Ruby 3.0.5p211 (2022-11-24 revision ba5cf0f7c52d4d35cc6a173c89eda98ceffa2dcf)
bits.h
Go to the documentation of this file.
1#ifndef INTERNAL_BITS_H /*-*-C-*-vi:se ft=c:*/
2#define INTERNAL_BITS_H
30#include <limits.h> /* for CHAR_BITS */
31#include <stdint.h> /* for uintptr_t */
32#include "internal/compilers.h" /* for MSC_VERSION_SINCE */
33
34#if MSC_VERSION_SINCE(1310)
35# include <stdlib.h> /* for _byteswap_uint64 */
36#endif
37
38#if defined(HAVE_X86INTRIN_H) && ! defined(MJIT_HEADER)
39# /* Rule out MJIT_HEADER, which does not interface well with <immintrin.h> */
40# include <x86intrin.h> /* for _lzcnt_u64 */
41#elif MSC_VERSION_SINCE(1310)
42# include <intrin.h> /* for the following intrinsics */
43#endif
44
45#if defined(_MSC_VER) && defined(__AVX__)
46# pragma intrinsic(__popcnt)
47# pragma intrinsic(__popcnt64)
48#endif
49
50#if defined(_MSC_VER) && defined(__AVX2__)
51# pragma intrinsic(__lzcnt)
52# pragma intrinsic(__lzcnt64)
53#endif
54
55#if MSC_VERSION_SINCE(1310)
56# pragma intrinsic(_rotl)
57# pragma intrinsic(_rotr)
58# ifdef _WIN64
59# pragma intrinsic(_rotl64)
60# pragma intrinsic(_rotr64)
61# endif
62#endif
63
64#if MSC_VERSION_SINCE(1400)
65# pragma intrinsic(_BitScanForward)
66# pragma intrinsic(_BitScanReverse)
67# ifdef _WIN64
68# pragma intrinsic(_BitScanForward64)
69# pragma intrinsic(_BitScanReverse64)
70# endif
71#endif
72
73#include "ruby/ruby.h" /* for VALUE */
74#include "internal/static_assert.h" /* for STATIC_ASSERT */
75
76/* The most significant bit of the lower part of half-long integer.
77 * If sizeof(long) == 4, this is 0x8000.
78 * If sizeof(long) == 8, this is 0x80000000.
79 */
80#define HALF_LONG_MSB ((SIGNED_VALUE)1<<((SIZEOF_LONG*CHAR_BIT-1)/2))
81
82#define SIGNED_INTEGER_TYPE_P(T) (0 > ((T)0)-1)
83
84#define SIGNED_INTEGER_MIN(T) \
85 ((sizeof(T) == sizeof(int8_t)) ? ((T)INT8_MIN) : \
86 ((sizeof(T) == sizeof(int16_t)) ? ((T)INT16_MIN) : \
87 ((sizeof(T) == sizeof(int32_t)) ? ((T)INT32_MIN) : \
88 ((sizeof(T) == sizeof(int64_t)) ? ((T)INT64_MIN) : \
89 0))))
90
91#define SIGNED_INTEGER_MAX(T) ((T)(SIGNED_INTEGER_MIN(T) ^ ((T)~(T)0)))
92
93#define UNSIGNED_INTEGER_MAX(T) ((T)~(T)0)
94
95#if __has_builtin(__builtin_mul_overflow_p)
96# define MUL_OVERFLOW_P(a, b) \
97 __builtin_mul_overflow_p((a), (b), (__typeof__(a * b))0)
98#elif __has_builtin(__builtin_mul_overflow)
99# define MUL_OVERFLOW_P(a, b) \
100 __extension__ ({ __typeof__(a) c; __builtin_mul_overflow((a), (b), &c); })
101#endif
102
103#define MUL_OVERFLOW_SIGNED_INTEGER_P(a, b, min, max) ( \
104 (a) == 0 ? 0 : \
105 (a) == -1 ? (b) < -(max) : \
106 (a) > 0 ? \
107 ((b) > 0 ? (max) / (a) < (b) : (min) / (a) > (b)) : \
108 ((b) > 0 ? (min) / (a) < (b) : (max) / (a) > (b)))
109
110#if __has_builtin(__builtin_mul_overflow_p)
111/* __builtin_mul_overflow_p can take bitfield */
112/* and GCC permits bitfields for integers other than int */
113# define MUL_OVERFLOW_FIXNUM_P(a, b) \
114 __extension__ ({ \
115 struct { long fixnum : sizeof(long) * CHAR_BIT - 1; } c = { 0 }; \
116 __builtin_mul_overflow_p((a), (b), c.fixnum); \
117 })
118#else
119# define MUL_OVERFLOW_FIXNUM_P(a, b) \
120 MUL_OVERFLOW_SIGNED_INTEGER_P(a, b, FIXNUM_MIN, FIXNUM_MAX)
121#endif
122
123#ifdef MUL_OVERFLOW_P
124# define MUL_OVERFLOW_LONG_LONG_P(a, b) MUL_OVERFLOW_P(a, b)
