1/*===---- bmiintrin.h - BMI intrinsics -------------------------------------=== 2 * 3 * Permission is hereby granted, free of charge, to any person obtaining a copy 4 * of this software and associated documentation files (the "Software"), to deal 5 * in the Software without restriction, including without limitation the rights 6 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 7 * copies of the Software, and to permit persons to whom the Software is 8 * furnished to do so, subject to the following conditions: 9 * 10 * The above copyright notice and this permission notice shall be included in 11 * all copies or substantial portions of the Software. 12 * 13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 15 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 16 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 17 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 18 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 19 * THE SOFTWARE. 20 * 21 *===-----------------------------------------------------------------------=== 22 */ 23 24#if !defined __X86INTRIN_H && !defined __IMMINTRIN_H 25#error "Never use <bmiintrin.h> directly; include <x86intrin.h> instead." 26#endif 27 28#ifndef __BMIINTRIN_H 29#define __BMIINTRIN_H 30 31/// \brief Counts the number of trailing zero bits in the operand. 32/// 33/// \headerfile <x86intrin.h> 34/// 35/// \code 36/// unsigned short _tzcnt_u16(unsigned short a); 37/// \endcode 38/// 39/// This intrinsic corresponds to the \c TZCNT instruction. 40/// 41/// \param a 42/// An unsigned 16-bit integer whose trailing zeros are to be counted. 43/// \returns An unsigned 16-bit integer containing the number of trailing zero 44/// bits in the operand. 45#define _tzcnt_u16(a) (__tzcnt_u16((a))) 46 47/// \brief Performs a bitwise AND of the second operand with the one's 48/// complement of the first operand. 49/// 50/// \headerfile <x86intrin.h> 51/// 52/// \code 53/// unsigned int _andn_u32(unsigned int a, unsigned int b); 54/// \endcode 55/// 56/// This intrinsic corresponds to the \c ANDN instruction. 57/// 58/// \param a 59/// An unsigned integer containing one of the operands. 60/// \param b 61/// An unsigned integer containing one of the operands. 62/// \returns An unsigned integer containing the bitwise AND of the second 63/// operand with the one's complement of the first operand. 64#define _andn_u32(a, b) (__andn_u32((a), (b))) 65 66/* _bextr_u32 != __bextr_u32 */ 67/// \brief Clears all bits in the source except for the least significant bit 68/// containing a value of 1 and returns the result. 69/// 70/// \headerfile <x86intrin.h> 71/// 72/// \code 73/// unsigned int _blsi_u32(unsigned int a); 74/// \endcode 75/// 76/// This intrinsic corresponds to the \c BLSI instruction. 77/// 78/// \param a 79/// An unsigned integer whose bits are to be cleared. 80/// \returns An unsigned integer containing the result of clearing the bits from 81/// the source operand. 82#define _blsi_u32(a) (__blsi_u32((a))) 83 84/// \brief Creates a mask whose bits are set to 1, using bit 0 up to and 85/// including the least siginificant bit that is set to 1 in the source 86/// operand and returns the result. 87/// 88/// \headerfile <x86intrin.h> 89/// 90/// \code 91/// unsigned int _blsmsk_u32(unsigned int a); 92/// \endcode 93/// 94/// This intrinsic corresponds to the \c BLSMSK instruction. 95/// 96/// \param a 97/// An unsigned integer used to create the mask. 98/// \returns An unsigned integer containing the newly created mask. 99#define _blsmsk_u32(a) (__blsmsk_u32((a))) 100 101/// \brief Clears the least siginificant bit that is set to 1 in the source 102/// operand and returns the result. 103/// 104/// \headerfile <x86intrin.h> 105/// 106/// \code 107/// unsigned int _blsr_u32(unsigned int a); 108/// \endcode 109/// 110/// This intrinsic corresponds to the \c BLSR instruction. 