Page 1 of 1

Performance for an interval of a benchmark

PostPosted: Thu Mar 19, 2015 2:20 pm
by ayaz_89
I want to use PAPI to observe performance counter registers for a particular instruction patch of an SPEC benchmark. In other words i am only interested in cycles taken by Instruction number 10000000000 to 10500000000. I know we can use PAPI to find out the statistics for the whole benchmark. I was wondering if there is a way to do what I want, considering the fact that any parallel thread in my program would affect the counter register values. Any help will be highly appreciated !

Re: Performance for an interval of a benchmark

PostPosted: Mon Mar 30, 2015 11:07 am
by jagode00
In other words i am only interested in cycles taken by Instruction number 10000000000 to 10500000000.


It's not quite clear to us what you are trying to do. Are you interested in cycle count or instruction count? Can you please rephrase your question and provide some more details on what you want to do and with what objective.

Thanks,
Heike

Re: Performance for an interval of a benchmark

PostPosted: Fri Aug 28, 2015 1:26 pm
by ayazakram_89
Hi, I am interested in both. To accomplish that task, I am using attach3 test case in papi source rep. I run the benchmark as child process and from my parent process I keep on reading child's counters to know when particular number of instructions have been executed (starting point of interval of interest) and then I keep track of cycles count until executed instructions number reach close to the end point of interval of interest. Do you think, this is the right solution or is there a better one ?