125# define MUL_OVERFLOW_LONG_P(a, b) MUL_OVERFLOW_P(a, b)
126# define MUL_OVERFLOW_INT_P(a, b) MUL_OVERFLOW_P(a, b)
127#else
128# define MUL_OVERFLOW_LONG_LONG_P(a, b) MUL_OVERFLOW_SIGNED_INTEGER_P(a, b, LLONG_MIN, LLONG_MAX)
129# define MUL_OVERFLOW_LONG_P(a, b) MUL_OVERFLOW_SIGNED_INTEGER_P(a, b, LONG_MIN, LONG_MAX)
130# define MUL_OVERFLOW_INT_P(a, b) MUL_OVERFLOW_SIGNED_INTEGER_P(a, b, INT_MIN, INT_MAX)
131#endif
132
133#ifdef HAVE_UINT128_T
134# define bit_length(x) \
135 (unsigned int) \
136 (sizeof(x) <= sizeof(int32_t) ? 32 - nlz_int32((uint32_t)(x)) : \
137 sizeof(x) <= sizeof(int64_t) ? 64 - nlz_int64((uint64_t)(x)) : \
138 128 - nlz_int128((uint128_t)(x)))
139#else
140# define bit_length(x) \
141 (unsigned int) \
142 (sizeof(x) <= sizeof(int32_t) ? 32 - nlz_int32((uint32_t)(x)) : \
143 64 - nlz_int64((uint64_t)(x)))
144#endif
145
146#ifndef swap16
147# define swap16 ruby_swap16
148#endif
149
150#ifndef swap32
151# define swap32 ruby_swap32
152#endif
153
154#ifndef swap64
155# define swap64 ruby_swap64
156#endif
157
158static inline uint16_t ruby_swap16(uint16_t);
159static inline uint32_t ruby_swap32(uint32_t);
160static inline uint64_t ruby_swap64(uint64_t);
161static inline unsigned nlz_int(unsigned x);
162static inline unsigned nlz_long(unsigned long x);
163static inline unsigned nlz_long_long(unsigned long long x);
164static inline unsigned nlz_intptr(uintptr_t x);
165static inline unsigned nlz_int32(uint32_t x);
166static inline unsigned nlz_int64(uint64_t x);
167#ifdef HAVE_UINT128_T
168static inline unsigned nlz_int128(uint128_t x);
169#endif
170static inline unsigned rb_popcount32(uint32_t x);
171static inline unsigned rb_popcount64(uint64_t x);
172static inline unsigned rb_popcount_intptr(uintptr_t x);
173static inline int ntz_int32(uint32_t x);
174static inline int ntz_int64(uint64_t x);
175static inline int ntz_intptr(uintptr_t x);
176static inline VALUE RUBY_BIT_ROTL(VALUE, int);
177static inline VALUE RUBY_BIT_ROTR(VALUE, int);
178
179static inline uint16_t
180ruby_swap16(uint16_t x)
181{
182#if __has_builtin(__builtin_bswap16)
183 return __builtin_bswap16(x);
184
185#elif MSC_VERSION_SINCE(1310)
186 return _byteswap_ushort(x);
187
188#else
189 return (x << 8) | (x >> 8);
190
191#endif
192}
193
194static inline uint32_t
195ruby_swap32(uint32_t x)
196{
197#if __has_builtin(__builtin_bswap32)
198 return __builtin_bswap32(x);
199
200#elif MSC_VERSION_SINCE(1310)
201 return _byteswap_ulong(x);
202
203#else
204 x = ((x & 0x0000FFFF) << 16) | ((x & 0xFFFF0000) >> 16);
205 x = ((x & 0x00FF00FF) << 8) | ((x & 0xFF00FF00) >> 8);
206 return x;
207
208#endif
209}
210
211static inline uint64_t
212ruby_swap64(uint64_t x)
213{
214#if __has_builtin(__builtin_bswap64)
215 return __builtin_bswap64(x);
216
217#elif MSC_VERSION_SINCE(1310)
218 return _byteswap_uint64(x);
219
220#else
221 x = ((x & 0x00000000FFFFFFFFULL) << 32) | ((x & 0xFFFFFFFF00000000ULL) >> 32);
222 x = ((x & 0x0000FFFF0000FFFFULL) << 16) | ((x & 0xFFFF0000FFFF0000ULL) >> 16);
223 x = ((x & 0x00FF00FF00FF00FFULL) << 8) | ((x & 0xFF00FF00FF00FF00ULL) >> 8);
224 return x;
225
226#endif
227}
228
229static inline unsigned int
230nlz_int32(uint32_t x)
231{
232#if defined(_MSC_VER) && defined(__AVX2__)
233 /* Note: It seems there is no such thing like __LZCNT__ predefined in MSVC.
234 * AMD CPUs have had this instruction for decades (since K10) but for
235 * Intel, Haswell is the oldest one. We need to use __AVX2__ for maximum
236 * safety. */
237 return (unsigned int)__lzcnt(x);
238
239#elif defined(__x86_64__) && defined(__LZCNT__) && ! defined(MJIT_HEADER)
240 return (unsigned int)_lzcnt_u32(x);
241
242#elif MSC_VERSION_SINCE(1400) /* &&! defined(__AVX2__) */
243 unsigned long r;
244 return _BitScanReverse(&r, x) ? (31 - (int)r) : 32;
245
246#elif __has_builtin(__builtin_clz)
247 STATIC_ASSERT(sizeof_int, sizeof(int) * CHAR_BIT == 32);
248 return x ? (unsigned int)__builtin_clz(x) : 32;
249
250#else
251 uint32_t y;
252 unsigned n = 32;
253 y = x >> 16; if (y) {n -= 16; x = y;}
254 y = x >> 8; if (y) {n -= 8; x = y;}
255 y = x >> 4; if (y) {n -= 4; x = y;}
256 y = x >> 2; if (y) {n -= 2; x = y;}
257 y = x >> 1; if (y) {return n - 2;}
258 return (unsigned int)(n - x);
259#endif
260}
261
262static inline unsigned int
263nlz_int64(uint64_t x)
264{
265#if defined(_MSC_VER) && defined(__AVX2__)
266 return (unsigned int)__lzcnt64(x);
267
268#elif defined(__x86_64__) && defined(__LZCNT__) && ! defined(MJIT_HEADER)
269 return (unsigned int)_lzcnt_u64(x);
270
271#elif defined(_WIN64) && MSC_VERSION_SINCE(1400) /* &&! defined(__AVX2__) */
272 unsigned long r;
273 return _BitScanReverse64(&r, x) ? (63u - (unsigned int)r) : 64;
274
275#elif __has_builtin(__builtin_clzl)
276 if (x == 0) {
277 return 64;
278 }
279 else if (sizeof(long) * CHAR_BIT == 64) {
280 return (unsigned int)__builtin_clzl((unsigned long)x);
281 }
282 else if (sizeof(long long) * CHAR_BIT == 64) {
283 return (unsigned int)__builtin_clzll((unsigned long long)x);
284 }
285 else {
286 /* :FIXME: Is there a way to make this branch a compile-time error? */
288 }
289
290#else
291 uint64_t y;
292 unsigned int n = 64;
293 y = x >> 32; if (y) {n -= 32; x = y;}
294 y = x >> 16; if (y) {n -= 16; x = y;}
295 y = x >> 8; if (y) {n -= 8; x = y;}
296 y = x >> 4; if (y) {n -= 4; x = y;}
297 y = x >> 2; if (y) {n -= 2; x = y;}
298 y = x >> 1; if (y) {return n - 2;}
299 return (unsigned int)(n - x);
300
301#endif
302}
303
304#ifdef HAVE_UINT128_T
305static inline unsigned int
306nlz_int128(uint128_t x)
307{
308 uint64_t y = (uint64_t)(x >> 64);
309
310 if (x == 0) {
311 return 128;
312 }
313 else if (y == 0) {
314 return (unsigned int)nlz_int64(x) + 64;
315 }
316 else {
317 return (unsigned int)nlz_int64(y);
318 }
319}
320#endif
321
322static inline unsigned int
323nlz_int(unsigned int x)
324{
325 if (sizeof(unsigned int) * CHAR_BIT == 32) {
326 return nlz_int32((uint32_t)x);
327 }
328 else if (sizeof(unsigned int) * CHAR_BIT == 64) {
329 return nlz_int64((uint64_t)x);
330 }
331 else {
333 }
334}
335
336static inline unsigned int
337nlz_long(unsigned long x)
338{
339 if (sizeof(unsigned long) * CHAR_BIT == 32) {
340 return nlz_int32((uint32_t)x);
341 }
342 else if (sizeof(unsigned long) * CHAR_BIT == 64) {
343 return nlz_int64((uint64_t)x);
344 }
345 else {
347 }
348}
349
350static inline unsigned int
351nlz_long_long(unsigned long long x)
352{
353 if (sizeof(unsigned long long) * CHAR_BIT == 64) {
354 return nlz_int64((uint64_t)x);
355 }
