# US 2025/0004716 A1 · Systems, methods, and apparatuses for tile load

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2025/0004716 A1
- Document ID: US20250004716A1
- Status: Published application
- Kind: A1
- Application number: 18/654,951
- Filed: 2024-05-03
- Published: 2025-01-02
- Earliest claimed: 2019-08-21
- Assignees: Intel Corporation
- Applicants: Intel Corporation
- Inventors: Robert VALENTINE; Menachem ADELMAN; Milind B. GIRKAR; Zeev SPERBER; Mark J. CHARNEY; Bret L. TOLL; Rinat RAPPOPORT; Jesus Corbal; Stanislav SHWARTSMAN; Dan BAUM; Igor YANOVER; Alexander F. HEINECKE; Barukh ZIV; Elmoustapha OULD-AHMED-VALL; Yuri GEBIL; Raanan SADE
- Web page: https://patentgrep.com/patent/US20250004716A1

## CPC

- G06F 9/30036: {Instructions to perform operations on packed data, e.g. vector, tile or matrix operations}
- G06F 7/485: Adding; Subtracting {(G06F7/4833, G06F7/4836 take precedence)}
- G06F 7/4876: {Multiplying}
- G06F 7/5443: {Sum of products (for applications thereof, see the relevant places, e.g. G06F17/10, H03H17/00)}
- G06F 7/762: {having at least two separately controlled rearrangement levels, e.g. multistage interconnection networks (G06F7/764 - G06F7/768 take precedence)}
- G06F 9/3001: {Arithmetic instructions}
- G06F 9/30014: {with variable precision}
- G06F 9/30032: {Movement instructions, e.g. MOVE, SHIFT, ROTATE, SHUFFLE}
- G06F 9/30038: {using a mask}
- G06F 9/30043: {LOAD or STORE instructions; Clear instruction}
- G06F 9/30109: {having multiple operands in a single register}
- G06F 9/30112: {comprising data of variable length}
- G06F 9/30134: {Register stacks; shift registers}
- G06F 9/30145: {Instruction analysis, e.g. decoding, instruction word fields}
- G06F 9/30149: {of variable length instructions}
- G06F 9/3016: {Decoding the operand specifier, e.g. specifier format}
- G06F 9/30185: {according to one or more bits in the instruction, e.g. prefix, sub-opcode}
- G06F 9/30196: {using decoder, e.g. decoder per instruction set, adaptable or programmable decoders}
- G06F 9/3818: {Decoding for concurrent execution}
- G06F 9/3836: {Instruction issuing, e.g. dynamic instruction scheduling or out of order instruction execution}
- G06F 9/3861: {Recovery, e.g. branch miss-prediction, exception handling (error detection or correction G06F11/00)}
- G06F 12/0207: {with multidimensional access, e.g. row/column, matrix}
- G06F 17/16: Matrix or vector computation {, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization (matrix transposition G06F7/78)}
- G06F 2212/454: Vector or matrix data
- G06F 2212/455: Image or video data
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

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.

## Claims

### Claim 1

A processor comprising:
a register to store a configuration value for tiles;
decode circuitry to decode an instruction, the instruction having a field to identify a plurality of non-consecutive vector registers, a field to identify a register to store a base, and a field to identify a register to store an index, wherein the plurality of non-consecutive vector registers are to have a single-instruction, multiple-data (SIMD) dimension based on the configuration value for the tiles; and
execution circuitry coupled with the decode circuitry, the execution circuitry to perform operations corresponding to the instruction, including to:
load 64-bit data elements from memory locations generated using the base and the index in the plurality of non-consecutive vector registers.

### Claim 2 (depends on 1)

The processor of claim 1, wherein the plurality of non-consecutive vector registers is a configured number of non-consecutive vector registers.

### Claim 3 (depends on 1)

The processor of claim 1, wherein the plurality of non-consecutive vector registers is four non-consecutive vector registers.

### Claim 4 (depends on 1)

The processor of claim 1, further comprising a register to indicate whether the tiles are configured for use.

### Claim 5 (depends on 1)

The processor of claim 1, wherein the decode circuitry is to decode a second instruction, and further comprising execution circuitry to perform operations corresponding to the second instruction, including to configure the tiles for use.

### Claim 6 (depends on 1)

The processor of claim 1, wherein the instruction is to indicate a stride.

### Claim 7 (depends on 6)

The processor of claim 6, wherein the non-consecutive vector registers correspond to different rows delineated based on the stride.

### Claim 8 (depends on 1)

The processor of claim 1, wherein the processor has a reduced instruction set computing (RISC) architecture.

### Claim 9 (depends on 1)

The processor of claim 1, wherein the processor is a central processing unit (CPU), and wherein the CPU further comprises a reorder buffer and register renaming circuitry.

### Claim 10

A method comprising:
storing a configuration value for tiles in a register;
decoding an instruction, the instruction having a field identifying a plurality of non-consecutive vector registers, a field identifying a register storing a base, and a field identifying a register storing an index, wherein the plurality of non-consecutive vector registers have a single-instruction, multiple-data (SIMD) dimension based on the configuration value for the tiles; and
performing operations corresponding to the instruction, including:
generating addresses of memory locations using the base and the index; and
loading 64-bit data elements from the memory locations in the plurality of non-consecutive vector registers.

### Claim 11 (depends on 10)

The method of claim 10, further comprising:
configuring a number of the plurality of non-consecutive vector registers; and
indicating in a register whether tiles are configured for use.

