Abstract: A computing machine capable of performing multiple operations using a universal computing unit is provided. The universal computing unit maps an input signal to an output signal. The mapping is initiated using an instruction that includes the input signal, a weight matrix, and an activation function. Using the instruction, the universal computing unit may perform multiple operations using the same hardware configuration. The computation that is performed by the universal computing unit is determined by the weight matrix and activation function used. Accordingly, the universal computing unit does not require any programming to perform a type of computing operation because the type of operation is determined by the parameters of the instruction, specifically, the weight matrix and the activation function.

Abstract: In accordance with an embodiment of the present invention, a Viterbi decoder is described that operates on convolutional error correcting codes. The decoder allows for a pipelined architecture and a unique partitioning of survivor memory to maintain data integrity. Throughput rate is improved and stalling minimized by accessing memory words using a look-ahead function to fill the pipeline.

Abstract: The present invention provides an adaptive computing engine (ACE) that includes processing nodes having different capabilities such as arithmetic nodes, bit-manipulation nodes, finite state machine nodes, input/output nodes and a programmable scalar node (PSN). In accordance with one embodiment of the present invention, a common architecture is adaptable to function in either a kernel node, or k-node, or as general purpose RISC node. The k-node acts as a system controller responsible for adapting other nodes to perform selected functions. As a RISC node, the PSN is configured to perform computationally intensive applications such as signal processing. The present invention further provides an interconnection scheme so that a plurality of ACE devices operates under the control of a single k-node.

Abstract: The present invention concerns a new category of integrated circuitry and a new methodology for adaptive or reconfigurable computing. The preferred IC embodiment includes a plurality of heterogeneous computational elements coupled to an interconnection network. The plurality of heterogeneous computational elements include corresponding computational elements having fixed and differing architectures, such as fixed architectures for different functions such as memory, addition, multiplication, complex multiplication, subtraction, configuration, reconfiguration, control, input, output, and field programmability. In response to configuration information, the interconnection network is operative in real-time to configure and reconfigure the plurality of heterogeneous computational elements for a plurality of different functional modes, including linear algorithmic operations, non-linear algorithmic operations, finite state machine operations, memory operations, and bit-level manipulations.

Abstract: The present invention provides a system, method and software for programming and configuring an adaptive computing architecture or device. The invention utilizes program constructs which correspond to and map directly to the adaptive hardware having a plurality of reconfigurable nodes coupled through a reconfigurable matrix interconnection network. A first program construct corresponds to a selected node. A second program construct corresponds to an executable task of the selected node and includes one or more firing conditions capable of determining the commencement of the executable task of the selected node. A third program construct corresponds to at least one input port coupling the selected node to the matrix interconnect network for input data to be consumed by the executable task.

Abstract: The present invention provides a system and method for implementation and use of a shared memory. The shared memory may be accessed both independently and asynchronously by one or more processes at corresponding nodes, allowing data to be streamed to multiple processes and nodes without regard to synchronization of the plurality of processes. The various nodes may be adaptive computing nodes, kernel or controller nodes, or one or more host processor nodes. The present invention maintains memory integrity, not allowing memory overruns, underruns, or deadlocks. The present invention also provides for “push back” after a memory read, for applications in which it is desirable to “unread” some elements previously read from the memory.

Abstract: The present invention concerns data flow control in adaptive integrated circuitry which utilizes a data flow model for data processing. The present invention controls task initiation and execution based upon data consumption measured in data buffer units. In the various embodiments, when a first task of a plurality of tasks is initiated, buffer parameter is determined and a buffer count is initialized for the first task. For each iteration of the first task using a data buffer unit of input data, the buffer count is correspondingly adjusted, such as incremented or decremented. When the buffer count meets the buffer parameter requirements, the state of the first task is changed, which may including stopping the first task, and a next action is determined, such as initiating a second task. The various apparatus embodiments include a hardware task manager, a node sequencer, a programmable node, and use of a monitoring task within an adaptive execution unit.

Abstract: The present invention concerns internal synchronization in adaptive integrated circuitry which utilizes a data flow model for data processing. The present invention controls task initiation and execution based upon data consumption measured in data buffer units, with initiation of and transitions between tasks based on determined boundary condition within the data stream. In the various embodiments, when a data processing task is selected for synchronization, a boundary condition in a data stream is determined for commencement of the selected data processing task. From the boundary condition determination, a timing marker for the commencement of the selected data processing task is determined, relative to the data stream. The timing marker is dual-valued, providing a designated buffer unit and a designated byte or bit location within the designated buffer.

Abstract: A system for frequency-domain scaling for DCT computation. Scale factors are applied to coefficients during the final steps of composition of 2-point DCTs. The number of multiplications and required precision are reduced. Fixed values for various scale factors can be computed and stored prior to executing the DCT so that performance can be improved. The fixed values are derived by knowing the length of the time-domain sequence. Some fixed values can be derived independently of the length of the time-domain sequence. The approach of the invention can also reduce the number of multiplications to compute the transform, and allow smaller bit-width sizes by reducing the number of required high-precision calculations.

Abstract: The present invention provides a digital imaging apparatus having an optical sensor, an analog-to-digital converter, a plurality of computational elements, and an interconnection network. The optical sensor converts an object image into a detected image, which is then converted to digital image information by the analog-to-digital converter. The plurality of computational elements includes a first computational element having a first fixed architecture and a second computational element having a second, different fixed architecture. The interconnection network is capable of providing a processed digital image from the digital image information by configuring and reconfiguring the plurality of computational elements for performance of a plurality of different imaging functions. The invention may be embodied, for example, as a digital camera, a scanner, a printer, or a dry copier.

