How to guarantee 64-bit writes are atomic?

Question

When can 64-bit writes be guaranteed to be atomic, when programming in C on an Intel x86-based platform (in particular, an Intel-based Mac running MacOSX 10.4 using the Inte

Answer 1

On X86, the fastest way to atomically write an aligned 64-bit value is to use FISTP. For unaligned values, you need to use a CAS2 (_InterlockedExchange64). The CAS2 operation is quite slow due to BUSLOCK though so it can often be faster to check alignment and do the FISTP version for aligned addresses. Indeed, this is how the Intel Threaded building Blocks implements Atomic 64-bit writes.

Answer 2

If you want to do something like this for interthread or interprocess communication, then you need to have more than just an atomic read/write guarantee. In your example, it appears that you want the values written to indicate that some work is in progress and/or has been completed. You will need to do several things, not all of which are portable, to ensure that the compiler has done things in the order you want them done (the volatile keyword may help to a certain extent) and that memory is consistent. Modern processors and caches can perform work out of order unbeknownst to the compiler, so you really need some platform support (ie., locks or platform-specific interlocked APIs) to do what it appears you want to do.

"Memory fence" or "memory barrier" are terms you may want to research.