JIT: Replace fgMorphTree with fgMorphArgs for calls introduced after rationalization

[Copilot]

Post-rationalization phases (lowering, async transformation, GS cookie checks) were calling fgMorphTree on freshly created GenTreeCall* nodes. fgMorphTree routes through fgMorphCall, which performs tail-call analysis, inlining marking, and GC safe-point flagging — all irrelevant or already completed before rationalization. The correct operation for these sites is fgMorphArgs, which handles only ABI info and argument processing.

Description

Changed files: lower.cpp, async.cpp, gschecks.cpp, compiler.h

• lower.cpp — Replace fgMorphTree with fgMorphArgs at four call sites:

  • LowerVirtualVtableCall (interface lookup resolution call)
  • LowerCFGCall (CFG validate-indirect call)
  • InsertPInvokeCallProlog / InsertPInvokeCallEpilog (pinvoke helper calls)
  • For InsertPInvokeMethodProlog, the original fgMorphTree(store) (where store wraps the call) is replaced with fgMorphArgs(call) directly — the store wrapper needs no morphing in LIR mode, and the original was discarding the return value of fgMorphTree anyway
  • Update stale "A hacky way to safely call fgMorphTree in Lower" comments on existing fgMorphArgs sites to reflect the actual intent

• async.cpp — Replace fgMorphTree with fgMorphArgs for all seven call nodes created during async state-machine transformation (continuation allocation, context capture/restore, finish, rethrow)

• gschecks.cpp — Replace fgMorphTree with fgMorphArgs for the special-copy-helper call inserted for unsafe buffer GS checks

• compiler.h — Move fgMorphArgs declaration from private to public alongside fgMorphTree; AsyncTransformation is not a friend of Compiler so it requires public access

[dotnet-policy-service]

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