| 1 | /* |
| 2 | Simple DirectMedia Layer |
| 3 | Copyright (C) 1997-2025 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 | * # CategoryAtomic |
| 24 | * |
| 25 | * Atomic operations. |
| 26 | * |
| 27 | * IMPORTANT: If you are not an expert in concurrent lockless programming, you |
| 28 | * should not be using any functions in this file. You should be protecting |
| 29 | * your data structures with full mutexes instead. |
| 30 | * |
| 31 | * ***Seriously, here be dragons!*** |
| 32 | * |
| 33 | * You can find out a little more about lockless programming and the subtle |
| 34 | * issues that can arise here: |
| 35 | * https://learn.microsoft.com/en-us/windows/win32/dxtecharts/lockless-programming |
| 36 | * |
| 37 | * There's also lots of good information here: |
| 38 | * |
| 39 | * - https://www.1024cores.net/home/lock-free-algorithms |
| 40 | * - https://preshing.com/ |
| 41 | * |
| 42 | * These operations may or may not actually be implemented using processor |
| 43 | * specific atomic operations. When possible they are implemented as true |
| 44 | * processor specific atomic operations. When that is not possible the are |
| 45 | * implemented using locks that *do* use the available atomic operations. |
| 46 | * |
| 47 | * All of the atomic operations that modify memory are full memory barriers. |
| 48 | */ |
| 49 | |
| 50 | #ifndef SDL_atomic_h_ |
| 51 | #define SDL_atomic_h_ |
| 52 | |
| 53 | #include "SDL_stdinc.h" |
| 54 | #include "SDL_platform.h" |
| 55 | |
| 56 | #include "begin_code.h" |
| 57 | |
| 58 | /* Set up for C function definitions, even when using C++ */ |
| 59 | #ifdef __cplusplus |
| 60 | extern "C" { |
| 61 | #endif |
| 62 | |
| 63 | /** |
| 64 | * \name SDL AtomicLock |
| 65 | * |
| 66 | * The atomic locks are efficient spinlocks using CPU instructions, |
| 67 | * but are vulnerable to starvation and can spin forever if a thread |
| 68 | * holding a lock has been terminated. For this reason you should |
| 69 | * minimize the code executed inside an atomic lock and never do |
| 70 | * expensive things like API or system calls while holding them. |
| 71 | * |
| 72 | * The atomic locks are not safe to lock recursively. |
| 73 | * |
| 74 | * Porting Note: |
| 75 | * The spin lock functions and type are required and can not be |
| 76 | * emulated because they are used in the atomic emulation code. |
| 77 | */ |
| 78 | /* @{ */ |
| 79 | |
| 80 | typedef int SDL_SpinLock; |
| 81 | |
| 82 | /** |
| 83 | * Try to lock a spin lock by setting it to a non-zero value. |
| 84 | * |
| 85 | * ***Please note that spinlocks are dangerous if you don't know what you're |
| 86 | * doing. Please be careful using any sort of spinlock!*** |
| 87 | * |
| 88 | * \param lock a pointer to a lock variable. |
| 89 | * \returns SDL_TRUE if the lock succeeded, SDL_FALSE if the lock is already |
| 90 | * held. |
| 91 | * |
| 92 | * \since This function is available since SDL 2.0.0. |
| 93 | * |
| 94 | * \sa SDL_AtomicLock |
| 95 | * \sa SDL_AtomicUnlock |
| 96 | */ |
| 97 | extern DECLSPEC SDL_bool SDLCALL SDL_AtomicTryLock(SDL_SpinLock *lock); |
| 98 | |
| 99 | /** |
| 100 | * Lock a spin lock by setting it to a non-zero value. |
| 101 | * |
| 102 | * ***Please note that spinlocks are dangerous if you don't know what you're |
| 103 | * doing. Please be careful using any sort of spinlock!*** |
| 104 | * |
