The effectiveness of texture mapping in enhancing the realism of computer generated imagery has made support for real-time texture mapping a critical part of graphics pipelines. Despite a recent surge in interest in three-dimensional graphics from computer architects, high-quality high-speed texture mapping has so far been confined to costly hardware systems that use brute-force techniques to achieve high performance. One obstacle faced by designers of texture mapping systems is the requirement of extremely high bandwidth to texture memory. High bandwidth is necessary since there are typically tens to hundreds of millions of accesses to texture memory per second. In addition, to achieve the high clock rates required in graphics pipelines, low-latency access to texture memory is needed. In this paper, we propose the use of texture image caches to alleviate the above bottlenecks, and evaluate various tradeoffs that arise in such designs.We find that the factors important to cache behavior are (i) the representation of texture images in memory, (ii) the rasterization order on screen and (iii) the cache organization. Through a detailed investigation of these issues, we explore the best way to exploit locality of reference and determine whether this technique is robust with respect to different scenes and different amounts of texture. Overall, we observe that there is a significant amount of temporal and spatial locality and that the working set sizes are relatively small (at most 16KB) across all cases that we studied. Consequently, the memory bandwidth requirements of a texture cache system are substantially lower (at least three times and as much as fifteen times) than the memory bandwidth requirements of a system which achieves equivalent performance but does not utilize a cache. These results are very encouraging and indicate that caching is a promising approach to designing memory systems for texture mapping.

No abstract

We introduce hybrid rendering, a scheme that dynamically ray traces the local geometry of reflective and refractive objects, but approximates more distant geometry by hardwaresupported environment maps (EMs). To limit computation, we use a greedy ray path shading model that prunes the binary ray tree generated by refractive objects to form just two ray paths. We also restrict ray queries to triangle vertices, but perform adaptive tessellation to shoot additional rays where neighboring ray paths differ sufficiently. By using layered, parameterized EMs that are inferred over a set of viewpoint samples to match ray traced imagery, we accurately handle parallax and view-dependent shading in the environment. We increase robustness of EMs by inferring them simultaneously across multiple viewpoints and including environmental geometry that is occluded from the viewpoint sample but is revealed in nearby viewpoints. We demonstrate realistic shiny and glass objects with a user-controlled viewpoint.

We generalize image-based rendering by exploiting texture-mapping graphics hardware to decompress ray-traced "animations". Rather than 1D time, our animations are parameterized by two or more arbitrary variables representing view/lighting changes and rigid object motions. To best match the graphics hardware rendering to the input ray-traced imagery, we describe a novel method to infer parameterized texture maps for each object by modeling the hardware as a linear system and then performing least-squares optimization. The parameterized textures are compressed as a multidimensional Laplacian pyramid on fixed size blocks of parameter space. This scheme captures the coherence in parameterized animations and, unlike previous work, decodes directly into texture maps that load into hardware with a few, simple image operations. We introduce adaptive dimension splitting in the Laplacian pyramid and separate diffuse and specular lighting layers to further improve compression. High-quality results are demonstrated at compression ratios up to 800:1 with interactive playback on current consumer graphics cards.

Static environment maps fail to capture local reflections including effects like selfreflections and parallax in the reflected imagery. We instead propose parameterized environment maps (PEMs), a set of per-view environment maps which accurately reproduce local reflections at each viewpoint as computed by an offline ray tracer. Even with a small set of viewpoint samples, PEMs support plausible movement away from and between the pre-rendered viewpoint samples while maintaining local reflections. They also make use of environment maps supported in graphics hardware to provide real-time exploration of the pre-rendered space. In addition to parameterization by viewpoint, our notion of PEM extends to general, multidimensional parameterizations of the scene, including relative motions of objects and lighting changes.Our contributions include a technique for inferring environment maps providing a close match to ray-traced imagery. We also explicitly infer and encode all MIPMAP levels of the PEMs to achieve higher accuracy. We propose layered environment maps that separate local and distant reflected geometry. We explore several types of environment maps including finite spheres, ellipsoids, and boxes that better approximate the environmental geometry. We demonstrate results showing faithful local reflections in an interactive viewer.