Skip to content

[CIR][CodeGen] Introduce CIR CXXSpecialMember attribute #1711

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

Open
wants to merge 13 commits into
base: main
Choose a base branch
from

Conversation

bruteforceboy
Copy link
Contributor

I think this one is self-explanatory, so I will not write much 🙂‍

Adding this attribute helps in optimizations like #1653, and using the attribute it's easy to create operations like cir.std.vector.ctor/cir.std.vector.dtor by just modifying IdiomRecognizer a bit. I believe it will also be useful for future optimizations. Finally, I updated quite a number of tests so they now reflect this attribute.

Please, let me know if you see any issues.

Copy link
Member

@bcardosolopes bcardosolopes 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 adding more of the building blocks here. In the future it'd be nice to have a representation for struct/class more closer to C++ and it will probably be easier to extract this type of information, but to prevent over engineering early, it feels right to incrementally add pieces that can enable us to better analyze C++ code and make transformations easier to write.

Can you look into changing LifetimeChecker.cpp to use this attribute instead of the current AST approach?

@bcardosolopes
Copy link
Member

(@andykaylor @erichkeane @dkolsen-pgi in case you have any extra thoughts here)

Copy link
Collaborator

@andykaylor andykaylor left a comment

Choose a reason for hiding this comment

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

Do we also want to set this attribute on cir.func definitions/declarations?

Copy link

github-actions bot commented Jun 25, 2025

✅ With the latest revision this PR passed the C/C++ code formatter.

@bruteforceboy
Copy link
Contributor Author

A few updates:

  1. The attribute is now attached to the FuncOp (like ctor<type>) instead of the call
  2. The attribute now stores Type instead of a string
  3. I updated LifetimeChecker.cpp to make use of this attribute instead of the AST approach
  4. I added a couple of traits for the ctor attribute, and I think we can extend as necessary in the future
  5. CIR_CXXCtorDtor is now CIR_CXXSpecialMember
  6. Updated a couple of tests to reflect the ctor/dtor attribute.

cc: @bcardosolopes, @erichkeane, @andykaylor

Copy link
Collaborator

@erichkeane erichkeane left a comment

Choose a reason for hiding this comment

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

1 comment, else seems reasonable, I'm happy when the others are.

@bruteforceboy bruteforceboy changed the title [CIR][CodeGen] Introduce CIR CXXCtorDtor attribute [CIR][CodeGen] Introduce CIR CXXSpecialMember attribute Jun 25, 2025
Copy link
Member

@bcardosolopes bcardosolopes 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 update, one more round of suggestions

Copy link
Member

@bcardosolopes bcardosolopes 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 updates, one more round!

The `default` kind is used if the constructor is a default constructor.
The `copy` kind is used if the constructor is a copy constructor.
}];
let parameters = (ins "mlir::Type":$type,
Copy link
Collaborator

Choose a reason for hiding this comment

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

Why do you need type here?

If we want to tag that this is member function from specific class it might make more sense to be more generic and have attribute MemberFunctionOf that is not just tied to ctors/dtor.

Consequently these ctor/dtor attributes will be just enum attribute and unit attribute respectively.

@bcardosolopes what do you think?

Copy link
Member

@bcardosolopes bcardosolopes Jul 7, 2025

Choose a reason for hiding this comment

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

Why do you need type here?

The idea is to have the type around for grabbing any extra information we might need about the class (versus making assumptions on the type of the function this/first param). We don't have a use case for it just yet tho! That's usually a somewhat blocker for me, but I'm fine with keeping this one around.

If we want to tag that this is member function from specific class it might make more sense to be more generic and have attribute MemberFunctionOf that is not just tied to ctors/dtor.

That's the idea moving forward, but starting off with ctor/dtor and go incrementally.

what do you think?

I like the idea of a member_function_of<type, ctor<copy>>.

@@ -2798,6 +2813,15 @@ void cir::FuncOp::print(OpAsmPrinter &p) {
p.printAttribute(annotations);
}

if (getCxxSpecialMember()) {
p << " cxx_special_member<";
Copy link
Collaborator

Choose a reason for hiding this comment

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

Why do we need to wrap it in cxx_special_member? It is apparent from cir.cxx_ctor/dtor that this is the special member function.

Copy link
Member

@bcardosolopes bcardosolopes 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 your patience, more comments!

The `default` kind is used if the constructor is a default constructor.
The `copy` kind is used if the constructor is a copy constructor.
}];
let parameters = (ins "mlir::Type":$type,
Copy link
Member

@bcardosolopes bcardosolopes Jul 7, 2025

Choose a reason for hiding this comment

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

Why do you need type here?

The idea is to have the type around for grabbing any extra information we might need about the class (versus making assumptions on the type of the function this/first param). We don't have a use case for it just yet tho! That's usually a somewhat blocker for me, but I'm fine with keeping this one around.

If we want to tag that this is member function from specific class it might make more sense to be more generic and have attribute MemberFunctionOf that is not just tied to ctors/dtor.

That's the idea moving forward, but starting off with ctor/dtor and go incrementally.

what do you think?

I like the idea of a member_function_of<type, ctor<copy>>.

@@ -2618,6 +2620,37 @@ ParseResult cir::FuncOp::parse(OpAsmParser &parser, OperationState &state) {
state.addAttribute(annotationsNameAttr, annotations);
}

// Parse CXXSpecialMember attribute
if (mlir::succeeded(parser.parseOptionalKeyword("cxx_ctor"))) {
Copy link
Member

Choose a reason for hiding this comment

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

Still sounds simple enough that we shouldn't need these to be hand written, what specific problems are you hitting?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

I can update this once the syntax of the attribute is approved. I am also happy with member_function_of<type, ctor<copy>>

Copy link
Member

Choose a reason for hiding this comment

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

Thanks! I'll defer this to @xlauko cause I'm going to be out until next week. One important part here is not to have a custom parser/printer because this is simple enough that it doesn't need one (see more examples in the LLVM dialect for inspiration).

// CIR-NEXT: cir.func private @_ZN1EntEv(!cir.ptr<!rec_E>) -> !rec_E
// CIR-NEXT: cir.func private @_ZN1ED1Ev(!cir.ptr<!rec_E>) extra(#fn_attr)
// CIR-NEXT: cir.func private @_ZN1ED1Ev(!cir.ptr<!rec_E>) cxx_dtor<!rec_E> extra(#fn_attr)
Copy link
Member

Choose a reason for hiding this comment

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

Sounds like we could also drop "cxx": dtor<!rec_E> is probably good enough? No need to change this right now, let's converge on the syntax first (see other comment).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants