How does the IA64 increase the efficiency of software pipelining?

What will be an ideal response?

Consider a loop that evaluates P = Q + R, where P, Q and R are vectors. In a trip round the loop, the value of pi =
qi + ri is calculated. If software pipelining is used, the value of qi and ri are looked up, the value of qi?1 is added to
qi?1, and pi?2 is stored in memory. As you can see, we have registers qi, qi?1, and qi?2 where the subscripts
represent the registers at different times (i.e., iteration cycles). Efficient software pipelining requires an efficient
means of handling registers. The Itanium provides this by using automatic register re?numbering.
A problem with software pipelining is that the pipeline has to be filled initially and then emptied. For example,
if there are three steps in the pipeline, the last three cycles will perform three steps, two steps, and finally one
step as the pipeline drains. The Itanium is able to use its predicated operation mechanism to ensure that 3, 2
and 1 operations get performed in the final stages of the loop.