The processor company has designed a new $10$

stage pipeline $($shown below$).$ The new pipeline is

designed to perform better than the standard $5-$stage

pipeline. The useful results for the purpose of

forwarding are available at the last step in each stage

$($F$2,$ R$1,$ X$3,$ M$3,$ W$1).$ For example the ALU result

is not available until the end of the $x2$ stage.

F$1$ F$2$ R$1$ X$1$ X$2$ X$3$ M$1$ M$2$ M$3$ W$1$

$10\mathrm{.}1(5$pt$)$ Is it possible to eliminate data hazards

between R$-$type instructions in this pipeline by using

forwarding? Why or why not?

$10\mathrm{.}2(10$pt$)$ In the code fragments below, identify the

data hazards and describe the number of pipeline

stalls $($or bubbles$)$ required even with your full

bypassing.

$($a$)$

mul $$4,$$$5,$$$6$

add $$8,$$$9,$$$4$

$($b$)$

lw $$1,8($$$2)$

add $$3,$$$1,$$$4$

$10\mathrm{.}3(10$pt$)$ Suppose that branches are predicted as

not$-$taken until the actual branch decision is known.

Also, suppose that the branch target is known at the

end of the X$1$ stage. How many in$-$progress

instructions must be flushed $($invalidated$)$ when a

branch is taken? $($Or how many bubbles are required

for branch mis$-$prediction?$)$ Explain the reason.