Lazy Foo' Productions

Atomic Operations

Last Updated 6/02/14
Semaphores operate at an operating system level. Atomic operations are a way to lock data at a efficient CPU level. Here we'll be locking a critical section using GPU spinlocks.
//Data access spin lock SDL_SpinLock gDataLock = 0; //The "data buffer" int gData = -1;
Instead of a semaphore we'll be using a spinlock to protect our data buffer.
bool loadMedia() { //Loading success flag bool success = true; //Load splash texture if( !gSplashTexture.loadFromFile( "48_atomic_operations/splash.png" ) ) { printf( "Failed to load splash texture!\n" ); success = false; } return success; } void close() { //Free loaded images; //Destroy window SDL_DestroyRenderer( gRenderer ); SDL_DestroyWindow( gWindow ); gWindow = NULL; gRenderer = NULL; //Quit SDL subsystems IMG_Quit(); SDL_Quit(); }
Unlike semaphores, spin locks do not need to be allocated and deallocated.
int worker( void* data ) { printf( "%s starting...\n", data ); //Pre thread random seeding srand( SDL_GetTicks() ); //Work 5 times for( int i = 0; i < 5; ++i ) { //Wait randomly SDL_Delay( 16 + rand() % 32 ); //Lock SDL_AtomicLock( &gDataLock ); //Print pre work data printf( "%s gets %d\n", data, gData ); //"Work" gData = rand() % 256; //Print post work data printf( "%s sets %d\n\n", data, gData ); //Unlock SDL_AtomicUnlock( &gDataLock ); //Wait randomly SDL_Delay( 16 + rand() % 640 ); } printf( "%s finished!\n\n", data ); return 0; }
Here our critical section is protected by SDL_AtomicLock and SDL_AtomicUnlock.

In this case it may seem like semaphores and atomic locks are the same, but remember that semaphores can allow access beyond a single thread. Atomic operations are for when you want a strict locked/unlocked state.
Download the media and source code for this tutorial here.

Back to SDL Tutorials