I think it can handle it because the load and store are retired on the
same cycle. The dependency can be resolved during completion.
According to SimG5, this is exactly what SimG5 does:
Nope, I made a mistake. .align doesn't guarantee dispatch group
alignment, it just perturbs it. In this case, it got it just into the
right phase where the store was not in the same dispatch group as the
loads. Once this is rectified, then the simulator shows a dispatch
reject. So, if SimG5 is to be believed, we need to update the
documentation to remove the store forwarding comment.
