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 __BMI__
29 # error "BMI instruction set not enabled"
30 #endif /* __BMI__ */
31
32 #ifndef __BMIINTRIN_H
33 #define __BMIINTRIN_H
34
35 #define _tzcnt_u16(a) (__tzcnt_u16((a)))
36 #define _andn_u32(a, b) (__andn_u32((a), (b)))
37 /* _bextr_u32 != __bextr_u32 */
38 #define _blsi_u32(a) (__blsi_u32((a)))
39 #define _blsmsk_u32(a) (__blsmsk_u32((a)))
40 #define _blsr_u32(a) (__blsr_u32((a)))
41 #define _tzcnt_u32(a) (__tzcnt_u32((a)))
42
43 static __inline__ unsigned short __attribute__((__always_inline__, __nodebug__))
__tzcnt_u16(unsigned short __X)44 __tzcnt_u16(unsigned short __X)
45 {
46 return __X ? __builtin_ctzs(__X) : 16;
47 }
48
49 static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__))
__andn_u32(unsigned int __X,unsigned int __Y)50 __andn_u32(unsigned int __X, unsigned int __Y)
51 {
52 return ~__X & __Y;
53 }
54
55 /* AMD-specified, double-leading-underscore version of BEXTR */
56 static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__))
__bextr_u32(unsigned int __X,unsigned int __Y)57 __bextr_u32(unsigned int __X, unsigned int __Y)
58 {
59 return __builtin_ia32_bextr_u32(__X, __Y);
60 }
61
62 /* Intel-specified, single-leading-underscore version of BEXTR */
63 static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__))
_bextr_u32(unsigned int __X,unsigned int __Y,unsigned int __Z)64 _bextr_u32(unsigned int __X, unsigned int __Y, unsigned int __Z)
65 {
66 return __builtin_ia32_bextr_u32 (__X, ((__Y & 0xff) | ((__Z & 0xff) << 8)));
67 }
68
69 static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__))
__blsi_u32(unsigned int __X)70 __blsi_u32(unsigned int __X)
71 {
72 return __X & -__X;
73 }
74
75 static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__))
__blsmsk_u32(unsigned int __X)76 __blsmsk_u32(unsigned int __X)
77 {
78 return __X ^ (__X - 1);
79 }
80
81 static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__))
__blsr_u32(unsigned int __X)82 __blsr_u32(unsigned int __X)
83 {
84 return __X & (__X - 1);
85 }
86
87 static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__))
__tzcnt_u32(unsigned int __X)88 __tzcnt_u32(unsigned int __X)
89 {
90 return __X ? __builtin_ctz(__X) : 32;
91 }
92
93 #ifdef __x86_64__
94
95 #define _andn_u64(a, b) (__andn_u64((a), (b)))
96 /* _bextr_u64 != __bextr_u64 */
97 #define _blsi_u64(a) (__blsi_u64((a)))
98 #define _blsmsk_u64(a) (__blsmsk_u64((a)))
99 #define _blsr_u64(a) (__blsr_u64((a)))
100 #define _tzcnt_u64(a) (__tzcnt_u64((a)))
101
102 static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__))
__andn_u64(unsigned long long __X,unsigned long long __Y)103 __andn_u64 (unsigned long long __X, unsigned long long __Y)
104 {
105 return ~__X & __Y;
106 }
107
108 /* AMD-specified, double-leading-underscore version of BEXTR */
109 static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__))
__bextr_u64(unsigned long long __X,unsigned long long __Y)110 __bextr_u64(unsigned long long __X, unsigned long long __Y)
111 {
112 return __builtin_ia32_bextr_u64(__X, __Y);
113 }
114
115 /* Intel-specified, single-leading-underscore version of BEXTR */
116 static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__))
_bextr_u64(unsigned long long __X,unsigned int __Y,unsigned int __Z)117 _bextr_u64(unsigned long long __X, unsigned int __Y, unsigned int __Z)
118 {
119 return __builtin_ia32_bextr_u64 (__X, ((__Y & 0xff) | ((__Z & 0xff) << 8)));
120 }
121
122 static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__))
__blsi_u64(unsigned long long __X)123 __blsi_u64(unsigned long long __X)
124 {
125 return __X & -__X;
126 }
127
128 static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__))
__blsmsk_u64(unsigned long long __X)129 __blsmsk_u64(unsigned long long __X)
130 {
131 return __X ^ (__X - 1);
132 }
133
134 static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__))
__blsr_u64(unsigned long long __X)135 __blsr_u64(unsigned long long __X)
136 {
137 return __X & (__X - 1);
138 }
139
140 static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__))
__tzcnt_u64(unsigned long long __X)141 __tzcnt_u64(unsigned long long __X)
142 {
143 return __X ? __builtin_ctzll(__X) : 64;
144 }
145
146 #endif /* __x86_64__ */
147
148 #endif /* __BMIINTRIN_H */
149