Advanced parallel array processor (APAP)

• US 5,717,943 A
• Filed: 06/05/1995
• Issued: 02/10/1998
• Est. Priority Date: 11/13/1990
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Retter, Eric Eugene, Schoonover, Nicholas Jerome, Kogge, Peter Michael, Wilkinson, Paul Amba, Dieffenderfer, James Warren, Richardson, Robert Reist, Collins, Clive Allan, Rolfe, David Bruce, Dapp, Michael Charles, Barker, Thomas Norman, Lesmeister, Donald Michael, Kuchinski, David Christopher, Miles, Richard Ernest, Smoral, Vincent John, Knowles, Billy Jack, Nier, Richard Edward, Grice, Donald George, Stupp, James Robert
• Primary Examiner(s): Coleman, Eric
• Application Number:
  US08/465,926
• Time in Patent Office: 981 Days
• Field of Search: 395/800
• US Class Current: 712/20
• CPC Class Codes: F02B 2075/027 four
• CPC Class Codes: G06F 15/17337 Direct connection machines, e.g. completely connected computers, point to point communication networks coupling between buses G06F13/4004
• CPC Class Codes: G06F 15/17343 wherein the interconnection is dynamically configurable, e.g. having loosely coupled nearest neighbor architecture reconfigurable processors arrays G06F15/7867
• CPC Class Codes: G06F 15/17368 non hierarchical topologies
• CPC Class Codes: G06F 15/17381 Two dimensional, e.g. mesh, torus
• CPC Class Codes: G06F 15/8007 single instruction multiple data [SIMD] multiprocessors
• CPC Class Codes: G06F 15/8015 One dimensional arrays, e.g. rings, linear arrays, buses
• CPC Class Codes: G06F 15/803 Three-dimensional arrays or hypercubes
• CPC Class Codes: G06F 15/8092 Array of vector units
• CPC Class Codes: G06F 7/483 Computations with numbers represented by a non-linear combination of denominational numbers, e.g. rational numbers, logarithmic number system or floating-point numbers G06F7/4806, G06F7/4824, G06F7/49, G06F7/491, G06F7/544 take precedence
• CPC Class Codes: G06F 9/30036 Instructions to perform operations on packed data, e.g. vector, tile or matrix operations
• CPC Class Codes: G06F 9/30145 Instruction analysis, e.g. decoding, instruction word fields
• CPC Class Codes: G06F 9/30167 of immediate specifier, e.g. constants
• CPC Class Codes: G06F 9/30189 according to execution mode, e.g. mode flag
• CPC Class Codes: G06F 9/3834 Maintaining memory consistency
• CPC Class Codes: G06F 9/3851 from multiple instruction streams, e.g. multistreaming
• CPC Class Codes: G06F 9/3885 using a plurality of independent parallel functional units
• CPC Class Codes: G06F 9/3887 controlled by a single instruction for multiple data lanes [SIMD]
• CPC Class Codes: G06F 9/3889 controlled by multiple instructions, e.g. MIMD, decoupled access or execute
• CPC Class Codes: G06F 9/3891 organised in groups of units sharing resources, e.g. clusters

Abstract

A computer system having a plurality of processors and memory including a plurality of scalable nodes having multiple like processor memory elements. Each of the processor memory elements has a plurality of communication paths for communication within a node to other like processor memory elements within the node. Each of the processor memory elements also has a communication path for communication external to the node to another like scalable node of the computer system.

Claims

1. A computer system, comprising:

a plurality of processor memory elements on a common substrate each of said processor memory elements having internal and external communications paths for communication between a plurality of other processor memory elements on the substrate and for communication with other processing elements external to said substrate;

wherein the computer system provides a multi-processor memory system including a PME architecture multi-processor memory element on a single semiconductor substrate which functions as a system node, said multi-processor memory element including said plurality of processing memory elements, and means on said substrate for distributing interconnection and controls within the multi-processor memory system node enabling the system to perform SIMD/MIMD functions as a multi-processor memory system, wherein each dedicated local memory is independently accessible by the respectively coupled processor in both SIMD and MIMD modes exclusive of access by another processor.

2. A computer system according to claim 1 wherein a node of a processor array has multiple single processor elements comprising a network node of eight or more processors with their associated memory with fully distributed I/O routers and signal I/O ports at each of the single processor elements.

3. A computer system according to claim 1 wherein said processor memory elements common substrate is a chip and the internal and external communication paths plus the functional relationships between the processor memory elements and their communication paths are provided on said chip.

4. A computer system according to claim 1 wherein a node processing memory element has internal data flows using high speed hard registers to feed distributed ALU and I/O router registers and logic for all operations.

5. A computer system according to claim 1 wherein a node has multiple processor memory elements and is connected to other nodes in a cluster network with data routing distributed between hardware and software, with software controlling most of the task sequencing function.

6. A computer system according to claim 1 wherein a node has multiple processor memory elements and is connected to other nodes in a cluster network with data routing distributed between hardware and software, with hardware provided for performing inner PME-to-PME loop data transfers thereby minimizing overhead on the outer PME-to-PME loops of the node.

7. A computer system according to claim 1 wherein a node of a processor array has multiple single processor elements made up of 32K 16-bit words with a 16-bit processor for a network node of eight processors with their associated memory with their fully distributed I/O routers and signal I/O ports.

8. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein the system provides the control levels and features required to support distinct user levels ranging from completely unaware to fully involved in detailed tasking.

9. A computer system according to claim 1 organized as an n-dimensional network of chip processing memory elements having pins per chip can be well utilized by distributing I/O over as many pins as available while memory associated with each processor memory element is accessible to other processor memory elements providing scalable and large memory by replication of nodes of the system provided with parallel I/O bandwidths.

10. A computer system according to claim 1 further comprising means for enabling distributed fault tolerance by an n-redundancy mode, where data is given a redundancy number and replicated within an array of processing memory elements.

11. A computer system according to claim 1 wherein control tasks are distributed to multiple levels and highly repetitive tasks to be distributed to low levels of the system.

12. A computer system according to claim 1 further comprising within a processor memory element:

a native instruction set,

means for providing an expandable multiply function,

a programmer router for routing information alternatively left/right, NEWS matrix, NEWS/up-down, hypercube,

and wherein said programmable router is employs a hardwired distributed router provided by each processing memory element.

13. A computer system according to claim 1 wherein said processor memory elements are replicated on said common substrate as a single chip design that can be utilized as computational unit used as a main building block for system ranging from PC accelerators to super computer size machines.

14. A computer system according to claim 1 wherein each computer element (PME) has 2n processors, memory and routers within a single chip and internal and external communication paths which minimize delays due to chip crossings.

15. A computer system according to claim 1 wherein each computer element (PME) has a processor, memory and routers within a single chip and internal and external communication paths which minimize delays due to chip crossings, each processor memory element having means for fixed and floating point processing, routing and I/O control.

16. A computer system according to claim 1 wherein each processor memory element (PME) provides a processor, memory, and program interrupt capability for the system.

17. A computer system according to claim 1 wherein each processor memory element (PME) provides each of the PME'"'"'s 64K bytes of dedicated local memory, with a performance of 2.0 MIPS.

18. A computer system according to claim 1 a node architecture of plural processing memory elements on a common substrate provides the packaging of eight processing elements with their associated memory and communication logic encompassed into a single chip which enables a chip to be defined as a node in an array.

19. A computer system according to claim 1 wherein each processor memory element comprises a 16-bit processing memory element with full 16-bit internal control and data flow for PME to PME communication within a network defined for said common substrate as a node.

20. A computer system according to claim 1 wherein the common substrate is a chip and eight processing memory elements are provided on said chip, said chip providing a node of system composed of a plurality of computer systems, and wherein processor memory elements within the node are connected in a binary hypercube communication network.

21. A computer system according to claim 1 wherein multi-memory processor elements of the computer system are on a common silicon chip substrate and eight processing memory elements are provided on said chip, and communication between processing memory elements is controlled by in and out registers under control of each of processing elements.

