Extending width of performance monitor counters

Abstract

A performance monitor includes at least one Monitor Mode Control Register (MMCR) and plural Performance Control Monitors (PMCs). Each PMC is controlled by the MMCR to pair or group the PMCs so that the overflow from one PMC can be directed to its pair/group. By coupling the PMCs so that overflow from one can be directed to another, the effective size of the counters can be increased.

Description

BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention relates in general to a data processing system and, in particular, to a method and system for performance monitoring within a data processing system. Still more particularly, the present invention relates to a method and system for extending the width of performance monitoring counters in a processor.

2. Description of the Related Art

Within a state-of-the-art general purpose microprocessor facilities are often provided that enable the processor to count occurrences of selected events and thereby obtain a quantitative description of the operation of a data processing system. These facilities are generally referred to as performance monitors.

A conventional performance monitor includes at least one control element, such as a monitor mode control register (MMCR), and one or more counting elements, such as performance monitor counters (PMC's). The MMCR is typically comprised of a plurality of bit fields, which are set to specified values in order to select the events to be monitored and to specify the conditions under which the PMC's are enabled. Occurrences of the selected events can then be counted by the PMC'S.

Because both the number of events available for monitoring and the number of occurrences of monitored events may be large, it would be preferable for performance monitors to employ a large width MMCR and large width PMC's. In addition, because each PMC typically records occurrences of only a single specified event at any given time, it would be preferable to have a large number of PMC's in order to be able to provide a broad description of data processing system performance. However, because the added functionality provided by a large MMCR and multiple large PMC 's increases a processor's die size and therefore cost, the size and number of MMCR's and PMC's are generally somewhat restricted due to these economic and size considerations, and are typically 32 or 64 bits wide at their maximum.

After counting 32 bits, a 32-bit wide PMC is considered full. If a full PMC is allowed to continue counting, the PMC reverts to 0 and begins counting again. This process is known as “wrapping” and the PMC is described as “wrapping to 0.” Wrapping has the potential to lose data since any software configured to read the PMC (to allow evaluation of the state of the PMC) would not be able to determine if the PMC had wrapped or had simply not reached its capacity yet. To deal with this problem, prior art systems employ “interrupt handlers”. An interrupt handler is software written to handle conditions that cause interrupts and exceptions. Interrupt handlers can detect which PMC(s) cause an exception and then can maintain a “virtual” counter that records the overflow history. These interrupt handlers sense the transition of the left-most bit of a PMC from 0 to 1, which provides an indication that the PMC is almost full. The interrupt handler clears the data in the PMC by moving it to an accumulator, which is simply a software version of the PMC that can be arbitrarily large. Thus, the PMC's accumulate the data, dump the data to the software accumulator when full, and continue counting.

While this system functions sufficiently when the processor is fully operational (i.e., when the processor is running software that is capable of handling interrupts), during initial hardware testing of the processor, when the software is unavailable to perform the accumulation function, there is nowhere to move the stored data from a full PMC.

Accordingly, it would be desirable to have a hardware solution for increasing the available width of PMC 's during the initial hardware testing of the processor or when the processor is executing time-sensitive code that cannot be interrupted.

SUMMARY OF THE INVENTION

The above as well as additional objects, features, and advantages of the present invention will become apparent in the following detailed written description. In accordance with the present invention, in a performance monitor having plural performance monitor counters (PMC's) and at least on monitor mode control register (MMCR), each PMC is controlled by the MMCR to pair or group the PMCs so that the overflow from one PMC can be directed to its pair/group. By coupling the PMCs so that overflow from one can be directed to another, the effective size of the counters can be increased.

BRIEF DESCRIPTION OF THE DRAWINGS

The novel features believed characteristic of the invention are set forth in the appended claims. The invention itself however, as well as a preferred mode of use, further objects and advantages thereof, will best be understood by reference to the following detailed description of an illustrative embodiment when read in conjunction with the accompanying drawings, wherein:

FIG. 1 is a block diagram illustrating a typical processor environment in which a performance monitor monitors the operation of the processor;

FIG. 2 illustrates a performance monitor having eight performance monitor counters;

FIG. 3 is a flowchart illustrating an example of steps performed to allocate PMC's in accordance with a first embodiment of the present invention;

FIG. 4 is a flowchart illustrating an example of steps performed to allocate PMC's in accordance with a second embodiment of the present invention; and

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

With reference now to the figures and in particular with reference to FIG. 1, there is depicted a block diagram of an illustrative embodiment of a typical processor environment, indicated generally at 10, in which the invention recited within the appended claims can be utilized. In the depicted illustrative example, processor 10 comprises a single integrated circuit superscalar microprocessor. An example of processor 10 is the PowerPC™ line of microprocessors available from IBM Microelectronics; however, those skilled in the art will appreciate from the following description that the present invention could alternatively be incorporated within other suitable processors.

