Nice.. not sure why at the time I thought this was sufficient to do for external decls only..

I see my comment has not been addressed yet.

Further test reduction

Reduced test, and added asm roundtrip test

Forgot the Differential Revision: , it has been pushed to main.

This fixes https://github.com/llvm/llvm-project/issues/42472 btw

Addressed comment

thanks for the fix!

I tried, but this will take me longer to set up, which I can't do right now. I may be able to return to this later if I am still blocked by it.

It's local.

(it is apparently 128).

This assert is failing on VS2019 x64

This has landed, not sure why Phabricator doesn't notice..

Simplified using use_nodbg_empty

There might be a more elegant way to test if an instruction still has users for its def/reg, anyone know?

Removed unrelated fix and formatting

As for the changes to DW_AT_low/high_pc the code generated for this test is:

Apologies for pushing this to main, I did not see there was no "accepted" status yet, though did check with @dschuff offline that he was ok with this patch as-is.

Yes, it must always be run just before AsmPrinter.

@aeubanks looked up the use of this pass, and the comments next to its usage say:

Also there is no MachineModule.. MachineModuleInfo is the closest thing.. that may explain why there's no MachineModulePass. The code to obtain MachineModuleInfo is a bit clumsy though, but I've seen other passes use it this way too.

Well.. yes :) The code runs, and unlike before, causes all symbols for all functions to be collected before MCLowering.
It passes all existing tests that expect __stack_pointer, and the new test fails without this patch because the .globaltype is not emitted, much like in the reported bug.
Do you recommend anyone to add to this review to see if I am not committing a grave LLVM-no-no?

Fix type error; Fix original test had addr & value swapped for i32.store (discovered by the asm type checker in the 64-bit version :)

factored out repeating use of is64

Ok, I can land this as-is, with the few warts it has. Might be easier to address things in followups given the size.

Addressed code review.

Addressed most of sbc code review

This is a first version that passes all (modified) tests. There is still a few things wrong with this, but I thought I'd get some feedback on this direction as it stands.

reworked test for .s format and rebasing

fixed --check-prefixes

Addressed code review.

Yes, this makes sense to me. Another way to put it is that our br_if has the side effect of forcing termination of a "register" lifetime, which is uncommon.

Thanks for doing this, was sorely needed :)

Thanks, that would make sense, since this test was probably adapted from the 32-bit version.

I can land it.. let me pull it in locally and run tests to be sure.