xref: /freebsd/contrib/xz/src/common/tuklib_integer.h (revision 8ddb146abcdf061be9f2c0db7e391697dafad85c)
1 ///////////////////////////////////////////////////////////////////////////////
2 //
3 /// \file       tuklib_integer.h
4 /// \brief      Various integer and bit operations
5 ///
6 /// This file provides macros or functions to do some basic integer and bit
7 /// operations.
8 ///
9 /// Native endian inline functions (XX = 16, 32, or 64):
10 ///   - Unaligned native endian reads: readXXne(ptr)
11 ///   - Unaligned native endian writes: writeXXne(ptr, num)
12 ///   - Aligned native endian reads: aligned_readXXne(ptr)
13 ///   - Aligned native endian writes: aligned_writeXXne(ptr, num)
14 ///
15 /// Endianness-converting integer operations (these can be macros!)
16 /// (XX = 16, 32, or 64; Y = b or l):
17 ///   - Byte swapping: bswapXX(num)
18 ///   - Byte order conversions to/from native (byteswaps if Y isn't
19 ///     the native endianness): convXXYe(num)
20 ///   - Unaligned reads (16/32-bit only): readXXYe(ptr)
21 ///   - Unaligned writes (16/32-bit only): writeXXYe(ptr, num)
22 ///   - Aligned reads: aligned_readXXYe(ptr)
23 ///   - Aligned writes: aligned_writeXXYe(ptr, num)
24 ///
25 /// Since the above can macros, the arguments should have no side effects
26 /// because they may be evaluated more than once.
27 ///
28 /// Bit scan operations for non-zero 32-bit integers (inline functions):
29 ///   - Bit scan reverse (find highest non-zero bit): bsr32(num)
30 ///   - Count leading zeros: clz32(num)
31 ///   - Count trailing zeros: ctz32(num)
32 ///   - Bit scan forward (simply an alias for ctz32()): bsf32(num)
33 ///
34 /// The above bit scan operations return 0-31. If num is zero,
35 /// the result is undefined.
36 //
37 //  Authors:    Lasse Collin
38 //              Joachim Henke
39 //
40 //  This file has been put into the public domain.
41 //  You can do whatever you want with this file.
42 //
43 ///////////////////////////////////////////////////////////////////////////////
44 
45 #ifndef TUKLIB_INTEGER_H
46 #define TUKLIB_INTEGER_H
47 
48 #include "tuklib_common.h"
49 #include <string.h>
50 
51 // Newer Intel C compilers require immintrin.h for _bit_scan_reverse()
52 // and such functions.
53 #if defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 1500)
54 #	include <immintrin.h>
55 #endif
56 
57 
58 ///////////////////
59 // Byte swapping //
60 ///////////////////
61 
62 #if defined(HAVE___BUILTIN_BSWAPXX)
63 	// GCC >= 4.8 and Clang
64 #	define bswap16(n) __builtin_bswap16(n)
65 #	define bswap32(n) __builtin_bswap32(n)
66 #	define bswap64(n) __builtin_bswap64(n)
67 
68 #elif defined(HAVE_BYTESWAP_H)
69 	// glibc, uClibc, dietlibc
70 #	include <byteswap.h>
71 #	ifdef HAVE_BSWAP_16
72 #		define bswap16(num) bswap_16(num)
73 #	endif
74 #	ifdef HAVE_BSWAP_32
75 #		define bswap32(num) bswap_32(num)
76 #	endif
77 #	ifdef HAVE_BSWAP_64
78 #		define bswap64(num) bswap_64(num)
79 #	endif
80 
81 #elif defined(HAVE_SYS_ENDIAN_H)
82 	// *BSDs and Darwin
83 #	include <sys/endian.h>
84 
85 #elif defined(HAVE_SYS_BYTEORDER_H)
86 	// Solaris
87 #	include <sys/byteorder.h>
88 #	ifdef BSWAP_16
89 #		define bswap16(num) BSWAP_16(num)
90 #	endif
91 #	ifdef BSWAP_32
