[llvm] [AMDGPU][SROA] Expand cast chain handling to floating point types (PR #228113)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,131 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt -S -mtriple=amdgpu9.00-amd-amdhsa -passes=amdgpu-promote-alloca-to-vector,verify < %s | FileCheck %s
+
+; Promoting these allocas to a floating-point vector forces a conversion between
+; a pointer and a floating-point value or vector of the same size. A bitcast
+; cannot convert between pointers and non-pointers, so the conversion must
+; route through a pointer-sized integer via ptrtoint/inttoptr instead of
+; producing e.g. `bitcast ptr to <2 x float>`.
+
+target datalayout = "e-p:64:64-p5:32:32-A5"
----------------
arsenm wrote:

```suggestion
```

https://github.com/llvm/llvm-project/pull/228113