[llvm] [AMDGPU] don't mark control-flow intrinsics as convergent (PR #90026)

[Matt Arsenault via llvm-commits]

================
@@ -3143,25 +3143,37 @@ def int_amdgcn_cvt_sr_fp8_f32 : ClangBuiltin<"__builtin_amdgcn_cvt_sr_fp8_f32">,
 // Special Intrinsics for backend internal use only. No frontend
 // should emit calls to these.
 // ===----------------------------------------------------------------------===//
+//
+// This is really a workaround to allow control flow lowering in the presence of
----------------
arsenm wrote:

This comment starts with no context 

https://github.com/llvm/llvm-project/pull/90026