AbstractsComputer Science

Scattered data are unevenly distributed or randomly spread over the volume of interest. The random distribution of the data makes it hard to visualize since existing visualization algorithms are based on a 3D grid structure. Scattered data are commonly found in engineering applications. Thus quick interactive visualization of scattered data is in great demand. The most commonly used approach for scattered data visualization contains two steps. The first step involves converting the scattered sample data into a 3D uniform grid. The sample data consist of 3 values for the position and one data value. To form the grid we need to interpolate the data values onto each grid node. This modeling part has three steps including matrix inversion, interpolants calculation and grid value computation. Focusing on this part, our project aims to speed up the modeling on various platforms. We implement this approach on platforms including CPU, GPU, GPGPU and cloud-based GPGPU with different operating systems.For scattered data modeling, we need to treat the three steps differently to achieve good performance. For the step of sample data matrix inversion, it is only meaningful to use the GPU when the sample data is large. This is due to the overhead of synchronization and data transfer. For the step defining the interpolants, computing on the CPU is faster due to the simplicity of the computations. For the step of computing grid data values, the GPU is preferable for almost all grid sizes. Among the GPU computing overhead, data transfer back from the GPU to the host is more expensive than that from host to GPU. This is due to the fact that writing to the host memory is lock-stepped.