• State Resolved
  • Locked Locked
  • Replies 2 replies
  • Subscribers 53 subscribers
  • Views 942 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

inaccurate function profiler cycle counts

Bryan Brudevold
Prodigy 110 points

Hello,

I am running the DM6446 Device Cycle Accurate Simulator in Code Composer 5.2.0.00069. I tried running the function profiler to get cycle times for a few TI IMGLIB functions. The cycle times I observed were wildly inaccurate, around a factor of 60-100 times higher than the benchmark times listed in the IMGLIB documentation.

When I time the same functions using the Profile Clock, the cycle times do match the benchmarks nearly exactly. Thus the problem appears to be with the function profiler rather than my code. Does anyone have an idea why the function profiler results could be so far off? Thanks!

  • 0
    TI__Guru**** 450031 points

    Hi Bryan,

    Make sure you setup the function profiler properties to count the correct event. For example the default option for the profile clock is to count 'cycle.CPU' while the function profiler counts 'cycle.Total'. If the profile clock is giving you the expected results, the cycle.CPU may be what you are looking for.

    http://processors.wiki.ti.com/index.php/Simulator_Analysis_Event_Description#Cycle_Events

    Thanks

    ki

  • 0 in reply to Ki
    Prodigy 110 points

    Thanks Ki! The cycle.CPU counts do match the benchmark prediction. Previously I was looking at cycle.Total counts. For everyone's benefit, cycle.CPU counts can be enabled by going to Profile Setup -> function profiling Properties -> Events -> Cycle -> CPU.

    However now I wonder why my cycle.Total counts are so much higher. As I understand, total cycles = cpu cycles + stall cycles. I seem to have an excessive amount of stall cycles if the total counts are higher by 60-100x like I have seen.

    Bryan