US 11,341,210 B2 · Two-dimensional multi-layer convolution for deep learning

Abstract

This application relates to a multi-layer convolution operation. The multi-layer convolution operation is optimized for a vector processing unit having a number of data paths configured to operate on vector operands containing a number of elements processed in parallel by the data paths. The convolution operation specifies a convolution kernel utilized to filter a multi-channel input and generate a multi-channel output of the convolution operation. A number of threads are generated to process blocks of the multi-channel output, each block comprising a set of windows of a number of channels of the multi-channel output. Each window is a portion of the array of elements in a single layer of the multi-channel output. Each thread processes a block in accordance with an arbitrary width of the block, processing a set of instructions for each sub-block of the block having a well-defined width, the instructions optimized for the vector processing unit.

Claims20

Dependent claims

A method for performing a convolution operation, the method comprising:

receiving a multi-channel input for the convolution operation, wherein each channel of the multi-channel input comprises an array of elements;
and
executing, by a processor that includes a vector processing unit of a particular width, a number of threads to filter the multi-channel input by a convolution kernel that includes a set of coefficients, wherein:
each channel of a multi-channel output generated by the convolution operation has a width that is not equal to an integer multiple of the particular width of the vector processing unit, and
at least one thread in the number of threads is configured to perform an operation on a vector operand that includes one or more elements from two or more channels of the multi-channel input to generate elements for two or more channels of the multi-channel output.

2 more elements

A computing device configured to filter a multi-channel input signal according to a convolution kernel, the computing device comprising:

a memory storing the multi-channel input signal and a set of filter coefficients associated with the convolution kernel;
and
a vector processor that includes a vector processing unit of a particular width, the vector processor coupled to the memory and configured to:
execute a number of threads to filter the multi-channel input signal by a convolution kernel that includes a set of coefficients, wherein each channel of the multi-channel input signal comprises an array of elements and each thread in the number of threads is assigned to a block of a multi-channel output signal that includes elements from two or more channels of the multi-channel output signal,
wherein:
each channel of the multi-channel output signal has a width that is not equal to an integer multiple of the particular width of the vector processing unit, and
at least one thread in the number of threads is configured to perform an operation on a vector operand that includes one or more elements from two or more channels of the multi-channel input signal to generate elements for two or more channels of the multi-channel output signal.

4 more elements

Dependencies are inferred from the claim text.

Two-dimensional multi-layer convolution for deep learning

Life of this patent

Two-dimensional multi-layer convolution for deep learning

Life of this patent

Abstract

Claims20