[llvm] [TailDuplicator] Do not restrict the computed gotos (PR #114990)
via llvm-commits
llvm-commits at lists.llvm.org
Fri Mar 21 04:46:29 PDT 2025
dianqk wrote:
> It is very likely that it was a known fact that `tail-dup-pred-size` causes similar issue and that's why it was configured with a number.
That [PR](https://github.com/llvm/llvm-project/pull/78582) landed at LLVM 19 was also sent by me. w
I overlooked the difference between jump table and computed gotos in that PR.
So I think this regression should be caused by some changes between these two PRs.
> So circumventing the check for `tail-dup-pred-size` now opens the door to the same class of issues that the argument was supposed to control.
>
> Can you please add another feature to allow controlling this circumventing for computed gotos?
As I mentioned in the PR description, the key difference is that we expect the original intent and performance in computed gotos.
https://github.com/llvm/llvm-project/pull/114990
More information about the llvm-commits
mailing list