22. A computer system according to claim 1 wherein the common substrate is a chip and eight processing memory elements are provided on said chip, said chip providing a node of system composed of a plurality of computer systems, and wherein processor memory elements within the node are connected in a binary hypercube communication network, and each node has eight processing memory elements on said chip, and communication between processing memory elements is controlled by in and out registers under control of each of processing elements.

23. A computer system according to claim 1 further comprising a system controller, and a broadcast control interface, said broadcast control interface enabling the controller and the processor memory elements to interrogate a processor memory element for status information within the processor memory element even during operation of said processor memory element.

24. A computer system according to claim 1 further comprising a system controller, and a broadcast control interface, said broadcast control interface providing means for providing inter-processor busing protocols, said system controller functioning as an array controller for nodes of processing memory elements.

25. A computer system according to claim 1 further comprising a system controller, and a broadcast control interface, said broadcast control interface providing means for providing inter-processor busing protocols, said system functioning as an array accelerator for a conventional processor.

26. A computer system according to claim 1 further comprising means for enabling clustering of a set of processor memory elements within a node for sharing common broadcast and control interfaces.

27. A computer system according to claim 1 further comprising within a processing memory element:

multiple register sets for each interrupt level providing means for saving status when taking interrupts.

28. A computer system according to claim 1 further comprising within a processing memory element:

storage register elements mapped to memory for enabling hardware expansion of memory without a software impact.

29. A computer system according to claim 1 wherein the communication paths are wrapped to connects nodes of the computer system as cubes.

30. A computer system according to claim 1 wherein four interrupts levels are provided and two communication modes.

31. A computer system according to claim 1 wherein one of two communication modes permits direct memory addressing.

32. A computer system according to claim 1 wherein communication modes between processing memory elements are provided for handling contention with busy processing memory elements in a data stream.

33. A computer system according to claim 1 wherein connections between nodes of the system provides NEWS connections among nodes, 2-d torus connections, and 4-D wrapped torus connections among nodes of the computer system.

34. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional hypercube network with parallel communication paths between processor memory elements along said internal and external communication paths providing N-dimensional networking among nodes, the interconnection among nodes supporting a 4 dimensional hypercube.

35. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional hypercube network with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and further comprising a set of modules which serve as an modular array controller such that varying size systems can be configured by replication.

36. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and further comprising an array controller such that varying size systems can be configured by replication.

37. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein each node of a processing array has eight processing memory elements on a chip which are interconnected as part of a system network, and node to node control includes a broadcast bus and tag lines to provide a system interconnection among nodes.

38. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein each node interconnection is an eight dimensional low degree hypercube.

39. A computer system according to claim 1 wherein each processing memory element can implement floating point calculation.

40. A computer system according to claim 1 wherein each processing memory element can implement floating point calculation by emulating floating point on a processing memory element.

41. A computer system according to claim 1 wherein each processing memory element can implement floating point calculation by a hardware floating point accelerator section provided by logic.

42. A computer system according to claim 1 organized as an n-dimensional network of processing memory elements having the capability to perform MIMD operations and wherein an array of nodes is provided with directed inter-processor memory element communication for performing sparse to dense data structure transformations.

43. A computer system according to claim 1 organized as an n-dimensional network of processing memory elements having the capability to perform MIMD operations and wherein an array of nodes is provided with means for directed inter-processor memory element communication for preparing data for pipelined very high speed arithmetic and for mapping results back to a sparse data structure.

44. A computer system according to claim 1 organized as an n-dimensional network of processing memory elements having the capability to perform MIMD operations and wherein an array of nodes is provided with means for data packing/unpacking for high speed pipelined floating point operations.

45. A computer system according to claim 1 organized as an n-dimensional network of processing memory elements having the capability to perform MIMD operations and wherein an array of nodes is provided with means for performing scatter/gather operations in parallel with arithmetic operations.

46. A computer system according to claim 1 organized as an n-dimensional network of processing memory elements having the capability to perform MIMD operations and wherein an array of nodes is provided with means for performing programmable support support for sparse matrix encoding.

47. A computer system according to claim 1 organized as an n-dimensional network of processing memory elements having memory resident registers and floating point capability.

