Method and Apparatus for Computing User-Specified Cost Metrics in a Data Space Profiler

• US 20080127149A1
• Filed: 11/28/2006
• Published: 05/29/2008
• Est. Priority Date: 11/28/2006
• Status: Active Grant

• Current Assignee: Oracle America, Inc. (Oracle Corporation)
• Original Assignee: Oracle America, Inc. (Oracle Corporation)
• Inventors: Tracton, Kenneth, Kosche, Nicolai
• Granted Patent: US 8,032,875 B2
• Time in Patent Office: Days
• US Class Current: 717/158
• CPC Class Codes: G06F 8/443 Optimisation

Abstract

A system and method for profiling a software application may include means for defining a custom cost metric that includes a cost metric identifier and a cost function. The cost function may apply a mathematical formula to data extracted from an event set to calculate a respective cost metric value for each of one or more events in the event set. The data extracted from the event set may include one or more respective profiling object identifiers and one or more other respective costs associated with each of the one or more events. A cost associated with an event in the event space may be associated with a function or basic block of instructions. The cost function may include a distribution formula for attributing at least a portion of the cost associated with a function or basic block to each of the instructions comprising the function or basic block.

Claims

1. A method, comprising:

accessing an event set comprising a plurality of events, wherein each event represents a runtime event that occurred during execution of a software application and comprises a plurality of respective extended address element values; and

applying a user-specified cost function to the event set;

wherein said applying comprises;

determining a respective profiling object identifier value for each of one or more events in the event set dependent on one of the plurality of respective extended address element values comprised in the event, wherein the one of the plurality of respective extended address element values corresponds to a hardware component, a software construct, or a data allocation construct associated with the represented runtime event;

computing a respective cost metric value for each of the one or more events dependent on at least one extended address element value comprised in the event corresponding to a cost associated with the respective profiling object identifier and dependent on the user-specified cost function; and

returning the respective profiling object identifier value of each of the one or more events and the respective cost metric value of each of the one or more events.

2. The method of claim 1, wherein said computing a respective cost metric value for each of the one or more events comprises calculating the respective cost metric value according to a mathematical formula comprised in the user-specified cost function that is formatted according to a cost metric description language and is dependent on the at least one extended address element value corresponding to a cost.

3. The method of claim 2, wherein the mathematical formula operates on the at least one extended address element value corresponding to a cost using one or more of:

a bit-wise operator, an arithmetic operator, a comparator operator, a logical operator, a ternary operator, a union operator, an intersection operator, an inclusion operator, a terminator operator, a complement operator, and a relative complement operator.

4. The method of claim 1, further comprising, prior to said returning, applying sort criteria to the respective profiling object identifier value of each of the one or more events and the respective cost metric value computed for each of the one or more events, wherein said sort criteria specifies one or more of:

an ascending value order, a descending value order, an aggregation formula, a number of bins in which to distribute values, a cost metric on which to sort, and a profiling object on which to sort.

5. The method of claim 1, wherein the at least one extended address element value corresponding to a cost comprises one of:

a value of a hardware counter, a value of a software counter, a number of CPU cycles, an event count, an instruction count, an execution time of an instruction, a memory access time, an execution time of an instruction instance attributable to data transformation, an execution time of an instruction instance attributable to data movement, an access rate, and a memory stall time.

6. The method of claim 1, further comprising defining a second user-specified cost function, wherein the second user-specified cost function is dependent on one or more values computed using the user-specified cost function.

7. The method of claim 1, wherein the at least one extended address element value corresponding to a cost comprised in an event represents a cost associated with a block of instructions executing when the event was detected.

8. The method of claim 7, wherein said computing comprises attributing at least a portion of the cost associated with the block of instructions to each of a plurality of instructions comprising the block of instructions according to a distribution formula comprised in the user-specified cost function.

9. The method of claim 1, wherein the cost metric value represents one of:

an average cost, a median cost, an aggregated cost, an instruction rate, an event rate, and a total cost over a specified time period.

10. The method of claim 1, further comprising associating the respective cost metric value with each of the one or more events in the event set.

11. The method of claim 1, wherein the one of the plurality of respective extended address element values comprises one of:

a time, an identifier of a hardware component, an identifier of a software construct, an identifier of a data allocation, an identifier of a protocol, a physical address, a virtual address, and a register value.

12. The method of claim 1, wherein each of the runtime events represented in the event set comprises one of:

a data reference, a cache miss, a cache reference, a data translation buffer miss, a data translation buffer reference, a branch misprediction, and an event counter condition.

