[clang] [CIR][SYCL] Device kernel caller (PR #213771)

[Konstantinos Parasyris via cfe-commits]

koparasy wrote:

> @koparasy, it's cool to see you working on adding this support to CIR! I wasn't aware you were working on this. I just discovered commit [8ba3d8f](https://github.com/llvm/llvm-project/commit/8ba3d8f5992656764f404998926bb9e6bb842851). Please add me as a reviewer to future SYCL related PRs.
> 
> There will be significant changes coming soon that change the semantics of the `SYCLKernelCallStmt`. A PR for one of them has been posted as #208958. We've been holding off merging that one until we get another PR done that will conflict with it. That one will add support for SYCL "special types" and will modify the protocol used on both the host and device sides to coordinate with the SYCL runtime. You might want to wait for those to land before doing more CIR work for them.
> 
> I haven't reviewed these changes closely yet, but will try to do so early next week. I'm still pretty uneducated regarding CIR, so I might have questions.

Thanks for the heads-up, and happy to add you as a reviewer on future SYCL-related CIR PRs.

On #208958: I took a look at the diff. My two CIR tests (`clang/test/CIR/CodeGenSYCL/kernel-call-stmt.cpp` and `kernel-caller-entry-point.cpp`) will break once it lands. They declare the old `sycl_kernel_launch(const char *, Ts...)` signature and hard-code the corresponding launch mangling, and they also run an OGCG prefix that cross-checks classic CodeGen output, so they'd go red on your change. The fix will be mechanical.

https://github.com/llvm/llvm-project/pull/213771