48. A computer system according to claim 1 organized as an n-dimensional network of chip processing memory elements having chip I/O pins utilized for control.

49. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein said system has multiple clusters.

50. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein said system has quarter cluster XYWZ wiring.

51. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein said system has cluster XY internal wiring and WZ external wiring.

52. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein said system has cluster to cluster W Z wiring.

53. A computer system according to claim 1 wherein a node interconnection supports a 4 dimensional hypercube.

54. A computer system according to claim 1 wherein nodes form a processing array and set of modules and an array controller such that varying modular size systems can be designed or configured by replication.

55. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein said system is provided with an array director having overall control of the cluster providing a cluster controller for an array of nodes.

56. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein said system is provided with a plurality of clusters and an array director having overall control of the clusters providing a cluster controller for an array of nodes, and a cluster synchronizer for an array of clusters.

57. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein said system is provided with a plurality of clusters and an array director having overall control of the clusters providing a cluster controller for an array of nodes, and a cluster synchronizer for an array of clusters, and an I/O zipper along the edges of selected clusters.

58. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein said system is provided with a plurality of clusters and an array director having overall control of the clusters providing a cluster controller for an 8×

8 array of nodes, and a cluster synchronizer for an 8×

8 array of clusters, and an I/O zipper along the edges of selected clusters.

59. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein said system has rings for communication which can be broken to provide an effective interface to external I/O bus paths, and interfaces to attach external storage or additional display units to a subset of processing memory elements of the processing array of nodes.

60. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein any set of inter-processor memory element communication paths can be broken to add an external I/O.

61. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein any set of inter-processor memory element communication paths can be broken to add an external I/O interface with all processing memory elements of a node interfacing to the external interface.

62. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein processing memory elements of an array have a transparent mode utilized when routing data between processing memory elements within a chip set of processing memory elements for permitting reduction of the effective network diameter of a network of nodes.

63. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein simulation and users are supported with variable resolution simulator software tools and support software visualization tools for load balancing and performance evaluation.

64. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein an applications program interface is provided for coupling said processing array within a computer system, with a processor and software within the application processor interface enabling system utilization to be controlled.

65. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein an applications program interface is provided for coupling said processing array within a computer system, with a processor and software within the application processor interface enabling system utilization to be controlled in either a standalone mode or as an attached processor to one or more hosts.

66. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein an applications program interface is provided for coupling said processing array within a computer system, with a processor and software within the application processor interface enabling system utilization to be controlled when coupled to one or more hosts by either a tightly or loosely coupled attachment.

67. A computer system according to claim 1 wherein a node forms a node element of multiple processor memory element of an parallel array processor, incorporating up to 32,768 identical processor memory elements in an array cluster, said array processor further including an array director, said array director comprising an application processor interface, a cluster synchronizer and cluster controller.

68. A computer system according to claim 1 wherein a node forms a node element of multiple processor memory element of an parallel array processor incorporating up to 64 array clusters as the basic building block of a parallel array computer system having an array director.

69. A computer system according to claim 1 wherein a node forms a node element of multiple processor memory element of an parallel array processor incorporating array clusters as the basic building block of a parallel array computer system having an array director for controlling clusters, and an application processor interface that supports a test/debug unit for design, debug, and monitoring functions.

70. A computer system according to claim 1 wherein a node form part of a cluster, each node having a set of multiple processor memory elements, and wherein a complete set of processor memory element may be divided into 1 to many distinct subsets under control of an array director to interleave between subsets.

71. A computer system according to claim 1 wherein a node form part of a cluster, each node having a set of multiple processor memory elements, and wherein a complete set of processor memory element may be divided into 1 to many distinct subsets under control of an array director to interleave between subsets, the sequence of the interleave process and the amount of control exercised over each subset being program controlled by said array director.

72. A computer system according to claim 1 wherein a node form part of a cluster, each node having a set of multiple processor memory elements, and wherein a complete set of processor memory element may be divided into 1 to many distinct subsets under control of an array director to interleave between subsets to to operate distinct subsets of the array in one mode while other sets operate in tightly synchronized SIMD mode under array director control.

