Back to results
US 2019/0197718 A1Published application
Method to compute sliding window block sum using instruction based selective horizontal addition in vector processor
AssigneeTexas Instruments Incorporated
InventorsJayasree Sankaranarayanan, Dipan Kumar Mandal
Life of this patent
Estimating the term…
Abstract
Disclosed techniques relate to forming a block sum of picture elements employing a vector dot product instruction to sum packed picture elements and the mask producing a vector of masked horizontal picture element. The block sum is formed from plural horizontal sums via vector single instruction multiple data (SIMD) addition.
Claims20
A method for computing a sliding window block sum for a matrix of picture elements having a size M×N, the method comprising:
- using a processor to:
- for each picture element in the matrix:
- determine a block sum for a window having a size m×n around the picture element, wherein M>m and N>n, and wherein determining the block sum comprises:
- for each row of the window, performing a vector dot product operation based on a vector of packed picture elements corresponding to the row and a mask that is determined based at least partially on a horizontal size of the window to determine a respective vector of masked horizontal element sums;
- and
- after determining the respective vector of masked horizontal element sums for each row of the window, performing a vector single instruction multiple data (SIMD) addition operation to sum the respective vectors of masked horizontal element sums for each row of the window to determine the block sum of the window that corresponds to the picture element;
- and
- replace each picture element in the matrix with the block sum of the window corresponding to the picture element to obtain the sliding window block sum of the matrix.
5 more elements
An electronic device comprising:
- a processor;
- and
- a memory to store image data, the image data including a matrix of picture elements having a size M×N, and instructions for determining a sliding window block sum of the matrix that, when executed by the processor, causes the processor to:
- for each picture element in the matrix:
- determine a block sum for a window having a size m×n around the picture element, wherein M>m and N>n, and wherein determining the block sum comprises:
- for each row of the window, performing a vector dot product operation based on a vector of packed picture elements corresponding to the row and a mask that is determined based at least partially on a horizontal size of the window to determine a respective vector of masked horizontal element sums;
- and
- after determining the respective vector of masked horizontal element sums for each row of the window, performing a vector single instruction multiple data (SIMD) addition operation to sum the respective vectors of masked horizontal element sums for each row of the window to determine the block sum of the window that corresponds to the picture element;
- and
- replace each picture element in the matrix with the block sum of the window corresponding to the picture element to obtain the sliding window block sum of the matrix.
7 more elements
Dependencies are inferred from the claim text.