Bops, Incorporated

Pts Corporation

Billions of Operations Per Second

Other

Altera Corporation

Edwin Frank Barry

Gerald George Pechanek

Edward A Wolff

Patrick Rene Marchand

Nikos P Pitsianis

Juan Guillermo Revilla

David Carl Strube

Charles W Kurak, Jr

Grayson Morris

Thomas L Drabenstott

Larry D Larsen

Dale Edward Schneider

Carl Donald Busboom

Ricardo E Rodriguez

Marco C Jacobs

Kevin Coopman

Details of a highly cost effective and efficient implementation of a manifold array (ManArray) architecture and instruction syntax for use therewith are described herein. Various aspects of this approach include the regularity of the syntax, the relative ease with which the instruction set can be represented in database form, the ready ability with which tools can be created, the ready generation of self-checking codes and parameterized test cases. Parameterizations can be fairly easily mapped and system maintenance is significantly simplified.

System core for transferring data between an external device and memory

Port priorities are defined on a 32-bit word, 16-bit half-word, and 8-bit byte basis to control the write enable signals to a compute register file (CRF). With a manifold array (ManArray) reconfigurable register file, it is possible to have double-word 64-bit and single word 32-bit data-type instructions mixed with other double-word, single-word, half-word, or byte data-type instructions within the same very long instruction word (VLIW). By resolving a write priority conflict on the byte, half-word, or word that is in conflict during the VLIW execution, it is possible to have partial operations complete that provide a useful function. For example, a load half-word to the half-word Hportion of a 32-bit register Rcan have priority to complete its operation while a 64-bit shift of the register pair Rand Rwill complete its operation on the non-conflicting half-word portions of the 64-bit register Rand ROther unique capabilities result from the present approach to assigning port priorities that improve the performance of the ManArray indirect VLIW processor.

Methods and apparatus for establishing port priority functions in a VLIW processor

A variety of advantageous mechanisms for improved data transfer control within a data processing system are described. A DMA controller is described which is implemented as a multiprocessing transfer engine supporting multiple transfer controllers which may work independently or in cooperation to carry out data transfers, with each transfer controller acting as an autonomous processor, fetching and dispatching DMA instructions to multiple execution units. In particular, mechanisms for initiating and controlling the sequence of data transfers are provided, as are processes for autonomously fetching DMA instructions which are decoded sequentially but executed in parallel. Dual transfer execution units within each transfer controller, together with independent transfer counters, are employed to allow decoupling of source and destination address generation and to allow multiple transfer instructions in one transfer execution unit to operate in parallel with a single transfer instruction in the other transfer unit. Improved flow control of data between a source and destination is provided through the use of special semaphore operations, signals and message synchronization which may be invoked explicitly using SIGNAL and WAIT type instructions or implicitly through the use of special 'event-action' registers. Transfer controllers are also described which can cooperate to perform 'DMA-to-DMA' transfers. Message-level synchronization can be used by transfer controllers to synchronize with each other.

Methods and apparatus for providing data transfer control

The ManArray processor is a scalable indirect VLIW array processor that defines two preferred architectures for indirect VLIW memories. One approach treats the VIM as one composite block of memory using one common address interface to access any VLIW stored in the VIM. The second approach treats the VIM as made up of multiple smaller VIMs each individually associated with the functional units and each individually addressable for loading and reading during XV execution. The VIM memories, contained in each processing element (PE), are accessible by the same type of LV and XV Short Instruction Words (SIWs) as in a single processor instantiation of the indirect VLIW architecture. In the ManArray architecture, the control processor, also called a sequence processor (SP), fetches the instructions from the SIW memory and dispatches them to itself and the PEs. By using the LV instruction, VLIWs can be loaded into VIMs in the SP and the PEs. Since the LV instruction is supplied by the SP through the instruction stream, when VLIWs are being loaded into any VIM no other processing takes place. In addition, as defined in the ManArray architecture, when the SP is processing SIWs, such as control and other sequential code, the PE array is not executing any instructions. Techniques are provided herein to independently load the VIMs concurrent with SIW or iVLIW execution on the SP or on the PEs thereby allowing the load latency to be hidden by the computation.