92 #		define bswap32(num) BSWAP_32(num)
93 #	endif
94 #	ifdef BSWAP_64
95 #		define bswap64(num) BSWAP_64(num)
96 #	endif
97 #	ifdef BE_16
98 #		define conv16be(num) BE_16(num)
99 #	endif
100 #	ifdef BE_32
101 #		define conv32be(num) BE_32(num)
102 #	endif
103 #	ifdef BE_64
104 #		define conv64be(num) BE_64(num)
105 #	endif
106 #	ifdef LE_16
107 #		define conv16le(num) LE_16(num)
108 #	endif
109 #	ifdef LE_32
110 #		define conv32le(num) LE_32(num)
111 #	endif
112 #	ifdef LE_64
113 #		define conv64le(num) LE_64(num)
114 #	endif
115 #endif
116 
117 #ifndef bswap16
118 #	define bswap16(n) (uint16_t)( \
119 		  (((n) & 0x00FFU) << 8) \
120 		| (((n) & 0xFF00U) >> 8) \
121 	)
122 #endif
123 
124 #ifndef bswap32
125 #	define bswap32(n) (uint32_t)( \
126 		  (((n) & UINT32_C(0x000000FF)) << 24) \
127 		| (((n) & UINT32_C(0x0000FF00)) << 8) \
128 		| (((n) & UINT32_C(0x00FF0000)) >> 8) \
129 		| (((n) & UINT32_C(0xFF000000)) >> 24) \
130 	)
131 #endif
132 
133 #ifndef bswap64
134 #	define bswap64(n) (uint64_t)( \
135 		  (((n) & UINT64_C(0x00000000000000FF)) << 56) \
136 		| (((n) & UINT64_C(0x000000000000FF00)) << 40) \
137 		| (((n) & UINT64_C(0x0000000000FF0000)) << 24) \
138 		| (((n) & UINT64_C(0x00000000FF000000)) << 8) \
139 		| (((n) & UINT64_C(0x000000FF00000000)) >> 8) \
140 		| (((n) & UINT64_C(0x0000FF0000000000)) >> 24) \
141 		| (((n) & UINT64_C(0x00FF000000000000)) >> 40) \
142 		| (((n) & UINT64_C(0xFF00000000000000)) >> 56) \
143 	)
144 #endif
145 
146 // Define conversion macros using the basic byte swapping macros.
147 #ifdef WORDS_BIGENDIAN
148 #	ifndef conv16be
149 #		define conv16be(num) ((uint16_t)(num))
150 #	endif
151 #	ifndef conv32be
152 #		define conv32be(num) ((uint32_t)(num))
153 #	endif
154 #	ifndef conv64be
155 #		define conv64be(num) ((uint64_t)(num))
156 #	endif
157 #	ifndef conv16le
158 #		define conv16le(num) bswap16(num)
159 #	endif
160 #	ifndef conv32le
161 #		define conv32le(num) bswap32(num)
162 #	endif
163 #	ifndef conv64le
164 #		define conv64le(num) bswap64(num)
165 #	endif
166 #else
167 #	ifndef conv16be
168 #		define conv16be(num) bswap16(num)
169 #	endif
170 #	ifndef conv32be
171 #		define conv32be(num) bswap32(num)
172 #	endif
173 #	ifndef conv64be
174 #		define conv64be(num) bswap64(num)
175 #	endif
176 #	ifndef conv16le
177 #		define conv16le(num) ((uint16_t)(num))
178 #	endif
179 #	ifndef conv32le
180 #		define conv32le(num) ((uint32_t)(num))
181 #	endif
182 #	ifndef conv64le
183 #		define conv64le(num) ((uint64_t)(num))
184 #	endif
185 #endif
186 
187 
188 ////////////////////////////////
189 // Unaligned reads and writes //
190 ////////////////////////////////
191 
192 // The traditional way of casting e.g. *(const uint16_t *)uint8_pointer
193 // is bad even if the uint8_pointer is properly aligned because this kind
194 // of casts break strict aliasing rules and result in undefined behavior.
195 // With unaligned pointers it's even worse: compilers may emit vector
196 // instructions that require aligned pointers even if non-vector
197 // instructions work with unaligned pointers.
198 //
199 // Using memcpy() is the standard compliant way to do unaligned access.
200 // Many modern compilers inline it so there is no function call overhead.
