Intel Corporation

Krishnakumar Nair

Andrew Yang

Brian S Morris

Michael Rotzin

Dennis Ryan Bradford

Brian J Hickmann

Tony Lee Werner

Bharat Daga

Nitin N Garegrat

Pradeep Janedula

Bijoy Pazhanimala

Aravind Babu Srinivasan

Ambili Vengallur

Maciej Urbanski

Tom Schebye

Methods and apparatuses relating to performing vector multiplication are described. Hardware accelerators to perform vector multiplication are also described. In one embodiment, a combined fixed-point and floating-point vector multiplication circuit includes at least one switch to change the circuit between a first mode and a second mode, where in the first mode, each multiplier of a set of multipliers is to multiply mantissas from a same element position of a first floating-point vector and a second floating-point vector to produce a corresponding product, shift the corresponding products with a set of shift registers based on a maximum exponent of exponents for the corresponding products determined by a maximum exponent determiner to produce shifted products, perform an numeric conversion operation on the shifted products with a set of numeric conversion circuits based on sign bits from the same element position of the first floating-point vector and the second floating-point vector to produce signed representations of the shifted products, add the signed representations of the shifted products with a set of adders to produce a single product, and normalize the single product with a normalization circuit based on the maximum exponent into a single floating-point resultant, and in the second mode, each multiplier of the set of multipliers is to multiply values from a same element position of a first integer vector and a second integer vector to produce a corresponding product, and add each corresponding product with the set of adders to produce a single integer resultant.

Apparatuses and methods to accelerate matrix multiplication

Thus, the present disclosure is directed to systems and methods for training neural networks using a tensor that includes a plurality of FP16 values and a plurality of bits that define an exponent shared by some or all of the FP16 values included in the tensor. The FP16 values may include IEEE 754 format 16-bit floating point values and the tensor may include a plurality of bits defining the shared exponent. The tensor may include a shared exponent and FP16 values that include a variable bit-length mantissa and a variable bit-length exponent that may be dynamically set by processor circuitry. The tensor may include a shared exponent and FP16 values that include a variable bit-length mantissa; a variable bit-length exponent that may be dynamically set by processor circuitry; and a shared exponent switch set by the processor circuitry to selectively combine the FP16 value exponent with the shared exponent.

Artificial neural network training using flexible floating point tensors

An apparatus to facilitate execution of non-linear functions operations is disclosed. The apparatus comprises accelerator circuitry including a compute grid having a plurality of processing elements to execute neural network computations, store values resulting from the neural network computations, and perform piecewise linear (PWL) approximations of one or more non-linear functions using the stored values as input data.

Mechanism to perform non-linear functions in a machine learning accelerator

An apparatus and method for a converting tensor data. For example, one embodiment of a method comprises: fetching source tensor blocks of a source tensor data structure, each source tensor block comprising a plurality of source tensor data elements having a first numeric representation, wherein the source tensor data structure comprises a predefined structural arrangement of source tensor blocks; converting the one or more source tensor blocks into one or more destination tensor blocks comprising a plurality of destination tensor data elements having a second numeric representation different from the first numeric representation, wherein the sets of one or more source tensor blocks are converted to one or more corresponding destination tensor blocks in a specified order based on the first and second numeric representations; and storing each individual destination tensor block in a designated memory region to maintain coherency with the predefined structural arrangement of the source tensor blocks.

Apparatus and method for coherent, accelerated conversion between data representations

Growing community of inventors

San Jose, CA, United States of America

Krishnakumar Nair