Processor 10 includes various execution units 20, 22, 24, 26, 28, and 30; registers 32, 36, and 40; buffers 34 and 38; memories 14, 16, and 39; and other functional units (e.g., bus interface unit 12 and sequencer unit 18), all of which are formed from integrated circuitry. For a detailed description of the configuration and operation of such a processor reference can be made to U.S. Pat. No. 5,991,708 to Levine et al. (and with specific reference to FIG. 1 thereof, the description of which is incorporated herein by reference). Of specific interest relevant to the present invention, however, is performance monitor 50 of FIG. 1.

Performance monitor 50 is a software-accessible mechanism capable of providing detailed information descriptive of the utilization of instruction execution resources and storage control. Although not illustrated in FIG. 1, performance monitor 50 is coupled to each functional unit of processor 10 in order to permit the monitoring of all aspects of the operation of processor 10 including reconstructing the relationship between events, identifying false triggering, identifying performance bottlenecks, monitoring pipeline stalls, monitoring idle cycles, determining dispatch efficiency, determining branch efficiency, determining the performance penalty of misaligned data accesses, identifying the frequency of execution of serialization instructions, identifying inhibited interrupts, and determining performance efficiency. Performance monitor 50 includes an implementation-dependent number (e.g., 2–8) of PMC's. In FIG. 1, two PMC's 52 and 54, labelled PMC1 and PMC2, are shown which are utilized to count occurrences of selected events. Performance monitor 50 further includes at least one MMCR 56 that specifies the function of PMC's 52–54. PMC's 52–54 and MMCR 56 are preferably implemented as special purpose registers (SPRs) that are accessible for read or write via MFSPR (move from SPR) and MTSPR (move to SPR) instructions executable by CFXU 26. However, PMC's 52–54 and MMCR 56 may instead be implemented simply as addresses in I/O space.

FIG. 2 illustrates a configuration for performance monitor 50 which can be used to perform the novel allocation process in accordance with the present invention. For the purposes of explanation, performance monitor 50 illustrated in FIG. 2 includes eight PMC's 252, 254, 256, 258, 260, 262, 264, and 266, labeled PMC1, PMC2, PMC3, PMC4, PMC5, PMC6, PMC7, and PMC8, respectively. It should be understood, however, that performance monitor 50 could include more or less than eight PMC's.

Referring to FIG. 2, each of the PMC's 252–266 are coupled to MMCR 270 to control the operation of the PMC'S. In addition, performance monitor 50 is coupled to each functional unit of the processor 10 of FIG. 1 to permit monitoring of all aspects of the operation of the processor. In accordance with the present invention, the MMCR is configured to “pair off” or group sets of the PMC's so that overflow from the first PMC of the pair/group can be counted by other PMC's of the pair/group. For example, if it is assumed that MMCR 270 groups the PMC's in pairs, when MMCR 270 senses the transition of the left-most bit of PMC 252 from a 0 to a 1, MMCR 270 might automatically direct additional counts, previously being counted by PMC 252, to be counted by PMC 254, without interruption. This essentially doubles the size of the counters available for counting events being counted by PMC 252. Each of the remaining PMC's can be similarly paired (e.g., PMC 256 paired with PMC 258; PMC 260 paired with PMC 262; and PMC 264 paired with PMC 266). If each PMC is 32 bits wide, this pairing enables four events to be monitored by the equivalent of one 64-bit counter per event. Further, since the control of the overflow is performed by MMCR 270, the available space can be maximized to suit the needs of the system. The actual pairing off or grouping of the PMC's is performed through programming MMCR 270 using known programming techniques to coordinate the counting by the designated pairs.

FIG. 3 is a flowchart illustrating an example of steps performed to allocate PMC's in accordance with a first embodiment of the invention in which the PMC's are divided evenly among the events being monitored. When the number of PMC's and number of events being monitored cannot be divided evenly, one or more of the PMC's will have less than others. Referring to FIG. 3, at step 302, the number of events being monitored is determined. At step 304, the number of PMC's available for monitoring is determined, and at step 306, the number of PMC's available is divided by the number of events to determine the grouping of the PMC's (step 308). Finally, at step 310, all of the events are monitored on a continuous basis, and allocation of the storage of the counted events is conducted based upon the grouping of the PMC's as done in step 308. All of these actions are carried out based on control from the MMCR. If, as an example, it is assumed that there are two events that need to be monitored, then, using the performance monitor 50 illustrated in FIG. 2, four PMC's could be grouped per event (or any combination could be utilized to monitor the events, depending upon need, as discussed below). Indeed, if only a single event was being monitored, all eight PMC 's of the performance monitor 50 of FIG. 2 could be utilized, thereby giving the tester the equivalent of a single counter that is eight times the size of a single counter.

