Top-Down performance analysis methodology

Top-Down performance analysis methodology

First is the breakdown of metric levels in TMAM and second shows reasoning for a single uop:

We will run our app and collect specific metrics that will help us to characterize our application. In the column there is performance event that we can use to locate exact place in the code where the issue occurs. Let’s look at the code:

I allocate a big enough array to make it not fit in the L3 cache (L3 cache on the machine I was using is 38,5 MB – Intel(R) Xeon(R) Platinum 8180 CPU).

Source: dendibakh.github.io