[PATCH] D46912: StructurizeCFG: Adjust the loop depth for a subregion to order the nodes correctly

[Nicolai Hähnle via llvm-commits]

On 27.05.2018 19:09, Daniel Berlin wrote:
> On Wed, May 16, 2018 at 1:32 PM, Changpeng Fang via Phabricator via 
> llvm-commits <llvm-commits at lists.llvm.org 
> <mailto:llvm-commits at lists.llvm.org>> wrote:
> 
>     cfang marked 2 inline comments as done.
>     cfang added a comment.
> 
>     Right. Reverse Post Order (RPO) should be the fundamental order. But
>     apparently there are some cases that pure RPO does not work.
> 
> 
> These should be analyzed, because something is fundamentally wrong there :)
> (I understand being pragmatic and fixing bugs, but on the side we should 
> look deeper and understand what is going wrong)

For the purpose of AMDGPU, we are looking (long term, hopefully over the 
course of the year going into 2019) at an alternative approach for 
control flow translation which doesn't use the existing StructurizeCFG 
at all.

Whether this will be useful for other backends is unclear. Part of the 
problem is that the requirements that StructurizeCFG is supposed to 
fulfill are not super clearly defined. But rest assured that we're not 
just ignoring the fundamental issues here.

Cheers,
Nicolai

-- 
Lerne, wie die Welt wirklich ist,
Aber vergiss niemals, wie sie sein sollte.