• Locked Locked
  • Replies 3 replies
  • Subscribers 118 subscribers
  • Views 3023 views
  • Users 0 members are here
Options
Related
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

Profiling on Cortex A9 RTSM with cache enabled

Naveen G
Note: This was originally posted on 19th November 2009 at http://forums.arm.com

I am new to RVDS tools. I profiled my code on realview profiler. How reliable are these figures. Does these figures consider I-Cache and D-Cache effects. Is the cache enabled by default in RTSM or do I need to enable.
  • Note: This was originally posted on 20th November 2009 at http://forums.arm.com

    What are the other means by which I can profile on Cortex A9 accurately other than actual hardware.
  • Arm Employee Badge
    Note: This was originally posted on 20th November 2009 at http://forums.arm.com

    RTSM models are not accurate - they are high speed functional models, not performance models. ARM binaries get cross-compiled in to x86 code, and that is what gets executed - so you really see the performance characteristics of the x86 cache size =)

    The profiler itself has a simple pipeline model for identifying interlocks, although as A9 is an out-of-order core I'm not sure how much use this will be =)
  • Arm Employee Badge
    Note: This was originally posted on 20th November 2009 at http://forums.arm.com

    I'm not aware of any cycle-accurate or even cycle-approximate system models for a Cortex-A9 (could be wrong).

    Real hardware really is the best option if you want accuracy for high speed application cores - system behaviors such as L1 cache sizes, L2 size and latency, and L3 bus bandwidth / latency are really all critical to the performance of software running on a modern processor, and are all things which models tend to ignore.