C<ecb_expect_false> functions instead.
465
466 =item bool ecb_expect_true (cond)
467
468 =item bool ecb_expect_false (cond)
469
470 These two functions expect a expression that is true or false and return
471 C<1> or C<0>, respectively, so when used in the condition of an C<if> or
472 other conditional statement, it will not change the program:
473
474 /* these two do the same thing */
475 if (some_condition) ...;
476 if (ecb_expect_true (some_condition)) ...;
477
478 However, by using C<ecb_expect_true>, you tell the compiler that the
479 condition is likely to be true (and for C<ecb_expect_false>, that it is
480 unlikely to be true).
481
482 For example, when you check for a null pointer and expect this to be a
483 rare, exceptional, case, then use C<ecb_expect_false>:
484
485 void my_free (void *ptr)
486 {
487 if (ecb_expect_false (ptr == 0))
488 return;
489 }
490
491 Consequent use of these functions to mark away exceptional cases or to
492 tell the compiler what the hot path through a function is can increase
493 performance considerably.
494
495 You might know these functions under the name C<likely> and C<unlikely>
496 - while these are common aliases, we find that the expect name is easier
497 to understand when quickly skimming code. If you wish, you can use
498 C<ecb_likely> instead of C<ecb_expect_true> and C<ecb_unlikely> instead of
499 C<ecb_expect_false> - these are simply aliases.
500
501 A very good example is in a function that reserves more space for some
502 memory block (for example, inside an implementation of a string stream) -
503 each time something is added, you have to check for a buffer overrun, but
504 you expect that most checks will turn out to be false:
505
506 /* make sure we have "size" extra room in our buffer */
507 ecb_inline void
508 reserve (int size)
509 {
510 if (ecb_expect_false (current + size > end))
511 real_reserve_method (size); /* presumably noinline */
512 }
513
514 =item ecb_assume (cond)
515
516 Tries to tell the compiler that some condition is true, even if it's not
517 obvious. This is not a function, but a statement: it cannot be used in
518 another expression.
519
520 This can be used to teach the compiler about invariants or other
521 conditions that might improve code generation, but which are impossible to
522 deduce form the code itself.
523
524 For example, the example reservation function from the C<ecb_expect_false>
525 description could be written thus (only C<ecb_assume> was added):
526
527 ecb_inline void
528 reserve (int size)
529 {
530 if (ecb_expect_false (current + size > end))
531 real_reserve_method (size); /* presumably noinline */
532
533 ecb_assume (current + size <= end);
534 }
535
536 If you then call this function twice, like this:
537
538 reserve (10);
539 reserve (1);
540
541 Then the compiler I<might> be able to optimise out the second call
542 completely, as it knows that C<< current + 1 > end >> is false and the
543 call will never be executed.
544
545 =item ecb_unreachable ()
546
547 This function does nothing itself, except tell the compiler that it will
548 never be executed. Apart from suppressing a warning in some cases, this
549 function can be used to implement C<ecb_assume> or similar functionality.
550
551 =item ecb_prefetch (addr, rw, locality)
552
553 Tells the compiler to try to prefetch memory at the given I<addr>ess
554 for either reading (I<rw> = 0) or writing (I<rw> = 1). A I<locality> of
555 C<0> means that there will only be one access later, C<3> means that
556 the data will likely be accessed very often, and values in between mean
557 something... in between. The memory pointed to by the address does not
558 need to be accessible (it could be a null pointer for example), but C<rw>
559 and C<locality> must be compile-time constants.
560
561 This is a statement, not a function: you cannot use it as part of an
562 expression.
563
564 An obvious way to use this is to prefetch some data far away, in a big
565 array you loop over. This prefetches memory some 128 array elements later,
566 in the hope that it will be ready when the CPU arrives at that location.
567
568 int sum = 0;
569
570 for (i = 0; i < N; ++i)
571 {
572 sum += arr [i]
573 ecb_prefetch (arr + i + 128, 0, 0);
574 }
575
576 It's hard to predict how far to prefetch, and most CPUs that can prefetch
577 are often good enough to predict this kind of behaviour themselves. It
578 gets more interesting with linked lists, especially when you do some fair
579 processing on each list element:
580
581 for (node *n = start; n; n = n->next)
582 {
583 ecb_prefetch (n->next, 0, 0);
584 ... do medium amount of work with *n
585 }
586
587 After processing the node, (part of) the next node might already be in
588 cache.
589
590 =back
591
592 =head2 BIT FIDDLING / BIT WIZARDRY
593
594 =over
595
596 =item bool ecb_big_endian ()
597
598 =item bool ecb_little_endian ()
599
600 These two functions return true if the byte order is big endian
601 (most-significant byte first) or little endian (least-significant byte
602 first) respectively.
603
604 On systems that are neither, their return values are unspecified.
605
606 =item int ecb_ctz32 (uint32_t x)
607
608 =item int ecb_ctz64 (uint64_t x)
609
610 =item int ecb_ctz (T x) [C++]
611
612 Returns the index of the least significant bit set in C<x> (or
613 equivalently the number of bits set to 0 before the least significant bit
614 set), starting from 0. If C<x> is 0 the result is undefined.
