Image Processing Using CUDA: Designing an object oriented framework for CUDA based image processing (Paperback)

Nowadays, GPUs are multi-core parallel processors with very high memory bandwidth. Recently for last few years developers and researchers are offloading computationally intensive tasks to GPUs to gain significant speedup compared to the CPU. Most of the image processing algorithms perform same operation on a large number of pixels, which can be parallelized on single instruction multiple data GPU architecture. Initially shader based frameworks like GLSL, Cg etc have been used for GP-GPU. NVIDIA has developed CUDA to write scalable parallel programs in C like language. GLSL, Cg and CUDA provide significant speedup, however they offer complex integration frameworks and require specialized programming skills. Most of GPU based frameworks are developed using procedural programming, which limits the scope of flexibility, code reusability, information and complexity hiding. In this book, we bring in an object oriented framework for the CUDA based image processing. We demonstrate a set of design patterns exploiting programming advantages of an object oriented language, such as encapsulation, information hiding, code reusability, complexity hiding and extensibility.

如今，GPU已成為具有非常高記憶體頻寬的多核並行處理器。近年來，開發人員和研究人員將計算密集型任務轉移到GPU上，以實現與CPU相比的顯著加速。大多數圖像處理算法對大量像素執行相同操作，可以在單指令多數據的GPU架構上進行並行處理。最初，像GLSL、Cg等基於着色器的框架被用於通用GPU計算。NVIDIA開發了CUDA，用於使用類似C的語言編寫可擴展的並行程序。GLSL、Cg和CUDA提供了顯著的加速效果，但它們提供了複雜的集成框架，需要專門的編程技能。大多數基於GPU的框架都是使用程序化編程開發的，這限制了靈活性、代碼可重用性、信息和複雜性隱藏的範圍。在本書中，我們引入了一個基於CUDA的面向對象框架，用於圖像處理。我們展示了一組設計模式，利用面向對象語言的編程優勢，如封裝、信息隱藏、代碼可重用性、複雜性隱藏和可擴展性。