| 105 | * \param lock a pointer to a lock variable. |
| 106 | * |
| 107 | * \since This function is available since SDL 2.0.0. |
| 108 | * |
| 109 | * \sa SDL_AtomicTryLock |
| 110 | * \sa SDL_AtomicUnlock |
| 111 | */ |
| 112 | extern DECLSPEC void SDLCALL SDL_AtomicLock(SDL_SpinLock *lock); |
| 113 | |
| 114 | /** |
| 115 | * Unlock a spin lock by setting it to 0. |
| 116 | * |
| 117 | * Always returns immediately. |
| 118 | * |
| 119 | * ***Please note that spinlocks are dangerous if you don't know what you're |
| 120 | * doing. Please be careful using any sort of spinlock!*** |
| 121 | * |
| 122 | * \param lock a pointer to a lock variable. |
| 123 | * |
| 124 | * \since This function is available since SDL 2.0.0. |
| 125 | * |
| 126 | * \sa SDL_AtomicLock |
| 127 | * \sa SDL_AtomicTryLock |
| 128 | */ |
| 129 | extern DECLSPEC void SDLCALL SDL_AtomicUnlock(SDL_SpinLock *lock); |
| 130 | |
| 131 | /* @} *//* SDL AtomicLock */ |
| 132 | |
| 133 | |
| 134 | /** |
| 135 | * The compiler barrier prevents the compiler from reordering |
| 136 | * reads and writes to globally visible variables across the call. |
| 137 | */ |
| 138 | #if defined(_MSC_VER) && (_MSC_VER > 1200) && !defined(__clang__) |
| 139 | void _ReadWriteBarrier(void); |
| 140 | #pragma intrinsic(_ReadWriteBarrier) |
| 141 | #define SDL_CompilerBarrier() _ReadWriteBarrier() |
| 142 | #elif (defined(__GNUC__) && !defined(__EMSCRIPTEN__)) || (defined(__SUNPRO_C) && (__SUNPRO_C >= 0x5120)) |
| 143 | /* This is correct for all CPUs when using GCC or Solaris Studio 12.1+. */ |
| 144 | #define SDL_CompilerBarrier() __asm__ __volatile__ ("" : : : "memory") |
| 145 | #elif defined(__WATCOMC__) |
| 146 | extern __inline void SDL_CompilerBarrier(void); |
| 147 | #pragma aux SDL_CompilerBarrier = "" parm [] modify exact []; |
| 148 | #else |
| 149 | #define SDL_CompilerBarrier() \ |
| 150 | { SDL_SpinLock _tmp = 0; SDL_AtomicLock(&_tmp); SDL_AtomicUnlock(&_tmp); } |
| 151 | #endif |
| 152 | |
| 153 | /** |
| 154 | * Memory barriers are designed to prevent reads and writes from being |
| 155 | * reordered by the compiler and being seen out of order on multi-core CPUs. |
| 156 | * |
| 157 | * A typical pattern would be for thread A to write some data and a flag, and |
| 158 | * for thread B to read the flag and get the data. In this case you would |
| 159 | * insert a release barrier between writing the data and the flag, |
| 160 | * guaranteeing that the data write completes no later than the flag is |
| 161 | * written, and you would insert an acquire barrier between reading the flag |
| 162 | * and reading the data, to ensure that all the reads associated with the flag |
| 163 | * have completed. |
| 164 | * |
| 165 | * In this pattern you should always see a release barrier paired with an |
| 166 | * acquire barrier and you should gate the data reads/writes with a single |
| 167 | * flag variable. |
| 168 | * |
| 169 | * For more information on these semantics, take a look at the blog post: |
| 170 | * http://preshing.com/20120913/acquire-and-release-semantics |
| 171 | * |
| 172 | * \since This function is available since SDL 2.0.6. |
| 173 | */ |
| 174 | extern DECLSPEC void SDLCALL SDL_MemoryBarrierReleaseFunction(void); |
| 175 | extern DECLSPEC void SDLCALL SDL_MemoryBarrierAcquireFunction(void); |
| 176 | |
| 177 | #if defined(__GNUC__) && (defined(__powerpc__) || defined(__ppc__)) |