615
616 For smaller types than C<uint32_t> you can safely use C<ecb_ctz32>.
617
618 The overloaded C++ C<ecb_ctz> function supports C<uint8_t>, C<uint16_t>,
619 C<uint32_t> and C<uint64_t> types.
620
621 For example:
622
623 ecb_ctz32 (3) = 0
624 ecb_ctz32 (6) = 1
625
626 =item bool ecb_is_pot32 (uint32_t x)
627
628 =item bool ecb_is_pot64 (uint32_t x)
629
630 =item bool ecb_is_pot (T x) [C++]
631
632 Returns true iff C<x> is a power of two or C<x == 0>.
633
634 For smaller types than C<uint32_t> you can safely use C<ecb_is_pot32>.
635
636 The overloaded C++ C<ecb_is_pot> function supports C<uint8_t>, C<uint16_t>,
637 C<uint32_t> and C<uint64_t> types.
638
639 =item int ecb_ld32 (uint32_t x)
640
641 =item int ecb_ld64 (uint64_t x)
642
643 =item int ecb_ld64 (T x) [C++]
644
645 Returns the index of the most significant bit set in C<x>, or the number
646 of digits the number requires in binary (so that C<< 2**ld <= x <
647 2**(ld+1) >>). If C<x> is 0 the result is undefined. A common use case is
648 to compute the integer binary logarithm, i.e. C<floor (log2 (n))>, for
649 example to see how many bits a certain number requires to be encoded.
650
651 This function is similar to the "count leading zero bits" function, except
652 that that one returns how many zero bits are "in front" of the number (in
653 the given data type), while C<ecb_ld> returns how many bits the number
654 itself requires.
655
656 For smaller types than C<uint32_t> you can safely use C<ecb_ld32>.
657
658 The overloaded C++ C<ecb_ld> function supports C<uint8_t>, C<uint16_t>,
659 C<uint32_t> and C<uint64_t> types.
660
661 =item int ecb_popcount32 (uint32_t x)
662
663 =item int ecb_popcount64 (uint64_t x)
664
665 =item int ecb_popcount (T x) [C++]
666
667 Returns the number of bits set to 1 in C<x>.
668
669 For smaller types than C<uint32_t> you can safely use C<ecb_popcount32>.
670
671 The overloaded C++ C<ecb_popcount> function supports C<uint8_t>, C<uint16_t>,
672 C<uint32_t> and C<uint64_t> types.
673
674 For example:
675
676 ecb_popcount32 (7) = 3
677 ecb_popcount32 (255) = 8
678
679 =item uint8_t ecb_bitrev8 (uint8_t x)
680
681 =item uint16_t ecb_bitrev16 (uint16_t x)
682
683 =item uint32_t ecb_bitrev32 (uint32_t x)
684
685 =item T ecb_bitrev (T x) [C++]
686
687 Reverses the bits in x, i.e. the MSB becomes the LSB, MSB-1 becomes LSB+1
688 and so on.
689
690 The overloaded C++ C<ecb_bitrev> function supports C<uint8_t>, C<uint16_t> and C<uint32_t> types.
691
692 Example:
693
694 ecb_bitrev8 (0xa7) = 0xea
695 ecb_bitrev32 (0xffcc4411) = 0x882233ff
696
697 =item T ecb_bitrev (T x) [C++]
698
699 Overloaded C++ bitrev function.
700
701 C<T> must be one of C<uint8_t>, C<uint16_t> or C<uint32_t>.
702
703 =item uint32_t ecb_bswap16 (uint32_t x)
704
705 =item uint32_t ecb_bswap32 (uint32_t x)
706
707 =item uint64_t ecb_bswap64 (uint64_t x)
708
709 =item T ecb_bswap (T x)
710
711 These functions return the value of the 16-bit (32-bit, 64-bit) value
712 C<x> after reversing the order of bytes (0x11223344 becomes 0x44332211 in
713 C<ecb_bswap32>).
714
715 The overloaded C++ C<ecb_bswap> function supports C<uint8_t>, C<uint16_t>,
716 C<uint32_t> and C<uint64_t> types.
717
718 =item uint8_t ecb_rotl8 (uint8_t x, unsigned int count)
719
720 =item uint16_t ecb_rotl16 (uint16_t x, unsigned int count)
721
722 =item uint32_t ecb_rotl32 (uint32_t x, unsigned int count)
723
724 =item uint64_t ecb_rotl64 (uint64_t x, unsigned int count)
725
726 =item uint8_t ecb_rotr8 (uint8_t x, unsigned int count)
727
728 =item uint16_t ecb_rotr16 (uint16_t x, unsigned int count)
729
730 =item uint32_t ecb_rotr32 (uint32_t x, unsigned int count)
731
732 =item uint64_t ecb_rotr64 (uint64_t x, unsigned int count)
733
734 These two families of functions return the value of C<x> after rotating
735 all the bits by C<count> positions to the right (C<ecb_rotr>) or left
736 (C<ecb_rotl>). There are no restrictions on the value C<count>, i.e. both
737 zero and values equal or larger than the word width work correctly. Also,
738 notwithstanding C<count> being unsigned, negative numbers work and shift
739 to the opposite direction.
