svenvh added inline comments.

================
Comment at: clang/lib/Headers/opencl-c.h:17161
+#if (defined(__OPENCL_CPP_VERSION__) || __OPENCL_C_VERSION__ == 200)
+#undef __opencl_c_pipes
+#undef __opencl_c_generic_address_space
----------------
Anastasia wrote:
> Looping in @svenvh  - I don't mind if we define those macros in headers for 
> OpenCL 2.0. The only concern is that if we `undef` them here we will end up 
> with different behavior between `-finclude-default-header` and 
> `-fdeclare-opencl-builtins`. I would suggest not to `undef` them because it 
> is better if we have coherency. Alternatively we could also add a third 
> header with undefs that can be included at the end for both but it seems to 
> make things even more complicated.
> 
> FYI `__opencl_c_int64` is already added for all OpenCL versions.
> The only concern is that if we undef them here we will end up with different 
> behavior between -finclude-default-header and -fdeclare-opencl-builtins

This is a valid point.  Doing the undefs only in `opencl-c.h` will lead to a 
problem similar to https://reviews.llvm.org/D91429 .

A third header just for the undefs sounds like a bit of an overkill indeed, 
although having duplication isn't great either.  Not sure what's best to be 
honest.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D95776/new/

https://reviews.llvm.org/D95776

_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to