JIT: Bypass bulk write barrier wrapper for small struct copies

[EgorBo]

Bypass the managed bulk-write-barrier wrapper for GC-containing struct copies up to 128 bytes in JIT, ReadyToRun, and NativeAOT. Larger copies retain the existing helper.

        mov      r8d, 40
-       call     [CORINFO_HELP_BULK_WRITEBARRIER]
+       call     CORINFO_HELP_BULK_WRITEBARRIER_SMALL
        nop

        add      rsp, 40
        ret

Benchmark - shows 5-15% improvement for 5 refs struct.

[dotnet-policy-service]

Tagging @dotnet/jit-contrib for JIT-EE GUID update

[azure-pipelines]

Azure Pipelines:
Successfully started running 6 pipeline(s).
10 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[dotnet-policy-service]

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

[EgorBo]

@jkotas does this make sense as a micro-opt?

The current Bulk WB is calling a managed impl (very likely via an indirect call). the managed impl does a size check and calls FCall. Given JIT always knows the size of the layout and in most cases it's not too big, we can bypass that transition.

An additional version can be made for "definitely on the heap (aka unchecked)" to avoid the "is on the heap" check, but that is probably more churn for less benefits.

PS: not sure we need to abstract BULK_WRITEBARRIER_SMALL_SIZE under a JIT-EE. Presumably it's not a correctness issue if they don't match with whatever is used in the managed impl. It's just have to be not too big.

[Copilot]

Copilot review overview

🔵 Needs a closer look

Preserve GC polling for repeated small copies and add regression coverage for the 512-byte and overlapping-copy cases.

Review effort: Lite
Findings: None

What changed in this PR

Adds a raw write-barrier path for small GC-containing struct copies while preserving existing fallbacks.

Changes:

• Registers and versions the new helper.
• Routes copies up to 512 bytes through the raw worker where supported.
• Updates CoreCLR, NativeAOT, ReadyToRun, and WASM mappings.

File	Reviewed changes
src/​coreclr/​vm/​corelib.h	Registers the internal buffer helper.
src/​coreclr/​tools/​Common/​JitInterface/​CorInfoHelpFunc.cs	Adds the helper enum entry.
src/​coreclr/​tools/​aot/​ILCompiler.RyuJit/​JitInterface/​CorInfoImpl.RyuJit.cs	Uses the raw worker on 64-bit NativeAOT.
src/​coreclr/​tools/​aot/​ILCompiler.ReadyToRun/​JitInterface/​CorInfoImpl.ReadyToRun.cs	Preserves the wrapper fallback.
src/​coreclr/​jit/​utils.cpp	Marks the helper as heap-mutating.
src/​coreclr/​jit/​lower.cpp	Selects the raw helper for small copies.
src/​coreclr/​jit/​gentree.cpp	Handles VEX state for the helper.
src/​coreclr/​jit/​codegenwasm.cpp	Adds the WASM helper signature.
src/​coreclr/​jit/​codegencommon.cpp	Includes the helper in debug queries.
src/​coreclr/​inc/​jithelpers.h	Registers the dynamic helper.
src/​coreclr/​inc/​jiteeversionguid.h	Updates interface versioning.
src/​coreclr/​inc/​corinfo.h	Adds the native helper entry and contract.

[Copilot]

Copilot review overview

🟡 Changes recommended

lower.cpp has unresolved critical WASM GC-polling and moderate DEBUG GC-stress issues.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1

Open (1)

• Wasm bulk stores can bypass GC polling indefinitely · New

[Copilot]

Copilot review overview

🟡 Changes recommended

CoreLib ReadyToRun rooting, NativeAOT WASM entry-point, and no-GC metadata issues remain unresolved.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 2

Open (2)

• Add new helper root to ReadyToRunJitHelperRootProvider · New
• Add missing BulkMoveWithWriteBarrierInternal to WASM Test.CoreLib · New

Resolved since last review (1)

• Wasm bulk stores can bypass GC polling indefinitely

[Copilot]

Copilot review overview

🟡 Changes recommended

The helper’s GC-safepoint classification requires correction, and boundary plus AOT/ReadyToRun regression coverage is missing.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1

Open (1)

• Mark raw bulk write barrier helper as no-GC · New

Resolved since last review (2)

• Add missing BulkMoveWithWriteBarrierInternal to WASM Test.CoreLib
• Add new helper root to ReadyToRunJitHelperRootProvider

[Copilot]

Copilot review overview

🔵 Needs a closer look

Three unresolved moderate findings remain: incomplete ReadyToRun rooting, missing safepoint polling, and missing cutoff GC-stress coverage.

Review effort: Lite
Findings: None

Resolved since last review (1)

• Mark raw bulk write barrier helper as no-GC

[Copilot]

Copilot review overview

🔵 Needs a closer look

Address the missing boundary coverage and preserve a GC poll for repeated small copies.

Review effort: Lite
Findings: None

Previously missed (1)

In code that hasn't changed since last review

Add boundary and AOT coverage for 128-byte helper mapping

src/​coreclr/​jit/​lower.cpp:10402

This introduces a new 128-byte boundary and deliberately removes the GC poll for the <=128-byte path, but there is no targeted coverage for the exact 128/129 split or for the new helper through ReadyToRun/NativeAOT. Please add a focused runtime/codegen test covering both sides of the boundary so an off-by-one or incomplete helper mapping cannot silently change GC safepoint behavior.