740
741 Current GCC/clang versions understand these functions and usually compile
742 them to "optimal" code (e.g. a single C<rol> or a combination of C<shld>
743 on x86).
744
745 =item T ecb_rotl (T x, unsigned int count) [C++]
746
747 =item T ecb_rotr (T x, unsigned int count) [C++]
748
749 Overloaded C++ rotl/rotr functions.
750
751 C<T> must be one of C<uint8_t>, C<uint16_t>, C<uint32_t> or C<uint64_t>.
752
753 =item uint_fast8_t ecb_gray8_encode (uint_fast8_t b)
754
755 =item uint_fast16_t ecb_gray16_encode (uint_fast16_t b)
756
757 =item uint_fast32_t ecb_gray32_encode (uint_fast32_t b)
758
759 =item uint_fast64_t ecb_gray64_encode (uint_fast64_t b)
760
761 Encode an unsigned into its corresponding (reflective) gray code - the
762 kind of gray code meant when just talking about "gray code". These
763 functions are very fast and all have identical implementation, so there is
764 no need to use a smaller type, as long as your CPU can handle it natively.
765
766 =item T ecb_gray_encode (T b) [C++]
767
768 Overloaded C++ version of the above, for C<uint{8,16,32,64}_t>.
769
770 =item uint_fast8_t ecb_gray8_decode (uint_fast8_t b)
771
772 =item uint_fast16_t ecb_gray16_decode (uint_fast16_t b)
773
774 =item uint_fast32_t ecb_gray32_decode (uint_fast32_t b)
775
776 =item uint_fast64_t ecb_gray64_decode (uint_fast64_t b)
777
778 Decode a gray code back into linear index form (the reverse of
779 C<ecb_gray*_encode>. Unlike the encode functions, the decode functions
780 have higher time complexity for larger types, so it can pay off to use a
781 smaller type here.
782
783 =item T ecb_gray_decode (T b) [C++]
784
785 Overloaded C++ version of the above, for C<uint{8,16,32,64}_t>.
786
787 =back
788
789 =head2 HILBERT CURVES
790
791 These functions deal with (square, pseudo) Hilbert curves. The parameter
792 I<order> indicates the size of the square and is specified in bits, that
793 means for order C<8>, the coordinates range from C<0>..C<255>, and the
794 curve index ranges from C<0>..C<65535>.
795
796 The 32 bit variants of these functions map a 32 bit index to two 16 bit
797 coordinates, stored in a 32 bit variable, where the high order bits are
798 the x-coordinate, and the low order bits are the y-coordinate, thus,
799 these functions map 32 bit linear index on the curve to a 32 bit packed
800 coordinate pair, and vice versa.
801
802 The 64 bit variants work similarly.
803
804 The I<order> can go from C<1> to C<16> for the 32 bit curve, and C<1> to
805 C<32> for the 64 bit curve.
806
807 When going from one order to the next higher order, these functions
808 replace the curve segments by smaller versions of the generating shape,
809 while doubling the size (since they use integer coordinates), which is
810 what you would expect mathematically. This means that the curve will be
811 mirrored at the diagonal. If your goal is to simply cover more area while
812 retaining existing point coordinates you should increase or decrease the
813 I<order> by C<2> or, in the case of C<ecb_hilbert2d_index_to_coord>,
814 simply specify the maximum I<order> of C<16> or C<32>, respectively, as
815 these are constant-time.
816
817 =over
818
819 =item uint32_t ecb_hilbert2d_index_to_coord32 (int order, uint32_t index)
820
821 =item uint64_t ecb_hilbert2d_index_to_coord64 (int order, uint64_t index)
822
823 Map a point on a pseudo Hilbert curve from its linear distance from the
824 origin on the curve to a x|y coordinate pair. The result is a packed
825 coordinate pair, to get the actual x and < coordinates, you could do
826 something like this:
827
828 uint32_t xy = ecb_hilbert2d_index_to_coord32 (16, 255);
829 uint16_t x = xy >> 16;
830 uint16_t y = xy & 0xffffU;
831
832 uint64_t xy = ecb_hilbert2d_index_to_coord64 (32, 255);
833 uint32_t x = xy >> 32;
834 uint32_t y = xy & 0xffffffffU;
835
836 These functions work in constant time, so for many applications it is
837 preferable to simply hard-code the order to the maximum (C<16> or C<32>).