356#ifdef HAVE_UINT128_T
357 else if (sizeof(unsigned long long) * CHAR_BIT == 128) {
358 return nlz_int128((uint128_t)x);
359 }
360#endif
361 else {
363 }
364}
365
366static inline unsigned int
367nlz_intptr(uintptr_t x)
368{
369 if (sizeof(uintptr_t) == sizeof(unsigned int)) {
370 return nlz_int((unsigned int)x);
371 }
372 if (sizeof(uintptr_t) == sizeof(unsigned long)) {
373 return nlz_long((unsigned long)x);
374 }
375 if (sizeof(uintptr_t) == sizeof(unsigned long long)) {
376 return nlz_long_long((unsigned long long)x);
377 }
378 else {
380 }
381}
382
383static inline unsigned int
384rb_popcount32(uint32_t x)
385{
386#if defined(_MSC_VER) && defined(__AVX__)
387 /* Note: CPUs since Nehalem and Barcelona have had this instruction so SSE
388 * 4.2 should suffice, but it seems there is no such thing like __SSE_4_2__
389 * predefined macro in MSVC. They do have __AVX__ so use it instead. */
390 return (unsigned int)__popcnt(x);
391
392#elif __has_builtin(__builtin_popcount)
393 STATIC_ASSERT(sizeof_int, sizeof(int) * CHAR_BIT >= 32);
394 return (unsigned int)__builtin_popcount(x);
395
396#else
397 x = (x & 0x55555555) + (x >> 1 & 0x55555555);
398 x = (x & 0x33333333) + (x >> 2 & 0x33333333);
399 x = (x & 0x0f0f0f0f) + (x >> 4 & 0x0f0f0f0f);
400 x = (x & 0x001f001f) + (x >> 8 & 0x001f001f);
401 x = (x & 0x0000003f) + (x >>16 & 0x0000003f);
402 return (unsigned int)x;
403
404#endif
405}
406
407static inline unsigned int
408rb_popcount64(uint64_t x)
409{
410#if defined(_MSC_VER) && defined(__AVX__)
411 return (unsigned int)__popcnt64(x);
412
413#elif __has_builtin(__builtin_popcount)
414 if (sizeof(long) * CHAR_BIT == 64) {
415 return (unsigned int)__builtin_popcountl((unsigned long)x);
416 }
417 else if (sizeof(long long) * CHAR_BIT == 64) {
418 return (unsigned int)__builtin_popcountll((unsigned long long)x);
419 }
420 else {
421 /* :FIXME: Is there a way to make this branch a compile-time error? */
423 }
424
425#else
426 x = (x & 0x5555555555555555) + (x >> 1 & 0x5555555555555555);
427 x = (x & 0x3333333333333333) + (x >> 2 & 0x3333333333333333);
428 x = (x & 0x0707070707070707) + (x >> 4 & 0x0707070707070707);
429 x = (x & 0x001f001f001f001f) + (x >> 8 & 0x001f001f001f001f);
430 x = (x & 0x0000003f0000003f) + (x >>16 & 0x0000003f0000003f);
431 x = (x & 0x000000000000007f) + (x >>32 & 0x000000000000007f);
432 return (unsigned int)x;
433
434#endif
435}
436
437static inline unsigned int
438rb_popcount_intptr(uintptr_t x)
439{
440 if (sizeof(uintptr_t) * CHAR_BIT == 64) {
441 return rb_popcount64((uint64_t)x);
442 }
443 else if (sizeof(uintptr_t) * CHAR_BIT == 32) {
444 return rb_popcount32((uint32_t)x);
445 }
446 else {
448 }
449}
450
451static inline int
452ntz_int32(uint32_t x)
453{
454#if defined(__x86_64__) && defined(__BMI__) && ! defined(MJIT_HEADER)
455 return (unsigned)_tzcnt_u32(x);
456
457#elif MSC_VERSION_SINCE(1400)
458 /* :FIXME: Is there any way to issue TZCNT instead of BSF, apart from using
459 * assembly? Because issuing LZCNT seems possible (see nlz.h). */
460 unsigned long r;
461 return _BitScanForward(&r, x) ? (int)r : 32;
462
463#elif __has_builtin(__builtin_ctz)
464 STATIC_ASSERT(sizeof_int, sizeof(int) * CHAR_BIT == 32);
