1 /*************************************************************************/
2 /* safe_refcount.cpp */
3 /*************************************************************************/
4 /* This file is part of: */
5 /* GODOT ENGINE */
6 /* https://godotengine.org */
7 /*************************************************************************/
8 /* Copyright (c) 2007-2019 Juan Linietsky, Ariel Manzur. */
9 /* Copyright (c) 2014-2019 Godot Engine contributors (cf. AUTHORS.md) */
10 /* */
11 /* Permission is hereby granted, free of charge, to any person obtaining */
12 /* a copy of this software and associated documentation files (the */
13 /* "Software"), to deal in the Software without restriction, including */
14 /* without limitation the rights to use, copy, modify, merge, publish, */
15 /* distribute, sublicense, and/or sell copies of the Software, and to */
16 /* permit persons to whom the Software is furnished to do so, subject to */
17 /* the following conditions: */
18 /* */
19 /* The above copyright notice and this permission notice shall be */
20 /* included in all copies or substantial portions of the Software. */
21 /* */
22 /* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */
23 /* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */
24 /* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.*/
25 /* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */
26 /* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */
27 /* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */
28 /* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
29 /*************************************************************************/
30 #include "safe_refcount.h"
31
32 // Atomic functions, these are used for multithread safe reference counters!
33
34 #ifdef NO_THREADS
35
36 /* Bogus implementation unaware of multiprocessing */
37
38 template <class T>
_atomic_conditional_increment_impl(register T * pw)39 static _ALWAYS_INLINE_ T _atomic_conditional_increment_impl(register T *pw) {
40
41 if (*pw == 0)
42 return 0;
43
44 (*pw)++;
45
46 return *pw;
47 }
48
49 template <class T>
_atomic_decrement_impl(register T * pw)50 static _ALWAYS_INLINE_ T _atomic_decrement_impl(register T *pw) {
51
52 (*pw)--;
53
54 return *pw;
55 }
56
57 template <class T>
_atomic_increment_impl(register T * pw)58 static _ALWAYS_INLINE_ T _atomic_increment_impl(register T *pw) {
59
60 (*pw)++;
61
62 return *pw;
63 }
64
65 template <class T>
_atomic_sub_impl(register T * pw,register T val)66 static _ALWAYS_INLINE_ T _atomic_sub_impl(register T *pw, register T val) {
67
68 (*pw) -= val;
69
70 return *pw;
71 }
72
73 template <class T>
_atomic_add_impl(register T * pw,register T val)74 static _ALWAYS_INLINE_ T _atomic_add_impl(register T *pw, register T val) {
75
76 (*pw) += val;
77
78 return *pw;
79 }
80
81 template <class T>
_atomic_exchange_if_greater_impl(register T * pw,register T val)82 static _ALWAYS_INLINE_ T _atomic_exchange_if_greater_impl(register T *pw, register T val) {
83
84 if (val > *pw)
85 *pw = val;
86
87 return *pw;
88 }
89
90 #elif defined(__GNUC__)
91
92 /* Implementation for GCC & Clang */
93
94 // GCC guarantees atomic intrinsics for sizes of 1, 2, 4 and 8 bytes.
95 // Clang states it supports GCC atomic builtins.
96
97 template <class T>
_atomic_conditional_increment_impl(register T * pw)98 static _ALWAYS_INLINE_ T _atomic_conditional_increment_impl(register T *pw) {
99
100 while (true) {
101 T tmp = static_cast<T const volatile &>(*pw);
102 if (tmp == 0)
103 return 0; // if zero, can't add to it anymore
104 if (__sync_val_compare_and_swap(pw, tmp, tmp + 1) == tmp)
105 return tmp + 1;
106 }
107 }
108
109 template <class T>
_atomic_decrement_impl(register T * pw)110 static _ALWAYS_INLINE_ T _atomic_decrement_impl(register T *pw) {
111
112 return __sync_sub_and_fetch(pw, 1);
113 }
114
115 template <class T>
_atomic_increment_impl(register T * pw)116 static _ALWAYS_INLINE_ T _atomic_increment_impl(register T *pw) {
117
118 return __sync_add_and_fetch(pw, 1);
119 }
120
121 template <class T>
_atomic_sub_impl(register T * pw,register T val)122 static _ALWAYS_INLINE_ T _atomic_sub_impl(register T *pw, register T val) {
123
124 return __sync_sub_and_fetch(pw, val);
125 }
126
127 template <class T>
_atomic_add_impl(register T * pw,register T val)128 static _ALWAYS_INLINE_ T _atomic_add_impl(register T *pw, register T val) {
129
130 return __sync_add_and_fetch(pw, val);
131 }
132
133 template <class T>
_atomic_exchange_if_greater_impl(register T * pw,register T val)134 static _ALWAYS_INLINE_ T _atomic_exchange_if_greater_impl(register T *pw, register T val) {
135
136 while (true) {
137 T tmp = static_cast<T const volatile &>(*pw);
138 if (tmp >= val)
139 return tmp; // already greater, or equal
140 if (__sync_val_compare_and_swap(pw, tmp, val) == tmp)
141 return val;
142 }
143 }
144
145 #elif defined(_MSC_VER)
146
147 /* Implementation for MSVC-Windows */
148
149 // don't pollute my namespace!