As an alternative, the MMCR may be configured to determine, ahead of time, not only the number of events being counted, but also the potential frequency of counts for each event. This can be based on historical statistical information made available for use by the MMCR from a memory, or can be preset based on information manually input by a programmer. In this way, if one particular event occurs frequently while another event occurs infrequently, the MMCR can assign more PMC's to the first event and less to the second event. FIG. 4 is a flowchart illustrating an example of the steps to be performed in order to group the PMC's in accordance with this alternate method. Referring to FIG. 4, at step 402, the number of events being monitored by the PMC is determined, and at step 404, the number of PMC's available for doing the monitoring is determined. At step 406, the frequency of occurrence of each event being monitored is identified and, at step 408, based upon this determination, the PMC's are grouped so as to take advantage of the statistical data regarding frequency. Thus, events that occur more frequently will have more PMC 's allocated to them, and events that occur less frequently will have less PMC's allocated. Finally, at step 410, the events are monitored and the storage of the counted events is allocated based upon the grouping. Thus, for example, an event A, which is identified as being a frequently-occurring event, may be assigned six counters in an initialization stage, while event B, which is identified as happening very rarely, may be assigned only two counters during the initialization phase.

The techniques and methods for embodying the present invention in software program code to control the performance monitor are well-known and will not be further discussed herein.

Although the present invention has been described with respect to a specific preferred embodiment thereof, various changes and modifications may be suggested to one skilled in the art and it is intended that the present invention encompass such changes and modifications as fall within the scope of the appended claims.

Claims

1. A method for monitoring the occurrences of one or more events related to the operation of a processor, said processor including a performance monitor having a plurality of counting elements, said method comprising the steps of: identifying the number of events to be counted by said performance monitor;identifying the number of counting elements available to count incidences of said events; andassigning at least two of said counting elements to serially count incidences of at least one of said events, wherein when the number of events to be counted is less than the number of counting elements available to count incidences of said events, said assigning step comprising at least the steps of: dividing the number of available counting elements by the number of events to be counted;in a first assignment step, assigning a number of counting elements, said number equal to the integer resulting from said dividing step, to each of said events to be counted; andin a second assignment step, assigning any unassigned counting elements to at least one of said events.

2. A method as set forth in claim 1, wherein said performance monitor further includes at least one control element, said control element performing said steps of identifying the number of events, identifying the number of counting elements, and assigning of said counting elements.

3. A method as set forth in claim 2, wherein said assigning step further comprises at least the steps of: determining the historical frequency of occurrence of incidences of said events to be counted; andassigning said available counters to said events to be counted based upon said determined historical frequency.

4. A method as set forth in claim 2, wherein said counting elements each comprise a performance monitor counter, and wherein each control element comprises a monitor mode control register.

5. A computer program product in a computer-readable medium for monitoring the occurrences of one or more events related to the operation of a processor, said processor including a performance monitor having a plurality of counting elements, said computer program product comprising: first instructions for identifying the number of events to be counted by said performance monitor;second instructions for identifying the number of counting elements available to count incidences of said events; andthird instructions for assigning at least two of said counting elements to serially count incidences of at least one of said events, wherein when the number of events to be counted is less than the number of counting elements available to count incidences of said events, said third instructions for assigning include at least:fourth instructions for dividing the number of available counting elements by the number of events to be counted;fifth instructions for assigning a number of counting elements, said number equal to the integer resulting from the execution of said fourth instructions, to each of said events to be counted; andsixth instructions for assigning any unassigned counting elements to at least one of said events.

6. A computer program product as set forth in claim 5, wherein said performance monitor further includes at least one control element, said control element providing said first, second, and third instructions.

7. A computer program product as set forth in claim 6, wherein said third instructions further comprise at least: seventh instructions for determining the historical frequency of occurrence of incidences of said events to be counted; andeighth instructions for assigning said available counters to said events to be counted based upon said determined historical frequency.

8. A computer program product as set forth in claim 6, wherein said counting elements each comprise a performance monitor counter, and wherein each control element comprises a monitor mode control register.

9. A system for monitoring the occurrences of one or more events related to the operation of a processor, said processor including a performance monitor having a plurality of counting elements, said system comprising: means for identifying the number of events to be counted by said performance monitor;means for identifying the number of counting elements available to count incidences of said events; andmeans for assigning at least two of said counting elements to serially count incidences of at least one of said events, wherein the number of events to be counted is less than the number of counting elements available to count incidences of said events, said means for assigning comprising at least:means for dividing the number of available counting elements by the number of events to be counted;means for assigning a number of counting elements, said number equal to the integer resulting from operation of said dividing means, to each of said events to be counted; andmeans for assigning any unassigned counting elements to at least one of said events.

10. A system as set forth in claim 9, wherein said performance monitor further includes at least one control element, said control element identifying the number of events, identifying the number of counting elements, and assigning of said counting elements.

11. A system as set forth in claim 9, wherein said assigning means further comprises at least: means for determining the historical frequency of occurrence of incidences of said events to be counted; andmeans for assigning said available counters to said events to be counted based upon said determined historical frequency.

12. A system as set forth in claim 9, wherein said counting elements each comprise a performance monitor counter, and wherein each control element comprises a monitor mode control register.