838
839 This (production-ready, i.e. never run) example generates an SVG image of
840 an order 8 pseudo Hilbert curve:
841
842 printf ("<svg xmlns='http://www.w3.org/2000/svg' width='%d' height='%d'>\n", 64 * 8, 64 * 8);
843 printf ("<g transform='translate(4) scale(8)' stroke-width='0.25' stroke='black'>\n");
844 for (uint32_t i = 0; i < 64*64 - 1; ++i)
845 {
846 uint32_t p1 = ecb_hilbert2d_index_to_coord32 (6, i );
847 uint32_t p2 = ecb_hilbert2d_index_to_coord32 (6, i + 1);
848 printf ("<line x1='%d' y1='%d' x2='%d' y2='%d'/>\n",
849 p1 >> 16, p1 & 0xffff,
850 p2 >> 16, p2 & 0xffff);
851 }
852 printf ("</g>\n");
853 printf ("</svg>\n");
854
855 =item uint32_t ecb_hilbert2d_coord_to_index32 (int order, uint32_t xy)
856
857 =item uint64_t ecb_hilbert2d_coord_to_index64 (int order, uint64_t xy)
858
859 The reverse of C<ecb_hilbert2d_index_to_coord> - map a packed pair of
860 coordinates to their linear index on the pseudo Hilbert curve of order
861 I<order>.
862
863 They are an exact inverse of the C<ecb_hilbert2d_coord_to_index> functions
864 for the same I<order>:
865
866 assert (
867 u == ecb_hilbert2d_coord_to_index (32,
868 ecb_hilbert2d_index_to_coord32 (32,
869 u)));
870
871 Packing coordinates is done the same way, as well, from I<x> and I<y>:
872
873 uint32_t xy = ((uint32_t)x << 16) | y; // for ecb_hilbert2d_coord_to_index32
874 uint64_t xy = ((uint64_t)x << 32) | y; // for ecb_hilbert2d_coord_to_index64
875
876 Unlike C<ecb_hilbert2d_coord_to_index>, these functions are O(I<order>),
877 so it is preferable to use the lowest possible order.
878
879 =back
880
881 =head2 BIT MIXING, HASHING
882
883 Sometimes you have an integer and want to distribute its bits well, for
884 example, to use it as a hash in a hash table. A common example is pointer
885 values, which often only have a limited range (e.g. low and high bits are
886 often zero).
887
888 The following functions try to mix the bits to get a good bias-free
889 distribution. They were mainly made for pointers, but the underlying
890 integer functions are exposed as well.
891
892 As an added benefit, the functions are reversible, so if you find it
893 convenient to store only the hash value, you can recover the original
894 pointer from the hash ("unmix"), as long as your pointers are 32 or 64 bit
895 (if this isn't the case on your platform, drop us a note and we will add
896 functions for other bit widths).
897
898 The unmix functions are very slightly slower than the mix functions, so
899 it is equally very slightly preferable to store the original values wehen
900 convenient.
901
902 The underlying algorithm if subject to change, so currently these
903 functions are not suitable for persistent hash tables, as their result
904 value can change between different versions of libecb.
905
906 =over
907
908 =item uintptr_t ecb_ptrmix (void *ptr)
909
910 Mixes the bits of a pointer so the result is suitable for hash table
911 lookups. In other words, this hashes the pointer value.
912
913 =item uintptr_t ecb_ptrmix (T *ptr) [C++]
914
915 Overload the C<ecb_ptrmix> function to work for any pointer in C++.
916
917 =item void *ecb_ptrunmix (uintptr_t v)
918
919 Unmix the hash value into the original pointer. This only works as long
920 as the hash value is not truncated, i.e. you used C<uintptr_t> (or
921 equivalent) throughout to store it.
922
923 =item T *ecb_ptrunmix<T> (uintptr_t v) [C++]
924
925 The somewhat less useful template version of C<ecb_ptrunmix> for
926 C++. Example:
927
928 sometype *myptr;
929 uintptr_t hash = ecb_ptrmix (myptr);
930 sometype *orig = ecb_ptrunmix<sometype> (hash);
931
932 =item uint32_t ecb_mix32 (uint32_t v)
933
934 =item uint64_t ecb_mix64 (uint64_t v)
935
936 Sometimes you don't have a pointer but an integer whose values are very
937 badly distributed. In this case you can use these integer versions of the
938 mixing function. No C++ template is provided currently.
939
940 =item uint32_t ecb_unmix32 (uint32_t v)
941
942 =item uint64_t ecb_unmix64 (uint64_t v)
943
944 The reverse of the C<ecb_mix> functions - they take a mixed/hashed value
945 and recover the original value.
946
947 =back
948
949 =head2 HOST ENDIANNESS CONVERSION
950
951 =over
952
953 =item uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v)
954
955 =item uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v)
956
957 =item uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v)
958
959 =item uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v)
960
961 =item uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v)
962
963 =item uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v)
964
965 Convert an unsigned 16, 32 or 64 bit value from big or little endian to host byte order.
966
967 The naming convention is C<ecb_>(C<be>|C<le>)C<_u>C<16|32|64>C<_to_host>,
968 where C<be> and C<le> stand for big endian and little endian, respectively.
969
970 =item uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v)
971
972 =item uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v)
973
974 =item uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v)
975
976 =item uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v)
977
978 =item uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v)
979
980 =item uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v)
981
982 Like above, but converts I<from> host byte order to the specified
983 endianness.
984
985 =back
986
987 In C++ the following additional template functions are supported:
988
989 =over
990
991 =item T ecb_be_to_host (T v)
992
993 =item T ecb_le_to_host (T v)
994
995 =item T ecb_host_to_be (T v)
996
997 =item T ecb_host_to_le (T v)
998
999 =back
1000
1001 These functions work like their C counterparts, above, but use templates,
1002 which make them useful in generic code.
1003
1004 C<T> must be one of C<uint8_t>, C<uint16_t>, C<uint32_t> or C<uint64_t>
1005 (so unlike their C counterparts, there is a version for C<uint8_t>, which
1006 again can be useful in generic code).
1007
1008 =head2 UNALIGNED LOAD/STORE
1009
1010 These function load or store unaligned multi-byte values.
1011
1012 =over
1013
1014 =item uint_fast16_t ecb_peek_u16_u (const void *ptr)
1015
1016 =item uint_fast32_t ecb_peek_u32_u (const void *ptr)
1017
1018 =item uint_fast64_t ecb_peek_u64_u (const void *ptr)
1019
1020 These functions load an unaligned, unsigned 16, 32 or 64 bit value from
1021 memory.
1022
1023 =item uint_fast16_t ecb_peek_be_u16_u (const void *ptr)
1024
1025 =item uint_fast32_t ecb_peek_be_u32_u (const void *ptr)
1026
1027 =item uint_fast64_t ecb_peek_be_u64_u (const void *ptr)
1028
1029 =item uint_fast16_t ecb_peek_le_u16_u (const void *ptr)
1030
1031 =item uint_fast32_t ecb_peek_le_u32_u (const void *ptr)
1032
1033 =item uint_fast64_t ecb_peek_le_u64_u (const void *ptr)
1034
1035 Like above, but additionally convert from big endian (C<be>) or little
1036 endian (C<le>) byte order to host byte order while doing so.
1037
1038 =item ecb_poke_u16_u (void *ptr, uint16_t v)
1039
1040 =item ecb_poke_u32_u (void *ptr, uint32_t v)
1041
1042 =item ecb_poke_u64_u (void *ptr, uint64_t v)
1043
1044 These functions store an unaligned, unsigned 16, 32 or 64 bit value to
1045 memory.
1046
1047 =item ecb_poke_be_u16_u (void *ptr, uint_fast16_t v)
1048
1049 =item ecb_poke_be_u32_u (void *ptr, uint_fast32_t v)
1050
1051 =item ecb_poke_be_u64_u (void *ptr, uint_fast64_t v)
1052
1053 =item ecb_poke_le_u16_u (void *ptr, uint_fast16_t v)
1054
1055 =item ecb_poke_le_u32_u (void *ptr, uint_fast32_t v)
1056
1057 =item ecb_poke_le_u64_u (void *ptr, uint_fast64_t v)
1058
1059 Like above, but additionally convert from host byte order to big endian
1060 (C<be>) or little endian (C<le>) byte order while doing so.
1061
1062 =back
1063
1064 In C++ the following additional template functions are supported:
1065
1066 =over
1067
1068 =item T ecb_peek<T> (const void *ptr)
1069
1070 =item T ecb_peek_be<T> (const void *ptr)
1071
1072 =item T ecb_peek_le<T> (const void *ptr)
1073
1074 =item T ecb_peek_u<T> (const void *ptr)
1075
1076 =item T ecb_peek_be_u<T> (const void *ptr)
1077
1078 =item T ecb_peek_le_u<T> (const void *ptr)
1079
1080 Similarly to their C counterparts, these functions load an unsigned 8, 16,
1081 32 or 64 bit value from memory, with optional conversion from big/little
1082 endian.
1083
1084 Since the type cannot be deduced, it has to be specified explicitly, e.g.
1085
1086 uint_fast16_t v = ecb_peek<uint16_t> (ptr);
1087
1088 C<T> must be one of C<uint8_t>, C<uint16_t>, C<uint32_t> or C<uint64_t>.
1089
1090 Unlike their C counterparts, these functions support 8 bit quantities
1091 (C<uint8_t>) and also have an aligned version (without the C<_u> prefix),
1092 all of which hopefully makes them more useful in generic code.
1093
1094 =item ecb_poke (void *ptr, T v)
1095
1096 =item ecb_poke_be (void *ptr, T v)
1097
1098 =item ecb_poke_le (void *ptr, T v)
1099
1100 =item ecb_poke_u (void *ptr, T v)
1101
1102 =item ecb_poke_be_u (void *ptr, T v)
1103
1104 =item ecb_poke_le_u (void *ptr, T v)
1105
1106 Again, similarly to their C counterparts, these functions store an
1107 unsigned 8, 16, 32 or 64 bit value to memory, with optional conversion to
1108 big/little endian.
1109
1110 C<T> must be one of C<uint8_t>, C<uint16_t>, C<uint32_t> or C<uint64_t>.
1111
1112 Unlike their C counterparts, these functions support 8 bit quantities
1113 (C<uint8_t>) and also have an aligned version (without the C<_u> prefix),
1114 all of which hopefully makes them more useful in generic code.
