High-Performance Graphics 2016

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/15439

Browse

Now showing 1 - 18 of 18

Watertight Ray Traversal with Reduced Precision
(The Eurographics Association, 2016) Vaidyanathan, Karthik; Akenine-Möller, Tomas; Salvi, Marco; Ulf Assarsson and Warren Hunt
Reduced precision bounding volume hierarchies and ray traversal can significantly improve the efficiency of ray tracing through low-cost dedicated hardware. A key approach to enabling reduced precision computations during traversal is to translate the ray origin closer to the bounding volume hierarchy node after each traversal step. However, this approach precludes sharing of intersection computations between a parent node and its two children, which is an important optimization. In this paper, we introduce a novel traversal algorithm that addresses this limitation and achieves a significant reduction in the computational complexity of traversal compared to previous approaches.We also include an analysis that shows how our algorithm guarantees watertight intersections which is a key requirement for robust image quality, especially with reduced precision traversal where numerical errors can be large.
Masked Software Occlusion Culling
(The Eurographics Association, 2016) Hasselgren, Jon; Andersson, Magnus; Akenine-Möller, Tomas; Ulf Assarsson and Warren Hunt
Efficient occlusion culling in dynamic scenes is a very important topic to the game and real-time graphics community in order to accelerate rendering. We present a novel algorithm inspired by recent advances in depth culling for graphics hardware, but adapted and optimized for SIMD-capable CPUs. Our algorithm has very low memory overhead and is 3 faster than previous work, while culling 98% of all triangles culled by a full resolution depth buffer approach. It supports interleaving occluder rasterization and occlusion queries without penalty, making it easy to use in scene graph traversal or rendering code.
A Fast, Massively Parallel Solver for Large, Irregular Pairwise Markov Random Fields
(The Eurographics Association, 2016) Thuerck, Daniel; Waechter, Michael; Widmer, Sven; Buelow, Max von; Seemann, Patrick; Pfetsch, Marc E.; Goesele, Michael; Ulf Assarsson and Warren Hunt
Given the increasing availability of high-resolution input data, today's computer vision problems tend to grow beyond what has been considered tractable in the past. This is especially true for Markov Random Fields (MRFs), which have expanded beyond millions of variables with thousands of labels. Such MRFs pose new challenges for inference, requiring massively parallel solvers that can cope with large-scale problems and support general, irregular input graphs. We propose a block coordinate descent based solver for large MRFs designed to exploit many-core hardware such as recent GPUs. We identify tree-shaped subgraphs as a block coordinate scheme for irregular topologies and optimize them efficiently using dynamic programming. The resulting solver supports arbitrary MRF topologies efficiently and can handle arbitrary, dense or sparse label sets as well as label cost functions. Together with two additional heuristics for further acceleration, our solver performs favorably even compared to modern specialized solvers in terms of speed and solution quality, especially when solving very large MRFs.
Bandwidth-Efficient BVH Layout for Incremental Hardware Traversal
(The Eurographics Association, 2016) Liktor, Gabor; Vaidyanathan, Karthik; Ulf Assarsson and Warren Hunt
The memory footprint of bounding volume hierarchies (BVHs) can be significantly reduced using incremental encoding, which enables the coarse quantization of bounding volumes. However, this compression alone does not necessarily yield a comparable improvement in memory bandwidth. While the bounding volumes of the BVH nodes can be aggressively quantized, the size of the child node pointers remains a significant overhead. Moreover, as BVH nodes become comparably small to practical cache line sizes, the BVH is cached less efficiently. In this paper we introduce a novel memory layout and node addressing scheme and map it to a system architecture for fixed-function ray traversal. We evaluate this scheme using an architecture simulator and demonstrate a significant reduction in memory bandwidth, compared to previous approaches.
Comparison of Projection Methods for Rendering Virtual Reality
(The Eurographics Association, 2016) Toth, Robert; Nilsson, Jim; Akenine-Möller, Tomas; Ulf Assarsson and Warren Hunt
Virtual reality is rapidly gaining popularity, and may soon become a common way of viewing 3D environments. While stereo rendering has been performed on consumer grade graphics processors for a while now, the new wave of virtual reality display devices have two properties that typical applications have not needed to consider before. Pixels no longer appear on regular grids and the displays subtend a wide field-of-view. In this paper, we evaluate several techniques designed to efficiently render for head-mounted displays with such properties. We show that the amount of rendered pixels can be reduced down to 36% without compromising visual fidelity compared to traditional rendering, by rendering multiple optimized sub-projections.
