Platform Level Design Sample Clauses

Platform Level Design. In order to map an algorithm to hardware, several steps have to be followed. These steps have been presented extensively in D3.1, but for readability of this deliverable a brief summary will be presented. The steps that the designer follows entail algorithm study, design, and implementation. In order to study the algorithm the designer studies the nature of the algorithm and the size of inputs and outputs (called the dimensioning of the required implementation). Depending on these characteristics the designer should decide if some protocol implementation (e.g. for I/O) or any other module is needed. The designer also studies the size of inputs and outputs in order to decide how to process these data, e.g. if all the data are available on-chip or external memory is required, etc. Subsequently the designer decides on the data sets that are representative for the algorithm. Depending on the data set, algorithms may vary in their statistical characteristics, the size of their implementation, or their parallelization characteristics. When the data set characteristics are determined the designer can get the right decisions and validate his design properly. Using these data sets, the next step is algorithm profiling, where the designer tries to find the most compute- intensive part of the algorithm that it is going to be mapped to hardware. The designer also measures the ―memory footprint‖ of the algorithm (i.e. the memory requirements) in order to decide about data manipulation e.g. reorder data in main memory. The final step for the study procedure is to identify the important Data Structures and Operations. This step helps the designer to understand algorithmic parallelization and the type of the modules that are needed, e.g. 32- or 64- bits floating point arithmetic vs. fixed-point arithmetic. The design procedure follows, where the designer has to do Top Down analysis where he/she analyzes the algorithm in abstract blocks, giving functionality description and description for their interface. Bottom Up modeling follows where the designer models the functionality and interfaces for each abstract block to build the subsystem. Then the designer debugs the model starting from small and scaling to bigger system blocks, using data from the given data sets. The last step for the design procedure is to do verification of the designed model, in which the designer uses the representative datasets in order to verify that his/her model works exactly the same as the origi...