201 // For those compilers that don't handle the memcpy() method well, the
202 // old casting method (that violates strict aliasing) can be requested at
203 // build time. A third method, casting to a packed struct, would also be
204 // an option but isn't provided to keep things simpler (it's already a mess).
205 // Hopefully this is flexible enough in practice.
206 
207 static inline uint16_t
208 read16ne(const uint8_t *buf)
209 {
210 #if defined(TUKLIB_FAST_UNALIGNED_ACCESS) \
211 		&& defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING)
212 	return *(const uint16_t *)buf;
213 #else
214 	uint16_t num;
215 	memcpy(&num, buf, sizeof(num));
216 	return num;
217 #endif
218 }
219 
220 
221 static inline uint32_t
222 read32ne(const uint8_t *buf)
223 {
224 #if defined(TUKLIB_FAST_UNALIGNED_ACCESS) \
225 		&& defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING)
226 	return *(const uint32_t *)buf;
227 #else
228 	uint32_t num;
229 	memcpy(&num, buf, sizeof(num));
230 	return num;
231 #endif
232 }
233 
234 
235 static inline uint64_t
236 read64ne(const uint8_t *buf)
237 {
238 #if defined(TUKLIB_FAST_UNALIGNED_ACCESS) \
239 		&& defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING)
240 	return *(const uint64_t *)buf;
241 #else
242 	uint64_t num;
243 	memcpy(&num, buf, sizeof(num));
244 	return num;
245 #endif
246 }
247 
248 
249 static inline void
250 write16ne(uint8_t *buf, uint16_t num)
251 {
252 #if defined(TUKLIB_FAST_UNALIGNED_ACCESS) \
253 		&& defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING)
254 	*(uint16_t *)buf = num;
255 #else
256 	memcpy(buf, &num, sizeof(num));
257 #endif
258 	return;
259 }
260 
261 
262 static inline void
263 write32ne(uint8_t *buf, uint32_t num)
264 {
265 #if defined(TUKLIB_FAST_UNALIGNED_ACCESS) \
266 		&& defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING)
267 	*(uint32_t *)buf = num;
268 #else
269 	memcpy(buf, &num, sizeof(num));
270 #endif
271 	return;
272 }
273 
274 
275 static inline void
276 write64ne(uint8_t *buf, uint64_t num)
277 {
278 #if defined(TUKLIB_FAST_UNALIGNED_ACCESS) \
279 		&& defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING)
280 	*(uint64_t *)buf = num;
281 #else
282 	memcpy(buf, &num, sizeof(num));
283 #endif
284 	return;
285 }
286 
287 
288 static inline uint16_t
289 read16be(const uint8_t *buf)
290 {
291 #if defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
292 	uint16_t num = read16ne(buf);
293 	return conv16be(num);
294 #else
295 	uint16_t num = ((uint16_t)buf[0] << 8) | (uint16_t)buf[1];
296 	return num;
297 #endif
298 }
299 
300 
301 static inline uint16_t
302 read16le(const uint8_t *buf)
303 {
304 #if !defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
305 	uint16_t num = read16ne(buf);
306 	return conv16le(num);
307 #else
308 	uint16_t num = ((uint16_t)buf[0]) | ((uint16_t)buf[1] << 8);
309 	return num;
310 #endif
311 }
312 
313 
314 static inline uint32_t
315 read32be(const uint8_t *buf)
316 {
317 #if defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
318 	uint32_t num = read32ne(buf);
319 	return conv32be(num);
320 #else
321 	uint32_t num = (uint32_t)buf[0] << 24;
322 	num |= (uint32_t)buf[1] << 16;
323 	num |= (uint32_t)buf[2] << 8;
324 	num |= (uint32_t)buf[3];
325 	return num;
326 #endif
327 }
328 
329 
330 static inline uint32_t
331 read32le(const uint8_t *buf)
332 {
333 #if !defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
334 	uint32_t num = read32ne(buf);
335 	return conv32le(num);
336 #else
337 	uint32_t num = (uint32_t)buf[0];
338 	num |= (uint32_t)buf[1] << 8;
339 	num |= (uint32_t)buf[2] << 16;
340 	num |= (uint32_t)buf[3] << 24;
341 	return num;
342 #endif
343 }
344 
345 
346 // NOTE: Possible byte swapping must be done in a macro to allow the compiler
347 // to optimize byte swapping of constants when using glibc's or *BSD's
348 // byte swapping macros. The actual write is done in an inline function
349 // to make type checking of the buf pointer possible.