13. An apparatus, comprising:

means for defining a custom cost metric comprising a cost metric identifier and a user-specified cost function;

means for accessing an event set comprising a plurality of events, wherein each event represents a runtime event that occurred during execution of a software application and comprises a plurality of respective extended address element values; and

means for applying the user-specified cost function to the event set;

wherein said means for applying comprises;

means for determining a respective profiling object identifier value for each of one or more events in the event set dependent on one of the plurality of respective extended address element values comprised in the event, wherein the one of the plurality of respective extended address element values corresponds to a hardware component, a software construct, or a data allocation construct associated with the represented runtime event;

means for computing a respective cost metric value for each of the one or more events dependent on at least one extended address element value comprised in the event corresponding to a cost associated with the respective profiling object identifier and dependent on the user-specified cost function; and

means for returning the respective profiling object identifier value of each of the one or more events and the respective cost metric value of each of the one or more events.

14. The apparatus of claim 13,

wherein said means for computing comprises means for calculating the respective cost metric value according to a mathematical formula comprised in the user-specified cost function that is formatted according to a cost metric description language and is dependent on the at least one extended address element value corresponding to a cost;

and

wherein the mathematical formula is configured to operate on the at least one extended address element value corresponding to a cost using one or more of;

a bit-wise operator, an arithmetic operator, a comparator operator, a logical operator, a ternary operator, a union operator, an intersection operator, an inclusion operator, a terminator operator, a complement operator, and a relative complement operator.

15. The apparatus of claim 13,

wherein the at least one extended address element value corresponding to a cost comprised in an event represents a cost associated with a block of instructions executing when the event was detected;

and

wherein said means for computing comprises means for attributing at least a portion of the cost associated with the block of instructions to each of a plurality of instructions comprising the block of instructions according to a distribution formula comprised in the user-specified cost function.

16. The apparatus of claim 13,

wherein the at least one extended address element value corresponding to a cost comprises one of:

a value of a hardware counter, a value of a software counter, a number of CPU cycles, an event count, an instruction count, an execution time of an instruction, a memory access time, an execution time of an instruction instance attributable to data transformation, an execution time of an instruction instance attributable to data movement, an access rate, and a memory stall time; and

wherein the cost metric value represents one of;

an average cost, a median cost, an aggregated cost, an instruction rate, an event rate, and a total cost over a specified time period.

17. A computer-readable storage medium comprising program instructions computer-executable to implement:

accessing an event set comprising a plurality of events, wherein each event represents a runtime event that occurred during execution of a software application and comprises a plurality of respective extended address element values; and

applying a user-specified cost function to the event set;

wherein said applying comprises;

determining a respective profiling object identifier value for each of one or more events in the event set dependent on one of the plurality of respective extended address element values comprised in the event, wherein the one of the plurality of respective extended address element values corresponds to a hardware component, a software construct, or a data allocation construct associated with the represented runtime event;

computing a respective cost metric value for each of the one or more events dependent on at least one extended address element value comprised in the event corresponding to a cost associated with the respective profiling object identifier and dependent on the user-specified cost function; and

returning the respective profiling object identifier value of each of the one or more events and the respective cost metric value of each of the one or more events.

18. The storage medium of claim 17,

wherein said computing comprises calculating the respective cost metric value according to a mathematical formula comprised in the user-specified cost function that is formatted according to a cost metric description language and is dependent on the at least one extended address element value corresponding to a cost;

and

wherein the mathematical formula operates on the at least one extended address element value corresponding to a cost using one or more of;

a bit-wise operator, an arithmetic operator, a comparator operator, a logical operator, a ternary operator, a union operator, an intersection operator, an inclusion operator, a terminator operator, a complement operator, and a relative complement operator.

19. The storage medium of claim 17,

wherein the at least one extended address element value corresponding to a cost comprised in an event represents a cost associated with a block of instructions executing when the event was detected;

and

wherein said computing comprises attributing at least a portion of the cost associated with the block of instructions to each of a plurality of instructions comprising the block of instructions according to a distribution formula comprised in the user-specified cost function.

20. The storage medium of claim 17,

wherein the at least one extended address element value corresponding to a cost comprises one of:

a value of a hardware counter, a value of a software counter, a number of CPU cycles, an event count, an instruction count, an execution time of an instruction, a memory access time, an execution time of an instruction instance attributable to data transformation, an execution time of an instruction instance attributable to data movement, an access rate, and a memory stall time; and

wherein the cost metric value represents one of;

an average cost, a median cost, an aggregated cost, an instruction rate, an event rate, and a total cost over a specified time period.