a$)$ Consider the following loop

for $iin$ range $(256)$ :

$ifx[i]0$

$,x[i+1]=x[i]+Y[i]-z[i]$

Is the above loop vectorizable? If not, then show where the problem lies.

b$)$ Compare the Intel's SIMD extension instructions and RISC$-$V vector instructions.

c$)$ Consider the following code where $\text{'}$ and $B$ are m$$mory$-$based arrays and $\text{'}s\text{'}$ a double precision FP

number also stored in memory:

for $\{\begin{array}{ccc}\text{:}[i& in\text{r}\text{a}\text{n}\text{g}\text{e}& (200)\text{:}]\\ A[i]=A[i]**s\\ B[i]=B[i]+A[i]\end{array}$

The following table gives the stall cycles required between instruction producing the result and the

instruction using that result.

The corresponding assembly code without any scheduling is given as under:

Suppose a three$-$issue VLIW machine is designed that can issue one ALU$/$Branch$,$ one FP operation,

and one memory reference instruction in every clock cycle. Schedule the above code on this VLIW

processor so that the stalls are minimized and IPC is maximized. You can also unroll the loop if

required for better performance. All forwarding paths are available. What is the maximum IPC you

achieved? Compare it with the theoretical maximum.