Windows: atomic_fetch_sub()

[danturner]

There’s no implementation of atomic_fetch_sub() in Kokkos_Atomic_Windows.hpp.

[danturner]

Here's the implementation that I used in my local repo:

template < typename T >
KOKKOS_INLINE_FUNCTION
T atomic_fetch_sub(volatile T * const dest,
typename Kokkos::Impl::enable_if< sizeof(T) == sizeof(long), const T >::type val)
{
union { long i; T t; } assume, oldval, newval;

  oldval.t = *dest;

  do {
      assume.i = oldval.i;
      newval.t = assume.t - val;
      oldval.i = _InterlockedCompareExchange((long*)dest, newval.i, assume.i);
  } while (assume.i != oldval.i);

  return oldval.t;

}

[crtrott]

Pushed to master