JIT: Handle return buffer definitions in TreeLifeUpdater

[jakobbotsch]

Fix #116135

[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 addresses issue #116135 by handling return buffer definitions in TreeLifeUpdater.

• Removed outdated TODO-Bug comment regarding handling calls with return buffers.
• Added an else-if branch to obtain the return buffer local variable for call trees using gtCallGetDefinedRetBufLclAddr.

Comments suppressed due to low confidence (1)

src/coreclr/jit/treelifeupdater.cpp:307

• Ensure that existing tests cover scenarios where call trees include return buffer definitions. Adding dedicated tests would validate the new behavior introduced in this branch.

else if (tree->IsCall())

[Copilot]

[nitpick] Consider adding a comment explaining the purpose and expected behavior of gtCallGetDefinedRetBufLclAddr, to improve maintainability.

[jakobbotsch]

cc @dotnet/jit-contrib PTAL @AndyAyersMS

Minor asm diffs.

Some large Windows TP regressions. Detailed TP diff is:

Base: 15998499578, Diff: 16018881868, +0.1274%

63407795  : +6026.77% : 59.51% : +0.3963% : protected: void __cdecl CodeGenInterface::genUpdateLife(struct GenTree *)                                                  
-322342   : -2.68%    : 0.30%  : -0.0020% : private: void __cdecl emitter::emitInsLoadInd(enum instruction, enum emitAttr, enum _regNumber_enum, struct GenTreeIndir *)
-3453744  : -11.80%   : 3.24%  : -0.0216% : protected: void __cdecl CodeGen::genCodeForStoreLclVar(struct GenTreeLclVar *)                                             
-18334466 : -9.11%    : 17.21% : -0.1146% : protected: enum _regNumber_enum __cdecl CodeGen::genConsumeReg(struct GenTree *)                                           
-20884892 : -16.25%   : 19.60% : -0.1305% : protected: void __cdecl CodeGen::genProduceReg(struct GenTree *)                                                           

Appears to be because of different inlining decisions, so will leave that up to native PGO. linux-x64 TP diffs also do not show significant impact.