numericalalgorithmsgroup / pypop

Python Tools for the POP Metrics
https://numericalalgorithmsgroup.github.io/pypop/doc.html
Other
12 stars 5 forks source link

Don't calculate IPC scaling using stats["IPC"].mean() #17

Closed Jonathan3145 closed 4 years ago

Jonathan3145 commented 4 years ago

Instead use something like

(stats["Useful Instructions"].sum() / stats["Useful Cycles"].sum())/ (self._stats_dict[ref_key].stats["Useful Instructions"].sum()/self._stats_dict[ref_key].stats["Useful Cycles"].sum())

ptooley commented 4 years ago

Doing it this way ensures that the IPC is weighted by instructions per thread rather than being indifferent to relative workload