COMPUTE OPTIMIZATIONS FOR LOW PRECISION MACHINE LEARNING OPERATIONS

摘要

The present disclosure provides an interconnect fabric comprising one or more switches, a memory interface coupled to the interconnect fabric, an input/output (IO) interface coupled to the interconnect fabric and an array of processing clusters coupled to the interconnect fabric. The array of multiprocessors is to process mixed-precision instructions. At least one processing cluster comprises a plurality of registers to store a plurality of packed data elements at a first precision and an execution unit to execute mixed-precision dot-product instructions. The execution unit is to perform a plurality of multiplications of different pairs of the plurality of packed data elements to generate a corresponding plurality of products and to add the corresponding plurality of products to an accumulation value stored at a second precision greater than the first precision.