Improve UnwindInfoTable's performance

[eduardo-vp]

Place the entries of UnwindInfoTable in a buffer and flush such that we amortize the cost of the operations (pRtlAddGrowableFunctionTable + pRtlDeleteGrowableFunctionTable) to create the internal table of RUNTIME_FUNCTIONs.

Contributes to #123124.

[eduardo-vp]

Storing in a buffer and flushing with two locks actually led to better performance of the test case mentioned in #123124 (comment).

Change	Time
Main	~580 ms
Buffer + flush with 2 locks	~200 ms

[Copilot]

Pull request overview

This PR optimizes CoreCLR’s Windows x64 unwind-info publishing by buffering RUNTIME_FUNCTION entries and batch-merging them into the OS growable function table, aiming to reduce overhead from frequent register/unregister operations during JIT.

Changes:

• Introduces a per-UnwindInfoTable pending buffer and a FlushPendingEntries() path to batch-merge unwind entries.
• Splits the single unwind-info lock into publish vs pending locks to reduce contention.
• Updates Crst type definitions / generated mappings for the new lock types.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
src/coreclr/vm/codeman.h	Adds pending buffer fields and refactors APIs around buffered publishing.
src/coreclr/vm/codeman.cpp	Implements pending-buffer accumulation, flush/merge logic, and new lock usage.
src/coreclr/inc/CrstTypes.def	Defines UnwindInfoTablePublishLock and UnwindInfoTablePendingLock and their ordering.
src/coreclr/inc/crsttypes_generated.h	Updates generated Crst enum/name/level maps to include new locks.

Comments suppressed due to low confidence (1)

src/coreclr/vm/codeman.cpp:489

• PublishUnwindInfoForMethod calls AddToUnwindInfoTable(...) and then calls FlushPendingEntries(), but AddToUnwindInfoTable already flushes at the end of the batch. This adds an extra lock acquisition on the publish path; consider removing the extra flush or restructuring so flushing happens in only one place.

        unwindInfo->AddToUnwindInfoTable(methodUnwindData, methodUnwindDataCount);
    }
}

/*****************************************************************************/

[Copilot]

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

Comments suppressed due to low confidence (1)

src/coreclr/vm/codeman.cpp:1

• Removal only searches the published table and ignores entries that are still in the pending buffer. If a method is unpublished before its unwind data is flushed, the pending entry can later be published to the OS (stale unwind info), and removal will incorrectly report not found. Fix by also searching/removing matching entries from the pending buffer under s_pUnwindInfoTablePendingLock (and/or forcing a flush before attempting removal) so unpublish is correct regardless of whether the entry was already published.

// Licensed to the .NET Foundation under one or more agreements.

[Copilot]

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

[jkotas]

What does the perf look like with the latest delta?

[eduardo-vp]

These are the results with the latest changes

Branch	Test execution time
Main	~580 ms
Current	~198 ms

[AaronRobinsonMSFT]

Two nits I'd personally like to see fix, but other than that, LGTM. Thanks!

[Copilot]

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

[eduardo-vp]

/ba-g unrelated failure