Abstract: Systems and methods are provided to enable parallelized multiply-accumulate operations in a systolic array. Each row of the systolic array can include multiple busses enabling independent transmission of inputs along the respective bus. Each processing element can include a plurality of interconnects to receive a plurality of inputs corresponding to the multiple busses. Each processing element of a given row-oriented bus can receive an input from a prior element of the given row-oriented bus at an active bus position and perform arithmetic operations on the input. Each processing element can further receive a plurality of inputs at passive bus positions and provide the plurality of inputs to subsequent processing elements without the plurality of inputs being processed by the processing element. Use of row-oriented busses can enable parallelization to increase speed or enable increased latency at individual processing elements.

Abstract: Systems and methods are provided to enable parallelized multiply-accumulate operations in a systolic array. Each row of the systolic array can include multiple busses enabling independent transmission of inputs along the respective bus. Each processing element of a given row-oriented bus can receive an input from a prior element of the given row-oriented bus, and perform arithmetic operations on the input. Each processing element can generate an output partial sum based on the arithmetic operations, provide the input to a next processing element of the given row-oriented bus, without the input being processed by a processing element of the row located between the two processing elements that uses a different row-oriented bus. Use of row-oriented busses can enable parallelization to increase speed or enable increased latency at individual processing elements.

Abstract: Methods and hardware implementations for performing a motion search based on an estimated complexity of a block. The block is obtained from a first frame. A reference window is obtained from a second frame. For each partition type of N partition types, the block is partitioned into a plurality of partitioned blocks in accordance with the partition type. For each partitioned block of the plurality of partitioned blocks, a frequency transform of the partitioned block is computed, a complexity subvalue is computed for the partitioned block based on the frequency transform, and the complexity subvalue is accumulated into a complexity value for the partition type. M partition types are selected from the N partition types based on the N complexity values. A motion vector between the block and the reference window is determined for each partition type of the M partition types.

Abstract: Methods and hardware implementations for determining a motion vector between a first frame and a second frame. A block is obtained from the first frame. A reference window is obtained from the second frame. A set of lines are obtained from the block. The set of lines are delayed using a set of delay elements to produce a set of delayed lines. For each location of a search pattern comprising a plurality of locations arranged in a set of rows, a reference window line is obtained from the reference window, a block line is obtained from the set of delayed lines based on which row of the set of rows the location belongs to, and a similarity score of a plurality of similarity scores is accumulated based on the reference window line and the block line. A target location is identified by comparing the plurality of similarity scores.