11188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker/*===---- bmiintrin.h - BMI intrinsics -------------------------------------=== 21188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * 31188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * Permission is hereby granted, free of charge, to any person obtaining a copy 41188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * of this software and associated documentation files (the "Software"), to deal 51188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * in the Software without restriction, including without limitation the rights 61188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 71188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * copies of the Software, and to permit persons to whom the Software is 81188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * furnished to do so, subject to the following conditions: 91188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * 101188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * The above copyright notice and this permission notice shall be included in 111188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * all copies or substantial portions of the Software. 121188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * 131188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 141188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 151188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 161188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 171188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 181188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 191188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * THE SOFTWARE. 201188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker * 211188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker *===-----------------------------------------------------------------------=== 221188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker */ 231188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 241188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#if !defined __X86INTRIN_H && !defined __IMMINTRIN_H 251188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#error "Never use <bmiintrin.h> directly; include <x86intrin.h> instead." 261188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#endif 271188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 281188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#ifndef __BMIINTRIN_H 291188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#define __BMIINTRIN_H 301188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 311188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#define _tzcnt_u16(a) (__tzcnt_u16((a))) 321188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#define _andn_u32(a, b) (__andn_u32((a), (b))) 331188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker/* _bextr_u32 != __bextr_u32 */ 341188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#define _blsi_u32(a) (__blsi_u32((a))) 351188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#define _blsmsk_u32(a) (__blsmsk_u32((a))) 361188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#define _blsr_u32(a) (__blsr_u32((a))) 371188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#define _tzcnt_u32(a) (__tzcnt_u32((a))) 381188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 391188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker/* Define the default attributes for the functions in this file. */ 401188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("bmi"))) 411188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 421188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker/* Allow using the tzcnt intrinsics even for non-BMI targets. Since the TZCNT 431188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker instruction behaves as BSF on non-BMI targets, there is code that expects 441188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker to use it as a potentially faster version of BSF. */ 451188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#define __RELAXED_FN_ATTRS __attribute__((__always_inline__, __nodebug__)) 461188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 471188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned short __RELAXED_FN_ATTRS 481188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker__tzcnt_u16(unsigned short __X) 491188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 501188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return __X ? __builtin_ctzs(__X) : 16; 511188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 521188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 531188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned int __DEFAULT_FN_ATTRS 541188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker__andn_u32(unsigned int __X, unsigned int __Y) 551188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 561188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return ~__X & __Y; 571188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 581188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 591188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker/* AMD-specified, double-leading-underscore version of BEXTR */ 601188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned int __DEFAULT_FN_ATTRS 611188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker__bextr_u32(unsigned int __X, unsigned int __Y) 621188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 631188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return __builtin_ia32_bextr_u32(__X, __Y); 641188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 651188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 661188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker/* Intel-specified, single-leading-underscore version of BEXTR */ 671188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned int __DEFAULT_FN_ATTRS 681188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker_bextr_u32(unsigned int __X, unsigned int __Y, unsigned int __Z) 691188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 701188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return __builtin_ia32_bextr_u32 (__X, ((__Y & 0xff) | ((__Z & 0xff) << 8))); 711188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 721188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 731188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned int __DEFAULT_FN_ATTRS 741188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker__blsi_u32(unsigned int __X) 751188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 761188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return __X & -__X; 771188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 781188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 791188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned int __DEFAULT_FN_ATTRS 801188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker__blsmsk_u32(unsigned int __X) 811188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 821188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return __X ^ (__X - 1); 831188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 841188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 851188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned int __DEFAULT_FN_ATTRS 861188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker__blsr_u32(unsigned int __X) 871188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 881188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return __X & (__X - 1); 891188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 901188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 911188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned int __RELAXED_FN_ATTRS 921188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker__tzcnt_u32(unsigned int __X) 931188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 941188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return __X ? __builtin_ctz(__X) : 32; 951188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 961188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 971188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#ifdef __x86_64__ 981188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 991188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#define _andn_u64(a, b) (__andn_u64((a), (b))) 1001188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker/* _bextr_u64 != __bextr_u64 */ 1011188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#define _blsi_u64(a) (__blsi_u64((a))) 1021188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#define _blsmsk_u64(a) (__blsmsk_u64((a))) 1031188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#define _blsr_u64(a) (__blsr_u64((a))) 1041188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#define _tzcnt_u64(a) (__tzcnt_u64((a))) 1051188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 1061188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned long long __DEFAULT_FN_ATTRS 1071188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker__andn_u64 (unsigned long long __X, unsigned long long __Y) 1081188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 1091188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return ~__X & __Y; 1101188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 1111188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 1121188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker/* AMD-specified, double-leading-underscore version of BEXTR */ 1131188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned long long __DEFAULT_FN_ATTRS 1141188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker__bextr_u64(unsigned long long __X, unsigned long long __Y) 1151188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 1161188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return __builtin_ia32_bextr_u64(__X, __Y); 1171188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 1181188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 1191188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker/* Intel-specified, single-leading-underscore version of BEXTR */ 1201188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned long long __DEFAULT_FN_ATTRS 1211188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker_bextr_u64(unsigned long long __X, unsigned int __Y, unsigned int __Z) 1221188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 1231188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return __builtin_ia32_bextr_u64 (__X, ((__Y & 0xff) | ((__Z & 0xff) << 8))); 1241188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 1251188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 1261188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned long long __DEFAULT_FN_ATTRS 1271188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker__blsi_u64(unsigned long long __X) 1281188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 1291188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return __X & -__X; 1301188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 1311188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 1321188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned long long __DEFAULT_FN_ATTRS 1331188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker__blsmsk_u64(unsigned long long __X) 1341188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 1351188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return __X ^ (__X - 1); 1361188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 1371188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 1381188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned long long __DEFAULT_FN_ATTRS 1391188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker__blsr_u64(unsigned long long __X) 1401188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 1411188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return __X & (__X - 1); 1421188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 1431188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 1441188dcf30923cb444143ffa4b83dc951037e76agitbuildkickerstatic __inline__ unsigned long long __RELAXED_FN_ATTRS 1451188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker__tzcnt_u64(unsigned long long __X) 1461188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker{ 1471188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker return __X ? __builtin_ctzll(__X) : 64; 1481188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker} 1491188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 1501188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#endif /* __x86_64__ */ 1511188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 1521188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#undef __DEFAULT_FN_ATTRS 1531188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#undef __RELAXED_FN_ATTRS 1541188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker 1551188dcf30923cb444143ffa4b83dc951037e76agitbuildkicker#endif /* __BMIINTRIN_H */ 156