SYSTEM AND METHOD FOR PROCESSING ARRAYS
An example computing device includes: a set of processing elements; a controller interconnected with the set of processing elements, the controller configured to: divide an input array into a plurality of strips, each strip comprising at least one primary vector; define a plurality of strip representations, each strip representation comprising a 1-dimensional array representing a respective strip; assign each strip representation to one processing element in the set; control the set of processing elements to process the respective assigned strip representation to obtain a partial result for each element in the strip representation; and aggregate the partial results to obtain a final result representing a characteristic metric for the array.
1 . A computing device comprising:
a set of processing elements;
a controller interconnected with the set of processing elements, the controller configured to:
divide an input array into a plurality of strips, each strip comprising at least one primary vector;
define a plurality of strip representations, each strip representation comprising a 1-dimensional array representing a respective strip;
assign each strip representation to one processing element in the set;
control the set of processing elements to process the respective assigned strip representation to obtain a partial result for each element in the strip representation; and
aggregate the partial results to obtain a final result representing a characteristic metric for the array.
2 . The computing device of claim 1 , wherein each strip comprises one primary vector.
3 . The computing device of claim 1 , wherein the controller is configured to define the strip representation by concatenating secondary vectors of the strip, the secondary vectors being orthogonal to the at least one primary vector.
4 . The computing device of claim 1 , wherein each processing element in the set is configured to obtain a contributing element from an adjacent processing element to compute the partial result.
5 . The computing device of claim 1 , wherein each processing element in the set is configured to buffer elements from the strip representation to compute the partial result.
6 . The computing device of claim 1 , wherein to aggregate the partial results to obtain a final result, the controller is configured to control at least one subsequent set of processing elements to further process the partial results.
7 . The computing device of claim 6 , wherein the set of processing elements and the at least one subsequent set of processing elements each comprise a layer in a neural network.
8 . The computing device of claim 6 , further comprising residual paths between the set of processing elements and the at least one subsequent set of processing elements, the residual paths configured to buffer elements from the strip representations for input to the at least one subsequent set of processing elements.
9 . The computing device of claim 1 , wherein the set of processing elements corresponds to a row of processing elements in the computing device.
10 . A method comprising:
dividing an input array into a plurality of strips, each strip comprising at least one primary vector;
defining a plurality of strip representations, each strip representation comprising a 1-dimensional array representing a respective strip;
assigning each strip representation to one processing element in a set of processing elements;
processing, by the set of processing elements, the respective assigned strip representation to obtain a partial result for each element in the strip representation; and
aggregating the partial results to obtain a final result representing a characteristic metric for the array.
11 . The method of claim 10 , wherein each strip comprises one primary vector.
12 . The method of claim 10 , wherein defining the strip representation comprises: concatenating secondary vectors of the strip, the secondary vectors being orthogonal to the at least one primary vector.
13 . The method of claim 10 , wherein each processing element in the set is configured to obtain a contributing element from an adjacent processing element to compute the partial result.
14 . The method of claim 10 , wherein each processing element in the set is configured to buffer elements from the strip representation to compute the partial result.
15 . The method of claim 10 , wherein aggregating the partial results to obtain a final result, comprises further processing the partial results by at least one subsequent set of processing elements.
16 . The method of claim 15 , wherein the set of processing elements and the at least one subsequent set of processing elements each comprise a layer in a neural network.
17 . The method of claim 15 , further comprising buffering, by residual paths between the set of processing elements and the at least one subsequent set of processing elements, elements from the strip representations for input to the at least one subsequent set of processing elements.
18 . The method of claim 10 , wherein the set of processing elements corresponds to a row of processing elements in a computing device.