[clang] [llvm] [Offload][Driver] Split frontend CUDA/HIP language assumptions from backend device (PR #212373)

[Matt Arsenault via llvm-commits]

================
@@ -1439,13 +1457,11 @@ void CGNVCUDARuntime::emitOffloadProfilingSections() {
     OffloadProfSectionShadows.push_back({Shadow, DeviceName.str()});
   };
 
-  // Keep this order in sync with the runtime: data, counters, uniform counters,
-  // then names.
+  // Keep this order in sync with the runtime: data, counters, then names.
----------------
arsenm wrote:

Unrelated comment changes?

https://github.com/llvm/llvm-project/pull/212373