Graphics compute process scheduling

• US 9,176,794 B2
• Filed: 11/04/2011
• Issued: 11/03/2015
• Est. Priority Date: 12/13/2010
• Status: Active Grant

• Current Assignee: Advanced Micro Devices, Inc., ATI Technologies ULC (Advanced Micro Devices, Inc.)
• Original Assignee: Advanced Micro Devices, Inc., ATI Technologies ULC (Advanced Micro Devices, Inc.)
• Inventors: Cheng, Jeffrey Gongxian, Blinzer, Paul, Hummel, Mark, Van Doorn, Leendert Peter
• Primary Examiner(s): Crawford, Jacinta M
• Application Number:
  US13/289,260
• Publication Number:
  US 20120147021A1
• Time in Patent Office: 1,460 Days
• Field of Search: 345501-503, 345/522, 718100-103, 718/107, 718/108
• US Class Current: 1/1
• CPC Class Codes: G06F 2209/509 Offload
• CPC Class Codes: G06F 9/545 where tasks reside in different layers, e.g. user- and kernel-space

Abstract

A method, system, and computer program product are disclosed for providing improved access to accelerated processing device compute resources to user mode applications. The functionality disclosed allows user mode applications to provide commands to an accelerated processing device without the need for kernel mode transitions in order to access a unified ring buffer. Instead, applications are each provided with their own buffers, which the accelerated processing device hardware can access to process commands. With full operating system support, user mode applications are able to utilize the accelerated processing device in much the same way as a CPU.

Claims

1. A method comprising:

allocating a work queue by a kernel mode driver to a user mode application in response to a request by the application, wherein the work queue is directly accessible by an accelerated processing device;

selecting the application for scheduling in the accelerated processing device by an accelerated processing device scheduler, wherein at least non-dependent instructions are processed autonomously from a processor scheduler; and

allocating non-pageable memory allocations upon access by the accelerated processing device to support at least page fault handling by the accelerated processing device.

2. The method of claim 1, wherein the work queue comprises a compute command ring buffer for the application.

3. The method of claim 1, further comprising:

forwarding commands to the work queue from the application.

4. The method of claim 3, further comprising:

storing the commands in a command buffer, wherein forwarding the commands to the work queue from the application comprises forwarding the commands from the command buffer to the work queue.

5. The method of claim 1, further comprising:

registering the work queue associated with an application in a compute process list, wherein the work queue is selected by the accelerated processing device scheduler for execution from the compute process list.

6. The method of claim 1, further comprising:

storing accelerated processing device context data for a prior application in the prior application'"'"'s corresponding context switch block; and

restoring accelerated processing device context data for the application from the application'"'"'s corresponding context switch block.

7. The method of claim 1, further comprising:

dispatching one or more commands from the work queue referenced by a compute process control block to the accelerated processing device based on priority information stored in the compute process control block.

8. The method of claim 1, wherein the accelerated processing device scheduler is configured to operate as a component of the kernel mode driver.

9. The method of claim 1, wherein the accelerated processing device scheduler is configured to operate as microcode on a microprocessor.

10. The method of claim 1, wherein the accelerated processing device scheduler is configured to operate as hardware logical circuitry.

11. A non-transitory computer-usable medium having instructions recorded thereon that, if executed by a computing device, cause the computing device to perform a method comprising:

allocating a work queue by a kernel mode driver to a user mode application in response to a request by the application, wherein the work queue is directly accessible by an accelerated processing device;

selecting the application for scheduling in the accelerated processing device by an accelerated processing device scheduler, wherein at least non-dependent instructions are processed autonomously from a processor scheduler; and

allocating non-pageable memory allocations upon access by the accelerated processing device to support at least page fault handling by the accelerated processing device.

12. The non-transitory computer-usable medium of claim 11, the work queue comprises a compute command ring buffer for the application.

13. The non-transitory computer-usable medium of claim 11, further comprising:

forwarding commands to the work queue from the application.

14. The non-transitory computer-usable medium of claim 13, further comprising:

storing the commands in a command buffer, wherein forwarding the commands to the work queue from the application comprises forwarding the commands from the command buffer to the work queue.

15. The non-transitory computer-usable medium of claim 11, further comprising:

registering the work queue associated with an application in a compute process list, wherein the work queue is selected by the accelerated processing device scheduler for execution from the compute process list.

16. The non-transitory computer-usable medium of claim 11, further comprising:

storing accelerated processing device context data for a prior application in the prior application'"'"'s corresponding context switch block; and

restoring accelerated processing device context data for the application from the application'"'"'s corresponding context switch block.

17. The non-transitory computer-usable medium of claim 11, further comprising:

dispatching one or more commands from the work queue referenced by a compute process control block to the accelerated processing device based on priority information stored in the compute process control block.

18. The non-transitory computer usable medium of claim 11, wherein said instructions comprise hardware description language instructions adapted to configure a manufacturing process to fabricate the computing device.

19. A system comprising:

a memory;

a processor configured to allocate an accelerated processing device work queue for an application and to allocate a work queue by a kernel mode driver to a user mode application in response to a request by the application, wherein the work queue is directly accessible by an accelerated processing device; and

an accelerated processing device scheduler configured to select the application for scheduling in the accelerated processing device, wherein at least non-dependent instructions are processed autonomously from a processor scheduler; and

,

an accelerated processing device memory manager configured to allocate non-pageable memory allocations in the memory upon access by the accelerated processing device to support at least page fault handling by the accelerated processing device.

20. The system of claim 19, wherein the work queue comprises a compute command ring buffer for the application.

21. The system of claim 19, wherein the operating system is further configured to forward commands to the work queue from the application.

22. The system of claim 21, wherein the operating system is further configured to store the commands in a command buffer, wherein forwarding the commands to the work queue from the application comprises forwarding the commands from the command buffer to the work queue.

23. The system of claim 19, wherein the operating system is further configured to register the work queue associated with an application in a compute process list, wherein work queue is selected by the accelerated processing device scheduler for execution from the compute process list.

24. The system of claim 19, wherein the operating system is further configured to store accelerated processing device context data for a prior application in the prior application'"'"'s corresponding context switch block, and restore accelerated processing device context data for the application from the application'"'"'s corresponding context switch block.

25. The system of claim 19, wherein the accelerated processing device scheduler is further configured to dispatch one or more commands from the work queue referenced by a compute process control block to the accelerated processing device based on priority information stored in the compute process control block.

26. The system of claim 19, wherein the accelerated processing device scheduler is further configured to operate as a component of the kernel mode driver.

27. The system of claim 19, wherein the accelerated processing device scheduler is further configured to operate as microcode on a microprocessor.

28. The system of claim 19, wherein the accelerated processing device scheduler is further configured to operate as hardware logical circuitry.