Framework for user-directed profile-driven optimizations

Patent metadata

Field	Value
Publication number	US-9760351-B2
Application number	US-201313855557-A
Country	US
Kind code	B2
Filing date	Apr 2, 2013
Priority date	Apr 2, 2013
Publication date	Sep 12, 2017
Grant date	Sep 12, 2017

Abstract

A method for using profiling to obtain application-specific, preferred parameter values for an application is disclosed. First, a parameter for which to obtain an application-specific value is identified. Code is then augmented for application-specific profiling of the parameter. The parameter is profiled and profile data is collected. The profile data is then analyzed to determine the application's preferred parameter value for the profile parameter.

First claim

What is claimed is:

1. A method for using profiling to obtain application-specific, preferred parameter values for an application, the method comprising:
receiving identification of an application-specific parameter for which to obtain an application-specific value;
receiving augmented code for profiling the application-specific parameter;
profiling the parameter and collecting profile data by:
generating an instrumentation binary from an instrumentation build, the instrumentation binary containing at least one user-defined callback routine registered by the user in a profile initialization function;
executing a training run with one or more representative workloads using the instrumentation binary, the execution invoking the at least one user-defined callback routine to record at least one application-specific preferred parameter value for the application in the collected profile data;

analyzing the collected profile data using the at least one user-defined callback routine; and
generating a feedback-directed optimization (FDO) build using the collected profile data, the FDO build utilizing the at least one application-specific preferred parameter value recorded in the collected profile data for the profiled application-specific parameter.

2. The method of claim 1 wherein profiling is user-directed per-class global value profiling.

3. The method of claim 1 wherein profiling is user-directed per-site value profiling.

4. The method of claim 1 wherein profiling is user-directed per-site object value profiling.

5. The method of claim 1 wherein generating the FDO build, further comprises:
transforming, based on the profile data, an application-specific parameter value to a preferred value using a set of standard value profile transformations.

6. A method for user-directed per-class global value profiling, the method comprising:
receiving a user-defined profile instrumentation initialization routine, the profile instrumentation initialization routine initializing a counter to profile a parameter and registering a user-defined analysis call back routine, containing custom processing for profile data;
generating an instrumentation binary from an instrumentation build, the instrumentation binary containing at least one user-defined analysis callback routine registered by the user in a profile initialization function;
executing a training run with one or more representative workloads using the instrumentation binary, wherein the execution of the training run includes:
running a profile update function in a code location where the counter's value should be updated; and
executing a profile handler method, registered in the user-defined analysis callback routine, to process counter data and record a preferred parameter value for the profiled parameter; and

generating a feedback-directed optimization (FDO) build using profile data collected during execution of the training run.

7. The method of claim 1 further comprising defining a profile counter for the application-specific parameter to be profiled.

8. The method of claim 7 further comprising allocating one entry in the static counter array for the counter.

9. The method of claim 8 wherein allocation is done by calling a compiler extension for counter allocation.

10. The method of claim 1 further comprising allowing for profile counter address tracking by providing a special purpose declaration attribute for the profile counter.

11. The method of claim 1 further comprising allowing for profile initialization by providing a special purpose declaration attribute to designate a profile initialization function.

12. The method of claim 6 further comprising registering a user-defined analysis call back routine by using a GCC interface.

13. The method of claim 6 further comprising recording the preferred parameter value by using a GCC interface.

Assignees

• Google Inc

Inventors

• Johnson Teresa Louise
• Li Xinliang David

Classifications

• G06F8/4434 Primary

  Reducing the memory space required by the program code · CPC title

• G06F8/4441

  Reducing the execution time required by the program code · CPC title

• G06F8/443 Primary

  Optimisation · CPC title