high performance computing on graphics processing units: hgpu.org

This paper presents an accurate parallel implementation of unsteady flow line integral convolution (UFLIC) for high-performance visualization of large time-varying flows. Our approach differs from previous implementations by using a novel value scattering+gathering mechanism to parallelize UFLIC and designing a pathline reuse strategy to reduce the computational cost of pathline integration. By exploiting the massive parallelism of modern graphical processing units (GPU), the proposed method allows for real-time dense visualization of unsteady flows with high spatial-temporal coherence.