| When we meet some complex problems, the traditional mathematical modeling and experiment technology can not solve, such as large scale natural phenomenon, the physics process of transient, some of the problems with complex boundary. Due to the rapid development of computer in recent years, people has began to through the combination of computer and mathematical model to solve the similar problems, it has achieved remarkable result, because of the increase of the model leads to increased computation time, through the silicon technology to boost the frequency of the CPU has reached the bottleneck, who wanted to increase the computing speed by improving CPU frequency is very difficult, so, people sights on multi-core parallel and Intel multi-core processors have become increasingly mature, also invented the thread building blocks can be used in the multi-core programming, through the TBB(thread building blocks), will greatly improve the efficiency of some program.1D3V PIC simulation softtware BUMBLEBEE is about laser-plasma interaction, because of its running time is long, and it is suitable for parallel, so this article is based on Intel threading building blocks, parallel the serial program of 1D3 V PIC simulation software BUMBLEBEE, the main work of the article includes:1. Make a simple introduction to the background of TBB and it’s parallel environment Intel Parallel Studio, through the comparison of other programming languages such as MPI and OpenMP, describe the reason why do we choose TBB.2. Introduce the data structure of the serial code, including the function of each module and the execution sequence of each module, then analyzed the hot spots of serial code through tools in Intel Parallel Studio.3. Made detailed introduction of the TBB template what will be used in the article, including loop parallel template parallel_for and task scheduler templates, made a brief introduction to other function.4. Detailed describes the concrete steps of parallel, doing the analysis of the data competition problems in the process of parallel, detailed description of the solution about Data competition and the priciple, show the speed-up ratio of the parallel program relative to serial program and the scalability of parallel programs, the speed-up radio close to linear under the four nuclear and the extensible is good.5. Compared the parallel program and the operation result of serial program(run results including field changes, the change of particle position, etc.), the final result completely consistent, illustrates the parallel was successful, in the final analysis the changing process of the operation results over time. |
PDF Full Text Request