[CIR][CIRGen] Handle AS in temp Allocas and override AS methods for AMDGPU

[RiverDave]

This PR does a couple of things that are closely related:

1. Address space casting for temporary allocas: Inserts address space casts from the alloca address space to the flat/generic address space when compiling for targets that have a non-default alloca address space (e.g., AMDGPU where allocas are in addrspace(5)). Similar to the way this is done in OG. where CreateMemTemp/CreateTempAlloca calls MaybeCastStackAddressSpace.

2. Helper to retrieve underlying AllocaOp: In cases where an address space cast wraps the alloca, the defining op is no longer an AllocaOp but a CastOp. Introduced Address::getAllocaOp() to walk through any casts and retrieve the underlying AllocaOp. This is needed when calling allocaOp.setInit() or allocaOp.setConstant() on addresses that may have been cast.

3. Preserve address spaces at CIR level: Ensures AMDGPU address spaces are correctly represented in CIR for both allocas and globals, with proper lowering to LLVM IR.

Happy Holidays! 🎄

[RiverDave]

• [CIR][CIRGen] Emit Address Spaces on RTTI globals #2093
• [CIR][CIRGen] Handle AS on function return allocas for heterogeneous languages #2090
• [CIR][CIRGen] Handle AS in temp Allocas and override AS methods for AMDGPU #2088 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.