73. A computer system according to claim 1 wherein a node comprises processing memory elements having 16-bit wide dataflow by 16-bit memory, I/O ports and I/O switching paths, and the necessary control logic to permit any processing memory element to fetch, decode and execute a 16 bit reduced instruction set.

74. A computer system according to claim 1 wherein a node comprises processing memory elements having 16-bit wide dataflow by 16-bit memory, I/O ports and I/O switching paths, and the necessary control logic to permit any processing memory element to fetch, decode and execute a 16 bit reduced instruction set, and further logic enables a processing memory element to perform as both a processing unit and data router with specialized control means for rapid task switching and retention of program state information at interrupt execution levels.

75. A computer system according to claim 1 wherein a node processing memory element in SIMD mode mode receives decoded instruction information and executes that data in the next clock cycle.

76. A computer system according to claim 1 wherein a node processing memory element in MIMD mode maintains a program counter PC address and uses that as the address within its own memory to fetch an instruction.

77. A computer system according to claim 1 wherein a node processing memory element in has three input ports, and three output ports for on-chip communication plus an I/O port for off chip communications.

78. A computer system according to claim 1 wherein a node processing memory element in has an I/O port for off chip byte wide half duplex communication.

79. A computer system according to claim 1 wherein a node processing memory element in has an I/O port for off chip byte wide communication, and has input ports that are connected such that data may be routed from input to memory, or from an input address register to an output register via a direct parallel data path.

80. A computer system according to claim 1 wherein a node broadcast and control interface provided on each chip provides a parallel input interface and also provides two serial port interfaces providing each processor memory element with the capability to output a limited amount of status information for signaling an array director when the processor memory element has data that needs to be read.

81. A computer system according to claim 1 wherein a node broadcast and control interface provided on each chip provides a parallel input interface and also provides two serial port interfaces providing each processor memory element with the capability to output activity status information such that test and monitor displays can illustrate the status of the entire system of which a node forms a part.

82. A computer system according to claim 1 wherein a node broadcast and control interface provided on each chip provides a parallel input interface and also provides two serial port interfaces providing a serial port to permit an array director to input control data so as to direct the port to particular processor memory element and node internal registers and access points.

83. A computer system according to claim 1 wherein a node has multiple processing memory elements and each processing element is enabled to process data lengths which are multiples of 16 bits.

84. A computer system according to claim 1 wherein a node broadcast and control interface provided on each chip is coupled to a node controller coupled though a set of nodes with a serial scan loop for acquiring information on individual nodes or processing memory elements with a node.

85. A computer system according to claim 1 wherein a node broadcast and control interface provided on each chip is coupled to a node controller coupled though a set of nodes with a serial scan loop for acquiring information on individual nodes or processing memory elements with a node, with loops initially interfacing to the broadcast control interface which bridges to individual processor memory elements.

86. A computer system according to claim 1 wherein a node has multiple processor memory elements with single instructions which can perform adds/subtracts/compares/moves on operands of length 1 to 8 words where a length is a part of the instruction, enabling a programmer to view an operand length as 16 to 128 bits in length.

87. A computer system according to claim 1 wherein a node has multiple processor memory elements with single instructions which can perform adds/subtracts/compares/moves on operands of length 1 to 8 words where a length is a part of the instruction, enabling a programmer to view an operand length as 16 to 128 bits in length, an by using combinations of these instructions, operands of any length can be manipulated to add two numbers of up to 256 bits in length.

88. A computer system according to claim 1 wherein a node having multiple processor memory elements is configured within the node as a 2 by n array with each processor memory element interconnected with its three neighbors using a set of input/output ports, thus, providing full duplex communication capability between processor memory elements, with each processor memory element having an external input and output port connected to node I/O pins.

89. A computer system according to claim 1 wherein a node is provided with I/O pins for external to node communication with input and output ports connected to share pins for half-duplex communication.

90. A computer system according to claim 1 wherein a node is provided with I/O pins for external to node communication with input and output ports connected to separate pins for full-duplex communication.

91. A computer system according to claim 1 wherein a node has eight processing memory elements within the node which are connected in a binary hypercube communication network.

