Pipelining Coputing
- 1. COE 504 – Heterogeneous Computing Project Presentation Presenter: Mohammed Omer Assayony(201102150) 1
- 2. Pipelined Computation Pipelined Computation Applications ◦ Type 1 : Multiple instances of the complete problem ◦ Type 2 : Overlapping Pipeline Stages DDS Architecture for Pipelining Computation 2
- 3. Pipelined computation is a parallel technique used when the problem could be divided into series of tasks that have to be completed in sequential order. One form of functional decomposition: ◦ Each task (function) is implemented as a process, called a stage. ◦ Each stage contributes to the overall solution of the problem and passes on the required data for subsequent stage. 3
- 4. 4
- 5. Though the execution of pipelining seems sequential, it can provide reasonable speedup in TWO TYPES of computations: Type1: When more than one instance of the complete problem is to be executed. Type 2: When the data required starting the next state can be passed forward before the current stage has completed all its internal operation. 5
- 6. Pipelining increases the throughput (number of instances completed per time unit). Typical Application Examples: 1. Pipelining Addition. 2. Frequency Filtering. 3. Sound Digitizing……... 6
- 7. Time for m instances = (pipeline fill + number of instances) x stage delay = ( p- 1 + m ) x stage delay Throughput = m/((p- 1 +m) x stage delay) If m >> p Throughput = 1/stage delay 7
- 8. In this type of computation, pipelining improves the execution time of a single instance due to computation overlapping. The gained improvement depends on communication The fast the communication, the better the performance Typical Application Examples: ◦ Linear Algebra Algorithm: Gauss Elimination, Back-substitution ……. 8
- 9. 9
- 10. Report Result Subscriber Publisher Subscriber Publisher Subscriber Publisher Processin g Processing Processing Process 1 Process 1 General DDS Architecture for Pipelining Computation 10
- 11. Q&A 11