Intel Corporation

Debbie Marr

Lance E Hacking

Eriko Nurvitadhi

Srikanth T Srinivasan

Jonathan D Pearce

Jeffrey J Cook

Asit K Mishra

David B Sheffield

Abhijit Davare

Andrey Ayupov

Anton Alexandrovich Sorokin

Desmond A Kirkpatrick

Steven Burns

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

A method and apparatus for synchronizing load operations. In one embodiment, an apparatus includes a decode circuit to decode a load fence instruction. An execution unit executes the load fence instruction after it has been decoded by the decode circuit.

Method and apparatus for synchronizing load operations

A system includes a memory unit and a processor where the processor has a load buffer to store a first instruction and a cache controller to block the first instruction from dispatch into the cache controller until load data for load operations prior to the first instruction fetched from the memory unit are globally observed. The processor further may include a control register having a first mode storage to store a mode control selection for pre-serialization and a second mode storage to store a mode control selection for post-serialization to enable control of pre-serialization and post-serialization of load operations with respect to the first instruction. Other embodiments are described and claimed.

Globally observing load operations prior to fence instruction and post-serialization modes

A method and apparatus for synchronizing load operations. In one embodiment, an apparatus includes a decode circuit to decode a load fence instruction. An execution unit executes the load fence instruction after it as been decoded by the decode circuit. Furthermore, a control register is included to enable pre-serialization and post-serialization of operations appearing before and after the load fence instruction in program order, respectively.

Synchronization of load operations using load fence instruction in pre-serialization/post-serialization mode

Growing community of inventors

Portland, OR, United States of America

Debbie Marr