-
Notifications
You must be signed in to change notification settings - Fork 38
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Enabling ccache on github builds for Ubuntu #95
Conversation
WHat is the status of this? |
@XapaJIaMnu Cache (CCACHE based) is working, but I don't see any speed gains. I also had plans to cache the build folder (it's ready in theory, duplicate code). The host-os keeps changing between avx2 and avx512 so unsure how reusable this is. Is this related? |
No, this is completely unrelated. But since we're compiling with march native CCache would be unusable in between machines with different signatures. |
Shall we close this? |
@XapaJIaMnu No, we'll keep this. I'm reading more into this ccache business over multiple C++ projects which use GitHub CI, still hopeful something can be done. |
GitHub actions is this mysterious poorly documented thing to me. @graemenail said he does continuous distribution of his CV on his CV, so I'm passing the review over to him. |
What speedup are we seeing from CC cache? |
@XapaJIaMnu 5m, 3m for what was 30m, 17m previously if cache hit. They're in the screenshots. I think best for us is to drop march=native and fix the instruction to ssse3 or avx2 for cache advantages. |
I am absolutely in favour of keeping native because we have had breakages in the past the involve only newer ISA. We should indeed have an option that limits the architecture too. This is slightly more complicated because the way the architecture is set on Linux/Mac is different from the way its set on Windows. |
I'll wait for @graemenail 's review |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks good. Would be interested to see the cache usage if one were to re-run the job from UI.
@graemenail 10m, 20m vs 2m, 5m for minimal-marian and full-marian respectively, performed on |
…anslator into jp/brt-ghci-ccache
Primarily using the following for reference: