128- bit integer atomics Support?

I need that very mutch for a hash table!

You could try implementing your own atomic operations using atom_cmpxchg. For each 128-bit variable, you would need a 32-bit flag variable that indicates whether the variable is locked by a thread.

You could use atom_cmpxchg to test the flag variable and set it to the locked state if no other thread has locked it already.