The run-pass/issue-36856.rs fails on beta for mips{el}-linux-gnu

[draganmladjenovic]

The test starts failing after revert in #50137. This issue is still present in FastISel which incorrectly sign extends i1 constants to I32 when selecting binary ops link. I was able to trigger the issue reliably on mips32 https://godbolt.org/g/b3LBWo while other platforms seem to not use the FastISel consistently. For example for arm it needs to be forced https://godbolt.org/g/yWmPMu. Is it ok to revert the mentioned pull until the proper fix gets landed to LLVM repo @nox ?

[nox]

Cc @rust-lang/compiler

[eddyb]

We should revert, IMO, and properly track the FastISel issue, ideally in a way that we can easily check, without an ARM/MIPS system, whether the bug was fixed.

[draganmladjenovic]

Effectively this is the same issue as in https://bugs.llvm.org/show_bug.cgi?id=38019 from @alexcrichton. I wasn't able to find rust-lang issue tied to this. I've missed the date. There won't be any 1.27.* releases.

[draganmladjenovic]

Sorry for the late response. For reference this issue is fixed on LLVM side by https://reviews.llvm.org/rL337288 and https://reviews.llvm.org/rL336084 and that will be part of LLVM 7.0 release.

[Enselic]

Triage: I can't find any revert of #50137. Does anyone know if it ever was reverted? Since 5 years have passed I also wanted to check if the test still fails for MIPS? Maybe over time it began to pass again?