# US 2023/0418782 A1 · Enabling accelerated processing units to perform dataflow execution

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

- Number: US 2023/0418782 A1
- Document ID: US20230418782A1
- Status: Published application
- Kind: A1
- Application number: 17/852,306
- Filed: 2022-06-28
- Published: 2023-12-28
- Earliest claimed: 2022-06-28
- Assignees: Advanced Micro Devices, Inc.
- Applicants: Advanced Micro Devices, Inc.
- Inventors: Johnathan Robert Alsop; Karthik Ramu Sangaiah; Anthony T. Gutierrez
- Web page: https://patentgrep.com/patent/US20230418782A1

## CPC

- G06F 15/825: {Dataflow computers}
- Y02D 10/00: Energy efficient computing, e.g. low power processors, power management or thermal management
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- Y02D: CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE

## Abstract

Methods and systems are disclosed for performing dataflow execution by an accelerated processing unit (APU). Techniques disclosed include decoding information from one or more dataflow instructions. The decoded information is associated with dataflow execution of a computational task. Techniques disclosed further include configuring, based on the decoded information, dataflow circuitry, and, then, executing the dataflow execution of the computational task using the dataflow circuitry.

## Claims

### Claim 1

A method for performing dataflow execution by an accelerated processing unit (APU), comprising:
augmenting a sequence of instructions into one or more dataflow instructions, responsive to the sequence of instructions meeting its own dependencies;
configuring dataflow circuitry based on information from one or more dataflow instructions, the information being associated with dataflow execution of a computational task; and
executing the dataflow execution of the computational task using the dataflow circuitry.

### Claim 2 (depends on 1)

The method of claim 1, wherein the configuring further comprises:
dynamically configuring the dataflow circuitry by programmable components based on one or more counters.

### Claim 3 (depends on 2)

The method of claim 2, wherein the configuring further comprises:
controlling operational cycles of the dataflow circuitry by the programmable components.

### Claim 4 (depends on 2)

The method of claim 2, wherein the configuring further comprises:
controlling connectivity among components of the dataflow circuitry by the programmable components.

### Claim 5 (depends on 1)

The method of claim 1, further comprising:
before the executing of the dataflow execution, transitioning the APU into a dataflow execution mode, during which a circuitry of the APU that is not used for the dataflow execution is placed in a power-gated or a clock-gated state.

### Claim 6 (depends on 5)

The method of claim 5, further comprising:
after the executing of the dataflow execution, transitioning the APU into a normal compute mode, activating the circuitry of the APU that was placed in a power-gated or a clock-gated state.

### Claim 7 (depends on 1)

The method of claim 1, further comprising:
determining whether the configuring of the dataflow circuitry is valid; and
responsive to a determination of a non-valid configuration, executing the computational task in a normal compute mode.

### Claim 8 (depends on 1)

The method of claim 1, further comprising:
identifying, by a compiler, that a sequence of instructions, if scheduled sequentially, meets its own set of dependencies; and
augmenting the sequence of instructions into the one or more dataflow instructions.

### Claim 9

A system for performing dataflow execution by an accelerated processing unit (APU), comprising:
instruction augmenting circuitry configured to augment a sequence of instructions into one or more dataflow instructions, responsive to the sequence of instructions meeting its own dependencies;
a setup circuit which configures dataflow circuitry based on information from one or more dataflow instructions, the information being associated with dataflow execution of a computational task; and
execution circuitry configured to execute the dataflow execution of the computational task using the dataflow circuitry.

### Claim 10 (depends on 9)

The system of claim 9, wherein the setup circuit dynamically configures the dataflow circuitry by programmable components based on one or more counters.

### Claim 11 (depends on 10)

The system of claim 10, wherein the setup circuit controls operational cycles of the dataflow circuitry by the programmable components.

### Claim 12 (depends on 10)

The system of claim 10, wherein the setup circuit controls connectivity among components of the dataflow circuitry by the programmable components.

### Claim 13 (depends on 9)

The system of claim 9, wherein before executing, by the execution circuitry, the dataflow execution, the setup circuit transitions the APU into a dataflow execution mode, during which a circuitry of the APU that is not used for the dataflow execution is placed in a power-gated or a clock-gated state.

### Claim 14 (depends on 13)

The system of claim 13, wherein after executing, by the execution circuitry, the dataflow execution, the setup circuit transitions the APU into a normal compute mode, activating the circuitry of the APU that was placed in a power-gated or a clock-gated state.

### Claim 15 (depends on 9)

The system of claim 9, wherein the setup logic further determines whether the configuring of the dataflow circuitry is valid, and responsive to a determination of a non-valid configuration, the execution circuitry executes the computational task in a normal compute mode.

### Claim 16 (depends on 9)

The system of claim 9, further comprising:
a compiler, the compiler identifies that a sequence of instructions, if scheduled sequentially, meets its own set of dependencies, and augments the sequence of instructions into the one or more dataflow instructions.

### Claim 17

A non-transitory computer-readable medium comprising instructions, when executed on an accelerated processing unit (APU), perform a method for performing dataflow execution comprising:
augmenting a sequence of instructions into one or more dataflow instructions, responsive to the sequence of instructions meeting its own dependencies;
configuring dataflow circuitry based on information from one or more dataflow instructions, the information being associated with dataflow execution of a computational task; and
executing the dataflow execution of the computational task using the dataflow circuitry.

### Claim 18 (depends on 17)

The non-transitory computer-readable medium of claim 17, wherein the configuring further comprises:
dynamically configuring the dataflow circuitry by programmable components, based on one or more counters, including:
controlling operational cycles of the dataflow circuitry by the programmable components, and
controlling connectivity among components of the dataflow circuitry by the programmable components.

### Claim 19 (depends on 17)

The non-transitory computer-readable medium of claim 17, wherein the method further comprises:
before the executing of the dataflow execution, transitioning the APU into a dataflow execution mode, during which a circuitry of the APU that is not used for the dataflow execution is placed in a power-gated or a clock-gated state.

### Claim 20 (depends on 19)

The non-transitory computer-readable medium of claim 19, wherein the method further comprises:
after the executing of the dataflow execution, transitioning the APU into a normal compute mode, activating the circuitry of the APU that was placed in a power-gated or a clock-gated state.
