high performance computing on graphics processing units: hgpu.org

Direct visualization of higher-order data provides manifold advantages over the traditional approach, which is based on resampling and subsequent visualization by interpolation-based techniques. Most important, it avoids excessive computation and consumption of memory, and prevents artifacts by pixel-accurate visualization at interactive rates. This work addresses particle-partition of unity simulation data, where fields are modeled both using cell-based analytic representations together with enrichment functions centered at individual points. This combination of bases allows for superior simulation convergence rates and is able to capture high field variations with comparably small sets of basis functions. In this paper we propose direct visualization of such data from 2D simulations, providing accurate insight. We additionally visualize solver performance, allowing for more directed simulation design, and exemplify our technique using a GPU-based prototype on crack simulation examples.