Abstract: A system and method of dynamically provisioning virtualized computational resources in a networked computer architecture includes at least one client device operable to run one or more client applications, at least one server device and a resource controller. Each server device comprises one or more physical processors with local memory. Each server device provides a virtual resource layer through which one or more virtual processing resources can be defined and through which the physical processors of the server device can be assigned to the virtual processing resources. In use, one or more virtual processing resources is assigned to a client application for processing of data processing workloads. The resource controller then monitors the utilization of each virtual processing resource and/or any physical processor assigned to the virtual processing resource. The resource controller can dynamically adjust which, and how many, physical processors are assigned to the virtual processing resource.

Abstract: There is provided a method for processing multiple sets of data concurrently in a statically scheduled pipelined stream processor by allowing a data set to enter the pipeline while another data set is being processed. Dedicated logic units enable independent control of each of the data sets being processed.

Abstract: A system and method of dynamically provisioning virtualised computational resources in a networked computer architecture includes at least one client device operable to run one or more client applications, at least one server device and a resource controller. Each server device comprises one or more physical processors with local memory. Each server device provides a virtual resource layer through which one or more virtual processing resources can be defined and through which the physical processors of the server device can be assigned to the virtual processing resources. In use, one or more virtual processing resources is assigned to a client application for processing of data processing workloads. The resource controller then monitors the utilisation of each virtual processing resource and/or any physical processor assigned to the virtual processing resource. The resource controller can dynamically adjust which, and how many, physical processors are assigned to the virtual processing resource.

Abstract: A networked computational architecture for provisioning of virtualized computational resources. The architecture is accessible by a client application run on a client device. The architecture includes a hardware layer having a plurality of server devices, each server device having at least one physical processor having a local memory. A resource controller is provided and operable to allocate a plurality of server devices to a client application for data processing and to assign control information to the client application. The control information specifies the required allocation of a data processing workload to each server device allocated to the client application. The architecture is configured such that client applications send the data processing workload directly to each server in accordance with the control information. Thus, a networked architecture is load balanced indirectly without requiring a load balancer to be located in the data path between the client and the server.

Abstract: A system and method of provisioning virtualized computational resources in a networked computer architecture includes a client device to run a client application, a server device, and a resource controller. The server device includes one or more processors having a local memory, and provides a virtual resource layer through which one or more virtual processing resources can be defined and through which one or more physical processors of said server device can be assigned to one or more of said virtual processing resources. The physical processors process at least a part of a data processing workload from said one or more client applications, each workload including input data having a static data part and a dynamic data part. The resource controller assigns a virtual processing resource to a plurality of client applications, where the input data for the workload of each client application has the same static data part.

Abstract: A method of data compression includes obtaining a data set comprising a sequence of data blocks comprising a predetermined number of data items, partitioning said data set into one or more groups each comprising a predetermined number of data blocks, and performing data compression on one or more groups of data blocks. Data compression is performed by associating a control data item with each of said blocks, generating a control vector comprising the control data items assigned to each of said blocks within a group, removing data blocks comprising entirely data items having said specified value, compressing data blocks comprising at least one data item having a value different from said specified value using a fixed-rate compression scheme, providing a compressed data stream comprising said compressed data blocks, and providing an associated control vector stream to enable control of said compressed data stream.

Abstract: A method of configuring a hardware design for a pipelined parallel stream processor includes obtaining a scheduled graph representing a processing operation in the time domain as a function of clock cycles. The graph includes a data path to be implemented in hardware as part of the stream processor, an input, an output, and parallel branches to enable data values to be streamed therethrough from the input to the output as a function of increasing clock cycle. The data path is partitioned into a plurality of discrete regions, each region operating on a different clock phase and having discrete control logic elements. Phase transition registers to align data separated by a boundary between regions having different clock phases are introduced into the data path at the boundary. The graph and control logic elements define a hardware design for the pipelined parallel stream processor.

Abstract: A method of generating a hardware design for a stream processor. The method includes defining a graph representing a processing operation designating processes to be implemented in hardware as part of the stream processor. The graph represents the processing operation in the time domain as a function of clock cycles and includes at least one data path. At least one stream offset object is provided located at a particular point in the data path.