111/// 112/// \param a 113/// An unsigned integer containing the operand to be cleared. 114/// \returns An unsigned integer containing the result of clearing the source 115/// operand. 116#define _blsr_u32(a) (__blsr_u32((a))) 117 118/// \brief Counts the number of trailing zero bits in the operand. 119/// 120/// \headerfile <x86intrin.h> 121/// 122/// \code 123/// unsigned int _tzcnt_u32(unsigned int a); 124/// \endcode 125/// 126/// This intrinsic corresponds to the \c TZCNT instruction. 127/// 128/// \param a 129/// An unsigned 32-bit integer whose trailing zeros are to be counted. 130/// \returns An unsigned 32-bit integer containing the number of trailing zero 131/// bits in the operand. 132#define _tzcnt_u32(a) (__tzcnt_u32((a))) 133 134/* Define the default attributes for the functions in this file. */ 135#define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("bmi"))) 136 137/* Allow using the tzcnt intrinsics even for non-BMI targets. Since the TZCNT 138 instruction behaves as BSF on non-BMI targets, there is code that expects 139 to use it as a potentially faster version of BSF. */ 140#define __RELAXED_FN_ATTRS __attribute__((__always_inline__, __nodebug__)) 141 142/// \brief Counts the number of trailing zero bits in the operand. 143/// 144/// \headerfile <x86intrin.h> 145/// 146/// This intrinsic corresponds to the \c TZCNT instruction. 147/// 148/// \param __X 149/// An unsigned 16-bit integer whose trailing zeros are to be counted. 150/// \returns An unsigned 16-bit integer containing the number of trailing zero 151/// bits in the operand. 152static __inline__ unsigned short __RELAXED_FN_ATTRS 153__tzcnt_u16(unsigned short __X) 154{ 155 return __X ? __builtin_ctzs(__X) : 16; 156} 157 158/// \brief Performs a bitwise AND of the second operand with the one's 159/// complement of the first operand. 160/// 161/// \headerfile <x86intrin.h> 162/// 163/// This intrinsic corresponds to the \c ANDN instruction. 164/// 165/// \param __X 166/// An unsigned integer containing one of the operands. 167/// \param __Y 168/// An unsigned integer containing one of the operands. 169/// \returns An unsigned integer containing the bitwise AND of the second 170/// operand with the one's complement of the first operand. 171static __inline__ unsigned int __DEFAULT_FN_ATTRS 172__andn_u32(unsigned int __X, unsigned int __Y) 173{ 174 return ~__X & __Y; 175} 176 177/* AMD-specified, double-leading-underscore version of BEXTR */ 178/// \brief Extracts the specified bits from the first operand and returns them 179/// in the least significant bits of the result. 180/// 181/// \headerfile <x86intrin.h> 182/// 183/// This intrinsic corresponds to the \c BEXTR instruction. 184/// 185/// \param __X 186/// An unsigned integer whose bits are to be extracted. 187/// \param __Y 188/// An unsigned integer used to specify which bits are extracted. Bits [7:0] 189/// specify the index of the least significant bit. Bits [15:8] specify the 190/// number of bits to be extracted. 191/// \returns An unsigned integer whose least significant bits contain the 192/// extracted bits. 193static __inline__ unsigned int __DEFAULT_FN_ATTRS 194__bextr_u32(unsigned int __X, unsigned int __Y) 195{ 196 return __builtin_ia32_bextr_u32(__X, __Y); 197} 198 199/* Intel-specified, single-leading-underscore version of BEXTR */ 200/// \brief Extracts the specified bits from the first operand and returns them 201/// in the least significant bits of the result. 202/// 203/// \headerfile <x86intrin.h> 204/// 205/// This intrinsic corresponds to the \c BEXTR instruction. 206/// 207/// \param __X 208/// An unsigned integer whose bits are to be extracted. 209/// \param __Y 210/// An unsigned integer used to specify the index of the least significant 211/// bit for the bits to be extracted. Bits [7:0] specify the index. 212/// \param __Z 213/// An unsigned integer used to specify the number of bits to be extracted. 