Methods and apparatus for loading a very long instruction word memory

The ManArray core indirect VLIW processor consists of an array controller sequence processor (SP) merged with a processing element (PE) closely coupling the SP with the PE array and providing the capability to share execution units between the SP and PE. Consequently, in the merged SP/PEa single set of execution units are coupled with two independent register files. To make efficient use of the SP and PE resources, the ManArray architecture specifies a bit in the instruction format, the S/P-bit, to differentiate SP instructions from PE instructions. Multiple register contexts are obtained in the ManArray processor by controlling how the array S/P-bit in the ManArray instruction format is used in conjunction with a context switch bit (CSB) for the context selection of the PE register file or the SP register file. In arrays consisting of more than a single PE, the software controllable context switch mechanism is used to reconfigure the array to take advantage of the multiple context support the merged SP/PE provides. For example, a 1&#xd7;1 can be configured as a 1&#xd7;1 with context-0 and as a 1&#xd7;0 with context-1, a 1&#xd7;2 can be configured as a 1&#xd7;2 with context-0 and as a 1&#xd7;1 with context-1, and a 1&#xd7;5 can be configured as a 1&#xd7;5 with context-0 and as a 2&#xd7;2 with context-1. Other array configurations are clearly possible using the present techniques. In the 1&#xd7;5/2&#xd7;2 case, the two contexts could be a 1&#xd7;5 array (context-0) and a 2&#xd7;2 array (context-1).

Methods and apparatus for providing context switching between software tasks with reconfigurable control

A method and system are described which provide flexible coupling between a coprocessor and a control processor. The system includes a coprocessor and a system control bus connecting the coprocessor with the control processor. The coprocessor has two modes of access. In the first mode of access, the coprocessor retrieves an instruction stored in instruction memory and, in the second mode of access, the coprocessor retrieves an instruction from the control processor. The system control bus provides a path for loading an instruction to the coprocessor's shadow instruction register. The coprocessor, upon retrieving an entry in its instruction memory associated with the shadow instruction resigter, determines whether to load the instruction as an address in its program counter or to load the contents of the shadow instruction register into the instruction decode register.

Control processor dynamically loading shadow instruction register associated with memory entry of coprocessor in flexible coupling mode

Hardware and software techniques for interrupt detection and response in a scalable pipelined array processor environment are described. Utilizing these techniques, a sequential program execution model with interrupts can be maintained in a highly parallel scalable pipelined array processing containing multiple processing elements (PEs) and distributed memories and register files. When an interrupt occurs, interface signals are provided to all PEs to support independent interrupt operations in each PE dependent upon the local PE instruction sequence prior to the interrupt. Processing/element exception interrupts are supported and low latency interrupt processing is also provided for embedded systems where real time signal processing is required. Further, a hierarchical interrupt structure is used allowing a generalized debug approach using debug interrupts and a dynamic debuts monitor mechanism.

Methods and apparatus for scalable array processor interrupt detection and response

Techniques for providing improved data distribution to and collection from multiple memories are described. Such memories are often associated with and local to processing elements (PEs) within an array processor. Improved data transfer control within a data processing system provides support for radix and fast Fourier transform (FFT) algorithms through data reordering or bit-reversed addressing across multiple PEs, carried out concurrently with FFT computation on a digital signal processor (DSP) array by a DMA unit. Parallel data distribution and collection through forms of multicast and packet-gather operations are also supported.

