Remove ArrLen propagation from earlyprop

[EgorBo]

It looks unnecessary given all downstream phases are fine with just Value Numbering

Diffs

[dotnet-policy-service]

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

[hez2010]

I remember the ArrLen early propagation also contributes to loop optimizations like unrolling, which happens before VN.

[EgorBo]

I remember the ArrLen early propagation also contributes to loop optimizations like unrolling, which happens before VN.

Loop unrolling happens before early prop

[Copilot]

Pull Request Overview

This PR removes the array length propagation feature from the early propagation phase and cleans up related debugging and null-check folding logic. Key changes include:

• Removal of the SSA-based array length propagation and associated debugging/SSA update code.
• Updating constant range checks to use CORINFO_Array_MaxLength instead of INT32_MAX.
• Refactoring the null-check folding functions and the signature of optCanMoveNullCheckPastTree to remove an unused parameter.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/coreclr/jit/earlyprop.cpp	Removed array length propagation code and refactored debug output and null-check handling.
src/coreclr/jit/compiler.h	Updated function signature for optCanMoveNullCheckPastTree to match the refactored implementation.

Comments suppressed due to low confidence (2)

src/coreclr/jit/earlyprop.cpp:217

• The change from INT32_MAX to CORINFO_Array_MaxLength for the upper bound check is a critical update; please verify that CORINFO_Array_MaxLength is defined consistently and used appropriately across all relevant platforms.

if ((actualConstVal < 0) || (actualConstVal > CORINFO_Array_MaxLength))

src/coreclr/jit/earlyprop.cpp:262

• The removal of the tree cloning, SSA updates, and the associated constant propagation logic in optEarlyPropRewriteTree significantly changes the propagation strategy; please ensure that the updated value numbering phase fully addresses the necessary optimizations.

JITDUMP("optEarlyProp Rewriting " FMT_BB "\n", compCurBB->bbNum);

[EgorBo]

PTAL @jakobbotsch @dotnet/jit-contrib a few diffs

I've removed GT_ARR_LENGTH forward substitution (SSA-based) from early-prop, the diffs imply it was not useful anyway. Also, removed a few dead vars/args.

There is still a big chunk of code in the early prop responsible for constant-folding of GT_BOUNDS_CHECK (when index is constant and the length is constant at its SSA def).