The point
Produce a number you would be willing to defend to someone who wanted to disagree with it.
Do this
- Write a benchmark with a hot nested loop running a few thousand iterations.
- Code size. Record the TAC instruction count before and after optimization, plus the per-technique scorecard.
- Work done. Run both versions in SPIM and record instructions executed. To get an unoptimized
.s, point the back end at the unoptimized list. - Put the four numbers in a table and compute both percentages. They will not match — explain the gap in one sentence, in terms of what was inside the loop.
- Repeat three times. How much do the numbers move? Anything smaller than that spread is not a result.
- State a threshold. Given your machine and your measured noise, what reduction would you call significant? Justify it with the spread from step 5, not with a feeling.
Deliverable
A results table with all four numbers and both percentages; three repeated runs with the spread; the benchmark source; and a defended threshold. This is Lab Question 22 and a large part of the Project 6 performance section — do it properly now and reuse it in December.