Abstract: A method and processor comprising a command processing unit to receive, from a host processor, a sequence of commands to be executed; and generate based on the sequence of commands a plurality of tasks. The processor also comprises a plurality of compute units each having a first processing module for executing tasks of a first task type, a second processing module for executing tasks of a second task type, different from the first task type, and a local cache shared by at least the first processing module and the second processing module. The command processing unit issues the plurality of tasks to at least one of the plurality of compute units, and wherein at least one of the plurality of compute units is to process at least one of the plurality of tasks.

Abstract: Disclosed is a novel neural network architecture and methods for generating neural network-based models from such architecture. A first version of the neural network, that is used for training purposes, includes one or more blocks in a first format that can then be replaced with corresponding blocks in a second format for execution. An executable model can thus be provided comprising a second version of the neural network including the one or more blocks in the second format. This then allows the training to be performed in a first, e.g. expanded format, but with a second, e.g. reduced, format model then provided for execution.

Abstract: The present disclosure advantageously provides a system and a method for convolving data in a quantized convolutional neural network (CNN). The method includes selecting a set of complex interpolation points, generating a set of complex transform matrices based, at least in part, on the set of complex interpolation points, receiving an input volume from a preceding layer of the quantized CNN, performing a complex Winograd convolution on the input volume and at least one filter, using the set of complex transform matrices, to generate an output volume, and sending the output volume to a subsequent layer of the quantized CNN.

Abstract: A method of processing input data for a given layer of a neural network using a data processing system comprising compute resources for performing convolutional computations is described. The input data comprises a given set of input feature maps, IFMs, and a given set of filters. The method comprises generating a set of part-IFMs including pluralities of part-IFMs which correspond to respective IFMs, of the given set of IFMs. The method further includes grouping part-IFMs in the set of part-IFMs into a set of selections of part-IFMs. The method further includes convolving, by respective compute resources of the data processing system, the set of selections with the given set of filters to compute a set of part-output feature maps. A data processing system for processing input data for a given layer of a neural network is also described.

Abstract: When processing graphics primitives in a graphics processing system, the render output is divided into a plurality of regions for rendering, each region comprising a respective area of the render output. It is determined for which of the plurality of regions of the render output a primitive should be rendered for. Primitive data for rendering the primitive is then stored either in a combined data structure in memory that is associated with a plurality of different regions of the render output, or is stored in a respective data structure for each region of the render output it is determined the primitive should be rendered for. Which manner the primitive data is stored is determined in dependence on a property, e.g. a coverage, of the primitive.

Abstract: When processing graphics primitives in a graphics processing system, the render output is divided into a plurality of regions for rendering, each region comprising a respective area of the render output. It is determined for which of the plurality of regions of the render output a primitive should be rendered for. Associated state data for rendering the primitive is stored in a “state data” data structure in memory. For each region of the render output it is determined the primitive should be rendered for, a reference to the associated state data for rendering the primitive is stored in a respective, different data structure for each different region of the render output it is determined the primitive should be rendered for.

Abstract: When processing graphics primitives in a graphics processing system, the render output is divided into a plurality of regions for rendering, each region comprising a respective area of the render output. It is determined for which of the plurality of regions of the render output a primitive should be rendered for. For each region of the render output it is determined a primitive should be rendered for, geometry data for the primitive is stored in memory in a respective data structure for the region in a compressed form, such that the geometry data for the primitive to be rendered is stored in a compressed form, in a respective, different data structure for each different region of the render output it is determined the primitive should be rendered for.

Abstract: When processing graphics primitives in a graphics processing system, the render output is divided into a plurality of regions (40) for rendering, each region (40) comprising a respective area of the render output; and for sets of one or more primitives to be rendered, it is determined for which of the plurality of regions of the render output (40) the primitive(s) should be rendered; and for each region of the render output (40) it is determined the primitive(s) should be rendered for, geometry data for the primitive(s) is stored in memory in a respective data structure (42) along with an indication of state data that is to be used for rendering the primitive(s) for the region, such that the geometry data for the primitive(s) to be rendered is stored in a respective, different data structure (42) for each different region of the render output (40) it is determined the primitive(s) should be rendered for.

Abstract: A display controller 93 in a data processing system includes a timewarp module (transformation circuitry) 100 that is operable to perform timewarp processing of a rendered frame 92 generated by a graphics processor (GPU) 91 for provision to a display panel 94. The timewarp module (transformation circuitry) 100 operates to transform an input surface 92 read by the display controller 93 based on received view orientation data to provide an appropriately “timewarped” transformed version of the input surface as an output transformed surface for display on the display 94.

Abstract: When processing graphics primitives in a graphics processing system, the render output is divided into a plurality of regions (40) for rendering, each region (40) comprising a respective area of the render output; and for sets of one or more primitives to be rendered, it is determined for which of the plurality of regions of the render output (40) the primitive(s) should be rendered; and for each region of the render output (40) it is determined the primitive(s) should be rendered for, geometry data for the primitive(s) is stored in memory in a respective data structure (42) along with an indication of state data that is to be used for rendering the primitive(s) for the region, such that the geometry data for the primitive(s) to be rendered is stored in a respective, different data structure (42) for each different region of the render output (40) it is determined the primitive(s) should be rendered for.

