Lacroute P. G. Fast Volume Rendering Using a Share-Warp Factorization of the Viewing Transformation. PhD thesis, Stanford University, 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....up volume rendering to achieve the desired frame rates. Successful parallel implementations of ray casting renderers have been developed for both centralized and distributed shared memory multiprocessors [5] Parallel implementations of the faster shear warp method have also been developed [9, 4], but these have not scaled well beyond 8 to 12 processor systems. As programmers trying to improve parallel performances, we first attempt to understand why, studying the performance, scalability and characteristics of the parallel shear warp renderer through detailed measurement using both ....

....image pixels or object voxels. The shear warp algorithm is an object order algorithm. It operates by first streaming through the 3 D volume data and splatting voxels into a 2 D intermediate image, using a shearwarp factorization, and then warping the 2 D intermediate image into a 2 D final image [4]. The shear warp algorithm combines the advantages of both image order and objectorder algorithms, and is reported to be the fastest volume rendering algorithm so far that does not compromise quality on a serial workstation. 3.1 Common Optimizations For most volumes, particularly medical image, ....

[Article contains additional citation context not shown here]

Lacroute P. G. Fast Volume Rendering Using a Share-Warp Factorization of the Viewing Transformation. PhD thesis, Stanford University, 1995.

....5:1 6:1 Sequent NUMAQ 240 2500 N A 10:1 N A Table 1: Latencies and remote to local latency ratio on di#erent systems. 2. 2 Applications In addition to eight SPLASH 2 applications [20] the four new applications in this study: a Sample sorting kernel, a Shear Warp volume rendering application [15, 12], a probabilistic inference application for belief networks (Infer, applied to a medical diagnosis problem [11] and an application for protein structure determination in the presence of uncertainty (Protein) 4] Results for the last two are available only up to 64 processors so far. Taken ....

P. G. Lacroute. Fast Volume Rendering Using a Share-Warp Factorization of the Viewing Transformation. PhD thesis, Stanford University, 1995.

....a chunk of a task in the warping phase the warping phase partition in Figure 1.3: Shear warp volume rendering algorithm. Shear Warp renders three dimensional volume data into an image using a shearwarp factorization algorithm. There are two phases in the rendering (see Figure 1. 3) [50, 73]. First, the run length encoded volume (not shown) is composited into an intermediate image, by traversing the volume in scanline order slice by slice and writing the image. This is done by partitioning the intermediate image plane into chunks of scanlines assigned to processors in an interleaved ....

....with SVM page granularity to produce di#erent induced sharing patterns, as discussed in Section 1.3. Third, we choose applications from di#erent domains of computation. 3. 2 Methodology We begin with the applications as they appear in the SPLASH 2 suite (and for ShearWarp as it appears in [50]) They are all quite well tuned for hardware cache coherence. We perform data distribution on the SVM and DSM platforms as suggested in SPLASH 2. The partition methods are the best known so far and that perform well at CHAPTER 3. PERFORMANCE PORTABILITY TO CLUSTERS 62 moderate scale on the ....

[Article contains additional citation context not shown here]

P. G. Lacroute. Fast Volume Rendering Using a Share-Warp Factorization of the Viewing Transformation. PhD thesis, Stanford University, 1995. BIBLIOGRAPHY 173

....communication or coherence. Third, we choose applications from different domains of computation. Our application suite contains 7 applications, each with several versions. Six are originally from the SPLASH 2 [18] suite, and one is a recently published parallel shear warp volume rendering program [9, 15] that is also described in another paper in these proceedings [3] Let us briefly describe the access patterns in the original applications. Details of the applications themselves can be found in [18] and the references there. 2.2.1 Regular Applications LU performs the blocked LU factorization of ....

....volume and the write accesses to task queues and image data are fine grained, so it suffers both fragmentation and false sharing. Shear Warp renders three dimensional volume data into an image using a shear warp factorization algorithm. There are two phases in the rendering (see Figure 1) [15, 9]. First, the run length encoded volume (not shown) is composited warp phase intermediate image plane final image plane partition in warping phase a tile: a processor s warping a task in the compositing phase (a unit of stealing) scanlines: a chunk of a task in the Figure 1: Shear warp volume ....

[Article contains additional citation context not shown here]

Lacroute P. G. Fast Volume Rendering Using a ShareWarp Factorization of the Viewing Transformation. PhD thesis, Stanford University, 1995.

....makes them viable platforms for speeding up volume rendering. Successful parallel implementations of ray casting renderers have been developed for both centralized and distributed shared memory multiprocessors [8] Parallel implementations of the faster shear warp method have also been developed [4, 12], but these have not scaled well beyond 8 to 12 processor systems, or performed well on systems with physically distributed memory. We first attempt to understand why, studying the performance, scalability and characteristics of the existing parallel shear warp renderer. We do this through ....

....warp phase. The sizes of the intermediate and final images usually differ a lot due to the shear in the compositing phase. The compositing phase is O(n 3 ) for an n by n by n voxel volume, and is the dominant phase in sequential execution. A detailed description of this algorithm can be found in [4]. Viewing Rays Volume Slices Image Plane Final (Warped) Image Sheared Volume Slices Intermediate (Composited) Image (a) Physical Reality (b) Algorithm Phases Figure 1: The serial shear warp volume rendering algorithm. The shear warp algorithm is reported to be the fastest volume rendering ....

[Article contains additional citation context not shown here]

Lacroute P. G. Fast Volume Rendering Using a ShareWarp Factorization of the Viewing Transformation. PhD thesis, Stanford University, 1995.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC