Finding Code that Counts with Causal Profiling

Just in-case you didn't know about this, check out this new profiling technique and it's working implementation at; https://github.com/plasma-umass/coz/

Also a paper describing the work is available at; https://blog.acolyer.org/2015/10/14/coz-finding-code-that-counts-with-causal-profling/

Hi Mayuresh,

I skimmed the paper and it seems interesting. I have one high-level question: how relevant is this technique to conventional single-threaded compilers? There’s some mention of being able to account for system operations (e.g., IO), but It’s not clear to me if the tool is of any use for profiling LLVM-based compilers.

Thanks,
Jakub

Hi Jakub, thanks for taking the effort to skim through the paper.
I could describe it, but consider it better to hear the whole thing from the horse's mouth.
Check out the video at; https://youtu.be/r-TLSBdHe1A

~Mayuresh

I could describe it, but consider it better to hear the whole thing from
the horse’s mouth.

The issue I see with videos is that they are not searchable and require significant time commitment.

I assumed that since you posted cfe-dev this is related to clang/llvm, or uses clang/llvm in an interesting way. Providing a summary of the paper/video from the perspective of cfe-dev would be really helpful and potentially encourage folks to look more closely into the project.

Thanks,
Jakub

I could describe it, but consider it better to hear the whole thing from
the horse's mouth.

The issue I see with videos is that they are not searchable and require significant time commitment.

Previous email mentioned this link:
https://blog.acolyer.org/2015/10/14/coz-finding-code-that-counts-with-causal-profling/
Everything is explained, with link to the original paper:
http://www.sigops.org/sosp/sosp15/current/2015-Monterey/printable/090-curtsinger.pdf

Chris