Skip to content
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

bpo-44340: Add support for building with clang thin lto via --with-lto=thin #26585

Closed
wants to merge 4 commits into from

Conversation

holmanb
Copy link

@holmanb holmanb commented Jun 7, 2021

This adds support for building cpython with clang's --flto=thin option. Existing --with-lto behavior remains unchanged (default to no, with --with-lto currently using the default compiler lto option).

The tests (make test) currently pass for clang 11.1.0 for each of --with-lto and --with-lto=thin.

https://bugs.python.org/issue44340

@the-knights-who-say-ni
Copy link

Hello, and thanks for your contribution!

I'm a bot set up to make sure that the project can legally accept this contribution by verifying everyone involved has signed the PSF contributor agreement (CLA).

CLA Missing

Our records indicate the following people have not signed the CLA:

@holmanb

For legal reasons we need all the people listed to sign the CLA before we can look at your contribution. Please follow the steps outlined in the CPython devguide to rectify this issue.

If you have recently signed the CLA, please wait at least one business day
before our records are updated.

You can check yourself to see if the CLA has been received.

Thanks again for the contribution, we look forward to reviewing it!

Copy link
Member

@ned-deily ned-deily left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the PR. Without commenting on the desirability of the feature, be aware thatconfigure is a derived file produced from configure.ac by the autoconf tool. The PR needs to change configure.ac and then run autoconf and commit the resulting changes to configure too. See the devguide for more info.

@bedevere-bot
Copy link

A Python core developer has requested some changes be made to your pull request before we can consider merging it. If you could please address their requests along with any other requests in other reviews from core developers that would be appreciated.

Once you have made the requested changes, please leave a comment on this pull request containing the phrase I have made the requested changes; please review again. I will then notify any core developers who have left a review that you're ready for them to take another look at this pull request.

Copy link
Member

@corona10 corona10 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please add the NEWS.d with blurb tool

@corona10
Copy link
Member

@holmanb
I am a big fan of the thin LTO feature,
I will follow up on this PR. please ping me if you need my help cc @ned-deily

@corona10 corona10 self-assigned this Jul 18, 2021
@corona10
Copy link
Member

corona10 commented Jul 18, 2021

Let's run the https://github.com/python/pyperformance benchmark and compare it with full LTO
I would like to see how thin LTO executes aggressive optimization pass.
(Ref: http://blog.llvm.org/2016/06/thinlto-scalable-and-incremental-lto.html)

cc @vstinner @pablogsal @ambv

@corona10 corona10 changed the title bpo-44340 Add support for building with clang thin lto via --with-lto=thin bpo-44340: Add support for building with clang thin lto via --with-lto=thin Jul 18, 2021
@@ -1545,7 +1545,7 @@ Optional Packages:
--with-trace-refs enable tracing references for debugging purpose
(default is no)
--with-assertions build with C assertions enabled (default is no)
--with-lto enable Link-Time-Optimization in any build (default
--with-lto[=no|thin] enable Link-Time-Optimization in any build (default
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What about

--with-lto[=full|thin] enable Link-Time-Optimization in any build (default is full)

@@ -1545,7 +1545,7 @@ Optional Packages:
--with-trace-refs enable tracing references for debugging purpose
(default is no)
--with-assertions build with C assertions enabled (default is no)
--with-lto enable Link-Time-Optimization in any build (default
--with-lto[=no|thin] enable Link-Time-Optimization in any build (default
is no)
Copy link
Member

@corona10 corona10 Jul 18, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@holmanb

the current no means that whether we apply LTO or not.
So the configure should be updated if we also provide the thin LTO option.

Following options should be available.
A. no LTO (by default) ./configure
B. with LTO (default=full) ./configure --with-lto
C. with LTO (full LTO designated) ./configure --with-lto=full
D. with LTO (thin LTO designated) ./configure --with-lto=thin

Comment on lines +1415 to +1419
LTOFLAGS="-flto$LTO_ARG -Wl,-export_dynamic"
LTOCFLAGS="-flto$LTO_ARG"
;;
*)
LTOFLAGS="-flto"
LTOFLAGS="-flto$LTO_ARG"
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Are the flags (thing or otherwise) supported by all compilers that support LTO?

Copy link
Author

@holmanb holmanb Jul 18, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No. Thinlto is a more recent implementation than the default that is specific to clang.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copy link
Member

@pablogsal pablogsal Jul 18, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Then we should probably need show an error if the compiler doesn't support it or/and document the limitation, because otherwise, it seems that is a general option that we are exposing regardless of the compiler.

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@corona10 - I like it. That's more readable and provides a warning as @pablogsal suggested.

Copy link
Author

@holmanb holmanb Jul 18, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@corona10 Nitpick: gcc doesn't have a concept of full, and if --with-lto=full is configured with gcc, no warnings are thrown and -flto is assumed. Is this assumption sensible? Or would throwing an error on a nonsense input be better than assuming intent?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think given that these options are only available on clang we should just fail if the compiler is not clang. Adding checks to other compilers one by one is going to be suboptimal because CPython can be compiled with almost any C compiler: xcl, icc...

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

#27231

Let's discuss the supporting option with the attached PR.

  • We may check clang version
  • We may check other compiler support (clang, gcc supports LTO itself, others not)

@holmanb
Copy link
Author

holmanb commented Jul 18, 2021

Thanks for the PR. Without commenting on the desirability of the feature, be aware thatconfigure is a derived file produced from configure.ac by the autoconf tool. The PR needs to change configure.ac and then run autoconf and commit the resulting changes to configure too. See the devguide for more info.

@ned-deily The current configure file in main branch is not the product of running autoconf with configure.ac. What is the best way to proceed? 2fc857a is one example of a change that didn't update configure.ac.

I assume that getting this fixed upstream is a prerequisite for getting this PR merged, otherwise autoconf will generate a configure that rips out some recent updates to configure.

Perhaps a CI bot to verify that in future PRs autoconf(configure.ac) -> configure would be appropriate?

@corona10 corona10 closed this Jul 19, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

6 participants