权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

The Architecture of A Flexible Supercomputer Integrating A Realtime Visualization Mechanism with 3-Dimensional Memories

集成实时可视化机制和 3 维存储器的灵活超级计算机的体系结构

基本信息

批准号：
06402059
负责人：
TOMITA Shinji
金额：
$ 18.94万
依托单位：
KYOTO UNIVERSITY
依托单位国家：
日本
项目类别：
Grant-in-Aid for General Scientific Research (A)
财政年份：
1994
资助国家：
日本
起止时间：
1994 至 1995
项目状态：
已结题

项目摘要

We produced a prototype of ReVolver/C40, a parallel computer for volume rendering, following the research result of '94.The caharacteristic functions of ReVolver/C40 are realtime processing, perspective drawing, rendering of translucnet objects, and volume ray-tracing.ReVolver/C40 is consisted of macro-pipeline of 3-stages, ray-casting, pixel-calculation, and shading. Each stage render at high-speed by parallel processing of fine-grained pipeline processing depending on its form of processing. The performance is bounded by the pixel calculation stage. It is especially significant whether all the voxels crossed by a ray can be read or not. So, ReVolver/C40 adopts the triple volume memory, and the scheme to sample voxels in the direction of the main axis of a ray. This enables parallel volume access to any ray of arbitrary direction.From the design data at the point of PCB design, the rendering speed is 2.5 frames per second, in the case that the pixel calculation stage is consisted of 128 DSPs (TMS320C40,50MHz) . ReVolver/C40 can speeds up in proportion to a number of DSPs because of its scalable configuration.On the other hand, triple memory, the heart of ReVolver/C40, hold the problem about its big size. So, we also designed a single memory configuration which also enables parallel voxel access on a ray. It is achieved by limtting the angle of the viewing field to 54ﾟ.It is available because the result of scientific calculation or medical vision do not need a wide viewing field.

根据94年的研究成果，我们制作了一个用于体绘制的并行计算机ReVolver/C40的原型。ReVolver/C40的特征功能是实时处理、透视图绘制、半透明物体渲染和体射线追踪。ReVolver/C40由3个阶段的宏管道，光线投射，像素计算和着色组成。每个阶段根据其处理形式，采用细粒度流水线并行处理的方式高速渲染。性能受到像素计算阶段的限制。一条射线所穿过的所有体素能否被读取尤为重要。因此，ReVolver/C40采用三体积存储器，并采用沿射线主轴方向采样体素的方案。这使得平行体可以访问任意方向的任何射线。从PCB设计时的设计数据来看，在像素计算阶段由128个dsp （TMS320C40,50MHz）组成的情况下，渲染速度为2.5帧/秒。由于其可扩展的配置，ReVolver/C40可以与许多dsp成比例地加速。另一方面，作为左轮手枪/C40核心的三倍存储器存在体积过大的问题。因此，我们还设计了一个单一的内存配置，它也可以对射线进行并行体素访问。这是通过将视场的角度限制在54hz来实现的。它是可用的，因为科学计算或医学视觉的结果不需要一个宽的视野。