350 #if defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
351 #	define write16be(buf, num) write16ne(buf, conv16be(num))
352 #	define write32be(buf, num) write32ne(buf, conv32be(num))
353 #endif
354 
355 #if !defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
356 #	define write16le(buf, num) write16ne(buf, conv16le(num))
357 #	define write32le(buf, num) write32ne(buf, conv32le(num))
358 #endif
359 
360 
361 #ifndef write16be
362 static inline void
363 write16be(uint8_t *buf, uint16_t num)
364 {
365 	buf[0] = (uint8_t)(num >> 8);
366 	buf[1] = (uint8_t)num;
367 	return;
368 }
369 #endif
370 
371 
372 #ifndef write16le
373 static inline void
374 write16le(uint8_t *buf, uint16_t num)
375 {
376 	buf[0] = (uint8_t)num;
377 	buf[1] = (uint8_t)(num >> 8);
378 	return;
379 }
380 #endif
381 
382 
383 #ifndef write32be
384 static inline void
385 write32be(uint8_t *buf, uint32_t num)
386 {
387 	buf[0] = (uint8_t)(num >> 24);
388 	buf[1] = (uint8_t)(num >> 16);
389 	buf[2] = (uint8_t)(num >> 8);
390 	buf[3] = (uint8_t)num;
391 	return;
392 }
393 #endif
394 
395 
396 #ifndef write32le
397 static inline void
398 write32le(uint8_t *buf, uint32_t num)
399 {
400 	buf[0] = (uint8_t)num;
401 	buf[1] = (uint8_t)(num >> 8);
402 	buf[2] = (uint8_t)(num >> 16);
403 	buf[3] = (uint8_t)(num >> 24);
404 	return;
405 }
406 #endif
407 
408 
409 //////////////////////////////
410 // Aligned reads and writes //
411 //////////////////////////////
412 
413 // Separate functions for aligned reads and writes are provided since on
414 // strict-align archs aligned access is much faster than unaligned access.
415 //
416 // Just like in the unaligned case, memcpy() is needed to avoid
417 // strict aliasing violations. However, on archs that don't support
418 // unaligned access the compiler cannot know that the pointers given
419 // to memcpy() are aligned which results in slow code. As of C11 there is
420 // no standard way to tell the compiler that we know that the address is
421 // aligned but some compilers have language extensions to do that. With
422 // such language extensions the memcpy() method gives excellent results.
423 //
424 // What to do on a strict-align system when no known language extentensions
425 // are available? Falling back to byte-by-byte access would be safe but ruin
426 // optimizations that have been made specifically with aligned access in mind.
427 // As a compromise, aligned reads will fall back to non-compliant type punning
428 // but aligned writes will be byte-by-byte, that is, fast reads are preferred
429 // over fast writes. This obviously isn't great but hopefully it's a working
430 // compromise for now.
431 //
432 // __builtin_assume_aligned is support by GCC >= 4.7 and clang >= 3.6.
