We would like to execute the loop below as efficiently as possible. We have two different machines, a MIMD machine and a SIMD machine.

for (i=0; i<2000; i++)
	for (j=0; j<3000; j++)
		X_array[i][ j] = Y_array[j][i] + 200;

1.1 [10] <§6.3> For a four CPU MIMD machine, show the sequence of RISC-V instructions that you would execute on each CPU. What is the speed-up for this MIMD machine?

1.2 [20] <§6.3> For an eight-wide SIMD machine (i.e., eight parallel SIMD functional units), write an assembly program in using your own SIMD extensions to RISC-V to execute the loop. Compare the number of instructions executed on the SIMD machine to the MIMD machine.

A systolic array is an example of an MISD machine. A systolic array is a pipeline network or “wavefront” of data processing elements. Each of these elements does not need a program counter since execution is triggered by the arrival of data. Clocked systolic arrays compute in “lock-step” with each processor undertaking alternate compute and communication phases.

2.1 [10] <§6.3> Consider proposed implementations of a systolic array (you can find these on the Internet or in technical publications). Then attempt to program the loop provided in Problem 1 using this MISD model. Discuss any difficulties you encounter.

2.2 [10] <§6.3> Discuss the similarities and differences between an MISD and SIMD machines. Answer this question in terms of data-level parallelism.