Intel Corporation

Intel Corporatoin

Jesus Corbal

Robert Valentine

Elmoustapha Ould-Ahmed-Vall

Mark Jay Charney

Bret L Toll

Zeev Sperber

Amit Gradstein

Alexander F Heinecke

Milind Baburao Girkar

Venkateswara Rao Madduri

Simon Rubanovich

Raanan Sade

Menachem Adelman

Dan Baum

Christopher J Hughes

Thomas D Fletcher

Dennis Ryan Bradford

Binwei Yang

Edward Thomas Grochowski

Rinat Rappoport

Yuri Gebil

Stanislav Shwartsman

Barukh Ziv

Igor Yanover

Roger Gramunt

Roman S Dubtsov

Gilbert A Neiger

Galina Ryvchin

Piotr Majcher

Jason W Brandt

Andrew Thomas Forsyth

Jonathan Cannon Hall

Carl Murray

Paul Caprioli

Kent D Glossop

Jeffrey G Wiedemeier

Rajesh M Sankaran

Eriko Nurvitadhi

Koichi Yamada

Joseph Nuzman

Mikhail Plotnikov

Deborah T Marr

Stephen R Van Doren

Lokpraveen B Mosur

Narayan Ranganathan

Jonathan D Pearce

Niall D McDonnell

Richard J Greco

Cristina S Anderson

Jeffrey J Cook

Asit K Mishra

Federico Ardanaz

Roger Espasa

Tal Uliel

Dwight P Manley

Marius A Cornea-Hasegan

David B Sheffield

Tracy Garrett Drysdale

Nicholas P Carter

Ganesh Venkatesh

Martin Guy Dixon

Stephen H Gunther

Srinivas Chennupaty

Joel S Emer

David B Papworth

Dipankar Das

Robert S Chappell

Naveen K Mellempudi

Mikhail Smelyanskiy

Rajesh S Parthasarathy

Chris J Newburn

Mark J Buxton

Peter Lachner

Thomas R Huff

Dheevatsa Mudigere

James D Allen

Buford M Guy

Mayank Bomb

Lisa K Wu

Geoff Lowney

Lihu Rappoport

Ronak Singhal

George Zacharias Chrysos

Andrey Naraikin

Suleyman Sair

Brian J Hickmann

Mostafa Hagog

Chinmay Ashok

Simon C Steely, Jr

Yen-Kuang Chen

Eric A Sprangle

Jonathan M Eastep

Aditya Kesiraju

Ramacharan Sundararaman

Benjamin C Chaffin

Sridhar Samudrala

Dror Markovich

Ramon Matas

Igor Ermolaev

Hideki Ido

Described herein is a graphics processor including a processing resource including a multiplier configured to multiply input associated with the instruction at one of a first plurality of bit widths, an adder configured to add a product output from the multiplier with an accumulator value at one of a second plurality of bit widths, and circuitry to select a first bit width of the first plurality of bit widths for the multiplier and a second bit width of the second plurality of bit widths for the adder.

Optimized compute hardware for machine learning operations

Embodiments detailed herein relate to systems and methods to store a tile register pair to memory. In one example, a processor includes: decode circuitry to decode a store matrix pair instruction having fields for an opcode and source and destination identifiers to identify source and destination matrices, respectively, each matrix having a PAIR parameter equal to TRUE; and execution circuitry to execute the decoded store matrix pair instruction to store every element of left and right tiles of the identified source matrix to corresponding element positions of left and right tiles of the identified destination matrix, respectively, wherein the executing stores a chunk of C elements of one row of the identified source matrix at a time.

Systems and methods to store a tile register pair to memory

Embodiments detailed herein relate to matrix (tile) operations. For example, decode circuitry to decode an instruction having fields for an opcode and a memory address; and execution circuitry to execute the decoded instruction to set a tile configuration for the processor to utilize tiles in matrix operations based on a description retrieved from the memory address, wherein a tile a set of 2-dimensional registers are discussed.

Systems, methods, and apparatus for tile configuration

Embodiments detailed herein relate to matrix (tile) operations. For example, decode circuitry to decode an instruction having fields for an opcode and a memory address, and execution circuitry to execute the decoded instruction to store configuration information about usage of storage for two-dimensional data structures at the memory address.

Systems, methods, and apparatuses for matrix operations

Embodiments detailed herein relate to systems and methods to load a tile register pair. In one example, a processor includes: decode circuitry to decode a load matrix pair instruction having fields for an opcode and source and destination identifiers to identify source and destination matrices, respectively, each matrix having a PAIR parameter equal to TRUE; and execution circuitry to execute the decoded load matrix pair instruction to load every element of left and right tiles of the identified destination matrix from corresponding element positions of left and right tiles of the identified source matrix, respectively, wherein the executing operates on one row of the identified destination matrix at a time, starting with the first row.

Systems and methods to load a tile register pair

An apparatus is described that includes an execution unit to execute a first instruction and a second instruction. The execution unit includes input register space to store a first data structure to be replicated when executing the first instruction and to store a second data structure to be replicated when executing the second instruction. The first and second data structures are both packed data structures. Data values of the first packed data structure are twice as large as data values of the second packed data structure. The execution unit also includes replication logic circuitry to replicate the first data structure when executing the first instruction to create a first replication data structure, and, to replicate the second data structure when executing the second data instruction to create a second replication data structure. The execution unit also includes masking logic circuitry to mask the first replication data structure at a first granularity and mask the second replication data structure at a second granularity. The second granularity is twice as fine as the first granularity.

