New Derived Metrics: CPI and IPC

Performance Analyzer displays new derived metrics called CPI (Cycles Per Instruction) and IPC (Instructions Per Cycle) which can help you identify where your application is running efficiently or inefficiently. The CPI and IPC metrics are available when you perform hardware counter profiling on your application and specify the counters for cycles and instructions. For example, you could profile using the command collect -h cycles,on,insts,on to generate the metrics.

Inefficient areas display with a high CPI or low IPC. Efficient areas of your program show low CPI or high IPC.

The following figure shows the Source view with a high CPI metric on line 12, which indicates it is running inefficiently.

Figure 3-8  CPI Metric in the Source View