Thanks for your review, Penporn! Keep those useful comments coming!

rebased + addressed comments

For ease of reviewing, I split out the invariant generalization in its own review: https://reviews.llvm.org/D91985

Nice to see this in action! Right in time also for some parallel work in the sparse compiler!

renamed A BC into S AB

alloc -> alloca, added a few more comments in top description

addressed remaining comments (except builders)
also rebased with the new Affine utils that make for shorter code

I addressed your remaining comments, except that I am pushing back (for now) on the insertion point vs. builder issue. Most obvious, since we don't have such a builder for while yet. But also, I am not convinced it will necessarily make for clearer code. Right now, every method leaves the insertion "cursor" at the right place for the next, even if that cursor can be something different (inside for, while, or if). I am not sure builders will improve on that, but could be convinced otherwise by trying once we actually have those abstractions. But let's not keep stalling this CL by trying to obtain a perfection that is also not met in other parts yet ;-)

broke up large genStmt() method into many genXXX() methods

rebased with new built-in annotations

addressed comments

Thanks Medhi.

fixed typo

addressed Alex' comments

In D90994#2392659, @ftynse wrote:

Thanks Aart!

I did one pass, have a bunch of nits to make this look more MLIResque in style. I agree with Nicolas that we should think about how to connect this better to the infrastructure and avoid a monolithic codegen that will grow huge over time (the same rationale as for affine code generation could apply). My knowledge of the TACO paper is residual at this point, so it's not always easy to track what is happening in the code the high level. Maybe having a relatively short comment blob that explains the algorithm without referring to the paper could be a good way for us to crystallize the understanding on how this can be fitted into MLIR. That being said, I wouldn't oppose landing a version that is correct + tests, and then iterating on refactoring and generalization while keeping the tests green.

In D91316#2390343, @mehdi_amini wrote:

Do you have tests for all this?

addressed missing "for"

minor cleanup

addresses comments, no more magic strings

few more nits

In D90994#2383314, @nicolasvasilache wrote:

Thanks for pushing on this Aart!
I realize this may sound like a lot of things that have to come together to get to a similar functionality but written in a more progressive and composable fashion.

fixed lint issue

LGTM once the others are approved

lowering next? ;-)

Thanks for adding the while!

Overall looks good, two more comments. Please find out what causes windows to fail (I am curious too what makes the CHECK different)

Did you upload the changes? Revision seems the same?
Also, please see the failure in the report.

bufferize it is!