https://github.com/lenary approved this pull request.

> I think this change here just removing experimental from Zicfilp and Zicfiss. 
> If I understand correctly, the codegen for shadow stack and landing pads is 
> still affected by these features.

Sorry, of course it is. I went looking through the code, and I'm happy with 
this.

Emitting the shadow stack pseudo is now based on two things:
- one architectural: Zimop/Zcmop, as of 
9bbf22ccda848431da2393ff9a269af8301ea3aa (before yesterday was Zicfiss)
- one non-architectural: having a specific opt-in function attribute 
(`hw-shadow-stack`).

The same should be true for the Zicfilp work, which I realise is in progress.

This PR, as far as I can tell, only fundamentally affects assembly/disassembly.

My concern is just that it feels too early to mark all the complex codegen as 
non-experimental, but I agree the instruction definitions can be marked 
non-experimental.

https://github.com/llvm/llvm-project/pull/157015
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to