A method can be used to process an initial set of data through a convolutional neural network that includes a convolution layer followed by a pooling layer. The initial set is stored in an initial memory along first and second orthogonal directions. The method includes performing a first filtering of the initial set of data by the convolution layer using a first sliding window along the first direction. Each slide of the first window produces a first set of data. The method also includes performing a second filtering of the first sets of data by the pooling layer using a second sliding window along the second direction.

Techniques are disclosed for power conservation. A plurality of processing elements and a plurality of instructions are configured. The plurality of processing elements is controlled by instructions contained in a plurality of circular buffers. The plurality of processing elements can comprise a data flow processor. A first processing element, from the plurality of interconnected processing elements, is set into a sleep state by a first instruction from the plurality of instructions. The first processing element is woken from the sleep state as a result of valid data being presented to the first processing element. A subsection of the plurality of interconnected processing elements is also set into a sleep state based on the first processing element being set into a sleep state. At least one circular buffer from the plurality of circular buffers remains awake while the first processing element is in the sleep state, and the at least one circular buffer provides for data steering through a reconfigurable fabric.

Data processing apparatuses, methods of data processing, complementary instructions and programs related to ring buffer administration are disclosed. An enqueuing operation performs an atomic compare-and-swap oper-ation to store a first processed data item indication to an enqueuing-target slot in the ring buffer contingent on an in-order marker not being present there and, when successful, determines that a ready-to-dequeue condition is true for the first processed data item indication. A dequeuing operation, when the ready-to-de-queue condition for a dequeuing-target slot is true, comprises writing a null data item to the dequeuing-target slot and, when dequeuing in-order, further comprises, dependent on whether a next contiguous slot has null content, determining a retirement condition and, when the retirement condition is true, performing a retirement process on the next contiguous slot comprising making the next con-tiguous slot available to a subsequent enqueuing operation. Further subsequent slots may also be retired.

A modular system including a set of functionality tiles and a control tile storing a storage structure, an initialization file, and operation instructions. The method for operating a custom device made using the modular system includes: sending operation settings for each tile to the respective tile upon device initialization; operating each tile based on the operation settings; writing the output from each tile to the storage structure; monitoring data streams within the storage structure for a trigger event; reading data off the storage structure in response to occurrence of the trigger event; and processing the read data according to a processing function specified by the operation instructions.

Techniques are disclosed for power conservation. A plurality of processing elements and a plurality of instructions are configured. The plurality of processing elements is controlled by instructions contained in a plurality of circular buffers. The plurality of processing elements can comprise a data flow processor. A first processing element, from the plurality of interconnected processing elements, is set into a sleep state by a first instruction from the plurality of instructions. The first processing element is woken from the sleep state as a result of valid data being presented to the first processing element. A subsection of the plurality of interconnected processing elements is also set into a sleep state based on the first processing element being set into a sleep state. At least one circular buffer from the plurality of circular buffers remains awake while the first processing element is in the sleep state, and the at least one circular buffer provides for data steering through a reconfigurable fabric.

The present disclosure provides a speech data processing method and apparatus, a device and a storage medium, wherein the method comprises: a client obtains speech data, uses a speech wakeup engine to perform wakeup recognition for the obtained speech data, and stores the obtained speech data in a cyclic buffer queue; the client obtains a wakeup recognition result returned by the speech wakeup engine upon completion of the wakeup recognition, and if the wakeup recognition result is acknowledged wakeup and it is determined that there occurs a case of oneshot currently, determines an interception starting position and obtains buffer data in the cyclic buffer queue from the interception staring position to an end of the queue, the buffer data at least including the content to be recognized; the client sends the buffer data to a speech recognition engine so that the speech recognition engine performs speech recognition for the content to be recognized in the buffer data. The technical solution of the present disclosure can be applied to improve accuracy of the speech recognition result and the like.

A modular system including a set of functionality tiles and a control tile storing a storage structure, an initialization file, and operation instructions. The method for operating a custom device made using the modular system includes: sending operation settings for each tile to the respective tile upon device initialization; operating each tile based on the operation settings; writing the output from each tile to the storage structure; monitoring data streams within the storage structure for a trigger event; reading data off the storage structure in response to occurrence of the trigger event; and processing the read data according to a processing function specified by the operation instructions.

Techniques are disclosed for power conservation. A plurality of processing elements and a plurality of instructions are configured. The plurality of processing elements is controlled by instructions contained in a plurality of circular buffers. The plurality of processing elements can comprise a dataflow processor. A first processing element, from the plurality of interconnected processing elements, is set into a sleep state by a first instruction from the plurality of instructions. The first processing element is woken from the sleep state as a result of valid data being presented to the first processing element. A subsection of the plurality of interconnected processing elements is also set into a sleep state based on the first processing element being set into a sleep state. At least one circular buffer from the plurality of circular buffers remains awake while the first processing element is in the sleep state, and the at least one circular buffer provides for data steering through a reconfigurable fabric.

Techniques are described for logging communication traffic associated with one or more devices. For example, a system bus or other interface to a device may be monitored for traffic data elements. The traffic data elements may include, for example, transaction layer packets (TLPs) for communication across a PCI Express interface, or Ethernet packets for communication over a network. The traffic data elements can be processed by a classifier module and accordingly routed to one of a plurality of circular buffers. The circular buffers may maintain state (e.g., a head pointer and a tail pointer) that identify traffic data elements that are pending and those that are completed. Thus, the circular buffers can be inspected (such as after a crash) to determine recent activity.

A waveform generator includes a system control unit and signal channels controlled by the system control unit and configured to supply driving signals for driving a respective transducer of an array of transducers. Each signal channel includes a sequential access memory having rows, where each row contains an instruction word configured to generate a respective step of a waveform to be generated. A memory output of the sequential access memory is defined by an output row at a fixed location. The waveform to be generated is defined by a block of instruction words. Each signal channel also includes an internal control unit that is configured to sequentially move the content of the sequential access memory, based on the instruction word currently at the memory output, so that sequences of instruction words are provided at the output row.