JIT: Store unique switch successors in BBswtDesc

[amanasifkhalid]

• Rename BBehfDesc to BBJumpTable, and repurpose it for describing blocks with N successors. This type allows unique successor access, enumeration, and removal.
• Have BBswtDesc inherit from BBJumpTable. Through BBswtDesc's updated API surface, we can enumerate unique successors and switch cases.
• Remove Compiler's API surface for mapping switch blocks to their unique successors.

There are numerous points in the JIT where we want to iterate unique successors, but we tolerate iterating duplicates out of fear of needing to construct the switch successor map. We can now iterate unique successors at those sites without needing to precompute anything. I'll do that switchover in a follow-up PR.

This change is regrettably large for one PR, though most of the changes are trivial adjustments to match the new API surface.

[dotnet-policy-service]

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

[Copilot]

Pull Request Overview

This PR refactors switch and EH-finally successor tracking by introducing BBJumpTable for blocks with N successors and updating BBswtDesc to inherit from it, removing the old BBehfDesc and switch-descriptor map APIs.

• Replace BBehfDesc with BBJumpTable everywhere and migrate EH-finally successor handling.
• Update BBswtDesc to store both the full case list and a unique successor list via BBJumpTable.
• Remove compiler-level switch-desc map and enable direct unique-successor enumeration through the new API.

Reviewed Changes

Copilot reviewed 19 out of 19 changed files in this pull request and generated 1 comment.

File	Description
block.h	Added BBJumpTable, updated switch/EH descriptor types and APIs.
switchrecognition.cpp	Updated switch conversion to allocate and initialize BBJumpTable.
optimizer.cpp, morph.cpp, lower.cpp, importer.cpp, fg*.cpp, codegencommon.cpp, assertionprop.cpp	Replaced old descriptor field accesses with new GetSuccs/GetCases and API calls.

Comments suppressed due to low confidence (3)

src/coreclr/jit/block.h:2298

• Add a brief doc comment above BBJumpTable explaining its role as a generic descriptor for blocks with N successors, how it differs from BBswtDesc, and its lifecycle/ownership.

struct BBJumpTable

src/coreclr/jit/block.h:2319

• There are no existing unit tests that exercise the new BBJumpTable::GetSuccs/RemoveSucc/RemoveDefaultCase APIs. Consider adding tests to validate unique-successor enumeration and mutation under both switch and EH-finally scenarios.

    FlowEdge** GetSuccs() const

src/coreclr/jit/block.h:1412

• [nitpick] This C-style cast could be replaced with static_cast<BBJumpTable*> for clarity and safety.

    //    for (BasicBlock* const bTarget : block->SwitchCases()) ...

[amanasifkhalid]

@dotnet/jit-contrib PTAL. Small diffs. Some components of the JIT, such as our DFS computations, are sensitive to the order in which unique block successors are enumerated. Thus, I tried to preserve the current order of switch successors at each point where we compute them, but because we no longer need to recompute them when flow changes, the ordering can change if the switch's case labels are modified. This seems to rarely make a difference in codegen, so I think we ought to tolerate it.

[amanasifkhalid]

ping @dotnet/jit-contrib

[amanasifkhalid]

@AndyAyersMS could you PTAL? Thanks

[AndyAyersMS]

I wrote these up last week but apparently never submitted them.

[amanasifkhalid]

@AndyAyersMS thank you for the review -- I've addressed your feedback

[amanasifkhalid]

/ba-g build timeouts, #117724