465 return x ? (unsigned)__builtin_ctz(x) : 32;
466
467#else
468 return rb_popcount32((~x) & (x-1));
469
470#endif
471}
472
473static inline int
474ntz_int64(uint64_t x)
475{
476#if defined(__x86_64__) && defined(__BMI__) && ! defined(MJIT_HEADER)
477 return (unsigned)_tzcnt_u64(x);
478
479#elif defined(_WIN64) && MSC_VERSION_SINCE(1400)
480 unsigned long r;
481 return _BitScanForward64(&r, x) ? (int)r : 64;
482
483#elif __has_builtin(__builtin_ctzl)
484 if (x == 0) {
485 return 64;
486 }
487 else if (sizeof(long) * CHAR_BIT == 64) {
488 return (unsigned)__builtin_ctzl((unsigned long)x);
489 }
490 else if (sizeof(long long) * CHAR_BIT == 64) {
491 return (unsigned)__builtin_ctzll((unsigned long long)x);
492 }
493 else {
494 /* :FIXME: Is there a way to make this branch a compile-time error? */
496 }
497
498#else
499 return rb_popcount64((~x) & (x-1));
500
501#endif
502}
503
504static inline int
505ntz_intptr(uintptr_t x)
506{
507 if (sizeof(uintptr_t) * CHAR_BIT == 64) {
508 return ntz_int64((uint64_t)x);
509 }
510 else if (sizeof(uintptr_t) * CHAR_BIT == 32) {
511 return ntz_int32((uint32_t)x);
512 }
513 else {
515 }
516}
517
518static inline VALUE
519RUBY_BIT_ROTL(VALUE v, int n)
520{
521#if __has_builtin(__builtin_rotateleft32) && (SIZEOF_VALUE * CHAR_BIT == 32)
522 return __builtin_rotateleft32(v, n);
523
524#elif __has_builtin(__builtin_rotateleft64) && (SIZEOF_VALUE * CHAR_BIT == 64)
525 return __builtin_rotateleft64(v, n);
526
527#elif MSC_VERSION_SINCE(1310) && (SIZEOF_VALUE * CHAR_BIT == 32)
528 return _rotl(v, n);
529
530#elif MSC_VERSION_SINCE(1310) && (SIZEOF_VALUE * CHAR_BIT == 64)
531 return _rotl64(v, n);
532
533#elif defined(_lrotl) && (SIZEOF_VALUE == SIZEOF_LONG)
534 return _lrotl(v, n);
535
536#else
537 const int m = (sizeof(VALUE) * CHAR_BIT) - 1;
538 return (v << (n & m)) | (v >> (-n & m));
539#endif
540}
541
542static inline VALUE
543RUBY_BIT_ROTR(VALUE v, int n)
544{
545#if __has_builtin(__builtin_rotateright32) && (SIZEOF_VALUE * CHAR_BIT == 32)
546 return __builtin_rotateright32(v, n);
547
548#elif __has_builtin(__builtin_rotateright64) && (SIZEOF_VALUE * CHAR_BIT == 64)
549 return __builtin_rotateright64(v, n);
550
551#elif MSC_VERSION_SINCE(1310) && (SIZEOF_VALUE * CHAR_BIT == 32)
552 return _rotr(v, n);
553
554#elif MSC_VERSION_SINCE(1310) && (SIZEOF_VALUE * CHAR_BIT == 64)
555 return _rotr64(v, n);
556
557#elif defined(_lrotr) && (SIZEOF_VALUE == SIZEOF_LONG)
558 return _lrotr(v, n);
559
560#else
561 const int m = (sizeof(VALUE) * CHAR_BIT) - 1;
562 return (v << (-n & m)) | (v >> (n & m));
563#endif
564}
565
566#endif /* INTERNAL_BITS_H */
#define UNREACHABLE_RETURN
Definition: assume.h:31
Internal header absorbing C compipler differences.
Thin wrapper to ruby/config.h.
#define STATIC_ASSERT
Definition: static_assert.h:14
typedef int(ZCALLBACK *close_file_func) OF((voidpf opaque
Historical shim for <limits.h>.
#define CHAR_BIT
Definition: limits.h:44
unsigned int uint32_t
Definition: sha2.h:101
unsigned long long uint64_t
Definition: sha2.h:102
#define uint64_t
Definition: siphash.h:15
unsigned long VALUE
Definition: value.h:38
unsigned int uintptr_t
Definition: win32.h:106