JIT: Avoid unnecessary GTF_GLOB_REFs

[jakobbotsch]

• Avoid setting GTF_GLOB_REF on GT_FIELD_ADDR nodes
• Avoid setting GTF_GLOB_REF on GT_FIELD nodes off of implicit byrefs. This is ok now since implicit byref morphing indiscriminately sets GTF_GLOB_REF.
• Manually clone a "pointer to span" in span intrinsic expansion when it points to a local. Unfortunately this does not fall out from the above since gtClone does not handle FIELD_ADDR, and making it handle this needs some more work.

These changes are necessary to avoid address exposure in the two user benchmarks in #83388:

• JIT: Add a (disabled) prototype for a generalized promotion pass #83388 (comment) was address exposing a span field from the span intrinsic importation.
• JIT: Add a (disabled) prototype for a generalized promotion pass #83388 (comment) was spilling a field address around a call, also causing address exposure.

Fix#74563
Fix#856

[ghost]

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

Issue Details

---

• Avoid setting GTF_GLOB_REF on GT_FIELD_ADDR nodes
• Avoid setting GTF_GLOB_REF on GT_FIELD nodes off of implicit byrefs. This is ok now since implicit byref morphing indiscriminately sets GTF_GLOB_REF.

These changes are necessary to avoid address exposure in the two user benchmarks in #83388:

• JIT: Add a (disabled) prototype for a generalized promotion pass #83388 (comment) was address exposing a span field from this code:
  

  runtime/src/coreclr/jit/importercalls.cpp

  Lines 2884 to 2885 in 66211eb

  	ptrToSpan = impCloneExpr(ptrToSpan, &ptrToSpanClone, NO_CLASS_HANDLE, CHECK_SPILL_ALL,
  	nullptrDEBUGARG("Span.get_Item ptrToSpan"));

• JIT: Add a (disabled) prototype for a generalized promotion pass #83388 (comment) was spilling a field address around a call, also causing address exposure.

Both of these cases are fixed with this PR.

Fix #74563

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

[jakobbotsch]

/azp run Fuzzlyn, runtime-coreclr jitstress, runtime-coreclr libraries-jitstress

[azure-pipelines]

Azure Pipelines successfully started running 3 pipeline(s).

[jakobbotsch]

The hardware intrinsics test failure should be fixed by #84352. The Fuzzlyn failure is preexisting.

Diffs.

cc @dotnet/jit-contrib PTAL @SingleAccretion@EgorBo

[SingleAccretion]

Almost forgot - this also fixes #856.