• State Accepted Answer
  • +1 person also asked this people also asked this
  • Locked Locked
  • Replies 5 replies
  • Subscribers 138 subscribers
  • Views 2692 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

Mali Performance counters for L1

Swapnil Ghanshyala

Hi, can anyone help me find Mali Performance counters (direct or indirect) that can give information on 

1. L1 texture cache hits or misses

2. L1 load/store unit cache hits or misses

3. Number of requests made to L2 cache

4. Number of requests made to external memory

Here is the document I have been referring https://developer.arm.com/documentation/102730/latest/

Thank you

Parents
  • +1
    Swapnil Ghanshyala said:
    1. L1 texture cache hits or misses

    Not possible to count, and mostly a useless metrics for a GPU (cache misses are expected and in most cases do not actually cause lost cycles).

    The "Texture bytes read from L2 per texture cycle" and "Texture bytes read from external memory per texture cycle" counters will give you a feel for the number of bytes fetched per access, but what "good" looks like depends on your texture format. 

    Swapnil Ghanshyala said:
    2. L1 load/store unit cache hits or misses

    Not possible to count, and mostly a useless metrics for a GPU (cache misses are expected and in most cases do not actually cause lost cycles).

    The "Load/store bytes read from L2 per texture cycle" and "Load/store bytes read from external memory per texture cycle" counters will give you a feel for the number of bytes fetched per access, but what "good" looks like depends on your data structure. 

    Swapnil Ghanshyala said:
    3. Number of requests made to L2 cache

    See the shader core memory traffic section in the document you linked - it gives you the L2 traffic breakdown by source. 

    Swapnil Ghanshyala said:
    4. Number of requests made to external memory

    See the shader core memory traffic section in the document you linked - it gives you the external traffic breakdown by source. 

    HTH, 
    Pete

Reply
  • +1
    Swapnil Ghanshyala said:
    1. L1 texture cache hits or misses

    Not possible to count, and mostly a useless metrics for a GPU (cache misses are expected and in most cases do not actually cause lost cycles).

    The "Texture bytes read from L2 per texture cycle" and "Texture bytes read from external memory per texture cycle" counters will give you a feel for the number of bytes fetched per access, but what "good" looks like depends on your texture format. 

    Swapnil Ghanshyala said:
    2. L1 load/store unit cache hits or misses

    Not possible to count, and mostly a useless metrics for a GPU (cache misses are expected and in most cases do not actually cause lost cycles).

    The "Load/store bytes read from L2 per texture cycle" and "Load/store bytes read from external memory per texture cycle" counters will give you a feel for the number of bytes fetched per access, but what "good" looks like depends on your data structure. 

    Swapnil Ghanshyala said:
    3. Number of requests made to L2 cache

    See the shader core memory traffic section in the document you linked - it gives you the L2 traffic breakdown by source. 

    Swapnil Ghanshyala said:
    4. Number of requests made to external memory

    See the shader core memory traffic section in the document you linked - it gives you the external traffic breakdown by source. 

    HTH, 
    Pete

Children