| 178 | #define SDL_MemoryBarrierRelease() __asm__ __volatile__ ("lwsync" : : : "memory") |
| 179 | #define SDL_MemoryBarrierAcquire() __asm__ __volatile__ ("lwsync" : : : "memory") |
| 180 | #elif defined(__GNUC__) && defined(__aarch64__) |
| 181 | #define SDL_MemoryBarrierRelease() __asm__ __volatile__ ("dmb ish" : : : "memory") |
| 182 | #define SDL_MemoryBarrierAcquire() __asm__ __volatile__ ("dmb ish" : : : "memory") |
| 183 | #elif defined(__GNUC__) && defined(__arm__) |
| 184 | #if 0 /* defined(__LINUX__) || defined(__ANDROID__) */ |
| 185 | /* Information from: |
| 186 | https://chromium.googlesource.com/chromium/chromium/+/trunk/base/atomicops_internals_arm_gcc.h#19 |
| 187 | |
| 188 | The Linux kernel provides a helper function which provides the right code for a memory barrier, |
| 189 | hard-coded at address 0xffff0fa0 |
| 190 | */ |
| 191 | typedef void (*SDL_KernelMemoryBarrierFunc)(); |
| 192 | #define SDL_MemoryBarrierRelease() ((SDL_KernelMemoryBarrierFunc)0xffff0fa0)() |
| 193 | #define SDL_MemoryBarrierAcquire() ((SDL_KernelMemoryBarrierFunc)0xffff0fa0)() |
| 194 | #elif 0 /* defined(__QNXNTO__) */ |
| 195 | #include <sys/cpuinline.h> |
| 196 | |
| 197 | #define SDL_MemoryBarrierRelease() __cpu_membarrier() |
| 198 | #define SDL_MemoryBarrierAcquire() __cpu_membarrier() |
| 199 | #else |
| 200 | #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__) || defined(__ARM_ARCH_8A__) |
| 201 | #define SDL_MemoryBarrierRelease() __asm__ __volatile__ ("dmb ish" : : : "memory") |
| 202 | #define SDL_MemoryBarrierAcquire() __asm__ __volatile__ ("dmb ish" : : : "memory") |
| 203 | #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__) |
| 204 | #ifdef __thumb__ |
| 205 | /* The mcr instruction isn't available in thumb mode, use real functions */ |
| 206 | #define SDL_MEMORY_BARRIER_USES_FUNCTION |
| 207 | #define SDL_MemoryBarrierRelease() SDL_MemoryBarrierReleaseFunction() |
| 208 | #define SDL_MemoryBarrierAcquire() SDL_MemoryBarrierAcquireFunction() |
| 209 | #else |
| 210 | #define SDL_MemoryBarrierRelease() __asm__ __volatile__ ("mcr p15, 0, %0, c7, c10, 5" : : "r"(0) : "memory") |
| 211 | #define SDL_MemoryBarrierAcquire() __asm__ __volatile__ ("mcr p15, 0, %0, c7, c10, 5" : : "r"(0) : "memory") |
| 212 | #endif /* __thumb__ */ |
| 213 | #else |
| 214 | #define SDL_MemoryBarrierRelease() __asm__ __volatile__ ("" : : : "memory") |
| 215 | #define SDL_MemoryBarrierAcquire() __asm__ __volatile__ ("" : : : "memory") |
| 216 | #endif /* __LINUX__ || __ANDROID__ */ |
| 217 | #endif /* __GNUC__ && __arm__ */ |
| 218 | #else |
| 219 | #if (defined(__SUNPRO_C) && (__SUNPRO_C >= 0x5120)) |
| 220 | /* This is correct for all CPUs on Solaris when using Solaris Studio 12.1+. */ |
| 221 | #include <mbarrier.h> |
| 222 | #define SDL_MemoryBarrierRelease() __machine_rel_barrier() |
| 223 | #define SDL_MemoryBarrierAcquire() __machine_acq_barrier() |
| 224 | #else |
| 225 | /* This is correct for the x86 and x64 CPUs, and we'll expand this over time. */ |
| 226 | #define SDL_MemoryBarrierRelease() SDL_CompilerBarrier() |
| 227 | #define SDL_MemoryBarrierAcquire() SDL_CompilerBarrier() |
| 228 | #endif |
| 229 | #endif |
| 230 | |
| 231 | /* "REP NOP" is PAUSE, coded for tools that don't know it by that name. */ |
| 232 | #if (defined(__GNUC__) || defined(__clang__)) && (defined(__i386__) || defined(__x86_64__)) |
