This discussion has been locked.
You can no longer post new replies to this discussion. If you have a question you can start a new discussion

Have to use performance Monitors (for bench Marking) on Hardware?

Note: This was originally posted on 11th May 2009 at http://forums.arm.com

Hi all,

   I want to do benchmark my application in ARM Cortex A8. For that one i have downloaded some bench mark code using performance monitors from ARM website (see below link) .

[url="http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.faqs/ka4237.html"]http://infocenter.arm.com/help/index.jsp?t...aqs/ka4237.html[/url]

  Basing on the procedure provided in the above link i am bench marking my code in Cortex A8 using ISSM but it is giving zero... Even if i Used RTSM also it is giving very less value i.e., 0.8 MIPS... but same appication if i run on ARM Work bench profiller it is giving 28 MIPS...

  Is performance counters i have to use in Hardware or what?

  Please help me to do benchmarking a function with out using profilling but using performance monitors or other way?

            Thanks in Advance.

Regards,
Ranjith Kumar,
satyaranjith@gmail.com,
(09008787862)
  • Note: This was originally posted on 11th May 2009 at http://forums.arm.com

    Hi,

       Thanks for ur fast reply, 

       If i used performance Monitors to find bench mark information in Hardware, is it will give correct information or not?

       Is it correct way or is their any way to find the benchmark information in Hardware?

      I want to do all this for Cortex A8?

    Regards,
    Ranjith
  • Note: This was originally posted on 11th May 2009 at http://forums.arm.com

    Are you sure you are actually measuring MIPS?

    MIPS = millions of instructions per second.

    So a typical 800 MHz Cortex-A8 could theoretically be 1600 MIPS (it dual issues) assuming "perfect" code. MIPS figures are often normalized "MIPS per MHz" - so in the perfect case the Cortex-A8 will be 2 MIPS per MHz.

    As I have mentioned before RTSM is *not* a performance model - so don't expect a realistic benchmark directly from RTSM.

    The profiler should be able to give you fairly good results - but my personal opinion is that hardware is always the best way to get results because it provides the memory related side-effects which models often struggle to give.