- outlined benchmarking guidelines
→ all about time taken to do a task.
- We came up with an abstracted, idealized relationship for CPU time in terms of 3 properties
→ Clock period
→ Instruction count
⇒ once the compiler has done its thing, this is a structured path through the execution. not super dynamic.
→ Cycles per instruction
⇒ What is that? why is this one so important?
• this is a function of many different things, while the other two are pretty much fixed.
• What makes it so variable?
◇ ISA
◇ instruction type
▪ multiplication is longer than division, integer ops are faster than floating point, etc...
◇ sequencing
▪ we like to thnk of execution as sequential, until we come across a conditional which could send us to a different place, maybe somewhere that has already been executed
▪ spatial vs temporal locality.
- now we have to establish the properties of a “basic” cpu from the perspective of a dataflow.
- Assume a RISC architecture
Index