| 233 | #define SDL_CPUPauseInstruction() __asm__ __volatile__("pause\n") /* Some assemblers can't do REP NOP, so go with PAUSE. */ |
| 234 | #elif (defined(__arm__) && defined(__ARM_ARCH) && __ARM_ARCH >= 7) || defined(__aarch64__) |
| 235 | #define SDL_CPUPauseInstruction() __asm__ __volatile__("yield" ::: "memory") |
| 236 | #elif (defined(__powerpc__) || defined(__powerpc64__)) |
| 237 | #define SDL_CPUPauseInstruction() __asm__ __volatile__("or 27,27,27"); |
| 238 | #elif defined(_MSC_VER) && (defined(_M_IX86) || defined(_M_X64)) |
| 239 | #define SDL_CPUPauseInstruction() _mm_pause() /* this is actually "rep nop" and not a SIMD instruction. No inline asm in MSVC x86-64! */ |
| 240 | #elif defined(_MSC_VER) && (defined(_M_ARM) || defined(_M_ARM64)) |
| 241 | #define SDL_CPUPauseInstruction() __yield() |
| 242 | #elif defined(__WATCOMC__) && defined(__386__) |
| 243 | extern __inline void SDL_CPUPauseInstruction(void); |
| 244 | #pragma aux SDL_CPUPauseInstruction = ".686p" ".xmm2" "pause" |
| 245 | #else |
| 246 | #define SDL_CPUPauseInstruction() |
| 247 | #endif |
| 248 | |
| 249 | |
| 250 | /** |
| 251 | * A type representing an atomic integer value. |
| 252 | * |
| 253 | * It is a struct so people don't accidentally use numeric operations on it. |
| 254 | */ |
| 255 | typedef struct SDL_atomic_t { |
| 256 | int value; |
| 257 | } SDL_atomic_t; |
| 258 | |
| 259 | /** |
| 260 | * Set an atomic variable to a new value if it is currently an old value. |
| 261 | * |
| 262 | * ***Note: If you don't know what this function is for, you shouldn't use |
| 263 | * it!*** |
| 264 | * |
| 265 | * \param a a pointer to an SDL_atomic_t variable to be modified. |
| 266 | * \param oldval the old value. |
| 267 | * \param newval the new value. |
| 268 | * \returns SDL_TRUE if the atomic variable was set, SDL_FALSE otherwise. |
| 269 | * |
| 270 | * \since This function is available since SDL 2.0.0. |
| 271 | * |
| 272 | * \sa SDL_AtomicCASPtr |
| 273 | * \sa SDL_AtomicGet |
| 274 | * \sa SDL_AtomicSet |
| 275 | */ |
| 276 | extern DECLSPEC SDL_bool SDLCALL SDL_AtomicCAS(SDL_atomic_t *a, int oldval, int newval); |
| 277 | |
| 278 | /** |
| 279 | * Set an atomic variable to a value. |
| 280 | * |
| 281 | * This function also acts as a full memory barrier. |
| 282 | * |
| 283 | * ***Note: If you don't know what this function is for, you shouldn't use |
| 284 | * it!*** |
| 285 | * |
| 286 | * \param a a pointer to an SDL_atomic_t variable to be modified. |
| 287 | * \param v the desired value. |
| 288 | * \returns the previous value of the atomic variable. |
| 289 | * |
| 290 | * \since This function is available since SDL 2.0.2. |
| 291 | * |
| 292 | * \sa SDL_AtomicGet |
| 293 | */ |
| 294 | extern DECLSPEC int SDLCALL SDL_AtomicSet(SDL_atomic_t *a, int v); |
| 295 | |
| 296 | /** |
| 297 | * Get the value of an atomic variable. |
| 298 | * |
| 299 | * ***Note: If you don't know what this function is for, you shouldn't use |
| 300 | * it!*** |
| 301 | * |
| 302 | * \param a a pointer to an SDL_atomic_t variable. |
| 303 | * \returns the current value of an atomic variable. |
| 304 | * |
| 305 | * \since This function is available since SDL 2.0.2. |
| 306 | * |
| 307 | * \sa SDL_AtomicSet |
| 308 | */ |
| 309 | extern DECLSPEC int SDLCALL SDL_AtomicGet(SDL_atomic_t *a); |
| 310 | |
| 311 | /** |
