Moved SMP atomic operations from TBB to General.

Even when running with the serial backend, the compiler might enable SIMD vectorization when optimizations are turned on. When this occurs, we need to use properly atomic Add's and CAE's.