| 1 | /* |
| 2 | Simple DirectMedia Layer |
| 3 | Copyright (C) 1997-2018 Sam Lantinga <slouken@libsdl.org> |
| 4 | |
| 5 | This software is provided 'as-is', without any express or implied |
| 6 | warranty. In no event will the authors be held liable for any damages |
| 7 | arising from the use of this software. |
| 8 | |
| 9 | Permission is granted to anyone to use this software for any purpose, |
| 10 | including commercial applications, and to alter it and redistribute it |
| 11 | freely, subject to the following restrictions: |
| 12 | |
| 13 | 1. The origin of this software must not be misrepresented; you must not |
| 14 | claim that you wrote the original software. If you use this software |
| 15 | in a product, an acknowledgment in the product documentation would be |
| 16 | appreciated but is not required. |
| 17 | 2. Altered source versions must be plainly marked as such, and must not be |
| 18 | misrepresented as being the original software. |
| 19 | 3. This notice may not be removed or altered from any source distribution. |
| 20 | */ |
| 21 | |
| 22 | /** |
| 23 | * \file SDL_atomic.h |
| 24 | * |
| 25 | * Atomic operations. |
| 26 | * |
| 27 | * IMPORTANT: |
| 28 | * If you are not an expert in concurrent lockless programming, you should |
| 29 | * only be using the atomic lock and reference counting functions in this |
| 30 | * file. In all other cases you should be protecting your data structures |
| 31 | * with full mutexes. |
| 32 | * |
| 33 | * The list of "safe" functions to use are: |
| 34 | * SDL_AtomicLock() |
| 35 | * SDL_AtomicUnlock() |
| 36 | * SDL_AtomicIncRef() |
| 37 | * SDL_AtomicDecRef() |
| 38 | * |
| 39 | * Seriously, here be dragons! |
| 40 | * ^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
| 41 | * |
| 42 | * You can find out a little more about lockless programming and the |
| 43 | * subtle issues that can arise here: |
| 44 | * http://msdn.microsoft.com/en-us/library/ee418650%28v=vs.85%29.aspx |
| 45 | * |
| 46 | * There's also lots of good information here: |
| 47 | * http://www.1024cores.net/home/lock-free-algorithms |
| 48 | * http://preshing.com/ |
| 49 | * |
| 50 | * These operations may or may not actually be implemented using |
| 51 | * processor specific atomic operations. When possible they are |
| 52 | * implemented as true processor specific atomic operations. When that |
| 53 | * is not possible the are implemented using locks that *do* use the |
| 54 | * available atomic operations. |
| 55 | * |
| 56 | * All of the atomic operations that modify memory are full memory barriers. |
| 57 | */ |
| 58 | |
| 59 | #ifndef SDL_atomic_h_ |
| 60 | #define SDL_atomic_h_ |
| 61 | |
| 62 | #include "SDL_stdinc.h" |
| 63 | #include "SDL_platform.h" |
| 64 | |
| 65 | #include "begin_code.h" |
| 66 | |
| 67 | /* Set up for C function definitions, even when using C++ */ |
| 68 | #ifdef __cplusplus |
| 69 | extern "C" { |
| 70 | #endif |
| 71 | |
| 72 | /** |
| 73 | * \name SDL AtomicLock |
| 74 | * |
| 75 | * The atomic locks are efficient spinlocks using CPU instructions, |
| 76 | * but are vulnerable to starvation and can spin forever if a thread |
| 77 | * holding a lock has been terminated. For this reason you should |
| 78 | * minimize the code executed inside an atomic lock and never do |
| 79 | * expensive things like API or system calls while holding them. |
| 80 | * |
| 81 | * The atomic locks are not safe to lock recursively. |
| 82 | * |
| 83 | * Porting Note: |
| 84 | * The spin lock functions and type are required and can not be |
| 85 | * emulated because they are used in the atomic emulation code. |
| 86 | */ |
| 87 | /* @{ */ |
| 88 | |
| 89 | typedef int SDL_SpinLock; |
| 90 | |
| 91 | /** |
| 92 | * \brief Try to lock a spin lock by setting it to a non-zero value. |
| 93 | * |
| 94 | * \param lock Points to the lock. |
| 95 | * |
| 96 | * \return SDL_TRUE if the lock succeeded, SDL_FALSE if the lock is already held. |
| 97 | */ |
| 98 | extern DECLSPEC SDL_bool SDLCALL SDL_AtomicTryLock(SDL_SpinLock *lock); |
| 99 | |
| 100 | /** |
| 101 | * \brief Lock a spin lock by setting it to a non-zero value. |
