Use FIELD_ADDR for all ldfldas

[SingleAccretion]

I. e. add support for when the underlying address is local.

Diffs.

A small number of improvements due to:

1. Minor CSE changes.
2. Less aggressive promotion of implicit byrefs.

There are also some regressions in CoreCLR tests where we stop promoting locals with SIMD fields because FIELD_ADDRs don't resolve SIMD type handles. I think it is good TP tradeoff to keep it like that, considering the regressions are test-only and in non-test collections this leads to improvements.

There is also a nice TP win (~0.3%).

Fixes #77636.

[ghost]

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

Issue Details

---

I. e. add support for when the underlying address is local.

We're expecting a small number of improvements due to:

1. Minor CSE changes.
2. Less aggressive promotion of implicit byrefs.

There are also some regressions in CoreCLR tests where we stop promoting locals with SIMD fields because FIELD_ADDRs don't resolve SIMD type handles. I think it is good TP tradeoff to keep it like that, considering the regressions are test-only and in non-test collections this leads to improvements.

Author:	SingleAccretion
Assignees:	-
Labels:	area-CodeGen-coreclr, community-contribution
Milestone:	-

[SingleAccretion]

@dotnet/jit-contrib

[EgorBo]

@SingleAccretion can you post the link to the diff once it's ready, I see that previous diffs no longer available

[SingleAccretion]

Yes, will do.

[SingleAccretion]

Diffs have been obtained and still look like the ones in the description.

[EgorBo]

Nice TP wins!