Hi, > Even in fixed period mode, I still see a little randomization for the > sampling period. For example, here is the result on my box: > > $ ./task_smpl -p 200000 ls > period=200000 freq=0 > 17 PERF_COUNT_HW_CPU_CYCLES > 18 PERF_COUNT_HW_INSTRUCTIONS > IIP:0xffffffff81306980 PID:26702 TID:26702 TIME:717452755229999 > STREAM_ID:17 PERIOD:200000 ENA=0 RUN=0 NR=2 > 200244 PERF_COUNT_HW_CPU_CYCLES (17) > 86619 PERF_COUNT_HW_INSTRUCTIONS (18) > > IIP:0x0000003b49005c91 PID:26702 TID:26702 TIME:717452755484619 > STREAM_ID:17 PERIOD:200000 ENA=0 RUN=0 NR=2 > 400375 PERF_COUNT_HW_CPU_CYCLES (17) > 186010 PERF_COUNT_HW_INSTRUCTIONS (18) > > IIP:0x00000000004039f4 PID:26702 TID:26702 TIME:717452755716217 > STREAM_ID:17 PERIOD:200000 ENA=0 RUN=0 NR=2 > 600267 PERF_COUNT_HW_CPU_CYCLES (17) > 352344 PERF_COUNT_HW_INSTRUCTIONS (18) > Several points here:
- make sure you have the latest libpfm4 (git pull), I fixed some issues in task_smpl - there is a kernel bug related to how ENA, RUN are managed when sampling. I posted a patch to fix this yesterday. It is not in the kernel yet. If those timings are bogus, then scaling of events is bogus. - it is normal to have some (small) variations of the counts. ------------------------------------------------------------------------------ Nokia and AT&T present the 2010 Calling All Innovators-North America contest Create new apps & games for the Nokia N8 for consumers in U.S. and Canada $10 million total in prizes - $4M cash, 500 devices, nearly $6M in marketing Develop with Nokia Qt SDK, Web Runtime, or Java and Publish to Ovi Store http://p.sf.net/sfu/nokia-dev2dev _______________________________________________ perfmon2-devel mailing list perfmon2-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
