【作者】 刘峰；

【导师】 彭群生； 郑文庭；

【作者基本信息】 浙江大学 ， 计算机应用， 2005， 硕士

【摘要】 本文提出了一个利用图形处理器(GPU)加速分形布朗运动(FBM)算法的新方法，并成功地将其应用于自然场景中云的形状模拟。我们对原算法进行细化，将每一次迭代分为两次子迭代，以满足GPU中并行处理的要求，并进一步在子迭代中减少条件判断的次数。将新算法的计算任务动态划分给CPU和GPU，以充分发挥它们各自的长处。本文还提出了一种通过建立随机数序列的方法来有效地模拟云的生长、消散和飘动等动态效果。 在此基础上，本文将FBM算法生成的云分割成多个云层，从而将平面的云扩展到三维，并将上面提到的方法独立应用到每一云层上，实现三维云的动态变化。 最后，本文提出了一种将FBM算法和基于流体的方法相结合的方法，通过求解Navier-Stokes方程得到基于物理模型的云的动态效果。本文利用GPU来加速Navier-Stokes方程的求解，从而提高了云动态模拟的计算速度。更多还原

【Abstract】 This paper presents an algorithm for accelerating the generation of Fractal Brownian Motion (FBM) with Graphics Processing Unit (GPU). As an example, we successfully apply the algorithm to synthesize clouds in real world. By reformulating each pass in the original algorithm into two sub-passes, we make the new algorithm suitable to be implemented on GPU. Additionally, we reduce the number of the conditional branch instructions involved in the improved algorithm, taking advantage that there are a large number of similar conditional branch instructions between two sub-passes. The workload between CPU and GPU is dynamically balanced to make full use of the strength of GPU and CPU’s, high frame rate is therefore attained when generating clouds. In order to simulate dynamic clouds, we present a method based on created array of random floating numbers, and successfully simulate the growth, dissipation and movement of clouds.Based on above, this paper proposes a way to extend the planar clouds into three-dimensional clouds by dividing the original clouds into multiple layers. Further more, we apply dynamic simulation algorithm mentioned above on the multiple layers of clouds.In order to achieve better simulation of dynamic clouds, this paper proposes a way to integrate FBM and Fluid Mechanics based Navier-Stokes Equation. In addition, we use GPU to speed up solving Navier-Stokes Equation.更多还原