Electrical computers and digital processing systems: processing – Processing architecture – Vector processor
Reexamination Certificate
2000-02-11
2004-10-19
Coleman, Eric (Department: 2183)
Electrical computers and digital processing systems: processing
Processing architecture
Vector processor
Reexamination Certificate
active
06807620
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to computer system architectures. More particularly, the present invention relates to the architecture and use of a computer system optimized for efficient modeling of graphics.
BACKGROUND OF THE INVENTION
High resolution, real time computer graphics are an important aspect of computer systems, particularly simulators (such as flight simulators) and game machines. Computer games, in particular, involve a great deal of computer graphics. Computer systems used as game machines, therefore, must handle far more computer graphics than a standard business computer used primarily for word processing or similar applications.
The game developer is faced with many limitations. He or she often wants realistic, highly detailed graphics. Prior art game machines, however, make the implementation of such graphics difficult. High resolution graphics are computationally expensive and difficult to render in the time required by a fast moving game. Current graphics co-processors, if implemented at all in game consoles, have difficulty supplying the bandwidth necessary to render high resolution, real time graphics.
Prior art game machines also do not permit easy behavioral and physical modeling of game objects. Many objects in a game would be more realistically rendered if their position and shape could be calculated, or modeled, under a set of rules or equations. However, such modeling is computationally expensive, requiring many floating point operations, and the standard CPU is not optimized for such calculations.
Prior art game machines also cannot easily deal with compressed video data. As game developers code larger and larger game worlds, they are in danger of running out of space in removable media. The use of compression techniques to store various kinds of data, such as graphics data, is limited by the need to decompress such data quickly for use in a real time, interactive game.
Prior art game machines also are generally restricted to gaming applications. Given the increasing computational power of gaming systems, developers are looking at other applications for game consoles besides gaming. However, limitations in input and output interfaces render such applications difficult.
SUMMARY OF THE INVENTION
The present invention provides an improved computer system particularly suited for simulators and game machines. The system includes a new computer architecture for such devices. This architecture comprises a main processor and a graphics processor. The main processor contains two co-processors for geometry modeling and a central processing unit(CPU).
In one aspect, the present invention provides a frame buffer and rendering system on the same integrated chip. This structure enables the computer system to draw many pixels in parallel to the frame buffer at a very high fill rate (high band width). As a result, the computer system can provide quick renderings of screen images at a high resolution.
In another aspect, the present invention provides a main processor with a 128-bit bus throughout this processor connecting all co-processors and a memory system. This structure enables the passing of data and instructions quickly from component to component, thereby improving bandwidth resolution and speed.
In another aspect, the present invention provides sub-processors with four floating-point, multiply-add arithmetic logic units (ALUs). These four ALUs enable the processing of four 32-bit operations simultaneously from the data of two 128-bit registers. This structure, therefore, enables parallel, 128-bit floating point calculations through parallel pipelining of similar calculations to, e.g., assist in modeling and geometry transformations.
The present invention, in a preferred embodiment, further provides a multimedia instruction set using 128 bit wide integer registers in parallel. This structure enables the handling of different size integers in parallel (64-bits×2, or 32-bits×4, or 16-bits×8 or 8-bits×16).
In yet another aspect, the present invention provides two geometry engines feeding in parallel into one rendering engine. One geometry engine preferably consists of the CPU, for flexible calculations, tightly coupled to a vector operation unit as a co-processor, for complex irregular geometry processing such as modeling of physics or behavior. The second geometry engine preferably is a programmable vector operation unit for simple, repetitive geometry processing such as background and distant views (simple geometrical transformations).
In accordance with this aspect of the invention, each geometry engine preferably provides data (termed display lists) that are passed to the rendering engine. Arbitrator logic between the geometry engines and the rendering engine determines the order in which these data are passed to the rendering engine. The second geometry engine preferably is given priority over the first, as the second geometry engine generally has more data to send, and the first geometry engine is buffered in case of interruption. With this structure, the application programmer can, e.g., specify which geometry engine should do particular graphics processing, thereby enabling sophisticated behavioral and physical modeling in real time.
Also, in accordance with this aspect of the invention, the rendering engine remembers the data from each geometry engine and stores these data until deliberately changed. These data, therefore, do not require resending when the rendering engine begins receiving data from a different geometry engine, thereby enhancing speed.
In yet another aspect, the present invention provides a specialized decompression processor for decompressing high-resolution texture data from a compressed state as stored in main memory. This processor allows for more efficient use of memory.
In a preferred embodiment, the present invention provides a system for packing modeling data into optimal bit widths in data units in main memory. Unpacking logic in the vector processors automatically unpacks these data without sacrificing performance.
In yet another aspect, the present invention provides all processors with a local cache memory. This architecture reduces the amount of data that is required to be transmitted on the relevant buses. In accordance with this aspect of the invention, the cache of the CPU is divided into an instruction cache and a data cache. The data cache first loads a necessary word from a cache line (sub-block ordering) and permits a hazard-free, cache-line hit while a previous load is still in process (hit-under-miss). The output from the cache is also buffered in a write back buffer. This structure allows write requests to be stored until the main bus is free.
A particularly preferred embodiment of the invention provides a scratchpad RAM that works as a double buffer for the CPU. In an application dealing primarily with computer graphics, most of the data written out of the primary processor will be in the form of display lists, which contain the results of geometry calculations in the form of vertex information of primitive objects. These display lists, once generated, will not be needed again by the primary processor because they are a final result to be passed on to the geometry processor. Therefore, there is no benefit derived from caching these data in a traditional data cache when writing out this data (a write access scheme). However, most data read by such a computer graphics application are three-dimensional object data. A whole object must be cached in order to effect the speed of the CPU access to the object. The scratchpad allows a fast way to simultaneously write the display lists and read the object data without going through the standard data cache. Direct memory access (“DMA”) transfers between the main memory and the scratchpad allows data transfer without CPU overhead. Treating the scratchpad as a double buffer hides main memory latency from the CPU.
Another aspect of the present invention is the provision of common protocol data jacks for enabling multiple types
Hiroi Toshiyuki
Ohba Akio
Oka Masaaki
Okada Toyoshi
Suzuoki Masakazu
Coleman Eric
Lerner David Littenberg Krumholz & Mentlik LLP
Sony Computer Entertainment Inc.
LandOfFree
Game system with graphics processor does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Game system with graphics processor, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Game system with graphics processor will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3264852