Meta Platforms, Inc.

Facebook, Inc.

Rakesh Komuravelli

Krishnakumar Narayanan Nair

Abdulkadir Utku Diril

Yuchen Hao

Ehsan Khish Ardestani Zadeh

Thomas Mark Ulrich

Martin Schatz

Olivia Wu

Anup Ramesh Kadkol

Amin Firoozshahian

Aravind Kalaiah

Zhao Wang

Ehsan Khish Ardestani

The disclosed computer-implemented method may include (1) receiving, at a hardware accelerator that supports an ANN, an activation data set that is to undergo a convolution operation via a filter kernel of the ANN, (2) receiving, at the hardware accelerator, an argument indicating that the filter kernel exceeds at least one boundary of the activation data set when slid across a certain position during the convolution operation, (3) determining, based at least in part on the argument, that the hardware accelerator is to generate padding data at the boundary of the activation data set in connection with the certain position of the filter kernel, and then (4) performing, at the hardware accelerator, the convolution operation by processing a portion of the activation data set and the padding data when the filter kernel slides across the certain position. Various other systems and methods are also disclosed.

Systems and methods for reducing data movement during convolution operations in artificial neural networks

A computer-implemented method may include (1) maintaining (a) a filter matrix in a filter cache included in a local memory device (LMD) included in a hardware accelerator, and (b) a plurality of activation matrices corresponding to different rows of an activation volume in an activation cache included in the LMD, (2) for each activation matrix, directing a matrix multiplication unit (MMU) included in the hardware accelerator to execute a matrix multiplication operation (MMU) using the filter matrix and the activation matrix, (3) loading an additional filter matrix into the filter cache, and (4) directing the MMU to execute a plurality of additional MMOs, each additional MMO using one filter matrix included in the filter cache and one activation matrix included in the activation cache, such that the MMU reuses the filter matrix for at least one additional MMO and uses the additional filter matrix for a different additional MMO.

Systems and methods for reducing power consumption of convolution operations of artificial neural networks

A processor system comprises a plurality of processing elements. Each processing element includes a corresponding convolution processor unit configured to perform a portion of a groupwise convolution. The corresponding convolution processor unit determines multiplication results by multiplying each data element of a portion of data elements in a convolution data matrix with a corresponding data element in a corresponding groupwise convolution weight matrix. The portion of data elements in the convolution data matrix that are multiplied belong to different channels and different groups. For each specific channel of the different channels, the corresponding convolution processor unit sums together at least some of the multiplication results belonging to the same specific channel to determine a corresponding channel convolution result data element. The processing elements sum together a portion of the channel convolution result data elements from a group of different convolution processor units to determine a groupwise convolution result data element.

Grouped convolution using point-to-point connected channel convolution engines

A processor system comprises a first and second group of registers and a hardware channel convolution processor unit. The first group of registers is configured to store data elements of channels of a portion of a convolution data matrix. Each register stores at least one data element from each channel. The second group of registers is configured to store data elements of convolution weight matrices including a separate convolution weight matrix for each channel. Each register stores at least one data element from each convolution weight matrix. The hardware channel convolution processor unit is configured to multiply each data element in the first group of registers with a corresponding data element in the second group of registers and sum together the multiplication results for each specific channel to determine corresponding channel convolution result data elements in a corresponding channel convolution result matrix.

Mapping convolution to a channel convolution engine

A processor system comprises two groups of registers and a hardware channel convolution processor unit. The first group of registers is configured to store data elements of channels of a portion of a convolution data matrix. Each register stores at least one data element from each channel. The second group of registers is configured to store data elements of convolution weight matrices including a separate matrix for each channel. Each register stores at least one data element from each matrix. The hardware channel convolution processor unit is configured to multiply each data element in a first and second portion of the first group of registers with a corresponding data element in the second group of registers to determine corresponding multiplication results and sum together the multiplication results for each specific channel to determine two corresponding channel convolution result data elements in a corresponding channel convolution result matrix.

