JIT: revise may/must point to stack analysis

[AndyAyersMS]

Keep track of locals that may point to the heap as well as locals that may point to the stack.

The set of definitely stack-pointing locals is then the difference in these two sets. This handles the "cyclic" case where stack pointing locals are assigned to one another.

Fixes #115017

[Copilot]

Pull Request Overview

This PR improves JIT analysis by tracking locals that may point to both the heap and the stack to better compute the set of definitely stack-pointing locals, addressing an issue with cyclic assignments.

• Added a new test case to verify the updated connection cycles behavior.
• Updated object allocation logic in objectalloc.cpp to compute stack-pointing locals based on the difference with heap-pointing locals and added corresponding logging.
• Enhanced logging in assertionprop.cpp for better cross-block table insights.

Reviewed Changes

Copilot reviewed 3 out of 4 changed files in this pull request and generated 1 comment.

File	Description
src/tests/JIT/opt/ObjectStackAllocation/ConnectionCycles.cs	Added tests for verifying connection cycles via local pointer assignments.
src/coreclr/jit/objectalloc.cpp	Updated logic to distinguish stack- and heap-pointing locals with new logging messages.
src/coreclr/jit/assertionprop.cpp	Added a logging statement to report cross-block table size details.

Files not reviewed (1)

• src/tests/JIT/opt/ObjectStackAllocation/ConnectionCycles.csproj: Language not supported

Comments suppressed due to low confidence (1)

src/tests/JIT/opt/ObjectStackAllocation/ConnectionCycles.cs:20

• [nitpick] Consider renaming the test method 'Problem' to a more descriptive name that reflects the scenario under test, improving clarity for future readers.

public static int Problem()

[Copilot]

Consider adding an inline comment to clarify that subtracting the possibly heap-pointing pointers from the possibly stack-pointing pointers isolates the definitely stack-pointing locals, especially to support future maintenance.

[dotnet-policy-service]

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

[AndyAyersMS]

@jakobbotsch PTAL
cc @dotnet/jit-contrib

This reduces the number of ambient byrefs, and can also improve code quality.

Two bad diffs in x64 windows libraries SPMI (System.MemoryTests) where we now fill up the local AP table and so drop a couple of useful assertions.

[AndyAyersMS]

arm failure is related, there are diffs in VerifyUriNormalizationForEscapedCharacters

    System.PrivateUri.Tests.IriTest.Iri_UnicodePlane3_13 [FAIL]
      System.ArgumentNullException : Value cannot be null. (Parameter 'value')
      Stack Trace:
        /_/src/libraries/System.Private.CoreLib/src/System/ArgumentNullException.cs(97,0): at System.ArgumentNullException.Throw(String paramName)
        /_/src/libraries/System.Private.CoreLib/src/System/Globalization/CultureInfo.cs(374,0): at System.Globalization.CultureInfo.set_CurrentCulture(CultureInfo value)
        /_/src/libraries/System.Private.Uri/tests/FunctionalTests/IriTest.cs(331,0): at System.PrivateUri.Tests.IriTest.VerifyUriNormalizationForEscapedCharacters(String uriInput)
        /_/src/libraries/System.Private.Uri/tests/FunctionalTests/IriTest.cs(265,0): at System.PrivateUri.Tests.IriTest.EscapeUnescapeTestUnicodePlane(Int32 start, Int32 end, Int32 step)
        /_/src/libraries/System.Private.Uri/tests/FunctionalTests/IriTest.cs(246,0): at System.PrivateUri.Tests.IriTest.Iri_UnicodePlane3_13()
        /_/src/coreclr/System.Private.CoreLib/src/System/Reflection/MethodBaseInvoker.CoreCLR.cs(36,0): at System.Reflection.MethodBaseInvoker.InterpretedInvoke_Method(Object obj, IntPtr* args)
        /_/src/libraries/System.Private.CoreLib/src/System/Reflection/MethodBaseInvoker.cs(57,0): at System.Reflection.MethodBaseInvoker.InvokeWithNoArgs(Object obj, BindingFlags invokeAttr)
Fatal error.
System.AccessViolationException: Attempted to read or write protected memory. This is often an indication that other memory is corrupt.
   at System.Threading.AsyncLocal`1[[System.__Canon, System.Private.CoreLib, Version=10.0.0.0, Culture=neutral, PublicKeyToken=7cec85d7bea7798e]].System.Threading.IAsyncLocal.OnValueChanged(System.Object, System.Object, Boolean)
   at System.Threading.ExecutionContext.SetLocalValue(System.Threading.IAsyncLocal, System.Object, Boolean)
   at System.Threading.AsyncLocal`1[[System.__Canon, System.Private.CoreLib, Version=10.0.0.0, Culture=neutral, PublicKeyToken=7cec85d7bea7798e]].set_Value(System.__Canon)
   at System.Globalization.CultureInfo.set_CurrentCulture(System.Globalization.CultureInfo)
   at System.PrivateUri.Tests.IriTest.VerifyUriNormalizationForEscapedCharacters(System.String)

[AndyAyersMS]

We are missing a connection graph edge for the case where a local doesn't escape but we decide not to stack allocate it (say because the allocation site is in a loop).

[jakobbotsch]

I think you can use BitVecOps::Diff directly here to avoid manually writing the copy.

[AndyAyersMS]

Good point. Will clean this up in a future escape PR.