4 points, Due Tuesday, February 15
(1 point) Show 4 unrolled iterations of the MIPS loop from the web page A Comparison of Matrix Multipication Code.
(1 point) Suppose that the code is completely unrolled and executed on a single-issue processor that uses register renaming. The values produced by L.S, ADD.S, and MUL.S instructions are available 2, 5, and 7 cycles, respectively, after the instruction begins execution. An instruction can begin execution in the same cycle that its source operands are available. Why is it impossible for the processor to eventually achieve a CPI of 1.0? Which instruction do you need to speed up to achieve a CPI of 1.0? Give reasons for your answers.
(1 point) Rewrite the unrolled loop code so that you get a CPI of 1.0.
(1 point) Show 4 unrolled iterations of the MIPS loop in Exercise 4.9 in the text.