Method and apparatus for correlating profile data

• US 8,065,665 B1
• Filed: 05/06/2004
• Issued: 11/22/2011
• Est. Priority Date: 02/28/2004
• Status: Active Grant

• Current Assignee: Oracle America, Inc. (Oracle Corporation)
• Original Assignee: Oracle America, Inc. (Oracle Corporation)
• Inventors: Aoki, Christopher P., Itzkowitz, Martin S., Wylie, Brian J., Kosche, Nicolai
• Primary Examiner(s): Mitchell; Jason
• Application Number:
  US10/840,180
• Time in Patent Office: 2,756 Days
• Field of Search: 714 37- 39, 714/45, 717124-125, 717127-131, 717/151, 717/154, 717/158
• US Class Current: 717/127
• CPC Class Codes: G06F 11/3466 Performance evaluation by tracing or monitoring
• CPC Class Codes: G06F 11/3471 Address tracing
• CPC Class Codes: G06F 2201/86 Event-based monitoring
• CPC Class Codes: G06F 2201/865 Monitoring of software
• CPC Class Codes: G06F 8/443 Optimisation

Abstract

Correlating profile data facilitates sophisticated code optimization. Going beyond one to one relationships between code execution hindrances and single code behavior attributes provides insight into code behavior at a finer level of granularity. The capability to aggregate profile data based on multiple code behavior attributes and filter based on instances thereof, allows code optimization decisions to be made based on presentation of profile data from various perspectives. Profile data, which includes code behavior attributes correlated with code execution hindrances, is aggregated based at least in part on a first code behavior attribute. Code behavior attributes include one or more of memory references, memory reference objects, functions, time ranges, processors, processes, threads, and source-level data objects. The aggregated profile data is filtered based on an instance of the first code behavior attribute. The filtered profile data is then aggregated based on one or more additional code behavior attributes.

Claims

1. An apparatus providing a software tool encoded in at least one non-transitory machine-readable storage medium, comprising:

a first set of instructions, stored in at least one machine-readable storage medium and executable by at least one processing unit, that collects profile data during execution of code by;

incrementing a hardware counter when at least one runtime event associated with the hardware counter is detected during the execution of the code;

sampling the at least one runtime event when the hardware counter exceeds a threshold during the execution of the code;

determining an instruction instance that corresponds to the at least one runtime event by backtracking from a first point in the code that coincides with the detection of the at least one runtime event that caused the hardware counter to exceed the threshold to a second point in the code that is associated with the at least one runtime event;

decoding the corresponding instruction instance to determine a data address indicated by the corresponding instruction instance;

validating the data address; and

in response to validating the data address, recording the at least one runtime event, the associated instruction instance data address, and a plurality of code behavior attributes associated with the data address based on an inline annotation of the instruction instance comprising a context index of a construct table to identify a source-level language construct that represents a source-level data object associated with the at least one run-time event, the source level-language construct being one of the plurality of code behavior attributes;

a second set of instructions, stored in the at least one machine-readable storage medium and executable by the at least one processing unit, that aggregates code execution hindrances from the profile data collected during the execution of the code based, at least in part, on a first code behavior attribute of the plurality of code behavior attributes;

a third set of instructions, stored in the at least one machine-readable storage medium and executable by the at least one processing unit, that filters the aggregated code execution hindrances based on one or more instances of the first code behavior attribute; and

a fourth set of instructions, stored in the at least one machine-readable storage medium and executable by the at least one processing unit, that aggregates the filtered code execution hindrances based, at least in part, on a second code behavior attribute of the plurality of code behavior attributes to correlate code execution hindrances with the first code behavior attribute and the second code behavior attribute.

2. The apparatus of claim 1, wherein the first and second code behavior attributes include one or more of memory references, memory reference objects, source-level data objects, functions, time ranges, processors, processes, and threads.

3. The apparatus of claim 2, wherein the memory references include one or more of addresses, address groups, pages, and segments.

4. The apparatus of claim 3, wherein the addresses include one or more of virtual addresses and physical addresses.

5. The apparatus of claim 2, wherein source-level data object language constructs include one or more of data structures, members of data structures, data types, and data type definitions.

6. The apparatus of claim 2, wherein the memory reference objects include one or more of cache lines, cache sub-blocks, memory controllers, heap variables, stack variables, variable instances, cache levels, and source-level data objects.

7. The apparatus of claim 1, further comprising:

a fifth set of instructions, stored in the at least one machine-readable storage medium and executable by the at least one processing unit, that filters the filtered code execution hindrances based on one or more instances of the second code behavior attribute.

8. The apparatus of claim 1, wherein at least a subset of the execution hindrances are represented with one or more of time, cycles, stalls, cache references, data translation buffer misses, data translation buffer references, traps, event counter conditions, and percentages thereof.

9. The apparatus of claim 1, wherein the software tool associates one or more instruction instances, which correspond to the code execution hindrances, with the first and second code behavior attributes.

