Say I want to implement a software that uses hardware performance counters such as those for counting retired stores. Note that alternative solutions without the performance counters are possible but might be relatively a little inefficient. However, I can sacrifice performance a bit for portability and power efficiency. Also, note that the performance counters will be kept on the whole time.
How good hardware performance counters are in consuming power. Secondly, Are there popular platforms or processors, single or multicore, which don’t have performance counters. If so, could you kindly name them.
PAPI is a common profiler program that allows you to compile into the code and access the hardwrae counters. From my own experience, it doesn’t have a noticable effect on performance.
Although I don’t know for sure, I would assume that it will not increase power consumption because hardware counters are always enabled in the hardware. It’s just a matter of reading them.
As far as I know, I’m not aware of any modern non-embedded processors that don’t have performance counters. I may wrong. Someone care to correct me?