Deep G-Buffers for Stable Global Illumination Approximation
(The Eurographics Association, 2016) Mara, Michael; McGuire, Morgan; Nowrouzezahrai, Derek; Luebke, David; Ulf Assarsson and Warren Hunt
We introduce a new hardware-accelerated method for constructing Deep G-buffers that is 2x-8x faster than the previous depthpeeling method and produces more stable results. We then build several high-performance shading algorithms atop our representation, including dynamic diffuse interreflection, ambient occlusion (AO), and screen-space mirror reflection effects. Our construction method is order-independent, guarantees a minimum separation between layers, operates in a (small) bounded memory footprint, and does not require per-pixel sorting. Moreover, addressing the increasingly expensive cost of pre-rasterization, our approach requires only a single pass over the scene geometry. We include the pseudocode for our Deep G-buffer construction in the paper and the full source code of our technique in our supplemental document.
Lightcut Interpolation
(The Eurographics Association, 2016) Rehfeld, Hauke; Dachsbacher, Carsten; Ulf Assarsson and Warren Hunt
Many-light rendering methods replace multi-bounce light transport with direct lighting from many virtual point light sources to allow for simple and efficient computation of global illumination. Lightcuts build a hierarchy over virtual lights, so that surface points can be shaded with a sublinear number of lights while minimizing error. However, the original algorithm needs to run on every shading point of the rendered image. It is well known that the performance of Lightcuts can be improved by exploiting the coherence between individual cuts. We propose a novel approach where we invest into the initial lightcut creation at representative cache records, and then directly interpolate the input lightcuts themselves as well as per-cluster visibility for neighboring shading points. This allows us to improve upon the performance of the original Lightcuts algorithm by a factor of 4-8 compared to an optimized GPU-implementation of Lightcuts, while introducing only a small additional approximation error. The GPU-implementation of our technique enables us to create previews of Lightcuts-based global illumination renderings.
Infinite Resolution Textures
(The Eurographics Association, 2016) Reshetov, Alexander; Luebke, David; Ulf Assarsson and Warren Hunt
We propose a new texture sampling approach that preserves crisp silhouette edges when magnifying during close-up viewing, and benefits from image pre-filtering when minifying for viewing at farther distances. During a pre-processing step, we extract curved silhouette edges from the underlying images. These edges are used to adjust the texture coordinates of the requested samples during magnification. The original image is then sampled-only once!-with the modified coordinates. The new technique provides a resolution-independent image representation capable of billions of texels per second on a mid-range graphics card.
GVDB: Raytracing Sparse Voxel Database Structures on the GPU
(The Eurographics Association, 2016) Hoetzlein, Rama Karl; Ulf Assarsson and Warren Hunt
Simulation and rendering of sparse volumetric data have different constraints and solutions depending on the application area. Generating precise simulations and understanding very large data are problems in scientific visualization, whereas convincing simulations and realistic visuals are challenges in motion pictures. Both require volumes with dynamic topology, very large domains, and efficient high quality rendering.We present the GPU voxel database structure, GVDB, based on the voxel database topology of Museth [Mus13], as a method for efficient GPU-based compute and raytracing on a sparse hierarchy of grids. GVDB introduces an indexed memory pooling design for dynamic topology, and a novel hierarchical traversal for efficient raytracing on the GPU. Examples are provided for ray sampling of volumetric data, rendering of isosurfaces with multiple scattering, and raytracing of level sets. We demonstrate that GVDB can give large performance improvements over CPU methods with identical quality.
HPG 2016: Frontmatter
(Eurographics Association, 2016) Ulf Assarsson; Warren Hunt
Photon Splatting Using a View-Sample Cluster Hierarchy
(The Eurographics Association, 2016) Moreau, Pierre; Sintorn, Erik; Kämpe, Viktor; Assarsson, Ulf; Doggett, Michael; Ulf Assarsson and Warren Hunt
