Abstract: The present disclosure relates to a device for executing a convolutional neural network operation. The device comprises a first memory, a processing array comprising a plurality of processing strings, and a controller. The controller can be configured to fetch one or more batches of data into the first memory, regroup the one or more batches of data into multiple work items, wherein a first work item partially overlaps one or more work items among the multiple work items, and broadcast the multiple work items to the processing array, wherein the first work item is transferred to two or more processing strings of the processing array.

Abstract: The present disclosure relates to a device for executing a convolutional neural network operation. The device comprises a first memory, a processing array comprising a plurality of processing strings, and a controller. The controller can be configured to fetch one or more batches of data into the first memory, regroup the one or more batches of data into multiple work items, wherein a first work item partially overlaps one or more work items among the multiple work items, and broadcast the multiple work items to the processing array, wherein the first work item is transferred to two or more processing strings of the processing array.

Abstract: A programmable graphics processing unit (GPU) includes a first shader stage configured to receive slice data from a frame buffer and perform variable length decoding (VLD), wherein the first shader stage outputs data to a first buffer within the frame buffer; a second shader stage configured to receive the output data from the first shader stage and perform transformation and motion compensation on the slice data, wherein the second shader stage outputs decoded slice data to a second buffer within the frame buffer; a third shader stage configured to receive the decoded slice data and perform in-loop deblocking filtering (IDF) on the frame buffer; a fourth shader stage configured to perform post-processing on the frame buffer; and a scheduler configured to schedule execution of the shader stages, the scheduler comprising a plurality of counter registers; wherein execution of the shader stages is synchronized utilizing the counter registers.

Abstract: A multi-shader system in a programmable graphics processing unit (GPU) for processing video data, includes a first shader stage configured to receive slice data from a frame buffer and perform variable length decoding (VLD), wherein the first shader stage outputs data to a first buffer within the frame buffer; a second shader stage configured to receive the output data from the first shader stage and perform transformation and motion compensation on the slice data, wherein the second shader stage outputs decoded slice data to a second buffer within the frame buffer; a third shader stage configured to receive the decoded slice data and perform in-loop deblocking filtering (IDF) on the frame buffer; a fourth shader stage configured to perform post-processing on the frame buffer; and a scheduler configured to schedule execution of the shader stages, the scheduler comprising a plurality of counter registers; wherein execution of the shader stages is synchronized utilizing the counter registers.

Abstract: Methods and systems consistent with this invention comprise a control circuit for reading data from a memory comprising a plurality of data channels. Such control circuits comprise at least one multiplexer, wherein the at least one multiplexer is configured to route a data strobe signal to a first number of the plurality of data channels for reading the data from the memory when the at least one multiplexer is in a first selected state, and wherein the at least one multiplexer is configured to route the data strobe signal to a second number of the plurality of data channels, wherein the second number is greater than the first number, for reading the data from the memory when the at least one multiplexer is in a second selected state. Such methods and systems may also comprise a clock for generating a data strobe signal, and a flip-flop for latching the data from the memory into the control circuit with the data strobe signal, wherein the data strobe signal does not leave the control circuit.

Abstract: Methods and systems consistent with this invention comprise a control circuit for reading data from a memory comprising a plurality of data channels. Such control circuits comprise at least one multiplexer, wherein the at least one multiplexer is configured to route a data strobe signal to a first number of the plurality of data channels for reading the data from the memory when the at least one multiplexer is in a first selected state, and wherein the at least one multiplexer is configured to route the data strobe signal to a second number of the plurality of data channels, wherein the second number is greater than the first number, for reading the data from the memory when the at least one multiplexer is in a second selected state. Such methods and systems may also comprise a clock for generating a data strobe signal, and a flip-flop for latching the data from the memory into the control circuit with the data strobe signal, wherein the data strobe signal does not leave the control circuit.

Abstract: Methods and systems consistent with this invention comprise a control circuit for reading data from a memory comprising a plurality of data channels. Such control circuits comprise at least one multiplexer, wherein the at least one multiplexer is configured to route a data strobe signal to a first number of the plurality of data channels for reading the data from the memory when the at least one multiplexer is in a first selected state, and wherein the at least one multiplexer is configured to route the data strobe signal to a second number of the plurality of data channels, wherein the second number is greater than the first number, for reading the data from the memory when the at least one multiplexer is in a second selected state. Such methods and systems may also comprise a clock for generating a data strobe signal, and a flip-flop for latching the data from the memory into the control circuit with the data strobe signal, wherein the data strobe signal does not leave the control circuit.

Abstract: Methods and systems consistent with this invention comprise a control circuit for reading data from a memory comprising a plurality of data channels. Such control circuits comprise at least one multiplexer, wherein the at least one multiplexer is configured to route a data strobe signal to a first number of the plurality of data channels for reading the data from the memory when the at least one multiplexer is in a first selected state, and wherein the at least one multiplexer is configured to route the data strobe signal to a second number of the plurality of data channels, wherein the second number is greater than the first number, for reading the data from the memory when the at least one multiplexer is in a second selected state. Such methods and systems may also comprise a clock for generating a data strobe signal, and a flip-flop for latching the data from the memory into the control circuit with the data strobe signal, wherein the data strobe signal does not leave the control circuit.