Methods and apparatus for providing bit-reversal and multicast functions utilizing DMA controller

A ManArray processor pipeline design addresses an indirect VLIW memory access problem without increasing branch latency by providing a dynamically reconfigurable instruction pipeline for SIWs requiring a VLIW to be fetched. By introducing an additional cycle in the pipeline only when a VLIW fetch is required, the present invention solves the VLIW memory access problem. The pipeline stays in an expanded state, in general, until a branch type or load VLIW memory type operation is detected returning the pipeline to a compressed pipeline operation. By compressing the pipeline when a branch type operation is detected, the need for an additional cycle for the branch operation is avoided. Consequently, the shorter compressed pipeline provides more efficient performance for branch intensive control code as compared to a fixed pipeline with an expanded number of pipeline stages. In addition, the dynamic reconfigurable pipeline is scalable allowing each processing element (PE) in an array of PEs to expand and compress the pipeline in synchronism allowing iVLIW operations to execute independently in each PE. This is accomplished by having distributed pipelines in operation in parallel, one in each PE and in the controller sequence processor (SP).

Methods and apparatus to dynamically reconfigure the instruction pipeline of an indirect very long instruction word scalable processor

A manifold array topology includes processing elements, nodes, memories or the like arranged in clusters. Clusters are connected by cluster switch arrangements which advantageously allow changes of organization without physical rearrangement of processing elements. A significant reduction in the typical number of interconnections for preexisting arrays is also achieved. Fast, efficient and cost effective processing and communication result with the added benefit of ready scalability.

Methods and apparatus for manifold array processing

Details of a highly cost effective and efficient implementation of a manifold array (ManArray) architecture and instruction syntax for use therewith are described herein. Various aspects of this approach include the regularity of the syntax, the relative ease with which the instruction set can be represented in database form, the ready ability with which tools can be created, the ready generation of self-checking codes and parameterized testcases. Parameterizations can be fairly easily mapped and system maintenance is significantly simplified.

Constructing database representing manifold array architecture instruction set for use in support tool code creation

A processor with a generalized eventpoint architecture, which is scalable for use in a very long instruction word (VLIW) array processor, such as the manifold array (ManArray) processor is described. In one aspect, generalized processor event (p-event) detection facilities are provided by use of compares to check if an instruction address, a data memory address, an instruction, a data value, arithmetic-condition flags, or other processor change of state eventpoint has occurred. In another aspect, generalized processor action (p-action) facilities are provided to cause a change in the program flow by loading the program counter with a new instruction address, generate an interrupt, signal a semaphore, log or count the p-event, time stamp the event, initiate a background operation, or to cause other p-actions to occur. The generalized facilities are defined in the eventpoint architecture as consisting of a control register and three eventpoint parameters, namely at least one register to compare against, a register containing a second compare register, a vector address, or parameter to be passed, and a count or mask register. Based upon this generalized eventpoint architecture, new capabilities are enabled. For example, auto-looping with capabilities to branch out of a nested auto-loop upon detection of a specified condition, background DMA facilities, the ability to link a chain of p-events together for debug purposes, and others are all important capabilities which are readily obtained.

Specifying different type generalized event and action pair in a processor

A variety of advantageous mechanisms for improved data transfer control within a data processing system are described. A DMA controller is described which is implemented as a multiprocessing transfer engine supporting multiple transfer controllers which may work independently or in cooperation to carry out data transfers, with each transfer controller acting as an autonomous processor, fetching and dispatching DMA instructions to multiple execution units. In particular, mechanisms for initiating and controlling the sequence of data transfers are provided, as are processes for autonomously fetching DMA instructions which are decoded sequentially but executed in parallel. Dual transfer execution units within each transfer controller, together with independent transfer counters, are employed to allow decoupling of source and destination address generation and to allow multiple transfer instructions in one transfer execution unit to operate in parallel with a single transfer instruction in the other transfer unit. Improved flow control of data between a source and destination is provided through the use of special semaphore operations, signals and message synchronization which may be invoked explicitly using SIGNAL and WAIT type instructions or implicitly through the use of special &ldquo;event-action&rdquo; registers. Transfer controllers are also described which can cooperate to perform &ldquo;DMA-to-DMA&rdquo; transfers. Message-level synchronization can be used by transfer controllers to synchronize with each other.

Growing community of inventors

Cary, NC, United States of America

Edwin Frank Barry