433 #ifdef HAVE___BUILTIN_ASSUME_ALIGNED
434 #	define tuklib_memcpy_aligned(dest, src, size) \
435 		memcpy(dest, __builtin_assume_aligned(src, size), size)
436 #else
437 #	define tuklib_memcpy_aligned(dest, src, size) \
438 		memcpy(dest, src, size)
439 #	ifndef TUKLIB_FAST_UNALIGNED_ACCESS
440 #		define TUKLIB_USE_UNSAFE_ALIGNED_READS 1
441 #	endif
442 #endif
443 
444 
445 static inline uint16_t
446 aligned_read16ne(const uint8_t *buf)
447 {
448 #if defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING) \
449 		|| defined(TUKLIB_USE_UNSAFE_ALIGNED_READS)
450 	return *(const uint16_t *)buf;
451 #else
452 	uint16_t num;
453 	tuklib_memcpy_aligned(&num, buf, sizeof(num));
454 	return num;
455 #endif
456 }
457 
458 
459 static inline uint32_t
460 aligned_read32ne(const uint8_t *buf)
461 {
462 #if defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING) \
463 		|| defined(TUKLIB_USE_UNSAFE_ALIGNED_READS)
464 	return *(const uint32_t *)buf;
465 #else
466 	uint32_t num;
467 	tuklib_memcpy_aligned(&num, buf, sizeof(num));
468 	return num;
469 #endif
470 }
471 
472 
473 static inline uint64_t
474 aligned_read64ne(const uint8_t *buf)
475 {
476 #if defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING) \
477 		|| defined(TUKLIB_USE_UNSAFE_ALIGNED_READS)
478 	return *(const uint64_t *)buf;
479 #else
480 	uint64_t num;
481 	tuklib_memcpy_aligned(&num, buf, sizeof(num));
482 	return num;
483 #endif
484 }
485 
486 
487 static inline void
488 aligned_write16ne(uint8_t *buf, uint16_t num)
489 {
490 #ifdef TUKLIB_USE_UNSAFE_TYPE_PUNNING
491 	*(uint16_t *)buf = num;
492 #else
493 	tuklib_memcpy_aligned(buf, &num, sizeof(num));
494 #endif
495 	return;
496 }
497 
498 
499 static inline void
500 aligned_write32ne(uint8_t *buf, uint32_t num)
501 {
502 #ifdef TUKLIB_USE_UNSAFE_TYPE_PUNNING
503 	*(uint32_t *)buf = num;
504 #else
505 	tuklib_memcpy_aligned(buf, &num, sizeof(num));
506 #endif
507 	return;
508 }
509 
510 
511 static inline void
512 aligned_write64ne(uint8_t *buf, uint64_t num)
513 {
514 #ifdef TUKLIB_USE_UNSAFE_TYPE_PUNNING
515 	*(uint64_t *)buf = num;
516 #else
517 	tuklib_memcpy_aligned(buf, &num, sizeof(num));
518 #endif
519 	return;
520 }
521 
522 
523 static inline uint16_t
524 aligned_read16be(const uint8_t *buf)
525 {
526 	uint16_t num = aligned_read16ne(buf);
527 	return conv16be(num);
528 }
529 
530 
531 static inline uint16_t
532 aligned_read16le(const uint8_t *buf)
533 {
534 	uint16_t num = aligned_read16ne(buf);
535 	return conv16le(num);
536 }
537 
538 
539 static inline uint32_t
540 aligned_read32be(const uint8_t *buf)
541 {
542 	uint32_t num = aligned_read32ne(buf);
543 	return conv32be(num);
544 }
545 
546 
547 static inline uint32_t
548 aligned_read32le(const uint8_t *buf)
549 {
550 	uint32_t num = aligned_read32ne(buf);
551 	return conv32le(num);
552 }
553 
554 
555 static inline uint64_t
556 aligned_read64be(const uint8_t *buf)
557 {
558 	uint64_t num = aligned_read64ne(buf);
559 	return conv64be(num);
560 }
561 
562 
563 static inline uint64_t
564 aligned_read64le(const uint8_t *buf)
565 {
566 	uint64_t num = aligned_read64ne(buf);
567 	return conv64le(num);
568 }
569 
570 
571 // These need to be macros like in the unaligned case.
572 #define aligned_write16be(buf, num) aligned_write16ne((buf), conv16be(num))
573 #define aligned_write16le(buf, num) aligned_write16ne((buf), conv16le(num))
574 #define aligned_write32be(buf, num) aligned_write32ne((buf), conv32be(num))
575 #define aligned_write32le(buf, num) aligned_write32ne((buf), conv32le(num))
576 #define aligned_write64be(buf, num) aligned_write64ne((buf), conv64be(num))
577 #define aligned_write64le(buf, num) aligned_write64ne((buf), conv64le(num))
578 
579 
580 ////////////////////
581 // Bit operations //
582 ////////////////////
583 
584 static inline uint32_t
585 bsr32(uint32_t n)
586 {
587 	// Check for ICC first, since it tends to define __GNUC__ too.
