[llvm-dev] Issue with [InstCombine] Canonicalize range test idiom [D113366]

[Sanjay Patel via llvm-dev]

This problem was also reported as:
https://github.com/llvm/llvm-project/issues/52684
...and should be fixed with:
https://github.com/llvm/llvm-project/commit/bb2fc19c6355d990a4cfb94be20528fbe37950a8
<https://reviews.llvm.org/rGbb2fc19c6355>

On Sat, Dec 18, 2021 at 1:24 PM Sankisa, Krishna (Chaitanya) via llvm-dev <
llvm-dev at lists.llvm.org> wrote:

> [AMD Official Use Only]
>
> Hi,
>
> https://reviews.llvm.org/D113366 and
> https://github.com/llvm/llvm-project/commit/1376301c87274bccf5565eb892173442782b1de3
>
>
> The above change causes a soft hang for the below IR in InstCombine
> Transform.
>
> define i32 @p0(i32 %CF_init) {
>   %t0 = add i32 %CF_init, -3
>   %t1 = icmp ugt i32 %t0, 1
>   %t2 = select i1 %t1, i32 -3, i32 1
>   ret i32 %t2
> }
>
> tryToReuseConstantFromSelectInComparison() from InstCombineSelect.cpp will
> replace select with inverse icmp and select. 1376301c8727 change when it
> sees a icmp ugt will replace with icmp ult. InstCombinerImpl::run() in
> InstructionCombining.cpp, will run these transforms in a loop and it never
> ends. This is causing soft hang.
>
> icmp ugt is always transformed to icmp ult. Issue is with this transform.
> Commenting out the change fixes the soft hang issue.
> Any suggestion on other ways of fixing this in InstructionCombining
> transform pass?
>
>
> Regards,
> Chaitanya
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20211220/ae739526/attachment.html>