150 #include <windows.h>
151
152 #define ATOMIC_CONDITIONAL_INCREMENT_BODY(m_pw, m_win_type, m_win_cmpxchg, m_cpp_type) \
153 /* try to increment until it actually works */ \
154 /* taken from boost */ \
155 while (true) { \
156 m_cpp_type tmp = static_cast<m_cpp_type const volatile &>(*(m_pw)); \
157 if (tmp == 0) \
158 return 0; /* if zero, can't add to it anymore */ \
159 if (m_win_cmpxchg((m_win_type volatile *)(m_pw), tmp + 1, tmp) == tmp) \
160 return tmp + 1; \
161 }
162
163 #define ATOMIC_EXCHANGE_IF_GREATER_BODY(m_pw, m_val, m_win_type, m_win_cmpxchg, m_cpp_type) \
164 while (true) { \
165 m_cpp_type tmp = static_cast<m_cpp_type const volatile &>(*(m_pw)); \
166 if (tmp >= m_val) \
167 return tmp; /* already greater, or equal */ \
168 if (m_win_cmpxchg((m_win_type volatile *)(m_pw), m_val, tmp) == tmp) \
169 return m_val; \
170 }
171
_atomic_conditional_increment_impl(register uint32_t * pw)172 static _ALWAYS_INLINE_ uint32_t _atomic_conditional_increment_impl(register uint32_t *pw) {
173
174 ATOMIC_CONDITIONAL_INCREMENT_BODY(pw, LONG, InterlockedCompareExchange, uint32_t)
175 }
176
_atomic_decrement_impl(register uint32_t * pw)177 static _ALWAYS_INLINE_ uint32_t _atomic_decrement_impl(register uint32_t *pw) {
178
179 return InterlockedDecrement((LONG volatile *)pw);
180 }
181
_atomic_increment_impl(register uint32_t * pw)182 static _ALWAYS_INLINE_ uint32_t _atomic_increment_impl(register uint32_t *pw) {
183
184 return InterlockedIncrement((LONG volatile *)pw);
185 }
186
_atomic_sub_impl(register uint32_t * pw,register uint32_t val)187 static _ALWAYS_INLINE_ uint32_t _atomic_sub_impl(register uint32_t *pw, register uint32_t val) {
188
189 return InterlockedExchangeAdd((LONG volatile *)pw, -(int32_t)val) - val;
190 }
191
_atomic_add_impl(register uint32_t * pw,register uint32_t val)192 static _ALWAYS_INLINE_ uint32_t _atomic_add_impl(register uint32_t *pw, register uint32_t val) {
193
194 return InterlockedAdd((LONG volatile *)pw, val);
195 }
196
_atomic_exchange_if_greater_impl(register uint32_t * pw,register uint32_t val)197 static _ALWAYS_INLINE_ uint32_t _atomic_exchange_if_greater_impl(register uint32_t *pw, register uint32_t val) {
198
199 ATOMIC_EXCHANGE_IF_GREATER_BODY(pw, val, LONG, InterlockedCompareExchange, uint32_t)
200 }
201
_atomic_conditional_increment_impl(register uint64_t * pw)202 static _ALWAYS_INLINE_ uint64_t _atomic_conditional_increment_impl(register uint64_t *pw) {
203
204 ATOMIC_CONDITIONAL_INCREMENT_BODY(pw, LONGLONG, InterlockedCompareExchange64, uint64_t)
205 }
206
_atomic_decrement_impl(register uint64_t * pw)207 static _ALWAYS_INLINE_ uint64_t _atomic_decrement_impl(register uint64_t *pw) {
208
209 return InterlockedDecrement64((LONGLONG volatile *)pw);
210 }
211
_atomic_increment_impl(register uint64_t * pw)212 static _ALWAYS_INLINE_ uint64_t _atomic_increment_impl(register uint64_t *pw) {
213
214 return InterlockedIncrement64((LONGLONG volatile *)pw);
215 }
216
_atomic_sub_impl(register uint64_t * pw,register uint64_t val)217 static _ALWAYS_INLINE_ uint64_t _atomic_sub_impl(register uint64_t *pw, register uint64_t val) {