Splatting photons onto primary view samples, rather than gathering from a photon acceleration structure, can be a more efficient approach to evaluating the photon-density estimate in interactive applications, where the number of photons is often low compared to the number of view samples. Most photon splatting approaches struggle with large photon radii or high resolutions due to overdraw and insufficient culling. In this paper, we show how dynamic real-time diffuse interreflection can be achieved by using a full 3D acceleration structure built over the view samples and then splatting photons onto the view samples by traversing this data structure. Full dynamic lighting and scenes are possible by tracing and splatting photons, and rebuilding the acceleration structure every frame. We show that the number of view-sample/photon tests can be significantly reduced and suggest further culling techniques based on the normal cone of each node in the hierarchy. Finally, we present an approximate variant of our algorithm where photon traversal is stopped at a fixed level of our hierarchy, and the incoming radiance is accumulated per node and direction, rather than per view sample. This improves performance significantly with little visible degradation of quality.
Efficient Stackless Hierarchy Traversal on GPUs with Backtracking in Constant Time
(The Eurographics Association, 2016) Binder, Nikolaus; Keller, Alexander; Ulf Assarsson and Warren Hunt
The fastest acceleration schemes for ray tracing rely on traversing a bounding volume hierarchy (BVH) for efficient culling and use backtracking, which in the worst case may expose cost proportional to the depth of the hierarchy in either time or state memory. We show that the next node in such a traversal actually can be determined in constant time and state memory. In fact, our newly proposed parallel software implementation requires only a few modifications of existing traversal methods and outperforms the fastest stack-based algorithms on GPUs. In addition, it reduces memory access during traversal, making it a very attractive building block for ray tracing hardware.
Adaptive Sampling for On-The-Fly Ray Casting of Particle-based Fluids
(The Eurographics Association, 2016) Hochstetter, Hendrik; Orthmann, Jens; Kolb, Andreas; Ulf Assarsson and Warren Hunt
We present a fast and accurate ray casting technique for unstructured and dynamic particle sets. Our technique focuses on efficient, high quality volume rendering of fluids for computer animation and scientific applications. Our novel adaptive sampling scheme allows to locally adjust sampling rates both along rays and in lateral direction and is driven by a user-controlled screen space error tolerance. In order to determine appropriate local sampling rates, we propose a sampling error analysis framework based on hierarchical interval arithmetic. We show that our approach leads to significant rendering speed-ups with controllable screen space errors. Efficient particle access is achieved using a sparse view-aligned grid which is constructed on-the-fly without any pre-processing.
DIRT: Deferred Image-based Ray Tracing
(The Eurographics Association, 2016) Vardis, Konstantinos; Vasilakis, Andreas-Alexandros; Papaioannou, Georgios; Ulf Assarsson and Warren Hunt
We introduce a novel approach to image-space ray tracing ideally suited for the photorealistic synthesis of fully dynamic environments at interactive frame rates. Our method, designed entirely on the rasterization pipeline, alters the acceleration data structure construction from a per-fragment to a per-primitive basis in order to simultaneously support three important, generally conflicting in prior art, objectives: fast construction times, analytic intersection tests and reduced memory requirements. In every frame, our algorithm operates in two stages: A compact representation of the scene geometry is built based on primitive linked-lists, followed by a traversal step that decouples the ray-primitive intersection tests from the illumination calculations; a process inspired by deferred rendering and the path integral formulation of light transport. Efficient empty space skipping is achieved by exploiting several culling optimizations both in xy- and z-space, such as pixel frustum clipping, depth subdivision and lossless buffer down-scaling. An extensive experimental study is finally offered showing that our method advances the area of image-based ray tracing under the constraints posed by arbitrarily complex and animated scenarios.
Filtering Distributions of Normals for Shading Antialiasing
(The Eurographics Association, 2016) Kaplanyan, Anton S.; Hill, Stephen; Patney, Anjul; Lefohn, Aaron; Ulf Assarsson and Warren Hunt