| 312 | * Add to an atomic variable. |
| 313 | * |
| 314 | * This function also acts as a full memory barrier. |
| 315 | * |
| 316 | * ***Note: If you don't know what this function is for, you shouldn't use |
| 317 | * it!*** |
| 318 | * |
| 319 | * \param a a pointer to an SDL_atomic_t variable to be modified. |
| 320 | * \param v the desired value to add. |
| 321 | * \returns the previous value of the atomic variable. |
| 322 | * |
| 323 | * \since This function is available since SDL 2.0.2. |
| 324 | * |
| 325 | * \sa SDL_AtomicDecRef |
| 326 | * \sa SDL_AtomicIncRef |
| 327 | */ |
| 328 | extern DECLSPEC int SDLCALL SDL_AtomicAdd(SDL_atomic_t *a, int v); |
| 329 | |
| 330 | /** |
| 331 | * \brief Increment an atomic variable used as a reference count. |
| 332 | */ |
| 333 | #ifndef SDL_AtomicIncRef |
| 334 | #define SDL_AtomicIncRef(a) SDL_AtomicAdd(a, 1) |
| 335 | #endif |
| 336 | |
| 337 | /** |
| 338 | * \brief Decrement an atomic variable used as a reference count. |
| 339 | * |
| 340 | * \return SDL_TRUE if the variable reached zero after decrementing, |
| 341 | * SDL_FALSE otherwise |
| 342 | */ |
| 343 | #ifndef SDL_AtomicDecRef |
| 344 | #define SDL_AtomicDecRef(a) (SDL_AtomicAdd(a, -1) == 1) |
| 345 | #endif |
| 346 | |
| 347 | /** |
| 348 | * Set a pointer to a new value if it is currently an old value. |
| 349 | * |
| 350 | * ***Note: If you don't know what this function is for, you shouldn't use |
| 351 | * it!*** |
| 352 | * |
| 353 | * \param a a pointer to a pointer. |
| 354 | * \param oldval the old pointer value. |
| 355 | * \param newval the new pointer value. |
| 356 | * \returns SDL_TRUE if the pointer was set, SDL_FALSE otherwise. |
| 357 | * |
| 358 | * \since This function is available since SDL 2.0.0. |
| 359 | * |
| 360 | * \sa SDL_AtomicCAS |
| 361 | * \sa SDL_AtomicGetPtr |
| 362 | * \sa SDL_AtomicSetPtr |
| 363 | */ |
| 364 | extern DECLSPEC SDL_bool SDLCALL SDL_AtomicCASPtr(void **a, void *oldval, void *newval); |
| 365 | |
| 366 | /** |
| 367 | * Set a pointer to a value atomically. |
| 368 | * |
| 369 | * ***Note: If you don't know what this function is for, you shouldn't use |
| 370 | * it!*** |
| 371 | * |
| 372 | * \param a a pointer to a pointer. |
| 373 | * \param v the desired pointer value. |
| 374 | * \returns the previous value of the pointer. |
| 375 | * |
| 376 | * \since This function is available since SDL 2.0.2. |
| 377 | * |
| 378 | * \sa SDL_AtomicCASPtr |
| 379 | * \sa SDL_AtomicGetPtr |
| 380 | */ |
| 381 | extern DECLSPEC void* SDLCALL SDL_AtomicSetPtr(void **a, void* v); |
| 382 | |
| 383 | /** |
| 384 | * Get the value of a pointer atomically. |
| 385 | * |
| 386 | * ***Note: If you don't know what this function is for, you shouldn't use |
| 387 | * it!*** |
| 388 | * |
| 389 | * \param a a pointer to a pointer. |
| 390 | * \returns the current value of a pointer. |
| 391 | * |
| 392 | * \since This function is available since SDL 2.0.2. |
| 393 | * |
| 394 | * \sa SDL_AtomicCASPtr |
| 395 | * \sa SDL_AtomicSetPtr |
| 396 | */ |
| 397 | extern DECLSPEC void* SDLCALL SDL_AtomicGetPtr(void **a); |
| 398 | |
| 399 | /* Ends C function definitions when using C++ */ |
| 400 | #ifdef __cplusplus |
| 401 | } |
| 402 | #endif |
| 403 | |
| 404 | #include "close_code.h" |
| 405 | |
| 406 | #endif /* SDL_atomic_h_ */ |
| 407 | |
| 408 | /* vi: set ts=4 sw=4 expandtab: */ |
| 409 | |