Memory Barriers in multi-threaded applications

Sometimes, the compiler can get in the way and break your code because it makes single-thread assumptions about your code. In addition, the microprocessors can break your code because it makes single-core assumptions. The way to fix this is use a memory barrier.

Here is a very good description of the problem of multi-threading.


Here is how to do it on a PowerPC, using the lwsync instruction.

volatile unsigned variable1 = 0;

#define barrier() __asm__ volatile (”lwsync”)

#define ITERATIONS 50000000

void *writer(volatile unsigned *variable2) {
        utilBindThreadToCPU(0);
        for (;;) {
                variable1 = variable1 + 1;
                barrier();
                *variable2 = *variable2 + 1;
        }
        return NULL;
}