• State Accepted Answer
  • Locked Locked
  • Replies 2 replies
  • Subscribers 137 subscribers
  • Views 1146 views
  • Users 0 members are here
Options
Related
How was your experience today?
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

Streamline analyze performance

sindney

Hi, i'm using streamline to capture some cpu related counters of our game. The tool works fine.

But the analyze performance is quite slow, even when only 5.8s of data were capture, it still requires several minutes to finish generating the final report.

I'm not sure if it's related to stack frame's symbol translation or source code search or something else. (Both the symbol and source code has massive amount of data.) Is there any log or some sort of data i can provide to help you identify this performance bottle neck.

It will be great if streamline reports required data instantly.

Btw, now we have some degree of knowleage of how cache\branch\etc perform in function call level.

Is it possible to support sampling in source-code level in the future, so we know exactly which line of code produces the cache miss or miss predicted the branching.

Parents
  • +1

    What version are you using? Analysis and DWARF debug info parsing should be a lot faster in Mobile Studio 2023.0, and we have more improvements planned for later in the year.

    > Is it possible to support sampling in source-code level in the future, so we know exactly which line of code produces the cache miss or miss predicted the branching.

    Attribution should work today - both for source, and disassembly. Note that it has relatively large error bars on modern high-end cores - using 1 millisecond samples on hardware with a wide out-of-order instruction issue window on the hardware issue path means the data is relatively noisy. It should be indicative of problems, but please don't assume it's guaranteed to be 100% accurate on location.

    If you are able to share a capture we'd happily take a look - feel free to get in touch at mobilestudio@arm.com.

    Cheers,
    Pete

Reply
  • +1

    What version are you using? Analysis and DWARF debug info parsing should be a lot faster in Mobile Studio 2023.0, and we have more improvements planned for later in the year.

    > Is it possible to support sampling in source-code level in the future, so we know exactly which line of code produces the cache miss or miss predicted the branching.

    Attribution should work today - both for source, and disassembly. Note that it has relatively large error bars on modern high-end cores - using 1 millisecond samples on hardware with a wide out-of-order instruction issue window on the hardware issue path means the data is relatively noisy. It should be indicative of problems, but please don't assume it's guaranteed to be 100% accurate on location.

    If you are able to share a capture we'd happily take a look - feel free to get in touch at mobilestudio@arm.com.

    Cheers,
    Pete

Children