The class main object is to study and analyze the main techniques for the design of modern high-performance computers. Starting from the pre-requisites of the design of the modern instruction sets, the focus moves on identifying the main techniques for the realization of the processors. During the class, the approach followed assumes that instruction set and processor realization are highly correlated topics.
Main object of the techniques that are illustrated is to reduce as much as possible the penalization due to the processor hardware structure, aiming at an execution model where the execution order is no longer embedded in the source code, but more carried by the way the information is processed conforming to the elaboration flow design by the software designer.
In this way it is possible to realize systems able to automatically extract the intrinsic instruction level parallelism (ILP) by means of architectures that allow the simultaneous execution of multiple instruction during the same clock period.
The class will illustrate general purpose processors’ applications (RISC and Intel architectures), together with applications for specific processor (such as embedded processors, DSP, graphical and multimedia processors, I/O processors).
In the last par, the class will show the main lines along which the next generation processor will develop, as far as algorithmic approach is concerned (such as VLIW architectures) and also concerning non algorithmic architectures (such as neural and fuzzy processors).
Lectures and exercises |
hours |
Topics |
Specific contents |
|
The base processor architecture |
The ISA. Control and datapath blocks. Hardwired control. Micro-programmed control.
|
10
|
The pipeline architecture |
Pipeline structure. Pipeline hazards: Structural hazards, Data hazards, Control hazards. The pipeline for multi-cycle instructions.
|
10
|
Advanced techniques for Instruction Level Parallelism (ILP) |
Principal concepts. Dynamical scheduling: Tomasulo algorithm. Static and dynamic branch prediction. Multiple issue architectures. Evolution: superscalar architectures
Revolution: VLIW architectures. Register level parallelism
RISC and DSP applications. |
10
|
Non-conventional architectures |
Non-algorithmic computation architectures. Parallel architectures. High-level language processors.
|
10
|
Total hours for lectures and exercises |
40 |
for exercises only |
|
Further educational activities
|
hours
|
Labs |
|
Tutorials / Seminars |
|
Workshops |
|
Guided tours |
|
|
|
Total hours for further educational activities |
0 |
Total hours |
40
|