JIT: Consider removing GT_JMP

[jakobbotsch]

GT_JMP is an unfortunate node because it has implicit uses of all arguments, which requires a lot of special handling throughout the JIT (and causes bugs like #80731).

If we instead import jmp IL instructions as explicit tailcalls we should be able to get rid of this node and make the dependencies explicit. This would first require us to add fast tailcall support on x86 when the stack size of the caller and callee is the same. All other targets already have more general support for fast tailcalls.

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

Issue Details

---

GT_JMP is an unfortunate node because it has implicit uses of all arguments, which requires a lot of special handling throughout the JIT (and causes bugs like #80731).

If we instead import jmp IL instructions as explicit tailcalls we should be able to get rid of this node and makes the dependencies explicit. This would first require us to add fast tailcall support on x86 when the stack size of the caller and callee is the same.

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

[jakobbotsch]

This would first require us to add fast tailcall support on x86 when the stack size of the caller and callee is the same.

This is not strictly true, we could use the existing helper-based tailcall mechanism, but that would be a regression and give worse performance than normal calls which probably defeats half the purpose of the jmp instruction.