Computer Architecture

I hate ts

Pipeline depth tradeoffs?

More depth means more throughput but results in more stalls if hazard

Structural hazards

multiple inst need the same hardware resource at the same time

Data hazard

an inst needs a result that hasn’t been written back

Control hazard

doesn’t know if a branch will get taken or not

What does a BTB do

Stores target jump addy of a previous branch

BTB vs branch predictor

branch predictor - do I take it or not
BTB - where do I jump if take

Direct-mapped cache and pros/cons

Each addy maps to exactly one cache location

Pro: fast lookup

Con: thrashing

n-way set-associative cache

each addy has n spots it can map to

Pro: reduce thrashing

Con: complex

Full-associative cache

can go anywhere
Pro: no thrashing
Con: expensive, slow to search

Compulsory miss

first accessCap

Capacity

cache too small

Conflict

multiple addy mapping to the same space despite free space elsewhere