Abstract: The present invention concerns a new category of integrated circuitry and a new methodology for adaptive or reconfigurable computing. The preferred IC embodiment includes a plurality of heterogeneous computational elements coupled to an interconnection network. The plurality of heterogeneous computational elements include corresponding computational elements having fixed and differing architectures, such as fixed architectures for different functions such as memory, addition, multiplication, complex multiplication, subtraction, configuration, reconfiguration, control, input, output, and field programmability. In response to configuration information, the interconnection network is operative in real-time to configure and reconfigure the plurality of heterogeneous computational elements for a plurality of different functional modes, including linear algorithmic operations, non-linear algorithmic operations, finite state machine operations, memory operations, and bit-level manipulations.

Abstract: A reconfigurable bit-manipulation node is disclosed. The node includes an execution unit configured to perform a number of bit-oriented functions and a control unit configured to control the execution unit to allow one of the bit-oriented functions to be performed. The execution unit includes a number of elements interconnected with one another to allow the bit-oriented functions to be performed. The elements includes a programmable butterfly unit, a number of non-programmable butterfly units, a number of data path elements, a look-up-table memory and a reorder memory. The execution unit is capable of engaging in one of a number of operating modes to perform the bit-oriented functions. The operating modes include a programmable mode and a number of fixed operating modes including Viterbi decoding, turbo decoding and variable length encoding and decoding. The data path elements include a programmable shifter and a programmable combiner.

Abstract: A reconfigurable input/output controller (IOC) allows an adaptive computing engine (ACE) to communicate with external devices. The external devices can comprise a separate system on chip (SOC) or can be other devices or resources such as audio/visual output devices, memory, network or other communications, etc. The IOC allows different modes of transfer and performs necessary translation of input and output commands. In one embodiment, the IOC adheres to standard messaging and communication protocol used by other nodes in the ACE. This approach allows a uniform approach to the ACE design and provides advantages in scalability and adaptability of the ACE system. One feature of the invention provides a physical link adapter for accommodating different external communication types such as, RS231, optical, Firewire, universal synchronous bus (USB), etc.

Abstract: A memory controller to provide memory access services in an adaptive computing engine is provided. The controller comprises: a network interface configured to receive a memory request from a programmable network; and a memory interface configured to access a memory to fulfill the memory request from the programmable network, wherein the memory interface receives and provides data for the memory request to the network interface, the network interface configured to send data to and receive data from the programmable network.

Abstract: Input pipeline registers are provided at inputs to functional units and data paths in a adaptive computing machine. Input pipeline registers are used to hold last-accessed values and to immediately place commonly needed constant values, such as a zero or one, onto inputs and data lines. This approach can reduce the time to obtain data values and conserve power by avoiding slower and more complex memory or storage accesses. Another embodiment of the invention allows data values to be obtained earlier during pipelined execution of instructions. For example, in a three stage fetch-decode-execute type of reduced instruction set computer (RISC), a data value can be ready from a prior instruction at the decode or execute stage of a subsequent instruction.

Abstract: A system uses specialized software instructions for efficient management of freelists. In a preferred embodiment, special load and store instructions are provided. The load instruction is mapped to a register or memory location. When the load instruction is performed, hardware uses a bit-map free slot map to return an index of a free slot. Similarly, the store instruction is used to release, or free, a slot. The store instruction allows software to specify an index of a slot to be freed.

Abstract: The present invention provides an adaptive computing engine (ACE) that includes processing nodes having different capabilities such as arithmetic nodes, bit-manipulation nodes, finite state machine nodes, input/output nodes and a programmable scalar node (PSN). In accordance with one embodiment of the present invention, a common architecture is adaptable to function in either a kernel node, or k-node, or as general purpose RISC node. The k-node acts as a system controller responsible for adapting other nodes to perform selected functions. As a RISC node, the PSN is configured to perform computationally intensive applications such as signal processing.

Abstract: A system whereby a data flow language written in relatively high-level description is compiled to a hardware definition. The hardware definition is then used to configure data flow in a target processing system at execution time, or run time. In a preferred embodiment, the target processing system includes a Reduced-Instruction Set Computer (RISC) processor in communication with a finite state machine (FSM), shared memory, on-board memory, and other resources. The FSM is primarily used for accelerating matrix operations and is considered the target machine to be configured according to the dataflow definition. The RISC processor serves as a co-processor to an external central processing unit (CPU) that is a host processor for executing application code. Other embodiments can use aspects of the invention in any other processing architecture.

Abstract: The present invention includes a adaptable high-performance node (RXN) with several features that enable it to provide high performance along with adaptability. A preferred embodiment of the RXN includes a run-time configurable data path and control path. The RXN supports multi-precision arithmetic including 8, 16, 24, and 32 bit codes. Data flow can be reconfigured to minimize register accesses for different operations. For example, multiply-accumulate operations can be performed with minimal, or no, register stores by reconfiguration of the data path. Predetermined kernels can be configured during a setup phase so that the RXN can efficiently execute, e.g., discrete cosine transform (DCT), fast-Fourier transform (FFT) and other operations. Other features are provided.

Abstract: The present invention is a method, system, software and data structure for profiling programs, other code, and adaptive computing integrated circuit architectures, using a plurality of data parameters such as data type, input and output data size, data source and destination locations, data pipeline length, locality of reference, distance of data movement, speed of data movement, data access frequency, number of data load/stores, memory usage, and data persistence. The profiler of the invention accepts a data set as input, and profiles a plurality of functions by measuring a plurality of data parameters for each function, during operation of the plurality of functions with the input data set, to form a plurality of measured data parameters. From the plurality of measured data parameters, the profiler generates a plurality of data parameter comparative results corresponding to the plurality of functions and the input data set.