Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
14 | pmbaty | 1 | /*===----------------------- raointintrin.h - RAOINT ------------------------=== |
2 | * |
||
3 | * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
||
4 | * See https://llvm.org/LICENSE.txt for license information. |
||
5 | * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
||
6 | * |
||
7 | *===-----------------------------------------------------------------------=== |
||
8 | */ |
||
9 | |||
10 | #ifndef __X86GPRINTRIN_H |
||
11 | #error "Never use <raointintrin.h> directly; include <x86gprintrin.h> instead." |
||
12 | #endif // __X86GPRINTRIN_H |
||
13 | |||
14 | #ifndef __RAOINTINTRIN_H |
||
15 | #define __RAOINTINTRIN_H |
||
16 | |||
17 | #define __DEFAULT_FN_ATTRS \ |
||
18 | __attribute__((__always_inline__, __nodebug__, __target__("raoint"))) |
||
19 | |||
20 | /// Atomically add a 32-bit value at memory operand \a __A and a 32-bit \a __B, |
||
21 | /// and store the result to the same memory location. |
||
22 | /// |
||
23 | /// This intrinsic should be used for contention or weak ordering. It may |
||
24 | /// result in bad performance for hot data used by single thread only. |
||
25 | /// |
||
26 | /// \headerfile <x86intrin.h> |
||
27 | /// |
||
28 | /// This intrinsic corresponds to the \c AADD instruction. |
||
29 | /// |
||
30 | /// \param __A |
||
31 | /// A pointer to a 32-bit memory location. |
||
32 | /// \param __B |
||
33 | /// A 32-bit integer value. |
||
34 | /// |
||
35 | /// \code{.operation} |
||
36 | /// MEM[__A+31:__A] := MEM[__A+31:__A] + __B[31:0] |
||
37 | /// \endcode |
||
38 | static __inline__ void __DEFAULT_FN_ATTRS _aadd_i32(int *__A, int __B) { |
||
39 | __builtin_ia32_aadd32((int *)__A, __B); |
||
40 | } |
||
41 | |||
42 | /// Atomically and a 32-bit value at memory operand \a __A and a 32-bit \a __B, |
||
43 | /// and store the result to the same memory location. |
||
44 | /// |
||
45 | /// This intrinsic should be used for contention or weak ordering. It may |
||
46 | /// result in bad performance for hot data used by single thread only. |
||
47 | /// |
||
48 | /// \headerfile <x86intrin.h> |
||
49 | /// |
||
50 | /// This intrinsic corresponds to the \c AAND instruction. |
||
51 | /// |
||
52 | /// \param __A |
||
53 | /// A pointer to a 32-bit memory location. |
||
54 | /// \param __B |
||
55 | /// A 32-bit integer value. |
||
56 | /// |
||
57 | /// \code{.operation} |
||
58 | /// MEM[__A+31:__A] := MEM[__A+31:__A] AND __B[31:0] |
||
59 | /// \endcode |
||
60 | static __inline__ void __DEFAULT_FN_ATTRS _aand_i32(int *__A, int __B) { |
||
61 | __builtin_ia32_aand32((int *)__A, __B); |
||
62 | } |
||
63 | |||
64 | /// Atomically or a 32-bit value at memory operand \a __A and a 32-bit \a __B, |
||
65 | /// and store the result to the same memory location. |
||
66 | /// |
||
67 | /// This intrinsic should be used for contention or weak ordering. It may |
||
68 | /// result in bad performance for hot data used by single thread only. |
||
69 | /// |
||
70 | /// \headerfile <x86intrin.h> |
||
71 | /// |
||
72 | /// This intrinsic corresponds to the \c AOR instruction. |
||
73 | /// |
||
74 | /// \param __A |
||
75 | /// A pointer to a 32-bit memory location. |
||
76 | /// \param __B |
||
77 | /// A 32-bit integer value. |
||
78 | /// |
||
79 | /// \code{.operation} |
||
80 | /// MEM[__A+31:__A] := MEM[__A+31:__A] OR __B[31:0] |
||
81 | /// \endcode |
||
82 | static __inline__ void __DEFAULT_FN_ATTRS _aor_i32(int *__A, int __B) { |
||
83 | __builtin_ia32_aor32((int *)__A, __B); |
||
84 | } |
||
85 | |||
86 | /// Atomically xor a 32-bit value at memory operand \a __A and a 32-bit \a __B, |
||
87 | /// and store the result to the same memory location. |
||
88 | /// |
||
89 | /// This intrinsic should be used for contention or weak ordering. It may |
||
90 | /// result in bad performance for hot data used by single thread only. |
||
91 | /// |
||
92 | /// \headerfile <x86intrin.h> |
||
93 | /// |
||
94 | /// This intrinsic corresponds to the \c AXOR instruction. |
||
95 | /// |
||
96 | /// \param __A |
||
97 | /// A pointer to a 32-bit memory location. |
||
98 | /// \param __B |
||
99 | /// A 32-bit integer value. |
||
100 | /// |
||
