Can GCC use read-modify-write instructions to update volatile variables?

Question

Suppose you have C code for x86 systems code like this:

volatile uint32_t *reg = (volatile uint32_t *)0xCAFEBABE;
// ...
reg[0x10] |= 1;

Is GCC free to generate a read-modify-write instruction here? If so, will encapsulating the read and write to the volatile variable in functions make sure that GCC does not combine the accesses into a single RMW instruction?

I know that the C spec is intentionally vague about this.

Answer 1

The C standard does not specify what instruction is to be used. For a simple update like this gcc is most liklely to generate a single RMW instruction the style of

orl (%rdx), $1

The volatile keyword has nothing to do with it though. any combination of load, modify, store would have been valid too. The volatile keyword only tells the compiler to reload the value from memory on every use instead of using register caching optimisations.

If you want an atomic update then you need atomics, gcc provides for this _sync_fetch_and_or(type *, type)