SimpleOpenCL: desenvolupament i documentació d'una llibreria que facilita la programació paral·lela en OpenCL

Abstract

Computation always was related with big and expensive computers or clusters. But it changed the day that some people decide to use GPUs for general purpose computation instead of graphics rendering. GPUs are good for parallel computing because its high performance multi-core processors. It is like returns to the past, to coprocessors age but with GPUs. Main vendors implemented low level hardware programming interfaces (I.e. AMD close-to-metal and Nvidia CUDA) to take advantage of modern GPUs. But the (big)drawback about these solutions is that applications created with theses SDKs are only compatibles with their vendor hardware. This is where OpenCL comes in. OpenCL borns as an open standard. It let heterogeneous computing in a heterogeneous world. But programming with OpenCL is not easy at all. It is a low level API and after coding some applications you will see some patterns repeated. SimpleOpenCL is a library that simplifies programming with OpenCL. It is an open source project with GPL v3 license. The main idea behind this thesis is to analyze SimpleOpenCL for improving it and offer proposals for its future, but to achieve it I need to talk at first about OpenCL.