High-frequency illumination effects, such as highly glossy highlights on curved surfaces, are challenging to render in a stable manner. Such features can be much smaller than the area of a pixel and carry a high amount of energy due to high reflectance. These highlights are challenging to render in both offline rendering, where they require many samples and an outliers filter, and in real-time graphics, where they cause a significant amount of aliasing given the small budget of shading samples per pixel. In this paper, we propose a method for filtering the main source of highly glossy highlights in microfacet materials: the Normal Distribution Function (NDF). We provide a practical solution applicable for real-time rendering by employing recent advances in light transport for estimating the filtering region from various effects (such as pixel footprint) directly in the parallel-plane half-vector domain (also known as the slope domain), followed by filtering the NDF over this region. Our real-time method is GPU-friendly, temporally stable, and compatible with deferred shading, normal maps, as well as with filtering methods for normal maps.
Exploring and Expanding the Continuum of OIT Algorithms
(The Eurographics Association, 2016) Wyman, Chris; Ulf Assarsson and Warren Hunt
Order independent transparency (OIT) proves challenging for modern rasterization-based renderers. Rendering without transparency can limit the quality of visual effects, so researchers have proposed various algorithms enabling and approximating OIT. Unfortunately, this work generally has restrictions limiting its applicability. To identify directions for improvement, we performed an in-depth categorization of existing transparency techniques and placed them on a multi-dimensional continuum. This categorization shows how prior published techniques relate to each other and highlights unexposed parts of the continuum where further research may prove beneficial. We also describe two new OIT algorithms that occupy previously unexplored regions of this continuum. These novel algorithms include stochastic layered alpha blending (SLAB), which provides a parameter to explicitly transition along the continuum between stochastic transparency and k-buffers, and multi-layered coverage blending (MLCB), which explicitly decorrelates visibility and opacity in multi-layered alpha blending.
Local Shading Coherence Extraction for SIMD-Efficient Path Tracing on CPUs
(The Eurographics Association, 2016) Áfra, Attila T.; Benthin, Carsten; Wald, Ingo; Munkberg, Jacob; Ulf Assarsson and Warren Hunt
Accelerating ray traversal on data-parallel hardware architectures has received widespread attention over the last few years, but much less research has focused on efficient shading for ray tracing. This is unfortunate since shading for many applications is the single most time consuming operation. To maximize rendering performance, it is therefore crucial to effectively use the processor's wide vector units not only for the ray traversal step itself, but also during shading. This is non-trivial as incoherent ray distributions cause control flow divergence, making high SIMD utilization difficult to maintain. In this paper, we propose a local shading coherence extraction algorithm for CPU-based path tracing that enables efficient SIMD shading. Each core independently traces and sorts small streams of rays that fit into the on-chip cache hierarchy, allowing to extract coherent ray batches requiring similar shading operations, with a very low overhead. We show that operating on small independent ray streams instead of a large global stream is sufficient to achieve high SIMD utilization in shading (90% on average) for complex scenes, while avoiding unnecessary memory traffic and synchronization. For a set of scenes with many different materials, our approach reduces the shading time with 1.9–-3.4 compared to simple structure-of-arrays (SoA) based packet shading. The total rendering speedup varies between 1.2-3 , which is also determined by the ratio of the traversal and shading times.
SVGPU: Real Time 3D Rendering to Vector Graphics Formats
(The Eurographics Association, 2016) Ellis, Apollo I.; Hunt, Warren; Hart, John C.; Ulf Assarsson and Warren Hunt
We focus on the real-time realistic rendering of a 3-D scene to a 2-D vector image. There are several application domains which could benefit substantially from the compact and resolution independent intermediate format that vector graphics provides. In particular, cloud streaming services, which transmit large amounts of video data and notoriously suffer from low resolution and/or high latency. In addition, display resolutions are growing rapidly, exacerbating the issue. Raster images for large displays prove a significant bottleneck when being transported over communication networks. However the alternative of sending a full 3D scene worth of geometry is even more prohibitive. We implement a real time rendering pipeline that utilizes analytic visibility algorithms on the GPU to output a vector graphics representation of a 3D scene. Our system SVGPU (Scalable Vector on the GPU) is fast and efficient on modern hardware, and simple in design. As such we are making a much needed step towards enabling the benefits of vector graphics representations to be reaped by the real time community.

Browse

Browsing High-Performance Graphics 2016 by Issue Date

Results Per Page

Sort Options