218
219 return InterlockedExchangeAdd64((LONGLONG volatile *)pw, -(int64_t)val) - val;
220 }
221
_atomic_add_impl(register uint64_t * pw,register uint64_t val)222 static _ALWAYS_INLINE_ uint64_t _atomic_add_impl(register uint64_t *pw, register uint64_t val) {
223
224 return InterlockedAdd64((LONGLONG volatile *)pw, val);
225 }
226
_atomic_exchange_if_greater_impl(register uint64_t * pw,register uint64_t val)227 static _ALWAYS_INLINE_ uint64_t _atomic_exchange_if_greater_impl(register uint64_t *pw, register uint64_t val) {
228
229 ATOMIC_EXCHANGE_IF_GREATER_BODY(pw, val, LONGLONG, InterlockedCompareExchange64, uint64_t)
230 }
231
232 #else
233
234 //no threads supported?
235 #error Must provide atomic functions for this platform or compiler!
236
237 #endif
238
239 // The actual advertised functions; they'll call the right implementation
240
atomic_conditional_increment(register uint32_t * counter)241 uint32_t atomic_conditional_increment(register uint32_t *counter) {
242 return _atomic_conditional_increment_impl(counter);
243 }
244
atomic_decrement(register uint32_t * pw)245 uint32_t atomic_decrement(register uint32_t *pw) {
246 return _atomic_decrement_impl(pw);
247 }
248
atomic_increment(register uint32_t * pw)249 uint32_t atomic_increment(register uint32_t *pw) {
250 return _atomic_increment_impl(pw);
251 }
252
atomic_sub(register uint32_t * pw,register uint32_t val)253 uint32_t atomic_sub(register uint32_t *pw, register uint32_t val) {
254 return _atomic_sub_impl(pw, val);
255 }
256
atomic_add(register uint32_t * pw,register uint32_t val)257 uint32_t atomic_add(register uint32_t *pw, register uint32_t val) {
258 return _atomic_add_impl(pw, val);
259 }
260
atomic_exchange_if_greater(register uint32_t * pw,register uint32_t val)261 uint32_t atomic_exchange_if_greater(register uint32_t *pw, register uint32_t val) {
262 return _atomic_exchange_if_greater_impl(pw, val);
263 }
264
atomic_conditional_increment(register uint64_t * counter)265 uint64_t atomic_conditional_increment(register uint64_t *counter) {
266 return _atomic_conditional_increment_impl(counter);
267 }
268
atomic_decrement(register uint64_t * pw)269 uint64_t atomic_decrement(register uint64_t *pw) {
270 return _atomic_decrement_impl(pw);
271 }
272
atomic_increment(register uint64_t * pw)273 uint64_t atomic_increment(register uint64_t *pw) {
274 return _atomic_increment_impl(pw);
275 }
276
atomic_sub(register uint64_t * pw,register uint64_t val)277 uint64_t atomic_sub(register uint64_t *pw, register uint64_t val) {
278 return _atomic_sub_impl(pw, val);
279 }
280
atomic_add(register uint64_t * pw,register uint64_t val)281 uint64_t atomic_add(register uint64_t *pw, register uint64_t val) {
282 return _atomic_add_impl(pw, val);
283 }
284
atomic_exchange_if_greater(register uint64_t * pw,register uint64_t val)285 uint64_t atomic_exchange_if_greater(register uint64_t *pw, register uint64_t val) {
286 return _atomic_exchange_if_greater_impl(pw, val);
287 }
288