Intel Corporation

Andrey Ayupov

Srikanth T Srinivasan

Jonathan D Pearce

David B Sheffield

Jeffrey J Cook

Abhijit Davare

Eriko Nurvitadhi

Deborah T Marr

Asit K Mishra

Anton Alexandrovich Sorokin

Steven Burns

Desmond A Kirkpatrick

Jaewoong Sim

Debbie Marr

An apparatus and method for performing efficient, adaptable tensor operations. For example, one embodiment of a processor comprises: front end circuitry to schedule matrix operations responsive to a matrix multiplication instruction; a plurality of lanes to perform parallel execution of the matrix operations, wherein a lane comprises an arithmetic logic unit to multiply a block of a first matrix with a block of a second matrix to generate a product and to accumulate the product with a block of a third matrix, and wherein the matrix blocks are to be stored in registers within the lane; and broadcast circuitry to broadcast one or more invariant matrix blocks to at least one of different registers within the lane and different registers across different lanes.

Apparatus and method for adaptable and efficient lane-wise tensor processing

In one embodiment, an apparatus includes: a plurality of registers; a first instruction queue to store first instructions; a second instruction queue to store second instructions; a program order queue having a plurality of portions each associated with one of the plurality of registers, each of the portions having entries to store a state of an instruction, the state comprising an encoding of a use of the register by the instruction and a source instruction queue for the instruction; and a dispatcher to dispatch for execution the first and second instructions from the first and second instruction queues based at least in part on information stored in the program order queue, to manage instruction dependencies between the first instructions and the second instructions. Other embodiments are described and claimed.

System, apparatus and method for program order queue (POQ) to manage data dependencies in processor having multiple instruction queues

An apparatus and method for efficiently processing invariant operations on a parallel execution engine. For example, one embodiment of a processor comprises: a plurality of parallel execution lanes comprising execution circuitry and registers to concurrently execute a plurality of threads; front end circuitry coupled to the plurality of parallel execution lanes, the front end circuitry to arrange the threads into parallel execution groups and schedule operations of the threads to be executed across the parallel execution lanes, wherein the front end circuitry is to dynamically evaluate one or more variables associated with the operations to determine if one or more conditionally invariant operations will be invariant across threads of a parallel execution group and/or across the parallel execution lanes; a scheduler of the front end circuitry to responsively schedule a shared thread upon a determination that a conditionally invariant operation will be invariant across threads of a parallel execution group and/or across the parallel execution lanes.

Apparatus and method for gang invariant operation optimizations using dynamic evaluation

An apparatus and method for data parallel single program multiple data (SPMD) execution. For example, one embodiment of a processor comprises: instruction fetch circuitry to fetch instructions of one or more primary threads; a decoder to decode the instructions to generate uops; a data parallel cluster (DPC) to execute microthreads comprising a subset of the uops, the DPC further comprising: a plurality of execution lanes to perform parallel execution of the microthreads; an instruction decode queue (IDQ) to store the uops prior to execution; and a scheduler to evaluate the microthreads based on associated variables including instruction pointer (IP) values, the scheduler to gang microthreads into fragments for parallel execution on the execution lanes based on the evaluation.

Architecture and method for data parallel single program multiple data (SPMD) execution

An apparatus and method for performing efficient, adaptable tensor operations. For example, one embodiment of a processor comprises: front end circuitry to schedule a plurality of matrix operations responsive to a tensor matrix multiplication instruction; a plurality of lanes to perform parallel execution of the matrix operations, each lane comprising: first, second, and third tile registers to store blocks of a first matrix (A), second matrix (B), and third matrix (C), respectively; at least one tensor arithmetic logic unit (TALU) to multiply a block of elements of the first matrix with a block of elements of the second matrix to generate a product and to accumulate the product with a block of elements of the third matrix, wherein each lane is to multiply one or more different blocks of the first and second matrix and to accumulate the resulting one or more products with one or more different blocks of the third matrix; and broadcast circuitry to broadcast one or more invariant matrix blocks to different tile registers within a lane and/or different tile registers across different lanes.

Growing community of inventors

Santa Clara, CA, United States of America

Andrey Ayupov