Abstract: A graphics processor that rasterises input primitives to generate graphics fragments to be processed and renders the graphics fragments to generate a first, higher resolution version of a render output. When processing of a render output is stopped before the render output is finished, the first resolution version of the render output is downsampled to a second, lower resolution and the downsampled data elements at the second resolution are written out together with a set of difference values indicative of the differences between the data elements at the first resolution and the downsampled data elements at the second resolution. Then, when processing of the render output is resumed, these values can be loaded in and used to reconstruct the array of data elements at the first resolution for use when continuing processing of the render output.

Abstract: A graphics processor that rasterises input primitives to generate graphics fragments to be processed and renders the graphics fragments to generate a first, higher resolution version of a render output. When processing of a render output is stopped before the render output is finished, the first resolution version of the render output is downsampled to a second, lower resolution and the downsampled data elements at the second resolution are written out together with a set of difference values indicative of the differences between the data elements at the first resolution and the downsampled data elements at the second resolution. Then, when processing of the render output is resumed, these values can be loaded in and used to reconstruct the array of data elements at the first resolution for use when continuing processing of the render output.

Abstract: A graphics processor that performs early depth tests for primitives in respect of patches of a render output, and depth tests for sampling positions of the render output, maintains a per patch depth buffer that stores depth values for patches for use by the patch early depth test and a per sample depth buffer. When processing of a render output is stopped before the render output is finished, the per sample depth values in the per sample depth buffer are written to storage so that those values can be restored, but the per patch depth value information in the per patch depth buffer is discarded. Then, when processing of the render output is resumed, the per sample depth buffer values are loaded into a per sample depth buffer, and the loaded per sample depth buffer values are also used to restore the per patch depth buffer.

Abstract: When processing graphics primitives in a graphics processing system, the render output is divided into a plurality of regions (40) for rendering, each region (40) comprising a respective area of the render output; and for sets of one or more primitives to be rendered, it is determined for which of the plurality of regions of the render output (40) the primitive(s) should be rendered; and for each region of the render output (40) it is determined the primitive(s) should be rendered for, geometry data for the primitive(s) is stored in memory in a respective data structure (42) along with an indication of state data that is to be used for rendering the primitive(s) for the region, such that the geometry data for the primitive(s) to be rendered is stored in a respective, different data structure (42) for each different region of the render output (40) it is determined the primitive(s) should be rendered for.

Abstract: When processing graphics primitives in a graphics processing system, the render output is divided into a plurality of regions for rendering, each region comprising a respective area of the render output. It is determined for which of the plurality of regions of the render output a primitive should be rendered for. Primitive data for rendering the primitive is then stored either in a combined data structure in memory that is associated with a plurality of different regions of the render output, or is stored in a respective data structure for each region of the render output it is determined the primitive should be rendered for. Which manner the primitive data is stored is determined in dependence on a property, e.g. a coverage, of the primitive.

Abstract: When processing graphics primitives in a graphics processing system, the render output is divided into a plurality of regions for rendering, each region comprising a respective area of the render output. It is determined for which of the plurality of regions of the render output a primitive should be rendered for. Associated state data for rendering the primitive is stored in a “state data” data structure in memory. For each region of the render output it is determined the primitive should be rendered for, a reference to the associated state data for rendering the primitive is stored in a respective, different data structure for each different region of the render output it is determined the primitive should be rendered for.

Abstract: When processing graphics primitives in a graphics processing system, the render output is divided into a plurality of regions for rendering, each region comprising a respective area of the render output. It is determined for which of the plurality of regions of the render output a primitive should be rendered for. For each region of the render output it is determined a primitive should be rendered for, geometry data for the primitive is stored in memory in a respective data structure for the region in a compressed form, such that the geometry data for the primitive to be rendered is stored in a compressed form, in a respective, different data structure for each different region of the render output it is determined the primitive should be rendered for.

Abstract: The present disclosure advantageously provides a system and a method for convolving data in a quantized convolutional neural network (CNN). The method includes selecting a set of complex interpolation points, generating a set of complex transform matrices based, at least in part, on the set of complex interpolation points, receiving an input volume from a preceding layer of the quantized CNN, performing a complex Winograd convolution on the input volume and at least one filter, using the set of complex transform matrices, to generate an output volume, and sending the output volume to a subsequent layer of the quantized CNN.

Abstract: A data processing apparatus processes a set of weight values for an artificial neural network by representing the set of weight values in the form of an array of weight values and by using an image compression scheme to provide compressed weight data for the artificial neural network. The data processing apparatus uses an image decompression scheme to derive decompressed weight values from the compressed weight data and applies the decompressed weight values when producing a result from an input to the artificial neural network. The data processing apparatus can provide for efficient storage and processing of the weight values for the artificial neural network.

Abstract: A data processing apparatus implements an artificial neural network to generate a result that indicates one or more encoding options to use when encoding a set of data elements using an encoding scheme. The data processing apparatus can provide an efficient way of selecting between possible encoding options that can be used to encode a set of data elements.