92. A computer system according to claim 1 wherein a node has eight processing memory elements within the node which are connected in a binary hypercube communication network which binary hypercube displays three intra connections between processor memory element, and wherein communication among processor memory elements is controlled by in and out registers under control of the processing element.

93. A computer system according to claim 1 wherein a node has multiple processor memory elements With an inter-PME network that is a 4d torus mechanism used for inter-PME communication whereby any processor memory element can reach any other processor memory elements configured in an array processor.

94. A computer system according to claim 1 wherein a node is a chip having

a. eight processor memory elements forming each a 16-bit processor and 32K words of in 64K bytes memory,

b. a broadcast and control interface enabling an external controller to operate all or subsets of the processor memory elements and to accumulate processor memory element requests, and having plural interconnection levels wherebyeach processor memory elements supports four 8-bit wide inter-PME communication paths which connect to three neighboring processor memory elements on the chip and one off chip processing memory element,

broadcast-to-PME busing, which makes data or instructions available to processor memory elements,service request lines that permit any processor memory element to send a code to an external controller, andserial service loops permitting the controller to read detail with an interconnection level which extends from the node to all processor memory elements within the node.

95. A computer system according to claim 1 wherein a node has multiple processor memory elements on a chip with interconnections to their immediate neighbors with an interconnection pattern that results in a three dimensional cube structure.

96. A computer system according to claim 1 wherein a node has multiple processor memory elements on a chip with interconnections to their immediate neighbors with an interconnection pattern that results in a three dimensional cube structure, and wherein each processor memory element within the cube has a dedicated bidirectional byte port to the chip'"'"'s pins.

97. A computer system according to claim 1 wherein a node forms part of the computer system in which an n by n array of nodes is a cluster, and bridging between the + and -x ports and the + and -y ports provide the cluster node interconnections.

98. A computer system according to claim 1 wherein a node forms part of the computer system in which an n by n array of nodes is a cluster, and bridging between the + and -x ports and the + and -y ports provide the cluster node interconnections, and a node has multiple processor memory elements, each of which manages a single external port.

99. A computer system according to claim 1 wherein a node forms part of the computer system in which an n by n array of nodes is a cluster, and bridging between the + and -x ports and the + and -y ports provide the cluster node interconnections, and a node has multiple processor memory elements, each of which manages a single external port, and node bridging at the outer edges makes the cluster into a logical torus.

100. A computer system according to claim 1 wherein a node forms part of the computer system in which an n by n array of nodes is a cluster, and bridging between the + and -w ports and + and -z ports provides the 4d torus interconnections with each node within a cluster connected to an equivalent node in all adjacent clusters.

101. A computer system according to claim 1 wherein a node forms part of the computer system in which an n by n array of nodes is a cluster, and bridging between the + and -w ports and + and -z ports provides the 4d torus interconnections with each node within a cluster connected to an equivalent node in all adjacent clusters for providing 64 ports between two adjacent clusters of an array processor.

102. A computer system according to claim 1 wherein a node has multiple processor memory elements coupled by an optical path and an on-chip optical driver/receiver for coupling of nodes in an array of 4d toruses.

103. A computer system according to claim 1 wherein a node has twelve processor memory elements on a chip with an on-chip optical driver/receiver for coupling of nodes in an array of 4d toruses having multi-flop performance.

104. A computer system according to claim 1 wherein a node has multiple processor memory elements and is connected to other nodes in a cluster network with I/O programs at dedicated interrupt levels for managing the network.

105. A computer system according to claim 1 wherein a node has multiple processor memory elements and is connected to other nodes in a cluster network with I/O programs at dedicated interrupt levels for managing the network, each processor memory element having interrupt registers and dedicating four interrupt levels to receiving data from four neighbors, a buffer provided at each level by loading registers at the level, and having in and return instruction pairs using a buffer address and transfer count to enable the processor memory element to accept words from an input bus and to store them to the buffer.

