I have several concerns with this as is. Blocking while OOO.

Thanks for driving this!

Thanks!

I have concerns about several aspects of what you are doing here. Blocking until I have time to review.

Thanks! This all looks really really good. I'm extremely busy over the next few days, but I'll try to give it a run over this weekend. Happy to see this coming to fruition.

Please resolve the clang-tidy and clang-format issues.

Thanks!

Resolve comments

Please run clang-format, seems like there are a few issues sprinkled about.

Address comment

Rebase

Rebase

Resolve comments

Resolve comments

Address comments

Thanks.

Ah, I see that it already has an operator bool. Please still add the justification though. It's just as important to know why a change is happening as it is to know what is changing.

Please add justification in the description for why. Seems like this could be handled wherever Dialect is being constructed.

Rebase

Rebase

Rebase

format

format

Rebase

Fix accidental getSExtValue for unsigned

Thanks!!

Please run clang-format.

In D86592#2238849, @nihui wrote:

I do not have commit access, someone has to commit the change for me (with attribution). Thanks

Rebase

Address comments

In D85634#2238441, @frgossen wrote:

@rriddle, the mlir-cuda-runner tests are still failing because they materialize invalid casts. They did not fail earlier because there the verifier is not invoked when lowering to LLVM (?) and the the lowering for llvm.mlir.cast is actually able to deal with the index, it seems.

Has the fix you mentioned landed yet? I could not find it.

Address comments

The relationship between the op classes and registration feels under documented and easily error prone to new users. The fact that isa<AffineForOp>(op) will silently fail on an affine.for operation because the affine dialect isn't registered will likely feel wrong to many as an initial reaction. I'm not saying we should allow or encourage the use of the *Op classes when something is unregistered(we shouldn't), but it feels like we could make debugging/understanding a misconfiguration much easier; which if I recall is the original intention of the assert. E.g. as a bare bone hint, could we add a debug log that detects such situations and issues a message? e.g. something like: