Hi @hlanka3
There are too many events to report, hence right now we are reporting a selected few.
PCM customized reported events, we can obtain referring below points :
Go to AMDuProf/bin/Data/Config folder.
To check the config file name,
Check lspcu and note CPU family and Model number.
Example :
CPU family: 25
Model: 17
Convert these values to hex :
CPU Family : 0x19
Model : 0x11
So the config file name to modify would be : 0x19_0x11
If the model name after hex conversion has 2 digits , ignore the last digit.
Ex : 0x19_0x11 - Look for config file with name 0x19-0x1
If the Model number is single digit and no matching config file is present, that means PCM is not supported on that model.
Open the appropriate config file.
Though events can be customized, the no. of events in one block cannot be greater than 6.
If required create a new section : <core> --- </core>
Check for the parameter you need and modify the metric section.
Ex for L2 data is reported as below :
<metric name="L2 Access (pti)" expression="(($L2AccessWithoutPF + $L2PFHitinL2 + $L2PFMissL2HitinL3 + $L2PFMissL2L3) * 1000) / $IRPerfL2"> </metric>
<metric name="L2 Miss (pti)" expression="(($L2Miss * 1000 / $IRPerfL22) + ($L2PFMissinL2 * 1000 / $IRPerfL2))"> </metric>
If you need cache miss % (pti), you can add metrics L2Accesspti and L2Misspti as below :
*** Note the usage of "$" before the metric names
*** Ensure metric names have no special characters.
<metric name="$L2Accesspti" expression="(($L2AccessWithoutPF + $L2PFHitinL2 + $L2PFMissL2HitinL3 + $L2PFMissL2L3) * 1000) / $IRPerfL2"> </metric>
<metric name="$L2Misspti" expression="(($L2Miss * 1000 / $IRPerfL22) + ($L2PFMissinL2 * 1000 / $IRPerfL2))"> </metric>
<metric name="L2 Miss %" expression=""($L2Misspti * 100 ) / $L2Accesspti"> </metric>
Similarly you can check for L1 and customize the data for whichever cache you are looking for.
