Open deltaguo opened 1 year ago
@deltaguo Apologies for the lack of response. Do you still need assistance with this ticket? Thanks!
@ppanchad-amd of course, i am still curious about this.
Hi @deltaguo, Internal ticket has been created to assist with your issue. Thanks!
Hi @deltaguo, sorry for the delay, I've been reaching out to internal teams to see if we have specific guidance on this. In theory I think this should be possible with Analysis View (https://rocmdocs.amd.com/projects/rocprofiler/en/latest/how-to/rocprofv2-usage.html#analysis-view), which can be used to profile your code at the instruction level, but I haven't personally tested this. Looking at the MFMA and WMMA instructions should provide insight on when and how the matrix core is being used.
I read the documentation about hardware counters and found that there is no introduction on how to monitor the active status of the matrix core. Is there currently a way to achieve this?