# US 2023/0054059 A1 · Gateway Fabric Ports

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

- Number: US 2023/0054059 A1
- Document ID: US20230054059A1
- Status: Published application
- Kind: A1
- Application number: 17/823,237
- Filed: 2022-08-30
- Published: 2023-02-23
- Earliest claimed: 2018-07-04
- Assignees: Graphcore Limited
- Applicants: Graphcore Limited
- Inventors: Ola Torudbakken; Daniel John Pelham Wilkinson; Brian Manula
- Web page: https://patentgrep.com/patent/US20230054059A1

## CPC

- H04L 12/66: Arrangements for connecting between networks having differing types of switching systems, e.g. gateways
- G06F 9/223: {Execution means for microinstructions irrespective of the microinstruction function, e.g. decoding of microinstructions and nanoinstructions; timing of microinstructions; programmable logic arrays; delays and fan-out problems}
- G06F 9/522: {Barrier synchronisation}
- G06F 9/546: {Message passing systems or structures, e.g. queues}
- G06F 12/0813: with a network or matrix configuration
- G06F 12/0835: {for main memory peripheral accesses (e.g. I/O or DMA)}
- G06F 13/14: Handling requests for interconnection or transfer
- H04L 67/1097: for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS]
- H04W 88/16: Gateway arrangements
- H04W 92/06: between gateways and public network devices
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- H04W: WIRELESS COMMUNICATION NETWORKS (broadcast communication H04H; communication systems using wireless links for non-selective communication, e.g. wireless extensions H04M1/72)

## Abstract

A gateway for interfacing a host with a subsystem for acting as a work accelerator to the host. The gateway enables the transfer of batches of data to the subsystem at precompiled data exchange synchronisation points. The gateway acts to route data between accelerators which are connected in a scaled system of multiple gateways and accelerators using a global address space set up at compile time of an application to run on the computer system.

## Claims

### Claim 1

A computer system comprising:
a first accelerator configured to process data and to generate result data; and
a first gateway coupled to the first accelerator and a gateway interface for connection to a second gateway for conveying data for processing by a second accelerator, wherein the first accelerator and the second accelerator form a synchronisation zone in which a synchronisation barrier acts between a compute phase and an exchange phase in the synchronization zone,
wherein the first gateway is configured to receive a plurality of messages from the first accelerator, a first message of the plurality of messages comprising result data and a destination address indicating a storage location to receive the result data, the first gateway further configured to check the destination address and to route the first message to the gateway interface in response to the destination address indicating the storage location on the second accelerator.

### Claim 2 (depends on 1)

The computer system according to claim 1 wherein the plurality of messages are received in the exchange phase.

### Claim 3 (depends on 1)

The computer system according to claim 1, wherein the second accelerator is coupled to the second gateway.

### Claim 4 (depends on 1)

The computer system according to claim 1 wherein the gateway interface comprises at least one fabric port having an address range which corresponds to addresses of storage locations on the second gateway and the second accelerator.

### Claim 5 (depends on 1)

The computer system according to claim 1 wherein the computer system further comprises a third gateway coupled to the second gateway, the third gateway being coupled to a third accelerator.

### Claim 6 (depends on 5)

The computer system according to claim 5 wherein the first and second accelerators, and the first, second and third gateways form part of the synchronisation zone.

### Claim 7 (depends on 1)

The computer system according to claim 1 wherein the gateway interface comprises multiple fabric ports, each port for connection to a respective one of multiple second gateways.

### Claim 8 (depends on 1)

The computer system according to claim 1 wherein the first gateway is configured to augment or manipulate the result data prior to routing the first message containing the manipulated or augmented result data to the destination address.

### Claim 9 (depends on 8)

The computer system according to claim 8 wherein the first gateway is configured to manipulate or augment the data in the exchange phase.

### Claim 10 (depends on 1)

The computer system according to claim 1 wherein the destination address indicates a single storage location.

### Claim 11 (depends on 1)

The computer system according to claim 1 wherein the destination address comprises an implied broadcast address, and the first gateway is configured to route the first message to the gateway interface for transmission to the second accelerator and any further accelerators in the computer system.

### Claim 12 (depends on 1)

The computer system according to claim 1 wherein the first message indicates multiple destination addresses, wherein same result data is routed to each of multiple different destination address in the computer system.

### Claim 13 (depends on 1)

A computer system as claimed in claim 1, wherein the first gateway is configured to:
for each of the messages, use a destination address in the respective message to select between a plurality of interfaces for sending the respective message, the plurality of interfaces including the gateway interface and a further interface.

### Claim 14 (depends on 13)

The computer system as claimed in claim 13, wherein the first gateway is configured to:
in dependence upon the destination address in a second of the messages, route the second of the messages to the further interface.

### Claim 15 (depends on 1)

The computer system as claimed in claim 1, wherein the destination address comprises an index identifying a tile of the second accelerator.

### Claim 16

A method of exchanging data between first and second accelerators in a computer system, each accelerator configured to process data and to generate result data, wherein the first accelerator is coupled to a first gateway, wherein the first gateway has a gateway interface for connection to a second gateway for conveying data for processing by the second accelerator, wherein the first accelerator and the second accelerator form a synchronisation zone in which a synchronisation barrier acts between a compute phase and an exchange phase in the synchronisation zone, the method comprising:
receiving a plurality of messages from the first accelerator, a first message of the plurality of messages comprising result data and a destination address indicating a storage location to receive the result data; and
checking the destination address and routing the first message to the gateway interface in response to the destination address indicating the storage location on the second accelerator.

### Claim 17 (depends on 16)

The method according to claim 16 further comprising the gateway interface routing the first message to the second gateway, and the second gateway routing the first message to the second accelerator.

### Claim 18 (depends on 16)

The method according to claim 16 further comprising the first gateway routing a second message to a second gateway interface of the second gateway in response to a destination address of the second message indicating a second storage location on a third accelerator, the second gateway interface being connected to a third gateway which is connected to the third accelerator.

### Claim 19 (depends on 16)

The method according to claim 16 comprising the step of augmenting the result data prior to routing the first message containing the augmented result data to the destination address.

### Claim 20 (depends on 16)

The method according to claim 16 comprising the step of manipulating the result data at the first gateway prior to routing the first message containing the manipulated result data to the destination address.

### Claim 21

The computer program stored on transitory or non-transitory media which comprises a set of computer readable instructions which when executed cause a gateway to:
receive a plurality of messages from a first accelerator of the gateway, a first message of the plurality of messages comprising result data and a destination address indicating a storage location to receive the result data; and
check the destination address and route the first message to a gateway interface of the gateway in response to the destination address indicating the storage location on a second accelerator.
