Never set fgPgoSingleEdge for cctors or large single-edge blocks

[EgorBo]

Static cctors tend to be huge and single-edge (e.g. some List or Dictionary static field initialized with a lot of items inline) - if they're compiled with JIT optimizations (TC=0 or dynamic context) - they should not receive an extra inlining boost from #116054. Same for just large single-edge methods.

[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 refines the JIT’s optimization mask and enhances profile synthesis to avoid giving inlining boosts to static constructors or large single-edge methods.

• Reduced the set of flags in CLFLG_MAXOPT to a smaller subset.
• In ProfileSynthesis::Run, skip fgPgoSingleEdge for static constructors, size-optimized methods, or methods with many calls.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
src/coreclr/jit/jit.h	Updated CLFLG_MAXOPT to only include a few core optimizations
src/coreclr/jit/fgprofilesynthesis.cpp	Added checks to disable single-edge PGO for cctors, size-opt, and call-heavy methods

Comments suppressed due to low confidence (2)

src/coreclr/jit/jit.h:757

• Add or update a comment above this macro explaining why certain flags were removed and what the new target optimization level represents.

#define CLFLG_MAXOPT (CLFLG_REGVAR | CLFLG_TREETRANS | CLFLG_INLINING | CLFLG_STRUCTPROMOTE)

src/coreclr/jit/fgprofilesynthesis.cpp:256

• Consider adding tests or a profiling scenario to verify that static constructors and methods exceeding the call threshold are not granted fgPgoSingleEdge.

    m_comp->fgPgoSingleEdge = !isCctor && !preferSize;

[EgorBo]

@AndyAyersMS anything else? cc @dotnet/jit-contrib

[AndyAyersMS]

Did you try it out on methods like the .cctor that we looked at the other day?

From the diffs at least it looks like it is doing what we expect.

[EgorBo]

Did you try it out on methods like the .cctor that we looked at the other day?

From the diffs at least it looks like it is doing what we expect.

Yep, checked locally.