214/// Bits [7:0] specify the number of bits. 215/// \returns An unsigned integer whose least significant bits contain the 216/// extracted bits. 217static __inline__ unsigned int __DEFAULT_FN_ATTRS 218_bextr_u32(unsigned int __X, unsigned int __Y, unsigned int __Z) 219{ 220 return __builtin_ia32_bextr_u32 (__X, ((__Y & 0xff) | ((__Z & 0xff) << 8))); 221} 222 223/// \brief Clears all bits in the source except for the least significant bit 224/// containing a value of 1 and returns the result. 225/// 226/// \headerfile <x86intrin.h> 227/// 228/// This intrinsic corresponds to the \c BLSI instruction. 229/// 230/// \param __X 231/// An unsigned integer whose bits are to be cleared. 232/// \returns An unsigned integer containing the result of clearing the bits from 233/// the source operand. 234static __inline__ unsigned int __DEFAULT_FN_ATTRS 235__blsi_u32(unsigned int __X) 236{ 237 return __X & -__X; 238} 239 240/// \brief Creates a mask whose bits are set to 1, using bit 0 up to and 241/// including the least siginificant bit that is set to 1 in the source 242/// operand and returns the result. 243/// 244/// \headerfile <x86intrin.h> 245/// 246/// This intrinsic corresponds to the \c BLSMSK instruction. 247/// 248/// \param __X 249/// An unsigned integer used to create the mask. 250/// \returns An unsigned integer containing the newly created mask. 251static __inline__ unsigned int __DEFAULT_FN_ATTRS 252__blsmsk_u32(unsigned int __X) 253{ 254 return __X ^ (__X - 1); 255} 256 257/// \brief Clears the least siginificant bit that is set to 1 in the source 258/// operand and returns the result. 259/// 260/// \headerfile <x86intrin.h> 261/// 262/// This intrinsic corresponds to the \c BLSR instruction. 263/// 264/// \param __X 265/// An unsigned integer containing the operand to be cleared. 266/// \returns An unsigned integer containing the result of clearing the source 267/// operand. 268static __inline__ unsigned int __DEFAULT_FN_ATTRS 269__blsr_u32(unsigned int __X) 270{ 271 return __X & (__X - 1); 272} 273 274/// \brief Counts the number of trailing zero bits in the operand. 275/// 276/// \headerfile <x86intrin.h> 277/// 278/// This intrinsic corresponds to the \c TZCNT instruction. 279/// 280/// \param __X 281/// An unsigned 32-bit integer whose trailing zeros are to be counted. 282/// \returns An unsigned 32-bit integer containing the number of trailing zero 283/// bits in the operand. 284static __inline__ unsigned int __RELAXED_FN_ATTRS 285__tzcnt_u32(unsigned int __X) 286{ 287 return __X ? __builtin_ctz(__X) : 32; 288} 289 290/// \brief Counts the number of trailing zero bits in the operand. 291/// 292/// \headerfile <x86intrin.h> 293/// 294/// This intrinsic corresponds to the \c TZCNT instruction. 295/// 296/// \param __X 297/// An unsigned 32-bit integer whose trailing zeros are to be counted. 298/// \returns An 32-bit integer containing the number of trailing zero 299/// bits in the operand. 300static __inline__ int __RELAXED_FN_ATTRS 301_mm_tzcnt_32(unsigned int __X) 302{ 303 return __X ? __builtin_ctz(__X) : 32; 304} 305 306#ifdef __x86_64__ 307 308/// \brief Performs a bitwise AND of the second operand with the one's 309/// complement of the first operand. 310/// 311/// \headerfile <x86intrin.h> 312/// 313/// \code 314/// unsigned long long _andn_u64 (unsigned long long a, unsigned long long b); 315/// \endcode 316/// 317/// This intrinsic corresponds to the \c ANDN instruction. 318/// 319/// \param a 320/// An unsigned 64-bit integer containing one of the operands. 321/// \param b 322/// An unsigned 64-bit integer containing one of the operands. 323/// \returns An unsigned 64-bit integer containing the bitwise AND of the second 324/// operand with the one's complement of the first operand. 325#define _andn_u64(a, b) (__andn_u64((a), (b))) 326 327/* _bextr_u64 != __bextr_u64 */ 328/// \brief Clears all bits in the source except for the least significant bit 329/// containing a value of 1 and returns the result. 