# US 2021/0065425 A1 · Graphics Processor with Non-Blocking Concurrent Architecture

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

- Number: US 2021/0065425 A1
- Document ID: US20210065425A1
- Status: Published application
- Kind: A1
- Application number: 17/098,089
- Filed: 2020-11-13
- Published: 2021-03-04
- Earliest claimed: 2011-06-16
- Assignees: Imagination Technologies Limited
- Applicants: Imagination Technologies Limited
- Inventors: Luke T. Peterson; James A. McCombe; Steven J. Clohset; Jason R. Redgrave
- Web page: https://patentgrep.com/patent/US20210065425A1

## CPC

- G06T 15/005: {General purpose rendering architectures}
- G06F 9/5033: {considering data affinity}
- G06F 9/505: {considering the load}
- G06F 9/52: Program synchronisation; Mutual exclusion, e.g. by means of semaphores
- G06F 15/8007: {single instruction multiple data [SIMD] multiprocessors}
- G06T 1/20: Processor architectures; Processor configuration, e.g. pipelining
- G06T 1/60: Memory management
- G06T 15/06: Ray-tracing
- G06T 2200/28: involving image processing hardware
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

In some aspects, systems and methods provide for forming groupings of a plurality of independently-specified computation workloads, such as graphics processing workloads, and in a specific example, ray tracing workloads. The workloads include a scheduling key, which is one basis on which the groupings can be formed. Workloads grouped together can all execute from the same source of instructions, on one or more different private data elements. Such workloads can recursively instantiate other workloads that reference the same private data elements. In some examples, the scheduling key can be used to identify a data element to be used by all the workloads of a grouping. Memory conflicts to private data elements are handled through scheduling of non-conflicted workloads or specific instructions and/or deferring conflicted workloads instead of locking memory locations.

## Claims

### Claim 1

A computer-implemented method for processing workloads in a computer system comprising a plurality of computation elements, the method comprising:
creating one or more fibres, wherein each of the one or more fibres comprises:
a computation routine of computer executable instructions wherein the computer executable instructions are operable to:
read and write to fibre storage memory, and
instantiate additional fibres; and
a fibre key value; and
dispatching the fibres for execution on the computation elements at an indeterminate future time,
wherein the fibre key value is used to group fibres for efficient execution of the fibres.

### Claim 2 (depends on 1)

The method of claim 1, wherein the computer executable instructions of a fibre are further operable to write data to a memory other than the fibre storage memory.

### Claim 3 (depends on 1)

The method of claim 1, wherein the computer executable instructions are operable to read data from the fibre storage memory through a Non-Uniform Memory Access (NUMA) architecture.

### Claim 4 (depends on 1)

The method of claim 1, wherein the fibre key is based on a memory location from which the fibre routine is likely to read.

### Claim 5 (depends on 1)

The method of claim 1, wherein the fibres are scheduled for execution on the computation elements in an order such that data in a cache is shared among multiple fibres.

### Claim 6 (depends on 1)

The method of claim 1, wherein fibres referencing the same fibre storage memory are precluded from executing simultaneously.

### Claim 7 (depends on 1)

The method of claim 1, wherein the fibres are executed to perform ray tracing.

### Claim 8 (depends on 1)

The method of claim 1, wherein the fibres are executed to perform database traversal.

### Claim 9 (depends on 1)

The method of claim 1, wherein the fibres are executed to perform sorting.

### Claim 10 (depends on 1)

The method of claim 1, wherein the fibres are executed to perform spatial searching.

### Claim 11 (depends on 1)

The method of claim 1, wherein the computation routines of the fibres are executed on a GPU.

### Claim 12 (depends on 1)

The method of claim 1, wherein fibres referencing the same element of fibre storage are dispatched for execution on a computation element having that element of fibre storage locally cached.

### Claim 13 (depends on 1)

The method of claim 1, wherein a given element of fibre storage, for a given fibre or a given set of fibres, is readable and writable only by that given fibre or given set of fibres.

### Claim 14 (depends on 1)

The method of claim 1, wherein said creating one or more fibres comprises using an Application Programming Interface (API) to instantiate the one or more fibres.

### Claim 15 (depends on 14)

The method of claim 14, wherein the API provides a semantic for instantiating a plurality of fibres at the same time, wherein the plurality of fibres have a common fibre key.

### Claim 16 (depends on 1)

The method of claim 1, wherein the execution of a computation routine of a parent fibre creates a child fibre, the child fibre is dispatched for execution, and the computation routine of the parent fibre continues its execution,
wherein in response to the dispatch of the child fibre, the execution of the computation routine of the child fibre is scheduled for a future time rather than the dispatch of the child fibre causing a jump execution of the computation routine of the child fibre immediately.

### Claim 17 (depends on 16)

The method of claim 16, wherein the computation routine of the child fibre does not return a value for use by the computation routine of the parent fibre.

### Claim 18

A computing system configured to process workloads, wherein the computing system comprises:
a plurality of computation elements configured to execute computation routines of fibres, wherein each of the one or more fibres comprises:
a computation routine of computer executable instructions wherein the computer executable instructions are operable to:
read and write to fibre storage memory, and
instantiate additional fibres; and
a fibre key value; and
a computing module configured to:
create one or more fibres; and
dispatch the fibres for execution on the computation elements at an indeterminate future time,
wherein the fibre key value is used to group fibres for efficient execution of the fibres.

### Claim 19 (depends on 18)

The computing system of claim 18, further comprising a host processor which is configured to execute threads, and wherein the computing module is a fibre API which is further configured to interface with the host processor to allow threads running on the host processor to instantiate fibres for execution on one or more of the computation elements.

### Claim 20

A non-transitory computer readable storage medium having stored thereon computer executable instructions, which when executed on at least one processor in a computer system comprising a plurality of computation elements, causes the computer system to:
create one or more fibres, wherein each of the one or more fibres comprises:
a computation routine of computer executable instructions wherein the computer executable instructions are operable to:
read and write to fibre storage memory, and
instantiate additional fibres; and
a fibre key value; and
dispatch the fibres for execution on the computation elements at an indeterminate future time,
wherein the fibre key value is used to group fibres for efficient execution of the fibres.
