Careful review appreciated for this one in particular.
This changes the implementation of FEBs so that a memory fence occurs even when the output value passed to the FEB routine is null. This matters because prior to this change, there's not necessarily a guaranteed fence implied by calls like readFF and the tests frequently use readFF to decide that it's safe to read some value that may be completely distinct from the FEB that's been marked as full.
Careful review appreciated for this one in particular.
This changes the implementation of FEBs so that a memory fence occurs even when the output value passed to the FEB routine is null. This matters because prior to this change, there's not necessarily a guaranteed fence implied by calls like
readFF
and the tests frequently usereadFF
to decide that it's safe to read some value that may be completely distinct from the FEB that's been marked as full.