A Distributed-Memory Parallel Approach for Volume Rendering with Shadows

Abstract

We present a parallel, distributed-memory technique that enhances traditional ray-casting volume rendering of large data sets to highlight the depth and perception of interesting volumetric features. The technique introduces a lighting system that accounts for global shadows across distributed MPI nodes while using shared-memory parallelism within each node to compute shading information efficiently. The first stage of the approach involves estimating energy attenuation from a point light source through the global volume, using a reduced spatial resolution representation of the volume, with minimal global communication between nodes. It is then used in the second stage during volume rendering to shade sample points captured during ray-casting, generating a high-quality image. In this work, we study the technique's performance across varying spatial resolutions of the estimated light attenuation using synthetic and real-world volumetric data sets on distributed systems.