| 102 | * |
| 103 | * \param lock Points to the lock. |
| 104 | */ |
| 105 | extern DECLSPEC void SDLCALL SDL_AtomicLock(SDL_SpinLock *lock); |
| 106 | |
| 107 | /** |
| 108 | * \brief Unlock a spin lock by setting it to 0. Always returns immediately |
| 109 | * |
| 110 | * \param lock Points to the lock. |
| 111 | */ |
| 112 | extern DECLSPEC void SDLCALL SDL_AtomicUnlock(SDL_SpinLock *lock); |
| 113 | |
| 114 | /* @} *//* SDL AtomicLock */ |
| 115 | |
| 116 | |
| 117 | /** |
| 118 | * The compiler barrier prevents the compiler from reordering |
| 119 | * reads and writes to globally visible variables across the call. |
| 120 | */ |
| 121 | #if defined(_MSC_VER) && (_MSC_VER > 1200) && !defined(__clang__) |
| 122 | void _ReadWriteBarrier(void); |
| 123 | #pragma intrinsic(_ReadWriteBarrier) |
| 124 | #define SDL_CompilerBarrier() _ReadWriteBarrier() |
| 125 | #elif (defined(__GNUC__) && !defined(__EMSCRIPTEN__)) || (defined(__SUNPRO_C) && (__SUNPRO_C >= 0x5120)) |
| 126 | /* This is correct for all CPUs when using GCC or Solaris Studio 12.1+. */ |
| 127 | #define SDL_CompilerBarrier() __asm__ __volatile__ ("" : : : "memory") |
| 128 | #elif defined(__WATCOMC__) |
| 129 | extern _inline void SDL_CompilerBarrier (void); |
| 130 | #pragma aux SDL_CompilerBarrier = "" parm [] modify exact []; |
| 131 | #else |
| 132 | #define SDL_CompilerBarrier() \ |
| 133 | { SDL_SpinLock _tmp = 0; SDL_AtomicLock(&_tmp); SDL_AtomicUnlock(&_tmp); } |
| 134 | #endif |
| 135 | |
| 136 | /** |
| 137 | * Memory barriers are designed to prevent reads and writes from being |
| 138 | * reordered by the compiler and being seen out of order on multi-core CPUs. |
| 139 | * |
| 140 | * A typical pattern would be for thread A to write some data and a flag, |
| 141 | * and for thread B to read the flag and get the data. In this case you |
| 142 | * would insert a release barrier between writing the data and the flag, |
| 143 | * guaranteeing that the data write completes no later than the flag is |
| 144 | * written, and you would insert an acquire barrier between reading the |
| 145 | * flag and reading the data, to ensure that all the reads associated |
| 146 | * with the flag have completed. |
| 147 | * |
| 148 | * In this pattern you should always see a release barrier paired with |
| 149 | * an acquire barrier and you should gate the data reads/writes with a |
| 150 | * single flag variable. |
| 151 | * |
| 152 | * For more information on these semantics, take a look at the blog post: |
| 153 | * http://preshing.com/20120913/acquire-and-release-semantics |
| 154 | */ |
| 155 | extern DECLSPEC void SDLCALL SDL_MemoryBarrierReleaseFunction(void); |
| 156 | extern DECLSPEC void SDLCALL SDL_MemoryBarrierAcquireFunction(void); |
| 157 | |
| 158 | #if defined(__GNUC__) && (defined(__powerpc__) || defined(__ppc__)) |
| 159 | #define SDL_MemoryBarrierRelease() __asm__ __volatile__ ("lwsync" : : : "memory") |
| 160 | #define SDL_MemoryBarrierAcquire() __asm__ __volatile__ ("lwsync" : : : "memory") |
| 161 | #elif defined(__GNUC__) && defined(__aarch64__) |
| 162 | #define SDL_MemoryBarrierRelease() __asm__ __volatile__ ("dmb ish" : : : "memory") |
| 163 | #define SDL_MemoryBarrierAcquire() __asm__ __volatile__ ("dmb ish" : : : "memory") |
| 164 | #elif defined(__GNUC__) && defined(__arm__) |
| 165 | #if defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7EM__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) |
| 166 | #define SDL_MemoryBarrierRelease() __asm__ __volatile__ ("dmb ish" : : : "memory") |
| 167 | #define SDL_MemoryBarrierAcquire() __asm__ __volatile__ ("dmb ish" : : : "memory") |
| 168 | #elif defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6T2__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_5TE__) |
| 169 | #ifdef __thumb__ |
| 170 | /* The mcr instruction isn't available in thumb mode, use real functions */ |
| 171 | #define SDL_MemoryBarrierRelease() SDL_MemoryBarrierReleaseFunction() |
| 172 | #define SDL_MemoryBarrierAcquire() SDL_MemoryBarrierAcquireFunction() |
| 173 | #else |
