All collected values are are unsigned and the collection is stopped
at maximum unsigned value to prevent unnoticed value wrapping.
Yes understood - the main problem is the existence of these values at all - the test does not run long enough to cause capping everywhere (and they are spread almost equally across often-used and rarely-used code, mixed with many zeroes also). Bogus values must be arriving from the CPU core somehow to cause this. My results do not look at all like results posted previously on this forum (which to me looked quite sensible!)..
I'll look into that after I've gotten the callgraph stuff to working
state and we otherwise look more into cycles counts. For now cycle
counts are best ignored for CPU.
Ok no problem. I'm still not in a hurry for this fix but it seems to be the last obvious problem I see recurring in my tests...
Thanks
D.