Electrical computers and digital processing systems: processing – Processing control – Arithmetic operation instruction processing
Reexamination Certificate
1998-04-30
2001-07-24
Niebling, John F. (Department: 2812)
Electrical computers and digital processing systems: processing
Processing control
Arithmetic operation instruction processing
C345S520000
Reexamination Certificate
active
06266769
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates generally to the parallel manipulation of data and, more particularly, to the parallel conversion of data between data formats in different architectural registers.
BACKGROUND OF THE INVENTION
In typical computer systems, processors are implemented to operate on values represented by a large number of bits, for example, 32-bits, using instructions that produce one result. For example, the execution of an add instruction will add together a first 32-bit value and a second 32-bit value and store the result as a third 32-bit value. Some applications, however, require the manipulation of large amounts of data represented by fewer than 32 bits. Multi-media graphics, for instance, are typically generated by treating an image as a collection of small, independently controlled dots, or pixels. Position coordinates and color values corresponding to pixels are typically represented by fewer than 32 bits. The processing of the large amounts of data through a pipeline required by graphics applications can greatly increase processing time and slow graphics rendering correspondingly.
Multimedia graphics applications include, but are not limited to, applications targeted at computer supported cooperation (CSC), two-dimensional (2D) graphics, three-dimensional (3D) graphics, image processing, video compression/decompression, recognition algorithms and audio manipulation. As such, the data of multimedia applications typically comprises still images or video frames and sound data. The pixels of the still image or video data are typically represented using 8- or 16-bit data elements, and the sound data is typically represented using 8- or 16-bit data elements. When processing multimedia data comprising still images or video frames, the same operation is often performed repeatedly over all of the pixels of the image or of the frame. As each of these multimedia applications typically use one or more algorithms, and each algorithm typically uses a number of operations, multimedia extensions used to execute the same operations on 8-bit, 16-bit, or even 32-bit data while processing two, four, or eight data samples at a time speeds up computations that exhibit data parallelism.
To improve efficiency of multimedia applications, as well as other applications having similar characteristics, prior art processors use packed data formats. A packed data format is one in which a certain number of fixed sized data elements, each of which represents a separate value, are stored together. For example, a 64-bit register may be broken into two 32-bit elements, each of which represents a separate 32-bit value. In addition, these prior art processors provide instructions for separately manipulating each element in these packed data types in parallel. For example, a packed add instruction adds together corresponding data elements from a first packed data and a second packed data. Thus, if a multimedia algorithm requires a loop containing five operations that must be performed on a large number of data elements, it is desirable to pack the data and perform these operations in parallel using packed data instructions. In this manner, these processors can more efficiently process multimedia applications.
Therefore, in order to reduce the time required for graphics rendering in multimedia applications, parallel processing is used, wherein a single instruction operates on multiple elements of data; this process is typically referred to as Single Instruction Multiple Data (SIMD) processing. Typically, integer instructions operate on individual integer data elements (A+B). The SIMD instructions, however, operate on integer data arrays (A[1 . . . n]+B[1 . . . n]), where n is the number of elements in the array.
Typical prior art processing systems, in rendering 2D images, used only integer data in the geometry and rasterization phases because the smaller range of coordinate values did not necessitate the precision of floating point arithmetic. Therefore, the graphics data was rendered using SIMD processing of integer data, meaning that no conversion was typically required between the integer format and the floating point format.
However, in rendering 3D images, the data manipulations performed for the geometry phase are typically performed using floating point arithmetic because of the large range of values that define the coordinate space and because of the precision required within this range to accurately place the rendered images. Because the color component data is often stored and manipulated along with the corresponding position data it is convenient to perform operations on the rasterization data comprising color component data using floating point arithmetic. Upon completion of processing, the coordinates of the composited images are provided to the rasterization circuitry using the floating point format. In contrast, the color component data is provided to the rasterization circuitry using the integer format. Therefore, the color component data used to render the image is converted from the floating point format to the integer format in order to render an image display.
The problem in the prior art processors using SIMD processing of 3D graphic data is that, while parallel processing may be performed on floating point data, the conversion of the floating point data to integer data for rasterization creates a bottleneck in the processing pipeline because the prior art algorithms perform conversions sequentially. A prior art method of dealing with this problem duplicates the floating point execution resources of the processor. This duplication of resources allows for two floating point pipelines executing at the same time wherein the floating point data of each branch of the pipeline can be sequentially converted to integer format at the same time. While the delay due to the conversion execution bottleneck may be reduced with the use of the additional hardware, the additional hardware increases the cost and size of the system while increasing the overall complexity of the system.
SUMMARY OF THE INVENTION
A method and apparatus for converting a number between a floating point format and an integer format are described. Numbers are stored in the integer format in a register of a first set of architectural registers in a packed format. At least one of the numbers in the integer format is converted to at least one number in the floating point format. The numbers in the floating point format are placed in a register of a second set of architectural registers in a packed format.
Other features and advantages of the present invention will be apparent from the accompanying drawings and from the detailed description and appended claims that follow below.
REFERENCES:
patent: 4949291 (1990-08-01), Saini
patent: 5157388 (1992-10-01), Kohn
patent: 5257215 (1993-10-01), Poon
patent: 5303174 (1994-04-01), Okamoto
patent: 5404469 (1995-04-01), Chung et al.
patent: 5528741 (1996-06-01), Lucas
patent: 5627773 (1997-05-01), Wolrich et al.
patent: 5673407 (1997-09-01), Poland et al.
patent: 5696709 (1997-12-01), Smith, Sr.
patent: 5721855 (1998-02-01), Hinton et al.
patent: 5729724 (1998-03-01), Sharangpani et al.
patent: 5740093 (1998-04-01), Sharangpani
patent: 5761103 (2000-12-01), Oakland et al.
patent: 5764548 (1998-06-01), Keith et al.
patent: 5768170 (1998-06-01), Smith
patent: 5802219 (1998-09-01), Farkas et al.
patent: 5805486 (1998-09-01), Sharangpani
patent: 5848284 (1998-12-01), Sharangpani
patent: 5852726 (1998-12-01), Lin
patent: 5874969 (1999-02-01), Storm et al.
patent: 5933160 (1989-04-01), Kabir et al.
patent: 5968116 (1999-10-01), Day, II et al.
patent: 5978901 (1998-06-01), Luedtke et al.
patent: 5995122 (1999-11-01), Hsieh et al.
patent: 6058465 (2000-05-01), Nguyen
patent: 6061782 (2000-05-01), Elliot et al.
patent: WO99/23548 (2000-12-01), None
Digital Equipment Corp., “Alpha Architecture Handbook”, Oct. 1996. pp. 4-79—4-81, and 4-96—4-116.*
AMD, “AMD-K6 Processor Data Sheet”, AMD, pp. 21-38, Mar. 1998.*
“MIPS V Instruction Set”;
Abdallah Mohammad A. F.
Hsieh Hsien-Cheng E.
Huff Thomas R.
Pentkovski Vladimir
Roussel Patrice
Blakely , Sokoloff, Taylor & Zafman LLP
Intel Corporation
Niebling John F.
Whitmore Stacy
LandOfFree
Conversion between packed floating point data and packed... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Conversion between packed floating point data and packed..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Conversion between packed floating point data and packed... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2444069