EEC 272: High-Performance Computer Architecture

Prof. Vojin G. Oklobdzija

Grading:

Midterm: 30%

Project: 20%

Quizzes: 40%

Extras: 10%

Catalog Description:

Architectural issues in achieving high-performance in a single processor via concurrent execution of instructions that typically form a general-purpose computer as well as problems and fundamental limitations will be discussed. The course will also cover specialized architectures such as VLIW and Vectors.

Expanded Course Description:

1. Introduction to Super-scalar Concepts:

1.1 Beyond Pipelining, CISC and RISC

1.2 Instruction Issue and Machine Parallelism

1.3 Fundamental Limitations

1.3 Related Concepts: VLIW and Vectors

1.4 Unrelated Parallel Schemes

2. Developing and Execution Model:

2.1 Simulation Technique

2.2 Benchmarking Performance

2.3 Basic Observations on Hardware Design

2.4 The Design of the Standard Processor

2.5 Procedural Dependencies

3. Instruction Fetching and Decoding

3.1 Branches and Instruction-Fetch Inefficiencies

3.2 Improving Fetch Efficiency

3.3 Implementing Hardware Branch-Prediction

3.4 Implementing Four Instruction Decoder

3.5 Implementing Branches

3.6 Reducing the Penalty of Procedural Dependencies

4. Advanced Pipelining:

4.1 Pipeline Design

4.2 Pipeline Scheduling - reservation tables

4.3 Pipeline Hazards and conflict resolution

4.4 Multi-level pipelines

5. The Role of Exception Recovery:

5.1 Buffering State Information for Restart

5.2 Restart Implementation and Effect on Performance

5.3 Processor Restart

6. Register Dataflow:

6.1 Dependency Mechanisms

6.2 Result Buses and Arbitration

6.3 Result Forwarding

6.4 Supplying Instruction Operands

7. Out-of-Order Issue:

7.1 Reservation Stations

7.2 Implementing a Central Instruction Window

7.3 Out of Order Issue

8. Memory Dataflow:

8.1 Ordering of Loads and Stores

8.2 Addressing and Dependencies

8.3 What is more Load/Store Parallelism worth ?

8.4 Multiprocessing Considerations

8.5 Accessing External Data

9. Complexity and Controversy

9.1 Design Complexity

9.2 Major Hardware Features

9.3 Hardware Simplifications

9.4 Is the Complexity Worth ?

10. Evaluating Alternatives: A Perspective on Superscalar Processors

10.1 The Case for Software Solutions

10.2 The Case for Hardware Solutions

10.3 VLIW Machines

10.4 Vector Processors (Cray)

10.5 Massively Parallel Machines

Textbook:

1. Mike Johnson, "Superscalar Microprocessor Design", Prentice Hall 1991.

2. Hennessy-Patterson, "Computer Architecture: A Quantitative Approach", Morgan-Kaufman Publishing.

3. V. Oklobdzija, Set of selected papers and notes.