Please note: We are aware of an issue affecting replies on the Arm Community forums, which may not be loading as expected.

We apologize for any inconvenience and appreciate your patience while we investigate and work to resolve the issue.

Thank you for your understanding.

  • State Accepted Answer
  • Locked Locked
  • Replies 2 replies
  • Subscribers 137 subscribers
  • Views 2725 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

Clarification of external memory reads counter meaning

JPJ

Hi guys,

I'm having some trouble understanding the meaning of the $MaliCoreL2ReadsFragmentExternalReadBeats counter (in the context of a Mali G-76 at least). I have 3 external memory reading counters with the following values :

  • $MaliCoreL2ReadsFragmentExternalReadBeats * 16 = 119 mebi-bytes
  • $MaliCoreL2ReadsLoadStoreExternalReadBeats * 16  = 218 mebi-bytes
  • $MaliCoreL2ReadsTextureExternalReadBeats * 16 = 552 mebi-bytes

I can't find any reference to it in your documentation (alongside the other 2). The main reason I'm asking is that an optimization we've made reduced the remaining 2 (as we expected) but just slightly increased the first one - MaliCoreL2ReadsFragmentExternalReadBeats. The increase might just be noise (as it's insignificant) but it would be good to understand what kind of data it refers to.

Cheers!

  • +1 Arm Employee Badge

    Hi JPJ, 

    Sorry for the slow documentation update - as you've noticed we've added some more counters to the Streamline template, but we're delayed on updating the docs as we're in the middle of migrating our web docs to a new backend system. Hope to return to normal service with some new docs before the end of the year. 

    The MaliCoreL2ReadsFragment counter is the bandwidth loaded by the fragment front-end when setting up fragment workloads to be shaded. This includes:

    * Draw call descriptor loads
    * Uniform buffer preloads
    * Tile list metadata reads
    * Tile list vertex position reads

    I'd expect a little bit of variability in this counter due to changes in binning as triangles move around - the bin assignments can be quite sensitive to triangle size and their exact screen position.

    HTH, 
    Pete

  • 0 in reply to Peter Harris

    Thanks Pete! No problem with that, and thanks for the clarification. That makes sense as my comparison is across a long period of time as well.