Cleanup VisitOperandUses and GenTreeVisitor to check common kinds rather than be a big switch

[tannergooding]

This should notably reduce risk as well. Some handling had been missed before like GT_LZCNT not being handled and hitting the default binary path instead.

TP impact on Linux x64

linux x64
* Overall (-0.09% to -0.04%)
* MinOpts (-0.10% to +0.04%)
* FullOpts (-0.11% to -0.04%)

TP impact on Windows x64

linux arm64
* Overall (-0.21% to -0.15%)
* MinOpts (-0.14% to -0.04%)
* FullOpts (-0.26% to -0.16%)

linux x64
* Overall (-0.23% to -0.15%)
* MinOpts (-0.16% to -0.02%)
* FullOpts (-0.28% to -0.15%)

osx arm64
* Overall (-0.21% to -0.16%)
* MinOpts (-0.13% to -0.06%)
* FullOpts (-0.22% to -0.16%)

windows arm64
* Overall (-0.21% to -0.15%)
* MinOpts (-0.14% to -0.04%)
* FullOpts (-0.27% to -0.16%)

windows x64
* Overall (-0.23% to -0.15%)
* MinOpts (-0.16% to -0.02%)
* FullOpts (-0.29% to -0.15%)

Windows x86 has a TP hit, from -0.00% to +0.06% for MinOpts, but that should be fine given the wins we see elsewhere and the improved robustness for the visitors when new node kinds are introduced.

[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 GenTree::VisitOperandUses to use OperIsLeaf / OperIsBinary / OperIsUnary fast-path checks instead of an explicit per-op switch, leaving only GTK_SPECIAL nodes handled in a smaller switch.

Changes:

• Replace the large OperGet() switch with kind-based checks (leaf/binary/unary/special) in VisitOperandUses.
• Centralize operand walking for most nodes via AsOp() / AsUnOp() handling.
• Keep explicit operand iteration for special-structure nodes (e.g., call args, phi uses, field list, cmpxchg, arr elem).

Comments suppressed due to low confidence (1)

src/coreclr/jit/compiler.hpp:4438

• OperIsSpecial() includes ARM64 conditional nodes GT_SELECT_INC, GT_SELECT_INV, and GT_SELECT_NEG (see gtlist.h), but the OperGet() switch here only handles GT_SELECT. As a result, VisitOperandUses will hit the default case for these nodes and skip visiting their operands (and GT_SELECT_* can have an optional gtOp2 that still needs visiting when present). Please add explicit cases for GT_SELECT_INC/INV/NEG (and handle their optional gtOp2), so operand walking remains correct on ARM64.

    assert(OperIsSpecial());

    switch (OperGet())
    {
#if defined(FEATURE_HW_INTRINSICS)
        case GT_HWINTRINSIC:
            for (GenTree** use : this->AsMultiOp()->UseEdges())
            {

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.