Abstract: A method of generating a hardware design for a pipelined parallel stream processor, by defining a hardware processing operation; specifying at least one propagation rule; defining a graph representing the processing operation in the time domain, comprising at least one data path to be implemented as a hardware design and a plurality of parallel branches; each data path having: at least one data path input, output, and discrete object corresponding to a hardware element; each discrete object comprises an input for receiving an input variable, an operator for executing a function on said input variable, and an output variable; optimizing each output from each discrete object in dependence upon the propagation rule to produce an optimised graph; and utilizing the optimised graph to define an optimised hardware design for implementation in said pipelined parallel stream processor.

Abstract: Disclosed is a method of monitoring operation of programmable logic for a streaming processor, the method comprising: generating a graph representing the programmable logic to be implemented in hardware, the graph comprising nodes and edges connecting nodes in the graph; inserting, on each edge, monitoring hardware to monitor flow of data along the edge. Also disclosed is a method of monitoring operation of programmable logic for a streaming processor, the method comprising: generating a graph representing the programmable logic to be implemented in hardware, the graph comprising nodes and edges connecting the nodes in the graph; inserting, on at least one edge, data-generating hardware arranged to receive data from an upstream node and generate data at known values having the same flow control pattern as the received data for onward transmission to a connected node.

Abstract: Embodiments of the invention provide a method and apparatus for generating programmable logic for a hardware accelerator, the method comprising: generating a graph of nodes representing the programmable logic to be implemented in hardware; identifying nodes within the graph that affect external flow control of the programmable logic; retaining the identified nodes and removing or replacing all nodes which do not affect external flow control of the programmable logic in a modified graph; and simulating the modified graph or building a corresponding circuit of the retained nodes.

Abstract: There is provided embodiment of methods of generating a hardware design for a pipelined parallel stream processor.

Abstract: Embodiments of the invention provide a method of automatically generating a hardware stream processor design including plural processes and interconnect between the plural processes to provide data paths between the plural processes, the method comprising: providing an input designating processes to be performed by the stream processor; automatically optimizing parameters associated with the interconnect between processes within the design so as to minimise hardware requirements whilst providing the required functionality; and generating an optimized design in accordance with the optimization.

Abstract: The invention provides a method of transferring data from a data array within a main memory of a computer to an accelerator for processing, the embodiment of the method comprising: at the accelerator, requesting data from the main memory and generating a data stream between the main memory and the accelerator, the generated data stream including data from the data array; and, using an offset to determine the scheduling of array elements within the generated data stream.

Abstract: The invention provides a method of transferring data from a data array within a main memory of a computer to an accelerator for processing, the embodiment of the method comprising: at the accelerator, requesting data from the main memory and generating a data stream between the main memory and the accelerator, the generated data stream including data from the data array; and, using an offset to determine the scheduling of array elements within the generated data stream.

Abstract: There is provided a method of, and apparatus for, processing a computation on a computing device comprising at least one processor and a memory, the method comprising: storing, in said memory, plural copies of a set of data, each copy of said set of data having a different compression ratio and/or compression scheme; selecting a copy of said set of data; and performing, on a processor, a computation using said selected copy of said set of data. By providing such a method, different compression ratios and/or compression schemes can be selected as appropriate. For example, if high precision is required in a computation, a copy of the set of data can be chosen which has a low compression ratio at the expense of processing time and memory transfer time. In the alternative, if low precision is acceptable, then the speed benefits of a high compression ratio and/or lossy compression scheme may be utilised.

Abstract: There is provided a method of processing an iterative computation on a computing device comprising at least one processor. Embodiments of the method comprises performing, on a processor, an iterative calculation on data in a fixed point numerical format having a scaling factor, wherein the scaling factor is selectively variable for different steps of said calculation in order to prevent overflow and to minimise underflow. By providing such a method, the reliability, precision and flexibility of floating point operations can be achieved whilst using fixed point processing logic. The errors which fixed-point units are usually prone to generate if the range limits are exceeded can be mitigated, whilst still providing the advantage of a significantly reduced logic area to perform the calculations in fixed point.

Abstract: The invention provides a method of transferring data from a data array within a main memory of a computer to an accelerator for processing, the embodiment of the method comprising: at the accelerator, requesting data from the main memory and generating a data stream between the main memory and the accelerator, the generated data stream including data from the data array; and, using an offset to determine the scheduling of array elements within the generated data stream.