101 | /// \code{.operation} |
||
102 | /// MEM[__A+31:__A] := MEM[__A+31:__A] XOR __B[31:0] |
||
103 | /// \endcode |
||
104 | static __inline__ void __DEFAULT_FN_ATTRS _axor_i32(int *__A, int __B) { |
||
105 | __builtin_ia32_axor32((int *)__A, __B); |
||
106 | } |
||
107 | |||
108 | #ifdef __x86_64__ |
||
109 | /// Atomically add a 64-bit value at memory operand \a __A and a 64-bit \a __B, |
||
110 | /// and store the result to the same memory location. |
||
111 | /// |
||
112 | /// This intrinsic should be used for contention or weak ordering. It may |
||
113 | /// result in bad performance for hot data used by single thread only. |
||
114 | /// |
||
115 | /// \headerfile <x86intrin.h> |
||
116 | /// |
||
117 | /// This intrinsic corresponds to the \c AADD instruction. |
||
118 | /// |
||
119 | /// \param __A |
||
120 | /// A pointer to a 64-bit memory location. |
||
121 | /// \param __B |
||
122 | /// A 64-bit integer value. |
||
123 | /// |
||
124 | /// \code{.operation} |
||
125 | /// MEM[__A+63:__A] := MEM[__A+63:__A] + __B[63:0] |
||
126 | /// \endcode |
||
127 | static __inline__ void __DEFAULT_FN_ATTRS _aadd_i64(long long *__A, |
||
128 | long long __B) { |
||
129 | __builtin_ia32_aadd64((long long *)__A, __B); |
||
130 | } |
||
131 | |||
132 | /// Atomically and a 64-bit value at memory operand \a __A and a 64-bit \a __B, |
||
133 | /// and store the result to the same memory location. |
||
134 | /// |
||
135 | /// This intrinsic should be used for contention or weak ordering. It may |
||
136 | /// result in bad performance for hot data used by single thread only. |
||
137 | /// |
||
138 | /// \headerfile <x86intrin.h> |
||
139 | /// |
||
140 | /// This intrinsic corresponds to the \c AAND instruction. |
||
141 | /// |
||
142 | /// \param __A |
||
143 | /// A pointer to a 64-bit memory location. |
||
144 | /// \param __B |
||
145 | /// A 64-bit integer value. |
||
146 | /// |
||
147 | /// \code{.operation} |
||
148 | /// MEM[__A+63:__A] := MEM[__A+63:__A] AND __B[63:0] |
||
149 | /// \endcode |
||
150 | static __inline__ void __DEFAULT_FN_ATTRS _aand_i64(long long *__A, |
||
151 | long long __B) { |
||
152 | __builtin_ia32_aand64((long long *)__A, __B); |
||
153 | } |
||
154 | |||
155 | /// Atomically or a 64-bit value at memory operand \a __A and a 64-bit \a __B, |
||
156 | /// and store the result to the same memory location. |
||
157 | /// |
||
158 | /// This intrinsic should be used for contention or weak ordering. It may |
||
159 | /// result in bad performance for hot data used by single thread only. |
||
160 | /// |
||
161 | /// \headerfile <x86intrin.h> |
||
162 | /// |
||
163 | /// This intrinsic corresponds to the \c AOR instruction. |
||
164 | /// |
||
165 | /// \param __A |
||
166 | /// A pointer to a 64-bit memory location. |
||
167 | /// \param __B |
||
168 | /// A 64-bit integer value. |
||
169 | /// |
||
170 | /// \code{.operation} |
||
171 | /// MEM[__A+63:__A] := MEM[__A+63:__A] OR __B[63:0] |
||
172 | /// \endcode |
||
173 | static __inline__ void __DEFAULT_FN_ATTRS _aor_i64(long long *__A, |
||
174 | long long __B) { |
||
175 | __builtin_ia32_aor64((long long *)__A, __B); |
||
176 | } |
||
177 | |||
178 | /// Atomically xor a 64-bit value at memory operand \a __A and a 64-bit \a __B, |
||
179 | /// and store the result to the same memory location. |
||
180 | /// |
||
181 | /// This intrinsic should be used for contention or weak ordering. It may |
||
182 | /// result in bad performance for hot data used by single thread only. |
||
183 | /// |
||
184 | /// \headerfile <x86intrin.h> |
||
185 | /// |
||
186 | /// This intrinsic corresponds to the \c AXOR instruction. |
||
187 | /// |
||
188 | /// \param __A |
||
189 | /// A pointer to a 64-bit memory location. |
||
190 | /// \param __B |
||
191 | /// A 64-bit integer value. |
||
192 | /// |
||
193 | /// \code{.operation} |
||
194 | /// MEM[__A+63:__A] := MEM[__A+63:__A] XOR __B[63:0] |
||
195 | /// \endcode |
||
196 | static __inline__ void __DEFAULT_FN_ATTRS _axor_i64(long long *__A, |
||
197 | long long __B) { |
||
198 | __builtin_ia32_axor64((long long *)__A, __B); |
||
199 | } |
||
200 | #endif // __x86_64__ |
||
201 | |||
202 | #undef __DEFAULT_FN_ATTRS |
||
203 | #endif // __RAOINTINTRIN_H |