US 12,339,767 B2 · Topological scheduling

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing topological scheduling on a machine-learning accelerator having an array of tiles. One of the methods includes performing, at each time step of a plurality of time steps corresponding respectively to columns within each of a plurality of wide columns of the tile array, operations comprising: performing respective multiplications using tiles in a respective tile column for the time step, computing a respective output result for each respective tile column for the time step including computing a sum of results of the multiplications for the tile column, and storing the respective output result for the tile column in a particular output RAM having a location within the same tile column and on a row from which the output result will be read by a subsequent layer of the model.

Claims10

Dependent claims

A method performed by a device comprising a tile array comprising a plurality of tiles arranged in a plurality of tile columns, the method comprising:

obtaining an input data array, wherein the input data array comprises a plurality of pixels arranged in a plurality of data columns, each pixel comprising one or more activation values;
partitioning the input data array into one or more pixel wide columns, each pixel wide column comprising multiple data columns from the input data array;
partitioning the tile array into one or more tile wide columns, each tile wide column comprising multiple tile columns from the tile array;
assigning activation values from pixels of a first pixel wide column to a first tile wide column;
scheduling operations to be performed by the one or more tile wide columns.

2 more elements

Dependencies are inferred from the claim text.

Topological scheduling

Life of this patent

Topological scheduling

Life of this patent

Abstract

Claims10