- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I've been working with vtune lately and I'm very happy with what it can do.
However, I am wondering about the issue of generating large amounts of data. There is a 1000mb limit at which it stops profiling long processes.
The command I use to profile a process.
In practice, I only need the columns "CPU Time:Self, Clockticks:Self". Is it possible to limit the collected data to these columns only? I mainly use data stored in csv because it makes it easier for me to draw conclusions from multiple profiling runs.
Regards
Rafal
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
That isn't really possible at the function level, as VTune collects hardware events that are configured to calculate more meaningful data over time. But you can limit the microarchitecture exploration to only collect on retiring pipeline slots, which would be:
-collect uarch-exploration -knob collect-frontend-bound=false -knob collect-bad-speculation=false -knob collect-memory-bound=false -knob collect-core-bound=false
To limit it further you would probably need a custom analysis type that only collects the hardware events you need, but then VTune won't be able to do any of the post-processing calculations of higher-level metrics.
If you open the GUI, you can easily create more complex command lines by setting the WHERE panel to "Arbitrary Host (not connected)"
Then in the HOW pane you can change the defaults and generate the command line by pressing the >_ button at the bottom:
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Thanks for the very useful tips. I will apply them to the project.
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page