We introduce coroutines into GPU kernel programming, providing an automated solution for flexible splitting and scheduling of rendering tasks. This approach addresses a prevalent challenge in harnessing the power of modern GPUs for complex, imbalanced graphics workloads like path tracing. Usually, to accommodate the SIMT execution model and latency-hiding architecture, developers have to decompose a monolithic mega-kernel into smaller sub-tasks for improved thread coherence and reduced register pressure. However, involving the handling of intricate nested control flows and numerous interdependent program states, this process can be exceedingly tedious and error-prone when performed manually. Coroutines, a building block for asynchronous programming in many high-level CPU languages, exhibit untapped potential for restructuring GPU kernels due to their versatility in control representation. By extending Luisa [Zheng et al. 2022], we implement an asymmetric, stackless coroutine model with programming language support and multiple built-in schedulers for modern GPUs. To showcase the effectiveness of our model and implementation, we examine them in different application scenarios, including path tracing, SDF rendering, and incorporation with custom passes.

The advancements in hardware have drawn more attention than ever to high-quality offline rendering with modern stream processors, both in the industry and in research fields. However, the graphics APIs are fragmented and existing shading languages lack high-level constructs such as polymorphism, which adds complexity to developing and maintaining cross-platform high-performance renderers. We present LuisaRender 1 , a high-performance rendering framework for modern stream-architecture hardware. Our main contribution is an expressive C++-embedded DSL for kernel programming with JIT code generation and compilation. We also implement a unified runtime layer with resource wrappers and an optimized Monte Carlo renderer. Experiments on test scenes show that LuisaRender achieves much higher performance than existing research renderers on modern graphics hardware, e.g., 5--11× faster than PBRT-v4 and 4--16× faster than Mitsuba 3.

Physically correct, noise-free global illumination is crucial in physically-based rendering, but often takes a long time to compute. Recent approaches have exploited sparse sampling and filtering to accelerate this process but still cannot achieve interactive performance. It is partly due to the time-consuming ray sampling even at 1 sample per pixel, and partly because of the complexity of deep neural networks. To address this problem, we propose a novel method to generate plausible single-bounce indirect illumination for dynamic scenes in interactive framerates. In our method, we first compute direct illumination and then use a lightweight neural network to predict screen space indirect illumination. Our neural network is designed explicitly with bilateral convolution layers and takes only essential information as input (direct illumination, surface normals, and 3D positions). Also, our network maintains the coherence between adjacent image frames efficiently without heavy recurrent connections. Compared to state-of-the-art works, our method produces single-bounce indirect illumination of dynamic scenes with higher quality and better temporal coherence and runs at interactive framerates.

Various denoising methods have been proposed to clean up the noise in Monte Carlo (MC) renderings, each having different advantages, disadvantages, and applicable scenarios. In this paper, we present Ensemble Denoising , an optimization-based technique that combines multiple individual MC denoisers. The combined image is modeled as a per-pixel weighted sum of output images from the individual denoisers. Computation of the optimal weights is formulated as a constrained quadratic programming problem, where we apply a dual-buffer strategy to estimate the overall MSE. We further propose an iterative solver to overcome practical issues involved in the optimization. Besides nice theoretical properties, our ensemble denoiser is demonstrated to be effective and robust, and outperforms any individual denoiser across dozens of scenes and different levels of sample rates. We also perform a comprehensive analysis on the selection of individual denoisers to be combined, providing important and practical guides for users.