Abstract: In a data processing system comprising a first cache operable to store data for use when performing a data processing operation, and a second cache operable to store data required for fetching data into the first cache from memory, when it is determined that there is no entry for data for a data processing operation in the first cache, an entry in the first cache is allocated for the required data, and information that indicates an entry in the second cache for data required for fetching the required data is stored in the tag portion of the allocated entry. Then, once a request has been sent to a memory system for the required data, the information in the tag portion for the allocated entry in the first cache that indicates an entry in the second cache is replaced with information indicative of an address for the data required for fetching the required data.

Abstract: In a data processor, an input value having a sign, an exponent and a significand is encoded by determining an exponent difference between a base exponent and the exponent. When the exponent difference is not less than a first threshold, only the exponent difference, or a designated value, is encoded to a payload of the output value and one or more tag bits of the output value are set to a first value. When the exponent difference is less than the first threshold, the significand and exponent difference are encoded to the payload of an output value and, optionally, the one or more tag bits of the output value. A sign bit in the output value is set corresponding to the sign of the input value, and the output value is stored.

Abstract: In a data processor, an input datum, having a sign, a tag and a payload, is decoded by first determining a format of the payload based on the tag. For a first format, an exponent difference and an output fraction are decoded from the payload. For a second format, an exponent difference is decoded from the payload and the output fraction may be assumed to be zero. The exponent difference is subtracted from a shared exponent to produce the output exponent. The decoded output may be stored in a standard format for floating-point numbers.

Abstract: A data processing apparatus is configured to determine a product of two operands stored in an Extended Block Floating-Point format. The operands are decoded, based on their tags and payloads, to generate exponent differences and at least the fractional parts of significands. The significands are multiplied to generate an output significand and shared exponents and exponent differences of the operands are combined to generate an output exponent. Signs of the operands may also be combined to provide an output sign. The apparatus may be combined with an accumulator having one or more lanes to provide an apparatus for determining dot products.

Abstract: A non-volatile memory (NVM) crossbar for an artificial neural network (ANN) accelerator is provided. The NVM crossbar includes row signal lines configured to receive input analog voltage signals, multiply-and-accumulate (MAC) column signal lines, a correction column signal line, a MAC cell disposed at each row signal line and MAC column signal line intersection, and a correction cell disposed at each row signal line and correction column signal line intersection. Each MAC cell includes one or more programmable NVM elements programmed to an ANN unipolar weight, and each correction cell includes one or more programmable NVM elements. Each MAC column signal line generates a MAC signal based on the input analog voltage signals and the respective MAC cells, and the correction column signal line generates a correction signal based on the input analog voltage signals and the correction cells. Each MAC signal is corrected based on the correction signal.

Abstract: A multiply-accumulate method and architecture are disclosed. The architecture includes a plurality of networks of non-volatile memory elements arranged in tiled columns. Logic digitally modulates the equivalent conductance of individual networks among the plurality of networks to map the equivalent conductance of each individual network to a single weight within the neural network. A first partial selection of weights within the neural network is mapped into the equivalent conductances of the networks in the columns to enable the computation of multiply-and-accumulate operations by mixed-signal computation. The logic updates the mappings to select a second partial selection of weights to compute additional multiply-and-accumulate operations and repeats the mapping and computation operations until all computations for the neural network are completed.

Abstract: A texture cache comprises at least two banks of cache storage to cache texels for processing in texture mapping operations. Access to the cached texels corresponding to a given chunk of texels of a given texture image is controlled according to a selected bank mapping selected from two or more bank mappings supported by the texture cache access control circuitry. Each bank mapping corresponds to a different mapping of the respective texels within the given chunk to the banks of cache storage. In at least one operating mode, the selected bank mapping is selected for the given chunk of texels of the given texture image depending on: at least one of first/second chunk position coordinates associated with the given chunk of texels; and at least one further texture attribute associated with the given texture image.

Abstract: Apparatuses and methods are provided, relating to the control of data processing in devices which comprise both decoupled access-execute processing circuitry and prefetch circuitry. Control of the access portion of the decoupled access-execute processing circuitry may be dependent on a performance metric of the prefetch circuitry. Alternatively or in addition, control of the prefetch circuitry may be dependent on a performance metric of the access portion.

Abstract: Circuitry comprises ray tracing circuitry comprising a plurality of floating-point circuitries to perform floating-point processing operations to detect intersection between a virtual ray defined by a ray direction and a test region, the floating-point circuitries operating to a given precision to generate an output floating-point value comprising a significand and an exponent; in which at least some of the plurality of floating-point circuitries are configured to round using a predetermined directed rounding mode any denormal floating-point value generated by operation of that circuitry so as to output normal values, a denormal floating-point value being a floating-point value in which the significand comprises one or more leading zeroes.

Abstract: Disclosed are methods, systems and devices for operation of memory device. In one aspect, volatile memory bitcells and non-volatile memory bitcells may be integrated to facilitate transfer of stored values between the volatile and non-volatile memory bitcells.

Abstract: Example methods, apparatuses, and/or articles of manufacture are disclosed that may be implemented, in whole or in part, techniques to process pixel values sampled from a multi color channel imaging device. In particular, methods and/or techniques to process pixel samples for non-visible light from pixels allocated to detection of infrared light are disclosed.

Abstract: A method of controlling an imaging system for a Head Mounted Display (HMD) device. The method comprises capturing an external scene, for example using a camera, determining an attenuation pattern, for rendering a filter area. The method also comprises determining, based on the captured external scene, a compensation pattern to for compensating at least part of the filter area, attenuating the external scene using the attenuation pattern and generating a holographic image of a virtual object, the holographic image including the compensation pattern.

Abstract: The present disclosure relates generally to multi-processor arrangements and, more particularly, to broadcast hubs for multi-processor arrangements. A processing tile may comprise a broadcast hub to obtain a plurality of parameters applicable in a particular operation from at least one of a plurality of processing tiles and initiate distribution of the plurality of parameters to the plurality of processing tiles, wherein the plurality of processing tiles may execute the particular operation based at least in part on the plurality of distributed parameters.

Abstract: Memory access circuitry enforces ownership rights for memory regions. A given memory region is associated with an owner realm specified from multiple realms, each realm corresponding to a portion of at least one software process executed by processing circuitry. A realm management unit (RMU) is provided to perform realm management operations for managing the realms. The memory access circuitry controls access to a given memory region in dependence on at least one status attribute specifying whether the given memory region is an RMU-private memory region reserved for exclusive access by the RMU.

Abstract: A method, system and apparatus provide bit-sparse neural network optimization. Rather than quantizing and pruning weight and activation elements at the word level, weight and activation elements are pruned at the bit level, which reduces the density of effective “set” bits in weight and activation data, which, advantageously, reduces the power consumption of the neural network inference process by reducing the degree of bit-level switching during inference.

Abstract: Various implementations described herein are directed to device having a memory array that operates with an applied core voltage. The device includes a power gating switch that receives a core supply voltage and provides the applied core voltage to the memory array. The device includes a biasing stage that selectively activates the power gating switch based on sensing a changing voltage level of the applied core voltage.

Abstract: An apparatus comprises an non-inclusive cache (14) configured to cache data and coherency control circuitry (16). The coherency control circuitry is configured to look up the non-inclusive cache in response to a coherent access request from a first requestor (4). In response to determining that the coherent access request can be serviced using data stored in a matching entry of the non-inclusive cache, the coherency control circuitry references snoop-filter information associated with the matching entry to determine whether the first requestor can use the data stored in the matching entry without waiting for a response to a snoop of a coherent cache (8).

Abstract: A method of operating a tile-based graphics processor that executes a graphics processing pipeline is disclosed. When there are no more primitives left to be provided for processing to the pipeline for a rendering tile, it is determined whether any remaining processing steps for the rendering tile can be omitted, e.g. because they will not affect a buffer that will be output when the rendering tile is complete. When it is determined that a processing step can be omitted, that processing step is omitted.

Abstract: Circuitry comprises driver circuitry to control display of a prevailing display image by display elements of a display device, the driver circuitry generating a signal providing electrical charge for storage by display elements, in which an electrical charge stored by a display element controls a display output of that display element; detector circuitry to detect, for a display image transition from a current display image to a second, display image, a first set of one or more display elements which are in a respective first state controlled by a first stored electrical charge in the current display image and which are required to be in a respective second state controlled by a second electrical charge, in the second display image; switching circuitry, responsive to the detector circuitry, to divert electrical charge from the set of display elements to secondary charge store in response to initiation of the display image transition.

Abstract: Various implementations described herein refer to a device having an encoder coupled to memory. The ECC encoder receives input data from memory built-in self-test circuitry, generates encoded data by encoding the input data and by adding check bits to the input data, and writes the encoded data to memory. The device may have an ECC decoder coupled to memory. The ECC decoder reads the encoded data from memory, generates corrected data by decoding the encoded data and by extracting the check bits from the encoded data, and provides the corrected data and double-bit error flag as output. The ECC decoder has error correction logic that performs error correction on the decoded data based on the check bits, wherein if the error correction logic detects a multi-bit error in the decoded data, the error correction logic corrects the multi-bit error in the decoded data to provide the corrected data.