| 174 | #define SDL_MemoryBarrierRelease() __asm__ __volatile__ ("mcr p15, 0, %0, c7, c10, 5" : : "r"(0) : "memory") |
| 175 | #define SDL_MemoryBarrierAcquire() __asm__ __volatile__ ("mcr p15, 0, %0, c7, c10, 5" : : "r"(0) : "memory") |
| 176 | #endif /* __thumb__ */ |
| 177 | #else |
| 178 | #define SDL_MemoryBarrierRelease() __asm__ __volatile__ ("" : : : "memory") |
| 179 | #define SDL_MemoryBarrierAcquire() __asm__ __volatile__ ("" : : : "memory") |
| 180 | #endif /* __GNUC__ && __arm__ */ |
| 181 | #else |
| 182 | #if (defined(__SUNPRO_C) && (__SUNPRO_C >= 0x5120)) |
| 183 | /* This is correct for all CPUs on Solaris when using Solaris Studio 12.1+. */ |
| 184 | #include <mbarrier.h> |
| 185 | #define SDL_MemoryBarrierRelease() __machine_rel_barrier() |
| 186 | #define SDL_MemoryBarrierAcquire() __machine_acq_barrier() |
| 187 | #else |
| 188 | /* This is correct for the x86 and x64 CPUs, and we'll expand this over time. */ |
| 189 | #define SDL_MemoryBarrierRelease() SDL_CompilerBarrier() |
| 190 | #define SDL_MemoryBarrierAcquire() SDL_CompilerBarrier() |
| 191 | #endif |
| 192 | #endif |
| 193 | |
| 194 | /** |
| 195 | * \brief A type representing an atomic integer value. It is a struct |
| 196 | * so people don't accidentally use numeric operations on it. |
| 197 | */ |
| 198 | typedef struct { int value; } SDL_atomic_t; |
| 199 | |
| 200 | /** |
| 201 | * \brief Set an atomic variable to a new value if it is currently an old value. |
| 202 | * |
| 203 | * \return SDL_TRUE if the atomic variable was set, SDL_FALSE otherwise. |
| 204 | * |
| 205 | * \note If you don't know what this function is for, you shouldn't use it! |
| 206 | */ |
| 207 | extern DECLSPEC SDL_bool SDLCALL SDL_AtomicCAS(SDL_atomic_t *a, int oldval, int newval); |
| 208 | |
| 209 | /** |
| 210 | * \brief Set an atomic variable to a value. |
| 211 | * |
| 212 | * \return The previous value of the atomic variable. |
| 213 | */ |
| 214 | extern DECLSPEC int SDLCALL SDL_AtomicSet(SDL_atomic_t *a, int v); |
| 215 | |
| 216 | /** |
| 217 | * \brief Get the value of an atomic variable |
| 218 | */ |
| 219 | extern DECLSPEC int SDLCALL SDL_AtomicGet(SDL_atomic_t *a); |
| 220 | |
| 221 | /** |
| 222 | * \brief Add to an atomic variable. |
| 223 | * |
| 224 | * \return The previous value of the atomic variable. |
| 225 | * |
| 226 | * \note This same style can be used for any number operation |
| 227 | */ |
| 228 | extern DECLSPEC int SDLCALL SDL_AtomicAdd(SDL_atomic_t *a, int v); |
| 229 | |
| 230 | /** |
| 231 | * \brief Increment an atomic variable used as a reference count. |
| 232 | */ |
| 233 | #ifndef SDL_AtomicIncRef |
| 234 | #define SDL_AtomicIncRef(a) SDL_AtomicAdd(a, 1) |
| 235 | #endif |
| 236 | |
| 237 | /** |
| 238 | * \brief Decrement an atomic variable used as a reference count. |
| 239 | * |
| 240 | * \return SDL_TRUE if the variable reached zero after decrementing, |
| 241 | * SDL_FALSE otherwise |
| 242 | */ |
| 243 | #ifndef SDL_AtomicDecRef |
| 244 | #define SDL_AtomicDecRef(a) (SDL_AtomicAdd(a, -1) == 1) |
| 245 | #endif |
| 246 | |
| 247 | /** |
| 248 | * \brief Set a pointer to a new value if it is currently an old value. |
| 249 | * |
| 250 | * \return SDL_TRUE if the pointer was set, SDL_FALSE otherwise. |
| 251 | * |
| 252 | * \note If you don't know what this function is for, you shouldn't use it! |
| 253 | */ |
| 254 | extern DECLSPEC SDL_bool SDLCALL SDL_AtomicCASPtr(void **a, void *oldval, void *newval); |
| 255 | |
| 256 | /** |
| 257 | * \brief Set a pointer to a value atomically. |
| 258 | * |
| 259 | * \return The previous value of the pointer. |
| 260 | */ |
| 261 | extern DECLSPEC void* SDLCALL SDL_AtomicSetPtr(void **a, void* v); |
| 262 | |
| 263 | /** |
| 264 | * \brief Get the value of a pointer atomically. |
| 265 | */ |
| 266 | extern DECLSPEC void* SDLCALL SDL_AtomicGetPtr(void **a); |
| 267 | |
| 268 | /* Ends C function definitions when using C++ */ |
| 269 | #ifdef __cplusplus |
| 270 | } |
| 271 | #endif |
| 272 | |
| 273 | #include "close_code.h" |
| 274 | |
| 275 | #endif /* SDL_atomic_h_ */ |
| 276 | |
| 277 | /* vi: set ts=4 sw=4 expandtab: */ |
| 278 | |