[llvm] [AMDGPU] Support dynamically sized allocas (PR #121148)

[Matt Arsenault via llvm-commits]

================
@@ -1190,9 +1190,14 @@ bool AMDGPURegisterBankInfo::applyMappingDynStackAlloc(
 
   const RegisterBank *SizeBank = getRegBank(AllocSize, MRI, *TRI);
 
-  // TODO: Need to emit a wave reduction to get the maximum size.
-  if (SizeBank != &AMDGPU::SGPRRegBank)
-    return false;
+  if (SizeBank != &AMDGPU::SGPRRegBank) {
+    auto WaveReduction =
+        B.buildIntrinsic(Intrinsic::amdgcn_wave_reduce_umax,
+                         {LLT::scalar(MRI.getType(AllocSize).getSizeInBits())})
----------------
arsenm wrote:

```suggestion
                         {MRI.getType(AllocSize)})
```

This is extra steps, but you could also just hardcode this to S32

https://github.com/llvm/llvm-project/pull/121148