[llvm] [AMDGPU] Support divergent sized dynamic alloca (PR #121148)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,170 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX9 %s
----------------
arsenm wrote:

It's rare that it's useful to check the raw output of the selection, and this isn't one of those cases. At most we would want one function to show the temporary pseudo, but these really should show the alloca works end to end 

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