1115
1116 =back
1117
1118 =head2 FAST INTEGER TO STRING
1119
1120 Libecb defines a set of very fast integer to decimal string (or integer
1121 to ASCII, short C<i2a>) functions. These work by converting the integer
1122 to a fixed point representation and then successively multiplying out
1123 the topmost digits. Unlike some other, also very fast, libraries, ecb's
1124 algorithm should be completely branchless per digit, and does not rely on
1125 the presence of special CPU functions (such as C<clz>).
1126
1127 There is a high level API that takes an C<int32_t>, C<uint32_t>,
1128 C<int64_t> or C<uint64_t> as argument, and a low-level API, which is
1129 harder to use but supports slightly more formatting options.
1130
1131 =head3 HIGH LEVEL API
1132
1133 The high level API consists of four functions, one each for C<int32_t>,
1134 C<uint32_t>, C<int64_t> and C<uint64_t>:
1135
1136 Example:
1137
1138 char buf[ECB_I2A_MAX_DIGITS + 1];
1139 char *end = ecb_i2a_i32 (buf, 17262);
1140 *end = 0;
1141 // buf now contains "17262"
1142
1143 =over
1144
1145 =item ECB_I2A_I32_DIGITS (=11)
1146
1147 =item char *ecb_i2a_u32 (char *ptr, uint32_t value)
1148
1149 Takes an C<uint32_t> I<value> and formats it as a decimal number starting
1150 at I<ptr>, using at most C<ECB_I2A_I32_DIGITS> characters. Returns a
1151 pointer to just after the generated string, where you would normally put
1152 the terminating C<0> character. This function outputs the minimum number
1153 of digits.
1154
1155 =item ECB_I2A_U32_DIGITS (=10)
1156
1157 =item char *ecb_i2a_i32 (char *ptr, int32_t value)
1158
1159 Same as C<ecb_i2a_u32>, but formats a C<int32_t> value, including a minus
1160 sign if needed.
1161
1162 =item ECB_I2A_I64_DIGITS (=20)
1163
1164 =item char *ecb_i2a_u64 (char *ptr, uint64_t value)
1165
1166 =item ECB_I2A_U64_DIGITS (=21)
1167
1168 =item char *ecb_i2a_i64 (char *ptr, int64_t value)
1169
1170 Similar to their 32 bit counterparts, these take a 64 bit argument.
1171
1172 =item ECB_I2A_MAX_DIGITS (=21)
1173
1174 Instead of using a type specific length macro, you can just use
1175 C<ECB_I2A_MAX_DIGITS>, which is good enough for any C<ecb_i2a> function.
1176
1177 =back
1178
1179 =head3 LOW-LEVEL API
1180
1181 The functions above use a number of low-level APIs which have some strict
1182 limitations, but can be used as building blocks (studying C<ecb_i2a_i32>
1183 and related functions is recommended).
1184
1185 There are three families of functions: functions that convert a number
1186 to a fixed number of digits with leading zeroes (C<ecb_i2a_0N>, C<0>
1187 for "leading zeroes"), functions that generate up to N digits, skipping
1188 leading zeroes (C<_N>), and functions that can generate more digits, but
1189 the leading digit has limited range (C<_xN>).
1190
1191 None of the functions deal with negative numbers.
1192
1193 Example: convert an IP address in an C<uint32_t> into dotted-quad:
1194
1195 uint32_t ip = 0x0a000164; // 10.0.1.100
1196 char ips[3 * 4 + 3 + 1];
1197 char *ptr = ips;
1198 ptr = ecb_i2a_3 (ptr, ip >> 24 ); *ptr++ = '.';
1199 ptr = ecb_i2a_3 (ptr, (ip >> 16) & 0xff); *ptr++ = '.';
1200 ptr = ecb_i2a_3 (ptr, (ip >> 8) & 0xff); *ptr++ = '.';
1201 ptr = ecb_i2a_3 (ptr, ip & 0xff); *ptr++ = 0;
1202 printf ("ip: %s\n", ips); // prints "ip: 10.0.1.100"
