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

cortex-a8 : dual issue

Note: This was originally posted on 3rd November 2009 at http://forums.arm.com

Hi every1!

1>
I'm running a piece of code on target Cortex-a8 on ARM Profiler 2.0 using RTSM. The profile analysis that I get doesn't show any instance of dual issue. I'm aware of the bottlenecks in case of dual issue i.e. only certain combinations can be dual issued. But in my case I'm not able to find even a single instance of dual issue happening. Could any1 suggest the cause behind it?

2>RVD: When I select Cortex-a8.dll while selecting the target on rv debugger, multiple emulation baseboard gets displayed in the right pane. I'm confused out of those, which baseboard will be suitable for embedded applications. Could anybody clarify on this?

Rajeeva
Parents
  • Note: This was originally posted on 3rd November 2009 at http://forums.arm.com

    As ttfn says the RTSM models are functional model and do not model pipeline features.

    These models run fast because they sacrifice the pipeline model (cycle-accurate and cycle-approximate models which do provide this visibility are 10-1000 times slower than RTSM).

    RTSM models essentially cross-compile the ARM binary into functionally equivalent x86 code and runs this native code on the host machine - as you are executing x86 code on an x86 processor you obviously see x86 pipeline performance when benchmarking =)

    Performance counter events in RTSM, and cycles measured by the profiler are therefore only approximations - however command flow seen by the profiler will be correct.
Reply
  • Note: This was originally posted on 3rd November 2009 at http://forums.arm.com

    As ttfn says the RTSM models are functional model and do not model pipeline features.

    These models run fast because they sacrifice the pipeline model (cycle-accurate and cycle-approximate models which do provide this visibility are 10-1000 times slower than RTSM).

    RTSM models essentially cross-compile the ARM binary into functionally equivalent x86 code and runs this native code on the host machine - as you are executing x86 code on an x86 processor you obviously see x86 pipeline performance when benchmarking =)

    Performance counter events in RTSM, and cycles measured by the profiler are therefore only approximations - however command flow seen by the profiler will be correct.
Children
No data