vikramRH wrote:

> > This is specifically for "verify-machineinstrs" and cannot be used for the 
> > current issue. (target such as AMDGPU do run the verification.), instead I 
> > have added a new target option that caters to this NPM specific usecase
> 
> Why exactly can it not be used for this issue? It controls the addition of 
> the same pass (
> 
> https://github.com/llvm/llvm-project/blob/7b56bc85ca672d19237f2e8d4cc34ceeb6e7f50a/llvm/lib/CodeGen/TargetPassConfig.cpp#L802
> 
> ), although it looks like the original option only controls it with expensive 
> checks enabled? Maybe because it's added after each pass in that 
> configuration?
> Either way, I don't understand why we can't reuse that target hook given it 
> controls the same pass in the legacy PM.

Yes, the existing hook controls the verification with expensive checks enabled 
builds (atleast I see AMDGPU enabling it). I'm trying to control the verifier 
being added to the default NPM pipelines in release builds. I do see quite a 
few verifier failures with NPM during downstream testing and Probably the 
expensive checks tests aren't exhaustive enough.

https://github.com/llvm/llvm-project/pull/176693
_______________________________________________
llvm-branch-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits

Reply via email to