[llvm] [llvm] Preserve existing calling conv when auto-upgrading nvvm.annota… (PR #125568)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,14 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -mtriple=amdgcn-amd-amdhsa -O0 -S | FileCheck %s
----------------
arsenm wrote:

This does not belong in test/CodeGen/AMDGPU.

https://github.com/llvm/llvm-project/pull/125568