gcn: Fix CDNA3 atomics' buffer invalidation

gcn: Fix CDNA3 atomics' buffer invalidation

For device (agent) scope atomics - as needed when there is more than one teams,
a buffer_wbl2 followed by s_waitcnt is required. When doing the initial porting,
the pre-atomic instruction got accidentally replaced by buffer_inv sc1, which is
not quite the right instruction.

gcc/ChangeLog:

	* config/gcn/gcn.md (atomic_load, atomic_store, atomic_exchange):
	Fix CDNA3 L2 cache write-back before atomic instructions.