Vector width-aware synchronization-elision for vector processors

• US 8,966,461 B2
• Filed: 09/29/2011
• Issued: 02/24/2015
• Est. Priority Date: 09/29/2011
• Status: Active Grant

• Current Assignee: Advanced Micro Devices, Inc.
• Original Assignee: Advanced Micro Devices, Inc.
• Inventors: Gaster, Benedict R., Howes, Lee W., Hummel, Mark D.
• Primary Examiner(s): Wong, Don
• Assistant Examiner(s): Rivera, Anibal
• Application Number:
  US13/249,171
• Publication Number:
  US 20130086566A1
• Time in Patent Office: 1,244 Days
• US Class Current: 717/148
• CPC Class Codes: G06F 8/443 Optimisation
• CPC Class Codes: G06F 9/45525 Optimisation or modification within the same instruction set architecture, e.g. HP Dynamo

Abstract

A medium, method, and apparatus are disclosed for eliding superfluous function invocations in a vector-processing environment. A compiler receives program code comprising a width-contingent invocation of a function. The compiler creates a width-specific executable version of the program code by determining a vector width of a target computer system and omitting the function from the width-specific executable if the vector width meets one or more criteria. For example, the compiler may omit the function call if the vector width is greater than a minimum size.

Claims

1. A non-transitory computer-readable storage medium storing program instructions executable by a computer to implement a compiler configured to:

receive program code that specifies a barrier operation, wherein the barrier operation synchronizes execution of a plurality of work items, by pausing execution of one of the plurality of work items until others of the plurality of work items reach a same execution point;

create a width-specific executable version of the program code, wherein creating the width-specific executable version comprises;

determining a vector width of a target computer system, wherein the vector width is a number of work items that the target computer system is configured to execute in parallel; and

in response to the determined vector width meeting one or more criteria, omitting the barrier operation from the width-specific executable version.

2. The medium of claim 1, wherein the program code specifies that the barrier operation is to be omitted in response to the vector width of the target computer system meeting the one or more criteria.

3. The medium of claim 1, wherein the one or more criteria include a minimum vector-width for including the barrier operation in the width-specific executable.

4. The medium of claim 3, wherein the program code specifies the minimum vector width.

5. The medium of claim 1, wherein the compiler is a just-in-time compiler configured to create the width-specific executable version at runtime.

6. The medium of claim 1, wherein the compiler is configured to determine the vector width by querying a programmatic interface exposed by a runtime environment.

7. The medium of claim 1, wherein the compiler is configured to determine the vector width by receiving an indication of the width through configuration information.

8. A computer-implemented method of compiling program code, the method comprising:

analyzing program code that includes a function call for a synchronization operation of a plurality of work items, wherein the synchronization operation halts execution for ones of the plurality of work items until others of the plurality of work items reach a same point of execution;

producing a width-specific executable version of the program code, wherein producing the width-specific executable version comprises;

in response to a vector width of a target computer system satisfying a threshold indicated, eliding the function call from the width-specific executable version, wherein the vector width is a number of work items that the target computer system is configured to execute concurrently.

9. The method of claim 8, wherein the program code identifies the function call as a candidate for being elided.

10. The method of claim 8, further comprising:

compiling the program code with a just-in-time compiler such that the width-specific executable version is produced at program runtime.

11. The method of claim 8, wherein the producing comprises determining the vector width by querying a programmatic interface exposed by a runtime environment.

12. The method of claim 8, wherein the producing comprises determining the vector width by receiving an indication of the width through configuration information.

13. An apparatus, comprising:

a memory having a computer program stored therein;

a vector processor configured to elide a function call in the computer program in response to determining that the function call is superfluous, wherein determining that the function call is superfluous is based on a vector width of the vector processor, wherein the vector width is a number of work items that the vector processor is configured to execute in parallel, and wherein the function call is for a barrier operation to synchronize execution of a plurality of work items, by suspending execution of one of the plurality of work items until others of the plurality of work items reach a similar point in execution.

14. The apparatus of claim 13, wherein the vector processor is configured to elide the function call at runtime.

15. The apparatus of claim 13, wherein determining that the function call is superfluous comprises determining that the vector width satisfies a minimum vector width.

16. The apparatus of claim 15, wherein the minimum vector width is indicated by the function call.