SYSTEM AND METHOD FOR COMPILER SUPPORT FOR KERNEL LAUNCHES IN DEVICE CODE

• US 20130300752A1
• Filed: 01/07/2013
• Published: 11/14/2013
• Est. Priority Date: 05/10/2012
• Status: Active Grant

• Current Assignee: NVIDIA Corporation
• Original Assignee: NVIDIA Corporation
• Inventors: Grover, Vinod, Lee, Sean, Marathe, Jaydeep
• Granted Patent: US 10,025,643 B2
• Time in Patent Office: Days
• US Class Current: 345/520
• CPC Class Codes: G06F 8/41 Compilation
• CPC Class Codes: G06F 9/4843 by program, e.g. task dispatcher, supervisor, operating system
• CPC Class Codes: G06F 9/54 Interprogram communication

Abstract

A system and method for compiling source code (e.g., with a compiler). The method includes accessing a portion of device source code and determining whether the portion of the device source code comprises a piece of work to be launched on a device from the device. The method further includes determining a plurality of application programming interface (API) calls based on the piece of work to be launched on the device and generating compiled code based on the plurality of API calls. The compiled code comprises a first portion operable to execute on a central processing unit (CPU) and a second portion operable to execute on the device (e.g., GPU).

Claims

1. A method of compiling source code, said method comprising:

accessing a portion of device source code;

determining whether said portion of said device source code comprises a piece of work to be launched on a device from said device;

determining a plurality of application programming interface (API) calls based on said piece of work to be launched on said device; and

generating compiled code based on said plurality of API calls, wherein a first portion of said compiled code is operable to execute on a central processing unit (CPU) and a second portion of said compiled code is operable to execute on said device.

2. The method as described in claim 1 wherein said device is a graphics processing unit (GPU).

3. The method as described in claim 1 further comprising:

determining whether said piece of work to be launched on said device accesses a private resource; and

generating an error if said piece of work to be launched on said device accesses a private resource.

4. The method as described in claim 3 wherein said private resource comprises memory local to a portion of work launching said piece of work to be launched.

5. The method as described in claim 3 wherein said determining whether said piece of work to be launched on said device accesses a private resource comprises using data flow analysis.

6. The method as described in claim 1 further comprising:

determining whether said piece of work to be launched on said device is a launchable function; and

generating an error if said piece of work to be launched on said device is not a launchable function.

7. The method as described in claim 1 wherein said generating complied code comprises passing arguments of said piece of work to be launched on said device with a parameter buffer.

8. A system for compiling code, said system comprising:

a code access module operable to access source code from a datastore, wherein said source code comprises a call to launch a piece of work on a device from said device;

a mapping module operable to map said call to launch said piece of work on said device from said device to an API call to launch said piece of work on said device from said device; and

a compilation module operable to compile said source code into executable code, wherein a first portion of said executable code is operable to execute on a central processing unit (CPU) and a second portion of said executable code is operable to execute on said device.

9. The system as described in claim 8 wherein said device is a GPU.

10. The system as described in claim 8 further comprising:

a resource access module operable to determine whether said piece of work to be launched on said device from said device accesses a private resource.

11. The system as described in claim 10 wherein said resource access module is operable to generate an error if said piece of work to be launched accesses a private resource.

12. The system as described in claim 8 further comprising:

a launch checking module operable to determine whether said call to launch said piece of work on said device from said device is attempting to launch a launchable function.

13. The system as described in claim 12 wherein said launch checking module is operable to generate an error if said call to launch a piece of work on said device from said device is attempting to launch a function that is not a launchable function.

14. The system as described in claim 8 wherein said executable code comprises code for passing arguments of said piece of work to be launched on said device with a parameter buffer.

15. A method for compiling code comprising:

accessing a line of source code comprising a call to launch a piece of work on a graphics processor from said graphics processor;

determining whether said call to launch said piece of work on said graphics processor from said graphics processor comprises arguments;

evaluating arguments of said call to launch said piece of work on said graphics processor from said graphics processor;

generating a first code portion for allocating a parameter buffer;

generating a second code portion for storing said arguments into said parameter buffer; and

generating compiled code for launching said piece of work on said graphics processor from said graphics processor based on said first code portion and said second code portion.

16. The method as described in claim 15 wherein said generating compiled code for launching said piece of work on said graphics processor from said graphics processor comprises using an API call.

17. The method as described in claim 15 further comprising:

determining whether said piece of work to be launched on said graphics processor from said graphics processor accesses a private resource; and

generating an error if said piece of work to be launched on said graphics processor from said graphics processor accesses a private resource.

18. The method as described in claim 17 wherein said private resource comprises memory local to said line of said source code comprising said call to launch said piece of work on a graphics processor from said graphics processor.

19. The method as described in claim 17 wherein said determining whether said piece of work to be launched on said graphics processor accesses a private resource comprises using an analysis of a store instruction.

20. The method as described in claim 15 further comprising:

determining whether said piece of work to be launched on said graphics processor from said graphics processor is a launchable function; and

generating an error if said piece of work to be launched on said graphics processor from said graphics processor is not a launchable function.