Instruction execution that broadcasts and masks data values at different levels of granularity

Disclosed embodiments relate to computing dot products of nibbles in tile operands. In one example, a processor includes decode circuitry to decode a tile dot product instruction having fields for an opcode, a destination identifier to identify a M by N destination matrix, a first source identifier to identify a M by K first source matrix, and a second source identifier to identify a K by N second source matrix, each of the matrices containing doubleword elements, and execution circuitry to execute the decoded instruction to perform a flow K times for each element (M,N) of the identified destination matrix to generate eight products by multiplying each nibble of a doubleword element (M,K) of the identified first source matrix by a corresponding nibble of a doubleword element (K,N) of the identified second source matrix, and to accumulate and saturate the eight products with previous contents of the doubleword element (M,N).

Systems and methods for computing dot products of nibbles in two tile operands

Embodiments detailed herein relate to matrix operations. In particular, the loading of a matrix (tile) from memory. For example, support for a loading instruction is described in the form of decode circuitry to decode an instruction having fields for an opcode, a destination matrix operand identifier, and source memory information, and execution circuitry to execute the decoded instruction to load groups of strided data elements from memory into configured rows of the identified destination matrix operand to memory.

Systems, methods, and apparatuses for tile load, multiplication and accumulation

Embodiments detailed herein relate to matrix operations. In particular, matrix (tile) multiply accumulate and negated matrix (tile) multiply accumulate are discussed. For example, in some embodiments decode circuitry to decode an instruction having fields for an opcode, an identifier for a first source matrix operand, an identifier of a second source matrix operand, and an identifier for a source/destination matrix operand; and execution circuitry to execute the decoded instruction to multiply the identified first source matrix operand by the identified second source matrix operand, add a result of the multiplication to the identified source/destination matrix operand, and store a result of the addition in the identified source/destination matrix operand and zero unconfigured columns of identified source/destination matrix operand are detailed.

Systems, methods, and apparatuses for tile matrix multiplication and accumulation

Embodiments of systems, methods, and apparatuses for heterogeneous computing are described. In some embodiments, a hardware heterogeneous scheduler dispatches instructions for execution on one or more plurality of heterogeneous processing elements, the instructions corresponding to a code fragment to be processed by the one or more of the plurality of heterogeneous processing elements, wherein the instructions are native instructions to at least one of the one or more of the plurality of heterogeneous processing elements.

Systems, methods, and apparatuses for heterogeneous computing

Embodiments of an invention a processor architecture are disclosed. In an embodiment, a processor includes a decoder, an execution unit, a coherent cache, and an interconnect. The decoder is to decode an instruction to zero a cache line. The execution unit is to issue a write command to initiate a cache line sized write of zeros. The coherent cache is to receive the write command, to determine whether there is a hit in the coherent cache and whether a cache coherency protocol state of the hit cache line is a modified state or an exclusive state, to configure a cache line to indicate all zeros, and to issue the write command toward the interconnect. The interconnect is to, responsive to receipt of the write command, issue a snoop to each of a plurality of other coherent caches for which it must be determined if there is a hit.

Apparatuses and methods for a processor architecture

Embodiments of systems, apparatuses, and methods for fused multiple add. In some embodiments, a decoder decodes a single instruction having an opcode, a destination field representing a destination operand, and fields for a first, second, and third packed data source operand, wherein packed data elements of the first and second packed data source operand are of a first, different size than a second size of packed data elements of the third packed data operand. Execution circuitry then executes the decoded single instruction to perform, for each packed data element position of the destination operand, a multiplication of a M N-sized packed data elements from the first and second packed data sources that correspond to a packed data element position of the third packed data source, add of results from these multiplications to a full-sized packed data element of a packed data element position of the third packed data source, and storage of the addition result in a packed data element position destination corresponding to the packed data element position of the third packed data source, wherein M is equal to the full-sized packed data element divided by N.

Systems, apparatuses, and methods for addition of partial products

Systems, methods, and apparatuses relating to a configurable accelerator having dataflow execution circuits are described. In one embodiment, a hardware accelerator includes a plurality of dataflow execution circuits that each comprise a register file, a plurality of execution circuits, and a graph station circuit comprising a plurality of dataflow operation entries that each include a respective ready field that indicates when an input operand for a dataflow operation is available in the register file, and the graph station circuit is to select for execution a first dataflow operation entry when its input operands are available, and clear ready fields of the input operands in the first dataflow operation entry when a result of the execution is stored in the register file; a cross dependence network coupled between the plurality of dataflow execution circuits to send data between the plurality of dataflow execution circuits according to a second dataflow operation entry; and a memory execution interface coupled between the plurality of dataflow execution circuits and a cache bank to send data between the plurality of dataflow execution circuits and the cache bank according to a third dataflow operation entry.

Apparatuses, methods, and systems for a configurable accelerator having dataflow execution circuits

Embodiments of systems, apparatuses, and methods for chained fused multiply add. In some embodiments, an apparatus includes a decoder to decode a single instruction having an opcode, a destination field representing a destination operand, a first source field representing a plurality of packed data source operands of a first type that have packed data elements of a first size, a second source field representing a plurality of packed data source operands that have packed data elements of a second size, and a field for a memory location that stores a scalar value. A register file having a plurality of packed data registers includes registers for the plurality of packed data source operands that have packed data elements of a first size, the source operands that have packed data elements of a second size, and the destination operand. Execution circuitry executes the decoded single instruction to perform iterations of packed fused multiply accumulate operations by multiplying packed data elements of the sources of the first type by sub-elements of the scalar value, and adding results of these multiplications to an initial value in a first iteration and a result from a previous iteration in subsequent iterations.

Systems, apparatuses, and methods for chained fused multiply add

Growing community of inventors

King City, OR, United States of America

Jesus Corbal