588 #if defined(__INTEL_COMPILER)
589 	return _bit_scan_reverse(n);
590 
591 #elif TUKLIB_GNUC_REQ(3, 4) && UINT_MAX == UINT32_MAX
592 	// GCC >= 3.4 has __builtin_clz(), which gives good results on
593 	// multiple architectures. On x86, __builtin_clz() ^ 31U becomes
594 	// either plain BSR (so the XOR gets optimized away) or LZCNT and
595 	// XOR (if -march indicates that SSE4a instructions are supported).
596 	return (uint32_t)__builtin_clz(n) ^ 31U;
597 
598 #elif defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
599 	uint32_t i;
600 	__asm__("bsrl %1, %0" : "=r" (i) : "rm" (n));
601 	return i;
602 
603 #elif defined(_MSC_VER)
604 	unsigned long i;
605 	_BitScanReverse(&i, n);
606 	return i;
607 
608 #else
609 	uint32_t i = 31;
610 
611 	if ((n & 0xFFFF0000) == 0) {
612 		n <<= 16;
613 		i = 15;
614 	}
615 
616 	if ((n & 0xFF000000) == 0) {
617 		n <<= 8;
618 		i -= 8;
619 	}
620 
621 	if ((n & 0xF0000000) == 0) {
622 		n <<= 4;
623 		i -= 4;
624 	}
625 
626 	if ((n & 0xC0000000) == 0) {
627 		n <<= 2;
628 		i -= 2;
629 	}
630 
631 	if ((n & 0x80000000) == 0)
632 		--i;
633 
634 	return i;
635 #endif
636 }
637 
638 
639 static inline uint32_t
640 clz32(uint32_t n)
641 {
642 #if defined(__INTEL_COMPILER)
643 	return _bit_scan_reverse(n) ^ 31U;
644 
645 #elif TUKLIB_GNUC_REQ(3, 4) && UINT_MAX == UINT32_MAX
646 	return (uint32_t)__builtin_clz(n);
647 
648 #elif defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
649 	uint32_t i;
650 	__asm__("bsrl %1, %0\n\t"
651 		"xorl $31, %0"
652 		: "=r" (i) : "rm" (n));
653 	return i;
654 
655 #elif defined(_MSC_VER)
656 	unsigned long i;
657 	_BitScanReverse(&i, n);
658 	return i ^ 31U;
659 
660 #else
661 	uint32_t i = 0;
662 
663 	if ((n & 0xFFFF0000) == 0) {
664 		n <<= 16;
665 		i = 16;
666 	}
667 
668 	if ((n & 0xFF000000) == 0) {
669 		n <<= 8;
670 		i += 8;
671 	}
672 
673 	if ((n & 0xF0000000) == 0) {
674 		n <<= 4;
675 		i += 4;
676 	}
677 
678 	if ((n & 0xC0000000) == 0) {
679 		n <<= 2;
680 		i += 2;
681 	}
682 
683 	if ((n & 0x80000000) == 0)
684 		++i;
685 
686 	return i;
687 #endif
688 }
689 
690 
691 static inline uint32_t
692 ctz32(uint32_t n)
693 {
694 #if defined(__INTEL_COMPILER)
695 	return _bit_scan_forward(n);
696 
697 #elif TUKLIB_GNUC_REQ(3, 4) && UINT_MAX >= UINT32_MAX
698 	return (uint32_t)__builtin_ctz(n);
699 
700 #elif defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
701 	uint32_t i;
702 	__asm__("bsfl %1, %0" : "=r" (i) : "rm" (n));
703 	return i;
704 
705 #elif defined(_MSC_VER)
706 	unsigned long i;
707 	_BitScanForward(&i, n);
708 	return i;
709 
710 #else
711 	uint32_t i = 0;
712 
713 	if ((n & 0x0000FFFF) == 0) {
714 		n >>= 16;
715 		i = 16;
716 	}
717 
718 	if ((n & 0x000000FF) == 0) {
719 		n >>= 8;
720 		i += 8;
721 	}
722 
723 	if ((n & 0x0000000F) == 0) {
724 		n >>= 4;
725 		i += 4;
726 	}
727 
728 	if ((n & 0x00000003) == 0) {
729 		n >>= 2;
730 		i += 2;
731 	}
732 
733 	if ((n & 0x00000001) == 0)
734 		++i;
735 
736 	return i;
737 #endif
738 }
739 
740 #define bsf32 ctz32
741 
742 #endif
743