Mapping convolution to a partition channel convolution engine

A first group of elements is element-wise multiplied with a second group of elements using a plurality of multipliers belonging to a matrix multiplication hardware unit. Results of the plurality of multipliers are added together using a hierarchical tree of adders belonging to the matrix multiplication hardware unit and a final result of the hierarchical tree of adders or any of a plurality of intermediate results of the hierarchical tree of adders is selectively provided for use in determining an output result matrix. A control unit is used to instruct the matrix multiplication hardware unit to perform a plurality of different matrix multiplications in parallel by using a combined matrix that includes elements of a plurality of different operand matrices and utilize one or more selected ones of the intermediate results of the hierarchical tree of adders for use in determining the output result matrix that includes different groups of elements representing different multiplication results corresponding to different ones of the different operand matrices.

Support for different matrix multiplications by selecting adder tree intermediate results

A disclosed computer-implemented method may include maintaining, within a local memory device (LMD) included in a hardware accelerator (1) a filter matrix corresponding to a filter location included in each of a set of filters of a convolutional layer of an artificial neural network (ANN), and (2) a set of activation vectors corresponding to an active region of an activation volume input into the convolutional layer. The method may also include determining that the active region of the activation volume is contiguous with a padding region associated with at least a portion of the activation volume. The method may further include directing a matrix multiplication unit (MMU) included in the hardware accelerator to execute a matrix multiplication operation (MMO) using the filter matrix and an activation matrix that may include (1) the set of activation vectors, and (2) at least one padding vector corresponding to the padding region.

Systems and methods for handling padding regions in convolution operations

A processor system comprises a hardware channel convolution processor unit and dot product processor unit. The channel convolution processor unit is configured to perform depthwise convolution, including by multiplying each data element of a first group of data elements of a convolution data matrix with a corresponding data element of a second group of data elements of a plurality of depthwise convolution weight matrices and summing together, for each specific channel, multiplication results corresponding to the specific channel to determine one corresponding result data element in a corresponding channel convolution result matrix to calculate a portion of depthwise convolution results. The dot product processor unit is configured to perform pointwise convolution, including applying pointwise weight matrices to the portion of depthwise convolution results to determine a portion of separable convolution results while at least another portion of the depthwise convolution results is being calculated by the processor system.

Pipelined pointwise convolution using per-channel convolution operations

A system comprises a data input vector unit, a weight input vector unit, and a plurality of calculation units of a matrix processor unit. The data input vector unit is configured to concurrently receive elements of different rows of a first and second data matrix. The weight input vector unit is configured to receive a combined weight vector and at least in part concurrently provide obtained weight elements of a first and second weight matrix to a corresponding first and second group of calculation units. Each calculation unit of the first and second group of calculation units is configured to multiply elements from the data input vector unit with elements of the corresponding weight matrix from the weight input vector unit and sum together multiplication results of the corresponding calculation unit to at least in part determine a corresponding element in a first or second convolution result matrix.

High throughput matrix processor with support for concurrently processing multiple matrices

A floating-point number in a first format representation is received. Based on an identification of a floating-point format type of the floating-point number, different components of the first format representation are identified. The different components of the first format representation are placed in corresponding components of a second format representation of the floating-point number, wherein a total number of bits of the second format representation is larger than a total number of bits of the first format representation. At least one of the components of the second format representation is padded with one or more zero bits. The floating-point number in the second format representation is stored in a register. A multiplication using the second format representation of the floating-point number is performed.

Hardware for floating-point arithmetic in multiple formats

A processor system comprises one or more logic units configured to receive a processor instruction identifying a first floating point number to be multiplied with a second floating point number. The floating point numbers are each decomposed into a group of a plurality of component numbers, wherein a number of bits used to represent each floating point number is greater than a number of bits used to represent any component number in each group of the plurality of component numbers. The component numbers of the first group are multiplied with the component numbers of the second group to determine intermediate multiplication results that are summed together to determine an effective result that represents a result of multiplying the first floating point number with the second floating point number.

Growing community of inventors

Fremont, CA, United States of America

Rakesh Komuravelli