Electrical computers and digital data processing systems: input/ – Input/output data processing – Direct memory accessing
Reexamination Certificate
1999-07-16
2002-09-10
Ray, Gopal C. (Department: 2781)
Electrical computers and digital data processing systems: input/
Input/output data processing
Direct memory accessing
C709S253000, C370S408000, C712S011000, C712S029000
Reexamination Certificate
active
06449667
ABSTRACT:
INCORPORATION BY REFERENCE
Guy E. Blelloch, Scan Primitives and Parallel Vector Models, (Ph.D. Dissertation, Massachusetts Institute of Technology: 1988), incorporated herein by reference.
U.S. patent application Ser. No. 07/489,079, filed Mar. 5, 1990, in the name of W. Daniel Hillis, et al., entitled Digital Clock Buffer Circuit Providing Controllable Delay, and assigned to the assignee of the present application, incorporated herein by reference.
FIELD OF THE INVENTION
The invention relates generally to the field of digital computer systems, and more particularly to massively parallel computing systems. The invention particularly provides arrangements for controlling processors in a computing system having a large number of processors, for facilitating transfer of data among the processors and for facilitating diagnosis of faulty components in the computing system.
BACKGROUND OF THE INVENTION
A digital computer system generally comprises three basic elements, namely, a memory element, an input/output element and a processor element. The memory element stores information in addressable storage locations. This information includes data and instructions for processing the data. The processor element fetches information from the memory element, interprets the information as either an instruction or data, processes the data in accordance with the instructions, and returns the processed data to the memory element. The input/output element, under control of the processor element, also communicates with the memory element to transfer information, including instructions and the data to be processed, to the memory, and to obtain processed data from the memory.
Most modern computing systems are considered “von Neumann” machines, since they are generally constructed according to a paradigm attributed to John von Neumann. Von Neumann machines are characterized by having a processing element, a global memory which stores all information in the system, and a program counter that identifies the location in the global memory of the instruction being executed. The processing element executes one instruction at a time, that is, the instruction identified by the program counter. When the instruction is executed, the program counter is advanced to identify the location of the next instruction to be processed. (In many modern systems, the program counter is actually advanced before the processor has finished processing the current instruction.)
Von Neumann systems are conceptually uncomplicated to design and program, since they do only one operation at a time. A number of advancements have been made to the original von Neumann paradigm to permit the various parts of the system, most notably the various components of the processor, to operate relatively independently and achieve a significant increase in processing speed. One such advancement is pipelining of the various steps in executing an instruction, including instruction fetch, operation code decode (a typical instruction includes an operation code which identifies the operation to be performed, and in most cases one or more operand specifiers, which identify the location in memory of the operands, or data, to be used in executing the instruction), operand fetch, execution (that is, performing the operation set forth in the operation code on the fetched operands), and storing of processed data, which steps are performed relatively independently by separate hardware in the processor. In a pipelined processor, the processor's instruction fetch hardware may be fetching one instruction while other hardware is decoding the operation code of another instruction, fetching the operands of still another instruction, executing yet another instruction, and storing the processed data of a fifth instruction. Since the five steps are performed sequentially, pipelining does not speed up processing of an individual instruction. However, since the processor begins processing of additional instructions before it has finished processing a current instruction, it can speed up processing of a series of instructions.
A pipelined processor is obviously much more complicated than a simple processor in a von Neumann system, as it requires not only the various circuits to perform each of the operations (in a simple von Neumann processor, many circuits could be used to perform several operations), but also control circuits to coordinate the activities of the various operational circuits. However, the speed-up of the system can be dramatic.
More recently, some processors have been provided with execution hardware which includes multiple functional units each being optimized to perform a certain type of mathematical operation. For example, some processors have separate functional units for performing integer arithmetic and floating point arithmetic, since they are processed very differently. Some processors have separate hardware functional units each of which performs one or only several types of mathematical operations, including addition, multiplication, and division operations, and other operations such as branch control and logical operations, all of which can be operating concurrently. This can be helpful in speeding up certain computations, most particularly those in which several functional units may be used concurrently for performing parts of a single computation.
In a von Neumann processor, including those which incorporate pipelining or multiple functional units (or both, since both may be incorporated into a single processor), a single instruction stream operates on a single data stream. That is, each instruction operates on data to enable one calculation at a time. Such processors have been termed “SISD,” for single-instruction/single-data.” If a program requires a segment of a program to be used to operate on a number of diverse elements of data to produce a number of calculations, the program causes the processor to loop through that segment for each calculation. In some cases, in which the program segment is short or there are only a few data elements, the time required to perform such a calculation may not be unduly long.
However, for many types of such programs, SISD processors would require a very long time to perform all of the calculations required. Accordingly, processors have been developed which incorporate a large number of processing elements all of which may operate concurrently on the same instruction stream, but with each processing element processing a separate data stream. These processors have been termed “SIMD” processors, for “single-instruction/multiple-data.”
SIMD processors are useful in a number of applications, such as image processing, signal processing, artificial intelligence, database operations, and computer simulation of a number of things, such as electronic circuits and fluid dynamics. In image processing, each processing element may be used to perform processing on a pixel (“picture element”) of the image to enhance the overall image. In signal processing, the processors concurrently perform a number of the calculations required to perform such computations as the “Fast Fourier transform” of the data defining the signal. In artificial intelligence, the processors perform searches on extensive rule bases representing the stored knowledge of the particular application. Similarly, in database operations, the processors perform searches on the data in the database, and may also perform sorting and other operations. In computer simulation of, for example, electronic circuits, each processor may represent one part of the circuit, and the processor's iterative computations indicate the response of the part to signals from other parts of the circuit. Similarly, in simulating fluid dynamics, which can be useful in a number of applications such as weather predication and airplane design, each processor is associated with one point in space, and the calculations provide information about various factors such as fluid flow, temperature, pressure and so forth.
Typical SIMD systems include a SIMD array, which includes the a
Douglas David C.
Ganmukhi Mahesh N.
Hill Jeffrey V.
Wong-Chan Monica C.
Cesari and McKenna LLP
Ray Gopal C.
T. M. Patents, L.P.
LandOfFree
Tree network including arrangement for establishing sub-tree... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Tree network including arrangement for establishing sub-tree..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Tree network including arrangement for establishing sub-tree... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2869943