...

I have noticed that for Windows, atomic_write32 is defined thus:

inline void atomic_write32(volatile boost::uint32_t *mem, boost::uint32_t val) { winapi::interlocked_exchange(reinterpret_cast(mem), val); { }

Whereas otherwise like this:

inline void atomic_write32(volatile boost::uint32_t *mem, boost::uint32_t val) { *mem = val; }

the reads are both: return *mem;

Windows documentation states that: 'Simple reads and writes to properly-aligned 32-bit variables are atomic operations.' [http://msdn.microsoft.com/en- us/library/windows/desktop/ms684122%28v=vs.85%29.aspx]

Is there a reason for this?

Those operations require also memory barriers. I'm not an expert on atomic operations so the code might have several errors. My mid-term plan is to port it to Boost.Atomic.

btw, helge was planning to provide a separate implementation of boost.atomic for interprocess, because his blocking emulation of atomics depends on a per-process spinlock pool. tim