10. A method of providing profile data, the method comprising:

collecting profile data during executing of code, utilizing at least one processing unit, by;

incrementing a hardware counter when at least one runtime event associated with the hardware counter is detected during the execution of the code;

sampling the at least one runtime event during the execution of the code utilizing the at least one processing unit when the hardware counter exceeds a threshold;

determining an instruction instance that corresponds to the at least one runtime event utilizing the at least one processing unit by backtracking from a first point in the code that coincides with the detection of the at least one runtime event that that caused the hardware counter to exceed the threshold to a second point in the code that is associated with the at least one runtime event;

decoding the corresponding instruction instance utilizing the at least one processing unit to determine a data address indicated by the corresponding instruction instance;

validating the data address; and

in response to validating the data address, recording the at least one runtime event, the associated instruction instance data address, and a plurality of code behavior attributes associated with the data address based on an inline annotation of the instruction instance comprising a context index for of a construct table to identify a source-level language construct that represents a source-level data object associated with the at least one run-time event, the source level-language construct being one of the plurality of code behavior attributes;

aggregating the profile data collected during the execution of the code, utilizing the at least one processing unit, the profile data including the code behavior attributes correlated with execution hindrances of the code, based at least in part on a first code behavior attribute of the plurality of code behavior attributes;

filtering the aggregated profile data, utilizing the at least one processing unit, based, at least in part, on a first instance of the first code behavior attribute; and

aggregating the filtered profile data, utilizing the at least one processing unit, based at least in part on a second code behavior attribute of the plurality of code behavior attributes to correlate code execution hindrances with the first code behavior attribute and the second code behavior attribute.

11. The method of claim 10, wherein the plurality of code behavior attributes include one or more of memory references, memory reference objects, source-level data objects, functions, time ranges, processors, processes, and threads.

12. The method of claim 10, wherein the memory reference objects include one or more of cache lines, cache sub-blocks, memory controllers, heap variables, stack variables, variable instances, cache levels, and source-level data objects.

13. The method of claim 10, further comprising:

further filtering the second code behavior attribute aggregated profile data, utilizing the at least one processing unit, based on the first instance of the second code behavior attribute.

14. The method of claim 13, further comprising:

aggregating the profile data, utilizing the at least one processing unit, based on a third code behavior attribute of the plurality of code behavior attributes.

15. The method of claim 10, further comprising:

sorting the filtered profile data utilizing the at least one processing unit.

16. The method of claim 10, wherein at least a subset of the execution hindrances are represented with one or more of time, cycles, stalls, cache references, data translation buffer misses, data translation buffer references, traps, event counter conditions, and percentages thereof.

17. A computer program product for providing profile data, encoded in one or more non-transitory machine-readable storage medium, the computer program product comprising:

a first sequence of instructions, stored in at least one machine-readable storage medium and executable by at least one processing unit, for collecting profile data during execution of code by;

incrementing a hardware counter when at least one runtime event associated with the hardware counter is detected during the execution of the code;

sampling the at least one runtime event when the hardware counter exceeds a threshold during the execution of the code;

determining an instruction instance that corresponds to the at least one runtime event by backtracking from a first point in the code that coincides with the detection of the at least one runtime event that caused the hardware counter to exceed the threshold to a second point in the code that is associated with the at least one runtime event;

decoding the corresponding instruction instance to determine a data address indicated by the corresponding instruction instance;

validating the data address; and

associating the at least one runtime event with an instruction instance and a plurality of code behavior attributes that correspond to the at least one runtime event, and

in response to validating the data address, recording the at least one runtime event, the associated instruction instance data address, and a plurality of code behavior attributes associated with the data address based on an inline annotation of the instruction instance in the code comprising a context index of a construct table to identify a source-level language construct that represents a source-level data object associated with the at least one run-time event, the source level-language construct being one of the plurality of code behavior attributes;

a second sequence of instructions, stored in the at least one machine-readable storage medium and executable by the at least one processing unit, for aggregating the profile data collected during the execution of the code, the profile data including code execution hindrances, by a first code behavior attribute of the plurality of code behavior attributes, wherein the code execution hindrances are correlated with instances of the first code behavior attribute;

a third sequence of instructions, stored in the at least one machine-readable storage medium and executable by the at least one processing unit, for filtering the aggregated profile data by a first of the first code behavior attribute instances; and

a fourth sequence of instructions, stored in the at least one machine-readable storage medium and executable by the at least one processing unit, for aggregating the filtered profile data by a second code behavior attribute of the plurality of code behavior attributes, wherein instances of the second code behavior attribute are correlated with the code execution hindrances to correlate code execution hindrances with the first code behavior attribute and the second code behavior attribute.

18. The computer program product of claim 17, wherein the plurality of code behavior attributes include one or more of memory references, memory reference objects, source-level data objects, functions, time ranges, processors, processes, and threads.

19. The computer program product of claim 18, wherein the memory references include one or more of addresses, address groups, pages, and segments.

20. The computer program product of claim 18, wherein the memory reference objects include one or more of cache lines, cache sub-blocks, memory controllers, heap variables, stack variables, variable instances, cache levels, and source-level data objects.

21. The computer program product of claim 17, further comprising:

a fifth set of instructions, stored in the at least one machine-readable storage medium and executable by the at least one processing unit, for further filtering the profile data by one or more instances of the second code behavior attribute.

22. The computer program product of claim 21, further comprising:

a sixth set of instructions, stored in the at least one machine-readable storage medium and executable by the at least one processing unit, for aggregating the further filtered profile data by a third code behavior attribute of the plurality of code behavior attributes.

23. The computer program product of claim 17 wherein at least a subset of the execution hindrances are represented with one or more of time, cycles, stalls, cache references, data translation buffer misses, data translation buffer references, traps, event counter conditions, and percentages thereof.

24. The computer program product of claim 17, wherein the filtering comprises:

selecting a first instance of the first code behavior attribute; and

retrieving those of the code execution hindrances associated with the selected first code behavior attribute instance.

25. The computer program product of claim 24, wherein the execution hindrances are correlated with the first code behavior attribute instances based, at least in part, on associations between one or more instruction instances and runtime events that correspond to the code execution hindrances, wherein the instructions instances are associated with the first code behavior attribute instances.

26. The computer program product of claim 17, further comprising:

a fifth set of instructions for ordering the filtered profile data by one or more of the plurality of code behavior attributes and the execution hindrances.