[clang] [SYCL] The sycl_kernel_entry_point attribute. (PR #111389)

[Tom Honermann via cfe-commits]

================
@@ -407,7 +407,8 @@ def MicrosoftExt : LangOpt<"MicrosoftExt">;
 def Borland : LangOpt<"Borland">;
 def CUDA : LangOpt<"CUDA">;
 def HIP : LangOpt<"HIP">;
-def SYCL : LangOpt<"SYCLIsDevice">;
+def SYCLHost : LangOpt<"SYCLIsHost">;
----------------
tahonermann wrote:

I've never loved those names either. I don't think it makes sense to change them as part of this PR though. https://github.com/llvm/llvm-project/pull/107493 is introducing new `-fsycl-device-only` and `-fsycl-host-only` options as aliases of `--offload-device-only` and `--offload-host-only` (also aliased by `--cuda-device-only` and `--cuda-host-only`). It might make sense to argue for a change in that PR. Perhaps there is an opportunity to remove the `-fsycl-is-device` and `-fsycl-is-host` options in favor of a common offload indicator of host vs device compilation.

For the purposes of this PR, I think it makes sense to stick with the status quo pending a proposal for a better, perhaps unified, way of handling host vs device compilation.

https://github.com/llvm/llvm-project/pull/111389