Intel Corporation

Apple Inc.

Asaf Hargil

Doron Orenstein

Ali Sazegari

Doron Orenstien

Artiom Myaskouvskey

Evgeny Fiksman

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.

Two-dimensional multi-layer convolution for deep learning

Receiving an instruction indicating first and second operands. Each of the operands having packed data elements that correspond in respective positions. A first subset of the data elements of the first operand and a first subset of the data elements of the second operand each corresponding to a first lane. A second subset of the data elements of the first operand and a second subset of the data elements of the second operand each corresponding to a second lane. Storing result, in response to instruction, including: (1) in first lane, only lowest order data elements from first subset of first operand interleaved with corresponding lowest order data elements from first subset of second operand; and (2) in second lane, only highest order data elements from second subset of first operand interleaved with corresponding highest order data elements from second subset of second operand.

Unpacking packed data in multiple lanes

A technique to accelerate range detection in a spline calcuation. In one embodiment, an instruction and corresponding logic are provided to perform range detection within a computer or processor.

Growing community of inventors

Binyamina, Israel

Asaf Hargil