ITR/SY (Revolutionary Computing) Multipass Programming for Personal High-Performance Computing

图形处理器的性能正在超越通用处理器的性能，实际上是在加速增长曲线。最近已经表明，一些图形算法可以在图形处理器上以新的方式执行，使用称为多遍编程的新技术。多通道编程将图形处理器视为SIMD处理器，作用于存储在屏幕像素中的数据阵列，每个操作都作为图像处理通道应用。拟议的研究将进一步扩展这一概念，不仅在图形社区，但也为科学计算和一般计算社区，通过开发一个多通道编程语言一般足以支持图形和非图形程序。现有的数值范围和精度的问题将被克服，并得到数值实现的保真度分析。工作在程序着色的多通道算法将继续我们，和多通道编程应用于光能传递和基于物理的动画。多遍编程亦会扩展至计算机图形学以外的应用，包括解算线性系统、有限元素法和计算流体动力学。如果图形处理器继续以现时的速度增长，在建议的资助期结束时，其性能将可达4万亿次。在这一增长时期，将开发一个编程技术和应用程序库，以利用这一尚未开发的高性能计算能力。其结果将是更丰富的图形，更逼真的虚拟环境，更自然的模拟运动，以及在消费级个人电脑上执行复杂的超级计算模拟的能力。

The performance of graphic processors is outpacing the performance of general processors, in fact cubing the growth curve. It has been recently shown that some graphics algorithms can be performed in new ways on the graphics processor, using a novel technique known as multipass programming. Multipass programming treats the graphics processor as a SIMD processor acting on a data array stored in the screen's pixels, with each operation applied as an image-processing pass. The proposed research will extend this concept further not only within the graphics community, but also for the scientific computing and general computing communities, by developing a multipass programming language general enough to support both graphical and non-graphical programs. Existing problems of numerical range and precision will be overcome, and the fidelity of the resulting numerical implementations analyzed.Work in multipass algorithms for procedural shading will be continued our, and multipass programming applied to radiosity and physically-based animation. Multipass programming will also be extended to applications outside of computer graphics, including solving linear systems, the finite element method and computational fluid dynamics.If graphics processors continue to grow at the current rate, they will attain a four teraflop performance at the end of the proposed funding period. During this growth period a library of programming techniques and applications that will harness this otherwise untapped source of high-performance computing power will be developed. The result will be richer graphics, more realistic virtual environments, more natural simulated motion, and the ability to perform sophisticated supercomputing simulations on consumer-level personal computers.