330/// 331/// \headerfile <x86intrin.h> 332/// 333/// \code 334/// unsigned long long _blsi_u64(unsigned long long a); 335/// \endcode 336/// 337/// This intrinsic corresponds to the \c BLSI instruction. 338/// 339/// \param a 340/// An unsigned 64-bit integer whose bits are to be cleared. 341/// \returns An unsigned 64-bit integer containing the result of clearing the 342/// bits from the source operand. 343#define _blsi_u64(a) (__blsi_u64((a))) 344 345/// \brief Creates a mask whose bits are set to 1, using bit 0 up to and 346/// including the least siginificant bit that is set to 1 in the source 347/// operand and returns the result. 348/// 349/// \headerfile <x86intrin.h> 350/// 351/// \code 352/// unsigned long long _blsmsk_u64(unsigned long long a); 353/// \endcode 354/// 355/// This intrinsic corresponds to the \c BLSMSK instruction. 356/// 357/// \param a 358/// An unsigned 64-bit integer used to create the mask. 359/// \returns A unsigned 64-bit integer containing the newly created mask. 360#define _blsmsk_u64(a) (__blsmsk_u64((a))) 361 362/// \brief Clears the least siginificant bit that is set to 1 in the source 363/// operand and returns the result. 364/// 365/// \headerfile <x86intrin.h> 366/// 367/// \code 368/// unsigned long long _blsr_u64(unsigned long long a); 369/// \endcode 370/// 371/// This intrinsic corresponds to the \c BLSR instruction. 372/// 373/// \param a 374/// An unsigned 64-bit integer containing the operand to be cleared. 375/// \returns An unsigned 64-bit integer containing the result of clearing the 376/// source operand. 377#define _blsr_u64(a) (__blsr_u64((a))) 378 379/// \brief Counts the number of trailing zero bits in the operand. 380/// 381/// \headerfile <x86intrin.h> 382/// 383/// \code 384/// unsigned long long _tzcnt_u64(unsigned long long a); 385/// \endcode 386/// 387/// This intrinsic corresponds to the \c TZCNT instruction. 388/// 389/// \param a 390/// An unsigned 64-bit integer whose trailing zeros are to be counted. 391/// \returns An unsigned 64-bit integer containing the number of trailing zero 392/// bits in the operand. 393#define _tzcnt_u64(a) (__tzcnt_u64((a))) 394 395/// \brief Performs a bitwise AND of the second operand with the one's 396/// complement of the first operand. 397/// 398/// \headerfile <x86intrin.h> 399/// 400/// This intrinsic corresponds to the \c ANDN instruction. 401/// 402/// \param __X 403/// An unsigned 64-bit integer containing one of the operands. 404/// \param __Y 405/// An unsigned 64-bit integer containing one of the operands. 406/// \returns An unsigned 64-bit integer containing the bitwise AND of the second 407/// operand with the one's complement of the first operand. 408static __inline__ unsigned long long __DEFAULT_FN_ATTRS 409__andn_u64 (unsigned long long __X, unsigned long long __Y) 410{ 411 return ~__X & __Y; 412} 413 414/* AMD-specified, double-leading-underscore version of BEXTR */ 415/// \brief Extracts the specified bits from the first operand and returns them 416/// in the least significant bits of the result. 417/// 418/// \headerfile <x86intrin.h> 419/// 420/// This intrinsic corresponds to the \c BEXTR instruction. 421/// 422/// \param __X 423/// An unsigned 64-bit integer whose bits are to be extracted. 424/// \param __Y 425/// An unsigned 64-bit integer used to specify which bits are extracted. Bits 426/// [7:0] specify the index of the least significant bit. Bits [15:8] specify 427/// the number of bits to be extracted. 428/// \returns An unsigned 64-bit integer whose least significant bits contain the 429/// extracted bits. 430static __inline__ unsigned long long __DEFAULT_FN_ATTRS 431__bextr_u64(unsigned long long __X, unsigned long long __Y) 432{ 433 return __builtin_ia32_bextr_u64(__X, __Y); 434} 435 436/* Intel-specified, single-leading-underscore version of BEXTR */ 437/// \brief Extracts the specified bits from the first operand and returns them 438/// in the least significant bits of the result. 439/// 440/// \headerfile <x86intrin.h> 441/// 442/// This intrinsic corresponds to the \c BEXTR instruction. 443/// 444/// \param __X 445/// An unsigned 64-bit integer whose bits are to be extracted. 446/// \param __Y 447/// An unsigned integer used to specify the index of the least significant 448/// bit for the bits to be extracted. Bits [7:0] specify the index. 449/// \param __Z 450/// An unsigned integer used to specify the number of bits to be extracted. 451/// Bits [7:0] specify the number of bits. 452/// \returns An unsigned 64-bit integer whose least significant bits contain the 453/// extracted bits. 454static __inline__ unsigned long long __DEFAULT_FN_ATTRS 455_bextr_u64(unsigned long long __X, unsigned int __Y, unsigned int __Z) 456{ 457 return __builtin_ia32_bextr_u64 (__X, ((__Y & 0xff) | ((__Z & 0xff) << 8))); 458} 459 460/// \brief Clears all bits in the source except for the least significant bit 461/// containing a value of 1 and returns the result. 462/// 463/// \headerfile <x86intrin.h> 464/// 465/// This intrinsic corresponds to the \c BLSI instruction. 466/// 467/// \param __X 468/// An unsigned 64-bit integer whose bits are to be cleared. 469/// \returns An unsigned 64-bit integer containing the result of clearing the 470/// bits from the source operand. 471static __inline__ unsigned long long __DEFAULT_FN_ATTRS 472__blsi_u64(unsigned long long __X) 473{ 474 return __X & -__X; 475} 476 477/// \brief Creates a mask whose bits are set to 1, using bit 0 up to and 478/// including the least siginificant bit that is set to 1 in the source 479/// operand and returns the result. 480/// 481/// \headerfile <x86intrin.h> 482/// 483/// This intrinsic corresponds to the \c BLSMSK instruction. 484/// 485/// \param __X 486/// An unsigned 64-bit integer used to create the mask. 487/// \returns A unsigned 64-bit integer containing the newly created mask. 488static __inline__ unsigned long long __DEFAULT_FN_ATTRS 489__blsmsk_u64(unsigned long long __X) 490{ 491 return __X ^ (__X - 1); 492} 493 494/// \brief Clears the least siginificant bit that is set to 1 in the source 495/// operand and returns the result. 496/// 497/// \headerfile <x86intrin.h> 498/// 499/// This intrinsic corresponds to the \c BLSR instruction. 500/// 501/// \param __X 502/// An unsigned 64-bit integer containing the operand to be cleared. 503/// \returns An unsigned 64-bit integer containing the result of clearing the 504/// source operand. 505static __inline__ unsigned long long __DEFAULT_FN_ATTRS 506__blsr_u64(unsigned long long __X) 507{ 508 return __X & (__X - 1); 509} 510 511/// \brief Counts the number of trailing zero bits in the operand. 512/// 513/// \headerfile <x86intrin.h> 514/// 515/// This intrinsic corresponds to the \c TZCNT instruction. 516/// 517/// \param __X 518/// An unsigned 64-bit integer whose trailing zeros are to be counted. 519/// \returns An unsigned 64-bit integer containing the number of trailing zero 520/// bits in the operand. 521static __inline__ unsigned long long __RELAXED_FN_ATTRS 522__tzcnt_u64(unsigned long long __X) 523{ 524 return __X ? __builtin_ctzll(__X) : 64; 525} 526 527/// \brief Counts the number of trailing zero bits in the operand. 528/// 529/// \headerfile <x86intrin.h> 530/// 531/// This intrinsic corresponds to the \c TZCNT instruction. 532/// 533/// \param __X 534/// An unsigned 64-bit integer whose trailing zeros are to be counted. 535/// \returns An 64-bit integer containing the number of trailing zero 536/// bits in the operand. 537static __inline__ long long __RELAXED_FN_ATTRS 538_mm_tzcnt_64(unsigned long long __X) 539{ 540 return __X ? __builtin_ctzll(__X) : 64; 541} 542 543#endif /* __x86_64__ */ 544 545#undef __DEFAULT_FN_ATTRS 546#undef __RELAXED_FN_ATTRS 547 548#endif /* __BMIINTRIN_H */ 549