perf

From perf stat PROGRAM We get information like:

Here, we can see the total execution time of the program. If we improve a program:

• Less page faults
• Less branch misses
• Less instructions per cycle (less work) The goal is to decrease the total time elapsed.

To get useful information, we need to add [profile.release] and debug=true to the toml. This means that when we build, we will compile with debug information. This will slow down the program. Probably re-establish a baseline? Then we run the program using perf record to sample the execution of the program to produce a data set. Then we can look at the code using:

• perf report
• perf annotate
• flamegraph