Managing Coherency in Real-TimeRay Tracing

Abstract: In recent years, ray tracing’s popularity in real-time rendering wasboosted by the introduction of GPUs which supported hardwareacceleration for ray tracing routines, which sped up the computationtime considerably. In spite of this, ray tracing is still expensive, andas such, most developers enhance their graphical applications with onlyselected raytraced effects, while still relying on rasterization for thecore visual component. Different ray tracing workloads pose different challenges from aperformance standpoint, which require careful analysis by theprogrammer. Incoherent rays often lead to poor utilization of the GPU,and is therefore a common bottleneck. However, current graphicsapplication programming interfaces (APIs) such as Microsoft's DirectX Raytracing (DXR) expose no functionality for measuring the incoherence of rays, or for manually re-ordering the execution of rays as a means of improving performance. This thesis introduces a framework for study in gray coherence and performing ray sorting in DXR. Performance degradationcaused by divergent rays is analyzed and evaluated using this framework against popular benchmark scenes.