Cache coherence

Cache coherence is the guarantee that all cores see a single, consistent view of memory despite each core maintaining its own private cache. Without it, one core's write to a cached variable could be invisible to another core's read, violating fundamental correctness.

Coherence protocols fall into two families: snooping (broadcast on a shared bus) and directory-based (track sharers in a directory lookup). Coherence differs from consistency, which governs ordering of writes to different locations.

This example demonstrates cache coherence by writing and reading a shared variable across threads.

// compile: gcc -std=c11 -pthread -o coherence coherence.c
// run: ./coherence
// description: show cache coherence maintaining visibility across cores
 
#include <stdio.h>
#include <pthread.h>
#include <stdatomic.h>
#include <unistd.h>
 
atomic_int x = 0;
 
void* writer_thread(void* arg) {
    sleep(1);
    atomic_store(&x, 42);
    printf("Writer: wrote x = 42\n");
    return NULL;
}
 
void* reader_thread(void* arg) {
    for (int i = 0; i < 10; i++) {
        int val = atomic_load(&x);
        if (val == 42) {
            printf("Reader: saw x = 42\n");
            return NULL;
        }
        usleep(100000);
    }
    return NULL;
}
 
int main() {
    pthread_t writer, reader;
    pthread_create(&writer, NULL, writer_thread, NULL);
    pthread_create(&reader, NULL, reader_thread, NULL);
    pthread_join(writer, NULL);
    pthread_join(reader, NULL);
    return 0;
}