Hardware Reference
In-Depth Information
P[i,j]
R 1
0
R 2
0
R 3
0
ready to grant
G 1
G 2
G 3
Fig. 4.9 The logic-level implementation of the arbitration logic that receives the input requests
and computes the necessary grants based on the relative priorities of the inputs. Priority update
logic (not shown in this figure) is designed according to the selected priority update policy
how the priority matrix is updated for the next arbitration round. The arbitration
logic remains exactly the same in all cases.
￿
Least recently granted: Once the i th request is granted, its priority is updated
and set to be the lowest among all requestors. This is performed at first by
clearing all bits of the i th row, e.g., setting PŒi; to 0, and secondly by setting
the bits of the i th column, e.g., PΠ;i D 1, so that all other requests will have
higher priority over request i .
￿
Most recently granted: Once the i th request is granted, its priority is updated
and set to be the highest among all requestors. This is performed at first by setting
all bits of the i th row, e.g., setting PŒi; to 1, and secondly by clearing the bits
of the i th column, e.g., PΠ;i D 0, so that all other requests will have lower
priority over request i .
￿
Incremental Round robin: Under round-robin policy the request that has been
granted in the current arbitration cycle should receive the lowest priority in the
next cycle. With 1D priority state this is performed in a relatively easy way but
becomes very complex in 2D priority representation. Round-robin like operation
(or incremental round-robin) proposed in Satpathy et al. ( 2012 ) can be achieved
by downgrading the position with the highest priority irrespective if it received
 
Search WWH ::




Custom Search