1203
1204 =over
1205
1206 =item char *ecb_i2a_02 (char *ptr, uint32_t value) // 32 bit
1207
1208 =item char *ecb_i2a_03 (char *ptr, uint32_t value) // 32 bit
1209
1210 =item char *ecb_i2a_04 (char *ptr, uint32_t value) // 32 bit
1211
1212 =item char *ecb_i2a_05 (char *ptr, uint32_t value) // 64 bit
1213
1214 =item char *ecb_i2a_06 (char *ptr, uint32_t value) // 64 bit
1215
1216 =item char *ecb_i2a_07 (char *ptr, uint32_t value) // 64 bit
1217
1218 =item char *ecb_i2a_08 (char *ptr, uint32_t value) // 64 bit
1219
1220 =item char *ecb_i2a_09 (char *ptr, uint32_t value) // 64 bit
1221
1222 The C<< ecb_i2a_0I<N> >> functions take an unsigned I<value> and convert
1223 them to exactly I<N> digits, returning a pointer to the first character
1224 after the digits. The I<value> must be in range. The functions marked with
1225 I<32 bit> do their calculations internally in 32 bit, the ones marked with
1226 I<64 bit> internally use 64 bit integers, which might be slow on 32 bit
1227 architectures (the high level API decides on 32 vs. 64 bit versions using
1228 C<ECB_64BIT_NATIVE>).
1229
1230 =item char *ecb_i2a_2 (char *ptr, uint32_t value) // 32 bit
1231
1232 =item char *ecb_i2a_3 (char *ptr, uint32_t value) // 32 bit
1233
1234 =item char *ecb_i2a_4 (char *ptr, uint32_t value) // 32 bit
1235
1236 =item char *ecb_i2a_5 (char *ptr, uint32_t value) // 64 bit
1237
1238 =item char *ecb_i2a_6 (char *ptr, uint32_t value) // 64 bit
1239
1240 =item char *ecb_i2a_7 (char *ptr, uint32_t value) // 64 bit
1241
1242 =item char *ecb_i2a_8 (char *ptr, uint32_t value) // 64 bit
1243
1244 =item char *ecb_i2a_9 (char *ptr, uint32_t value) // 64 bit
1245
1246 Similarly, the C<< ecb_i2a_I<N> >> functions take an unsigned I<value>
1247 and convert them to at most I<N> digits, suppressing leading zeroes, and
1248 returning a pointer to the first character after the digits.
1249
1250 =item ECB_I2A_MAX_X5 (=59074)
1251
1252 =item char *ecb_i2a_x5 (char *ptr, uint32_t value) // 32 bit
1253
1254 =item ECB_I2A_MAX_X10 (=2932500665)
1255
1256 =item char *ecb_i2a_x10 (char *ptr, uint32_t value) // 64 bit
1257
1258 The C<< ecb_i2a_xI<N> >> functions are similar to the C<< ecb_i2a_I<N> >>
1259 functions, but they can generate one digit more, as long as the number
1260 is within range, which is given by the symbols C<ECB_I2A_MAX_X5> (almost
1261 16 bit range) and C<ECB_I2A_MAX_X10> (a bit more than 31 bit range),
1262 respectively.
1263
1264 For example, the digit part of a 32 bit signed integer just fits into the
1265 C<ECB_I2A_MAX_X10> range, so while C<ecb_i2a_x10> cannot convert a 10
1266 digit number, it can convert all 32 bit signed numbers. Sadly, it's not
1267 good enough for 32 bit unsigned numbers.
1268
1269 =back
1270
1271 =head2 FLOATING POINT FIDDLING
1272
1273 =over
1274
1275 =item ECB_INFINITY [-UECB_NO_LIBM]
1276
1277 Evaluates to positive infinity if supported by the platform, otherwise to
1278 a truly huge number.
1279
1280 =item ECB_NAN [-UECB_NO_LIBM]
1281
1282 Evaluates to a quiet NAN if supported by the platform, otherwise to
1283 C<ECB_INFINITY>.
1284
1285 =item float ecb_ldexpf (float x, int exp) [-UECB_NO_LIBM]
1286
1287 Same as C<ldexpf>, but always available.
1288
1289 =item uint32_t ecb_float_to_binary16 (float x) [-UECB_NO_LIBM]
1290
1291 =item uint32_t ecb_float_to_binary32 (float x) [-UECB_NO_LIBM]
1292
1293 =item uint64_t ecb_double_to_binary64 (double x) [-UECB_NO_LIBM]
1294
1295 These functions each take an argument in the native C<float> or C<double>
1296 type and return the IEEE 754 bit representation of it (binary16/half,
1297 binary32/single or binary64/double precision).
1298
1299 The bit representation is just as IEEE 754 defines it, i.e. the sign bit
1300 will be the most significant bit, followed by exponent and mantissa.
1301
1302 This function should work even when the native floating point format isn't
1303 IEEE compliant, of course at a speed and code size penalty, and of course
1304 also within reasonable limits (it tries to convert NaNs, infinities and
1305 denormals, but will likely convert negative zero to positive zero).
1306
1307 On all modern platforms (where C<ECB_STDFP> is true), the compiler should
1308 be able to completely optimise away the 32 and 64 bit functions.
1309
1310 These functions can be helpful when serialising floats to the network - you
1311 can serialise the return value like a normal uint16_t/uint32_t/uint64_t.
1312
1313 Another use for these functions is to manipulate floating point values
1314 directly.
1315
1316 Silly example: toggle the sign bit of a float.
1317
1318 /* On gcc-4.7 on amd64, */
1319 /* this results in a single add instruction to toggle the bit, and 4 extra */
1320 /* instructions to move the float value to an integer register and back. */
1321
1322 x = ecb_binary32_to_float (ecb_float_to_binary32 (x) ^ 0x80000000U)
1323
1324 =item float ecb_binary16_to_float (uint16_t x) [-UECB_NO_LIBM]
1325
1326 =item float ecb_binary32_to_float (uint32_t x) [-UECB_NO_LIBM]
1327
1328 =item double ecb_binary64_to_double (uint64_t x) [-UECB_NO_LIBM]
1329
1330 The reverse operation of the previous function - takes the bit
1331 representation of an IEEE binary16, binary32 or binary64 number (half,
1332 single or double precision) and converts it to the native C<float> or
1333 C<double> format.
1334
1335 This function should work even when the native floating point format isn't
1336 IEEE compliant, of course at a speed and code size penalty, and of course
1337 also within reasonable limits (it tries to convert normals and denormals,
1338 and might be lucky for infinities, and with extraordinary luck, also for
1339 negative zero).
