System, method, and computer program product for simultaneous execution of compute and graphics workloads

Patent metadata

Field	Value
Publication number	US-10217183-B2
Application number	US-201314137818-A
Country	US
Kind code	B2
Filing date	Dec 20, 2013
Priority date	Dec 20, 2013
Publication date	Feb 26, 2019
Grant date	Feb 26, 2019

Abstract

A system, method, and computer program product are provided for allocating processor resources to process compute workloads and graphics workloads substantially simultaneously. The method includes the steps of allocating a plurality of processing units to process tasks associated with a graphics pipeline, receiving a request to allocate at least one processing unit in the plurality of processing units to process tasks associated with a compute pipeline, and reallocating the at least one processing unit to process tasks associated with the compute pipeline.

First claim

What is claimed is:

1. A method comprising:
allocating a plurality of processing units included in a parallel processor to process tasks associated with a graphics pipeline, wherein the plurality of processing units includes a first processing unit and additional processing units;
receiving, at a processing cluster manager included in the parallel processor, a request to allocate the first processing unit to process tasks associated with a compute pipeline;
transmitting a first packet to a work distribution unit that indicates the first processing unit is requested for allocation to the compute pipeline;
reallocating the first processing unit to process tasks associated with the compute pipeline; and
simultaneously executing the tasks associated with the graphics pipeline and the tasks associated with the compute pipeline, wherein the tasks associated with the compute pipeline are executed by the first processing unit and the tasks associated with the graphics pipeline are executed by the additional processing units.

2. The method of claim 1 , further comprising receiving a stream of commands associated with a processor context, wherein the stream of commands includes a plurality of channels including a synchronous channel and an asynchronous channel.

3. The method of claim 2 , wherein each task includes an identifier that indicates a particular channel in the plurality of channels associated with the task.

4. The method of claim 1 , wherein each processing unit in the plurality of processing units comprises at least one streaming multiprocessor configured to execute a plurality of threads in parallel.

5. The method of claim 1 , further comprising transmitting a second packet to the work distribution unit that indicates a mode of operation for the plurality of processing units, wherein the mode of operation indicates one of a mixed operations mode or an all-compute operations mode.

6. The method of claim 5 , further comprising:
detecting that the graphics pipeline is idle; and
transmitting a third packet to the work distribution unit to transition to the all-compute operations mode.

7. The method of claim 6 , further comprising:
receiving a task associated with the graphics pipeline; and
transmitting a fourth packet to the work distribution unit to transition to the mixed operations mode.

8. The method of claim 1 , wherein the request includes a mask having a number of bits equal to a number of processing units included in the plurality of processing units, each bit in the mask indicating whether a corresponding processing unit should be allocated to the compute pipeline or the graphics pipeline.

9. The method of claim 1 , wherein the work distribution unit is configured to generate the request based on a number of scheduled tasks associated with the compute pipeline.

10. The method of claim 1 , further comprising, in response to the request, causing any pending graphics tasks to be drained from the at least one processing unit.

11. The method of claim 1 , further comprising:
executing a wait-for-idle (WFI) command that is associated with a channel and a type of task in the channel; and
receiving an indication that all tasks of the type associated with the WFI command previously launched on a corresponding pipeline have been completed.

12. The method of claim 1 , wherein the tasks are associated with two or more processor contexts.

13. A method, comprising:
allocating a plurality of processing units included in a parallel processor to process tasks associated with a graphics pipeline, wherein the plurality of processing units includes a first processing unit and additional processing units;
receiving, at a processing cluster manager included in the parallel processor, a first request to allocate the first processing unit to process tasks associated with a compute pipeline;
reallocating the first processing unit to process tasks associated with the compute pipeline,
simultaneously executing the tasks associated with the graphics pipeline and the tasks associated with the compute pipeline, wherein the tasks associated with the compute pipeline are executed by the first processing unit and the tasks associated with the graphics pipeline are executed by the additional processing units;
receiving a second request to free the first processing unit; and
in response to receiving the second request, reallocating the first processing unit to process tasks associated with the graphics pipeline.

14. A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform steps comprising
allocating a plurality of processing units included in a parallel processor to process tasks associated with a graphics pipeline, wherein the plurality of processing units includes a first processing unit and additional processing units;
receiving, at a processing cluster manager included in the parallel processor, a request to allocate the first processing unit to process tasks associated with a compute pipeline;
transmitting a packet to a work distribution unit that indicates the first processing unit is requested for allocation to the compute pipeline;
reallocating the first processing unit to process tasks associated with the compute pipeline; and
simultaneously executing the tasks associated with the graphics pipeline and the tasks associated with the compute pipeline, wherein the tasks associated with the compute pipeline are executed by the first processing unit and the tasks associated with the graphics pipeline are executed by the additional processing units.

15. A system comprising:
a parallel processor including:
a plurality of processing units, and
a processing cluster manager coupled to the plurality of processing units and configured to:
allocate the plurality of processing units to process tasks associated with a graphics pipeline, wherein the plurality of processing units includes a first processing unit and additional processing units;
receive a request to allocate the first processing unit to process tasks associated with a compute pipeline,
transmit a packet to a work distribution unit that indicates the first processing unit is requested for allocation to the compute pipeline, and
reallocate the first processing unit to process tasks associated with the compute pipeline, wherein the tasks associated with the compute pipeline are executed by the first processing unit and the tasks associated with the graphics pipeline are simultaneously executed by the additional processing units.

16. The system of claim 15 , further comprising a host unit coupled to the processing cluster manager and configured to receive a stream of commands associated with a processor context, wherein the stream of commands includes a plurality of channels including a synchronous channel and one or more asynchronous channels.

17. The system of claim 15 , further comprising a host interface unit that is configured to execute a wait-for-idle (WFI) command associated with a channel and a type of task in the channel and receive an indication that all tasks of the type associated with the WFI command previously launched on a corresponding pipeline have been completed.

18. The system of claim 15 , wherein the tasks are associated with two or more processor contexts.

19. A system, comprising:
a parallel processor including:
a plurality of processing units, and
a processing cluster manager coupled to the plurality of processing units and configured to:
allocate the plurality of pro

Assignees

• Nvidia Corp

Inventors

• Palmer Gregory S
• Duluk Jr Jerome F
• Abdalla Karim Maher
• Evans Jonathon S
• Weitkemper Adam Clark
• Shah Lacky Vasant
• Johnson Philip Browning
• Hirota Gentaro

Classifications

• G06T1/20 Primary

  Processor architectures; Processor configuration, e.g. pipelining · CPC title

• G06F9/505

  considering the load · CPC title