Ensure CreateScalarUnsafe gets constant handling on Arm64

[tannergooding]

No description provided.

[ghost]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Issue Details

---

null

Author:	tannergooding
Assignees:	-
Labels:	area-CodeGen-coreclr
Milestone:	-

[tannergooding]

@kunalspathak, could you assist with this?

I was working on a different PR but was getting failures, so I've isolated it down to something that is trivial and "should" work. But which raises an assert in the Arm64 LSRA

[kunalspathak]

I will check it out @tannergooding

[kunalspathak]

@tannergooding - This should unblock you.

[kunalspathak]

Earlier, the interval corresponding to CreateScalarUnsafe was a regular one, but with this change, it is a "constant interval". If a register holds a constant value and if we ever have to save upper half of this vector register, we should clear the constant value from the register as well. We were not doing it and that caused the assert failure. The fix was to clear it as well.

[tannergooding]

Thanks much for the fix!

[tannergooding]

This should be ready for review. CC. @dotnet/jit-contrib

Notably SPMI diffs has marked everything as "missed". I've not seen this before

[jakobbotsch]

Notably SPMI diffs has marked everything as "missed". I've not seen this before

The missed contexts look to be just a small fraction of the overall contexts. It is expected due to the revert #76235 that put us back at an old JIT-EE GUID, so the collection we are using now is quite old (from last Sunday) and JIT changes since then are causing some misses.

[tannergooding]

Thanks! That's a bit unfortunate as I was expecting a few diffs to show up, but I think that's fine overall

[kunalspathak]

LGTM