Block Parallel Programming.pdf

Block parallel programming is a technique used for real-time applications on multi-core processors. A dissertation by David Black-Schaffer presents a streaming block-parallel programming language for describing applications with hard real-time constraints. The language parameterizes data movement within the application, enabling simple analysis and rapid software development. Key features include native two-dimensional data streams, flexible control distribution, explicit data dependency edges, and multiple computation methods per kernel.

The dissertation introduces a 2D streaming model and provides a block programming example. It also discusses contributions, including a new programming language and transformations for parallelizing and mapping applications to many-core architectures. The thesis overview outlines the structure of the dissertation, which includes background, related work, application model, application analysis, buffers and insets, and conclusions.

The application model consists of an application graph, data model, and computation model. The application graph represents the application as a graph of kernels and streams. The data model describes the data used by the application, including inputs, outputs, and tokens. The computation model specifies the computations performed by the kernels.

The dissertation also discusses related work, including StreamIt, StreamC, KernelC, Brook, Sequoia, and multi-dimensional synchronous data flow. It presents a summary of the related work and discusses the contributions of the dissertation.

Application analysis is performed using frame sizes, frame rates, and iteration sizes. Data flow analysis is used to analyze the application and determine the required buffers and insets. The dissertation provides an example of application analysis and discusses the results.

Buffers and insets are used to manage data flow and ensure that the application meets its real-time constraints. The dissertation discusses buffer sizing, implementation, and automatic insertion of buffers and insets. It also presents a data flow analysis for insets and discusses zero padding.

Overall, the dissertation presents a new programming language and techniques for block parallel programming on multi-core processors. It provides a detailed analysis of the application model, application analysis, and buffers and insets, and discusses the contributions and related work.