Intel IXP400 manual Document Number 252539, Revision

Intel® IXP400 Software

Access-Layer Components: Performance Profiling (IxPerfProfAcc) API

The number of events that can be monitored simultaneously range from zero to four at a time. When the number of event to monitor is set to 0, only clock counting is performed. The clock count can be set to be incremented by one at each 64th processor clock cycle or at every processor clock cycle.

The steps needed to run this utility are:

1. To begin the clock and event counting, call the start function with parameters:

ixPerfProfAccXscalePmuEventCountStart ( BOOL clkCntDiv,

UINT32 numEvents, IxPerfProfAccXscalePmuEvent pmuEvent1, IxPerfProfAccXscalePmuEvent pmuEvent2, IxPerfProfAccXscalePmuEvent pmuEvent3, IxPerfProfAccXscalePmuEvent pmuEvent4

—BOOL [in] clkCntDiv — Enables/disables the clock divider. When true, the divider is enabled and the clock count will be incremented by one at each 64th processor clock cycle. When false, the divider is disabled and the clock count will be incremented at every processor clock cycle.

—UINT32 [in] numEvents — Number of PMU events that are to be monitored as specified by the user. For clock counting only, this is set to zero.

—pmuEvent1, pmuEvent2, pmuEvent3, pmuEvent4 — The specific PMU events to be monitored by counters as described in section 14.2 and defined in IxPerfProfAccXscalePmuEvent:

IxPerfProfAccXscalePmuEvent {

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_CACHE_MISS = 0,

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_CACHE_INSTRUCTION,

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_STALL,

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_INST_TLB_MISS,

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_DATA_TLB_MISS,

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_BRANCH_EXEC,

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_BRANCH_MISPREDICT,

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_INST_EXEC,

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_FULL_EVERYCYCLE,

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_ONCE,

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_DATA_CACHE_ACCESS,

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_DATA_CACHE_MISS,

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_DATA_CACHE_WRITEBACK,

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_SW_CHANGE_PC,

IX_PERFPROF_ACC_XSCALE_PMU_EVENT_MAX }

2. To end the counting, call the stop function with parameters:

ixPerfProfAccXscalePmuEventCountStop ( ixPerfProfAccXscalePmuResults *eventCountStopResults)

This function can only be called once IxPerfProfAccEventCountStart has been called. It is the user's responsibility to allocate the memory for the results pointer before calling the function. The user may then read/print the values stored in this pointer to obtain the results of the clock/ event counting process. It contains all values of counters and associated overflows.

April 2005	IXP400 Software Version 2.0	Programmer’s Guide
254	Document Number: 252539, Revision: 007