### Claim 12 (depends on 10)

The method of claim 10, wherein the loading comprises loading the 64-bit data elements from the memory locations in four non-consecutive vector registers, and determining a stride indicated by the instruction.

### Claim 13 (depends on 10)

The method of claim 10, further comprising:
reordering operations with a reorder buffer; and
performing register renaming for a plurality of registers.

### Claim 14 (depends on 10)

The method of claim 10, further comprising:
decoding a second instruction; and
performing operations corresponding to the second instruction, including configuring the tiles for use.

### Claim 15

A system on a chip (SoC) comprising:
a memory controller; and
a processor coupled with the memory controller, the processor comprising:
a register to store a configuration value for tiles;
decode circuitry to decode an instruction, the instruction having a field to identify a plurality of non-consecutive vector registers, a field to identify a register to store a base, and a field to identify a register to store an index, wherein the plurality of non-consecutive vector registers are to have a single-instruction, multiple-data (SIMD) dimension based on the configuration value for the tiles; and
execution circuitry coupled with the decode circuitry, the execution circuitry to perform operations corresponding to the instruction, including to:
load 64-bit data elements from memory locations generated using the base and the index in the plurality of non-consecutive vector registers.

### Claim 16 (depends on 15)

The SoC of claim 15, wherein the SoC further comprises a bus controller unit, wherein the processor is a central processing unit (CPU), and wherein the CPU further comprises a reorder buffer and register renaming circuitry.

### Claim 17 (depends on 16)

The SoC of claim 16, wherein the plurality of non-consecutive vector registers is four non-consecutive vector registers.

### Claim 18 (depends on 17)

The SoC of claim 17, further comprising a register to indicate whether the tiles are configured for use.

### Claim 19 (depends on 17)

The SoC of claim 17, wherein the plurality of non-consecutive vector registers is a configured number of non-consecutive vector registers.

### Claim 20 (depends on 17)

The SoC of claim 17, wherein the decode circuitry is to decode a second instruction, and further comprising execution circuitry to perform operations corresponding to the second instruction, including to configure the tiles for use.

### Claim 21 (depends on 17)

The SoC of claim 17, wherein the SoC further comprises a display unit, and wherein the instruction is to indicate a stride.

### Claim 22 (depends on 17)

The SoC of claim 17, wherein the SoC further comprises direct memory access (DMA) unit, and wherein the processor has a reduced instruction set computing (RISC) architecture.

### Claim 23

A non-transitory machine-readable storage medium storing instructions that, when executed by a machine, are to cause the machine to perform operations, including to:
decode an instruction, the instruction having a field to identify a plurality of non-consecutive vector registers, a field to identify a register storing a base, and a field to identify a register storing an index, wherein the plurality of non-consecutive vector registers are to have a single-instruction, multiple-data (SIMD) dimension based on a configuration value for tiles stored in a register; and
perform operations corresponding to the instruction, including to:
generate addresses of memory locations using the base and the index; and
load 64-bit data elements from the memory locations in the plurality of non-consecutive vector registers.

### Claim 24 (depends on 23)

The non-transitory machine-readable storage medium of claim 23, wherein the operations further include to:
configure a number of the plurality of non-consecutive vector registers; and
indicate in a register whether tiles are configured for use.

### Claim 25 (depends on 23)

The non-transitory machine-readable storage medium of claim 23, wherein the operations further include to:
decode a second instruction; and
perform operations corresponding to the second instruction, including to configure the tiles for use.

### Claim 26 (depends on 23)

The non-transitory machine-readable storage medium of claim 23, wherein the operations further include to determine a stride indicated by the instruction.

### Claim 27

An apparatus comprising:
programmable configuration storage to store configuration information for a matrix, the configuration information including a first value corresponding to a number of rows for the matrix, and a second value corresponding to a number of columns for the matrix;
decode circuitry to decode a matrix load instruction, the matrix load instruction having one or more fields to specify one or more registers, the matrix load instruction having scale-index-base (SIB) addressing information, the SIB addressing information to indicate a scale, an index, a base, and a displacement, the base plus the displacement to reference an initial memory address of the matrix, the index shifted by the scale to represent a stride of the matrix; and
execution circuitry, coupled to the programmable configuration storage and coupled to the decode circuitry, the execution circuitry to perform operations corresponding to the matrix load instruction, including to load the matrix from memory, based on the initial memory address and the stride, to the one or more registers specified by the one or more fields.

### Claim 28 (depends on 27)

The apparatus of claim 27, wherein the configuration information includes a start row value corresponding to a row of the matrix at which to restart execution of the matrix load instruction.

### Claim 29 (depends on 27)

The apparatus of claim 27, further comprising a tile register including the one or more registers.

### Claim 30 (depends on 29)

The apparatus of claim 29, wherein the tile register is one of a plurality of tile registers, each of the plurality of tile registers to store one of a plurality of matrices, and the configuration information is to include a plurality of pairs of values, each of the plurality of pairs of values corresponding to a number of rows and a number of columns for one of the plurality of matrices.

### Claim 31 (depends on 27)

The apparatus of claim 27, wherein the decode circuitry is to decode a second instruction to load the configuration information from one or more memory locations referenced by one or more operands of the second instruction to the programmable configuration storage.

### Claim 32 (depends on 27)

The apparatus of claim 27, wherein the apparatus comprises a core coupled to at least one accelerator, the core including the decode circuitry and the at least one accelerator including the execution circuitry.

### Claim 33 (depends on 27)

The apparatus of claim 27. wherein the apparatus is a core comprising the decode circuitry and the execution circuitry.