1340
1341 On all modern platforms (where C<ECB_STDFP> is true), the compiler should
1342 be able to optimise away this function completely.
1343
1344 =item uint16_t ecb_binary32_to_binary16 (uint32_t x)
1345
1346 =item uint32_t ecb_binary16_to_binary32 (uint16_t x)
1347
1348 Convert a IEEE binary32/single precision to binary16/half format, and vice
1349 versa, handling all details (round-to-nearest-even, subnormals, infinity
1350 and NaNs) correctly.
1351
1352 These are functions are available under C<-DECB_NO_LIBM>, since
1353 they do not rely on the platform floating point format. The
1354 C<ecb_float_to_binary16> and C<ecb_binary16_to_float> functions are
1355 usually what you want.
1356
1357 =back
1358
1359 =head2 ARITHMETIC
1360
1361 =over
1362
1363 =item x = ecb_mod (m, n)
1364
1365 Returns C<m> modulo C<n>, which is the same as the positive remainder
1366 of the division operation between C<m> and C<n>, using floored
1367 division. Unlike the C remainder operator C<%>, this function ensures that
1368 the return value is always positive and that the two numbers I<m> and
1369 I<m' = m + i * n> result in the same value modulo I<n> - in other words,
1370 C<ecb_mod> implements the mathematical modulo operation, which is missing
1371 in the language.
1372
1373 C<n> must be strictly positive (i.e. C<< >= 1 >>), while C<m> must be
1374 negatable, that is, both C<m> and C<-m> must be representable in its
1375 type (this typically excludes the minimum signed integer value, the same
1376 limitation as for C</> and C<%> in C).
1377
1378 Current GCC/clang versions compile this into an efficient branchless
1379 sequence on almost all CPUs.
1380
1381 For example, when you want to rotate forward through the members of an
1382 array for increasing C<m> (which might be negative), then you should use
1383 C<ecb_mod>, as the C<%> operator might give either negative results, or
1384 change direction for negative values:
1385
1386 for (m = -100; m <= 100; ++m)
1387 int elem = myarray [ecb_mod (m, ecb_array_length (myarray))];
1388
1389 =item x = ecb_div_rd (val, div)
1390
1391 =item x = ecb_div_ru (val, div)
1392
1393 Returns C<val> divided by C<div> rounded down or up, respectively.
1394 C<val> and C<div> must have integer types and C<div> must be strictly
1395 positive. Note that these functions are implemented with macros in C
1396 and with function templates in C++.
1397
1398 =back
1399
1400 =head2 UTILITY
1401
1402 =over
1403
1404 =item element_count = ecb_array_length (name)
1405
1406 Returns the number of elements in the array C<name>. For example:
1407
1408 int primes[] = { 2, 3, 5, 7, 11 };
1409 int sum = 0;
1410
1411 for (i = 0; i < ecb_array_length (primes); i++)
1412 sum += primes [i];
1413
1414 =back
1415
1416 =head2 SYMBOLS GOVERNING COMPILATION OF ECB.H ITSELF
1417
1418 These symbols need to be defined before including F<ecb.h> the first time.
1419
1420 =over
1421
1422 =item ECB_NO_THREADS
1423
1424 If F<ecb.h> is never used from multiple threads, then this symbol can
1425 be defined, in which case memory fences (and similar constructs) are
1426 completely removed, leading to more efficient code and fewer dependencies.
1427
1428 Setting this symbol to a true value implies C<ECB_NO_SMP>.
1429
1430 =item ECB_NO_SMP
1431
1432 The weaker version of C<ECB_NO_THREADS> - if F<ecb.h> is used from
1433 multiple threads, but never concurrently (e.g. if the system the program
1434 runs on has only a single CPU with a single core, no hyper-threading and so
1435 on), then this symbol can be defined, leading to more efficient code and
1436 fewer dependencies.
1437
1438 =item ECB_NO_LIBM
1439
1440 When defined to C<1>, do not export any functions that might introduce
1441 dependencies on the math library (usually called F<-lm>) - these are
1442 marked with [-UECB_NO_LIBM].
1443
1444 =back
1445
1446 =head1 UNDOCUMENTED FUNCTIONALITY
1447
1448 F<ecb.h> is full of undocumented functionality as well, some of which is
1449 intended to be internal-use only, some of which we forgot to document, and
1450 some of which we hide because we are not sure we will keep the interface
1451 stable.
1452
1453 While you are welcome to rummage around and use whatever you find useful
1454 (we don't want to stop you), keep in mind that we will change undocumented
1455 functionality in incompatible ways without thinking twice, while we are
1456 considerably more conservative with documented things.
1457
1458 =head1 AUTHORS
1459
1460 C<libecb> is designed and maintained by:
1461
1462 Emanuele Giaquinta <e.giaquinta@glauco.it>
1463 Marc Alexander Lehmann <schmorp@schmorp.de>