106. A computer system according to claim 1 wherein a node has multiple processor memory elements and is connected to other nodes in a cluster network with I/O programs at dedicated interrupt levels for managing the network, each processor memory element having interrupt registers and dedicating four interrupt levels to receiving data from four neighbors, a buffer provided at each level by loading registers at the level, and having in and return instruction pairs using a buffer address and transfer count to enable the processor memory element to accept words from an input bus and to store them to the buffer, means enabling a buffer full condition to generate an interrupt and restore the program counter to the instruction after a return to enable I/O programs which do need to test what caused an interrupt to read data, return, and then continue directly into processing the data they read.

107. A computer system according to claim 1 organized as a massively parallel machine with nodes interconnected as a n dimensional network cluster with parallel communication paths between processor memory elements along said internal and external communication paths providing a processing array, and wherein loading or unloading the array is done by connecting the edges of a cluster to a system bus.

108. A computer system according to claim 1 wherein a node array allows data transfers into or out of the array to alternate between the two nodes directly attached to a zipper to permit different zipper sizes to satisfy a configuration'"'"'s particular application requirements.

109. A computer system according to claim 1 wherein a node array allows data transfers into or out of the array to alternate between the two nodes directly attached to a zipper width size can be varied depending on the configuration and the need of the program to roll data and program into and out of the individual processor memory elements.

110. A computer system according to claim 1 including a zipper connection for a processor array which exists as a separate interface into an array of processor memory elements, said zipper being established on one or two edges of a hypercube network by dotting onto a broadcast bus at multiple nodes in the array and in multiple directions.

111. A computer system according to claim 1 including an array director having:

a. an application processor interface,

b. a cluster synchronizer,

c. a cluster controller for an array of nodes, and

d. a fast I/O for a zipper connection.

112. A computer system according to claim 1 wherein an application processor interface is provided for managing the path to and from an I/O zipper, and wherein data received from a host system in attached modes, or from devices in standalone modes is forwarded to the array, and wherein prior to initiating this type of operation processing memory elements within a array which will be managing the I/O are initiated.

113. A multi-processor memory system comprising:

on a chip a plurality of processing elements with a network interface, said processing elements of said chip being intercoupled by an internal communication network for passing information between processing elements on the chip, and having a broadcast port for external communication from the chip, said processing elements on the chip have their own memory and they are coupled in a network as a torus, said system having a plurality of chips which are coupled chip to chip to form a parallel array of multiple nodes of chips, each node having a broadcast and control interface for communications between processing elements on a chip and between nodes.

114. A multi-processor memory system according to claim 113, wherein said nodes are coupled by a network which is a modified hypercube.

115. A multi-processor memory system according to claim 114 further comprising selection means for selection of coupling paths to nodes in said modified hypercube network of nodes to change a path of coupled nodes in the network for the purpose of loading all of the processing elements in a node in parallel.

116. A computer system, comprising a plurality of chips which are coupled to form a parallel array of multiple nodes, each node having a broadcast and control interface for communications between processing elements on a chip and between nodes, said nodes being coupled by a network which is hypercube, selection means for selection of coupling paths to nodes in said modified hypercube network of nodes to change a path of coupled nodes in the network for the purpose of loading a node in parallel;

wherein the computer system provides a multi-processor memory system including a PME architecture multi-processor memory element on a single semiconductor substrate which functions as a system node, said multi-processor memory element including a plurality of processing memory elements, and means on said substrate for distributing interconnection and controls within the multi-processor memory system node enabling the system to perform SIMD/MIMD functions as a multi-processor memory system, wherein each dedicated local memory is independently accessible by the respectively coupled processor in both SIMD and MIMD modes exclusive of access by another processor.

117. A parallel array computer system, comprising:

a plurality of processing elements each having accessible memory and organized as cluster of processing elements, each of said processing elements of a cluster having a fast I/O tri-state driver;

wherein the parallel array computer system provides a multi-processor memory system including a PME architecture multi-processor memory element on a single semiconductor substrate which functions as a system node, said multi-processor memory element including a plurality of processing memory elements, and means on said substrate for distributing interconnection and controls within the multi-processor memory system node enabling the system to perform SIMD/MIMD functions as a multi-processor memory system, wherein each dedicated local memory is independently accessible by the respectively coupled processor in both SIMD and MIMD modes exclusive of access by another processor.