====== Pipelining ======

This lecture covers the design and implementation of the MIPS CPU using a pipeline. This implementation addresses the two main faults of the single-cycle CPU //not// by eliminating hardware redundancy but by allowing more than one instruction to execute at the same time thereby no hardware unit will be idle. The high latency of the single-cycle approach is also //not// minimized but it is offset by high throughput. Stage balancing, pipeline structure, and hazard detection and avoidance are the key drivers of pipelining.

===== Outline =====

  * The pipelining concept, balancing, and latency versus throughput
  * The need for pipeline registers
  * The MIPS ISA is designed to eliminate structural hazards
  * Data hazards and forwarding
  * Inserting a bubble into a pipeline
  * Handling branch hazards
  * Delayed branching and populating the delay slot

===== Big Ideas =====

  * One can increase throughput despite an increase in latency
  * For a process to be pipelined, the stages must be balanced and no structural violations should be allowed
  * Dependencies tend to lower the pipeline's throughput

===== Slides from Lecture =====


  * [[http://www.cse.yorku.ca/~roumani/course/2021/PL.pdf|Pipelining]]





===== To Do =====

  * Read Sections 6.1 through 6.7 of the textbook.
  * Do the pipeline exercises (in the Resources page).