Consider the following code. Show how it might be complied into generic assembly language assuming that the compiler unrolls the loop three times. All values are integers. Assume that register indirect with offset addressing is available, but pointers have to be explicitly updated.



for (int i = 0; i < 30; i++) {

z[i] = 3 * (x[i] + y[i]);

}

Assume that r0, r1, and r2 have been set up to point to x, y, and z, respectively. One trip round the loop might

be




Loop LDR r3,[r0] ;get x[i]


ADD r0,r0,#4 ;update pointer


LDR r4,[r1] ;get y[i]


ADD r1,r1,#4 ;update pointer


ADD r4,r4,r3 ;x[i] + y[i]


ADD r4,r4,r4,lsl #1 ;3(x[i] + y[i])


STR r4,[r2] ;store z[i]


ADD r2,r2,#4 ;update pointer


SUBS r5,r5,#1 ;dec loop counter


BNE Loop

If we unroll the loop three times, we get




Loop LDR r3,[r0] ;get x[i]


LDR r4,[r0,4] ;get x[i+1]


LDR r5,[r0,8] ;get x[i+2]


ADD r0,r0,#12 ;update pointer by 3 elements


LDR r6,[r1] ;get y[i]


LDR r7,[r1,4] ;get y[i+1]


LDR r8,[r1,4] ;get y[i+2]


ADD r1,r1,#12 ;update pointer


ADD r6,r6,r3 ;x[i] + y[i]


ADD r7,r7,r4 ;x[i+1] + y[i+1]


ADD r8,r8,r5 ;x[i+2] + y[i+2]


ADD r6,r6,r6,lsl #1 ;3(x[i] + y[i])


ADD r7,r7,r7,lsl #1 ;3(x[i+1] + y[i+1])


ADD r8,r8,r8,lsl #1 ;3(x[i+2] + y[i+2])


STR r6,[r2] ;store z[i]


STR r7,[r2,4] ;store z[i]


STR r8,[r2,8] ;store z[i]


ADD r2,r2,#12 ;update pointer


SUBS r9,r9,#3 ;decrement counter


BNE Loop

We have to increment the pointer by 3 × 4 = 12 at each trip round the loop because we have dealt with three

elements. The following snapshot gives the assembly window from the ARM debugger (using 9?element arrays).