Method and apparatus for interprocessor communication employing modular space division
Novel method and system for distributed database ray-tracing is presented, based on modular mapping of scene-data among processors. Its inherent properties include scattering data among processors for improved load balancing, and matching between geographical proximity in the scene with communication proximity between processors. High utilization is enabled by unique mechanism of cache sharing. The resulting improved performance enables deep level of ray tracing for real time applications.
1. A multiprocessing computing system having reduced inter-processor communication, comprising:
an array of at least three general purpose processing elements interconnected by ring topology;
graphics data representing three dimensional scene of multiple objects;
plurality of uniform cells of data forming a grid;
and a modular mapping scheme;
wherein the data is mapped onto said grid of uniform cells according to said modular mapping scheme;
there is a need to exchange elements of said data between a cell and any of its 6-connected neighboring cells; and
the cells of data are designated to said processing elements according to said modular mapping scheme; thereby
when processing said data by said processing elements, any exchange of elements of said data between a cell and its 6-connected neighboring cells can be accomplished bi-directionally across at most one ring hop, reducing the overall inter-processor communication,
wherein the said computing system is used as a ray tracing system, wherein secondary ray at each cell can be communicated along a shortest distance to one or more of the cell's six connected neighboring cells.
2. The computing system of claim 1 wherein due to said modular mapping most of the ring communication occurs between two adjacent processors.
3. A method of reducing inter-processor communication in an array having at least three general purpose processing elements, when the processed graphics data represents a three dimensional scene of multiple objects, comprising the steps of:
connecting processing elements by ring interconnection;
subdividing the data onto uniform grid of cells;
mapping said cells according to modular mapping scheme;
assigning said cells to processing elements according to modular mapping scheme;
thereby when processing the data by said processing elements, any exchange of data between a cell and any of its 6-connected neighboring cells is accomplished bi-directionally across at most a single hop, reducing the overall inter-processor communication, wherein the said method is implemented on a ray tracing application,
wherein all exchange of secondary rays is done by two adjacent processing elements on the ring.
4. A method of reducing inter-processor communication in an array having at least three general purpose processing elements, when the processed graphics data represents a three dimensional scene of multiple objects, comprising the steps of:
connecting processing elements by ring interconnection;
subdividing the data onto uniform grid of cells;
mapping said cells according to modular mapping scheme;
assigning said cells to processing elements according to modular mapping scheme;
thereby when processing the data by said processing elements, any exchange of data between a cell and any of its 6-connected neighboring cells is accomplished bi-directionally across at most a single hop, reducing the overall inter-processor communication, wherein the said method is implemented on a ray tracing application, wherein each secondary ray at a cell may be communicated to one of six connected neighboring cells at most one ring hop.
5. A method of reducing inter-processor communication in an array having at least three general purpose processing elements, when the processed graphics data represents a three dimensional scene of multiple objects, comprising the steps of:
connecting processing elements by ring interconnection;
subdividing the data onto uniform grid of cells;
mapping said cells according to modular mapping scheme;
assigning said cells to processing elements according to modular mapping scheme;
thereby when processing the data by said processing elements, any exchange of data between a cell and any of its 6-connected neighboring cells is accomplished bi-directionally across at most a single hop, reducing the overall inter-processor communication, wherein the said method is implemented on a ray tracing application, wherein the exchanged data elements between a cell and its neighboring cells are secondary rays.