next up previous

Exercise 12: Registers and the performance of the SAXPY benchmark.

What is your understanding of the connection between scalar registers, vector registers, and floating point units for the performance of the SAXPY benchmark?

Does vector chaining improve performance on the SAXPY benchmark? What could you infer about the organization of the data path and the connection between scalar registers, vector registers, and floating point units if you measured the performance of SAXPY and found out that operands were indeed being chained between data units?