Wrapping code that depends on __cplusplus

[maleadt]

Another question relating to nasty CUDA headers 😅

I'm trying to wrap headers that make use of the half datatype, which is only really defined in C++. Consequently, these functions are guarded by an ifdef __cplusplus check:

#if defined(__cplusplus)
CUBLASAPI cublasStatus_t CUBLASWINAPI cublasHSHgemvBatched(cublasHandle_t handle,
                                                           cublasOperation_t trans,
                                                           int m,
                                                           int n,
                                                           const float* alpha, /* host or device pointer */
                                                           const __half* const Aarray[],
                                                           int lda,
                                                           const __half* const xarray[],
                                                           int incx,
                                                           const float* beta, /* host or device pointer */
                                                           __half* const yarray[],
                                                           int incy,
                                                           int batchCount);
#endif

Now, given that these are pointers to half, we can use these APIs from Julia and pass a Ptr{Float16}. Is there a way that I can have Clang.jl wrap these? Adding __cplusplus to the defines breaks all kinds of things, because in other headers that then introduces actual C++ code which we can't handle.

I know this is a bit of a long shot, but I'd like to avoid having to pre-patch the headers before processing them with Clang.jl (either manually, or with some patch invocations), so I'm curious about possible workarounds. Maybe it's possible to have a define for a single header file? or have a pre-processing step where I can have Julia code filter out these specific macros (if I detect that there's headers inside, and not other kinds of code)?

[Gnimuc]

That code snippet is dropped by C's preprocessor, so Clang's ASTUnit can't see it. As a result, we have no choice but to add C++ support in Clang.Generators, which needs a lot of refactoring work.

Maybe it's possible to have a define for a single header file?

As long as the header is a top-level independent header and can be parsed successfully(either being parsed standalone or using -isystem tricks), you can create a separate context with dedicated compilation flags.

[maleadt]

Sadly, it includes other headers, some of which become problematic in C++ mode.

[maleadt]

I'll do some preprocessing in Julia then; I could rewrite the headers to a temp file and parse that with Clang.jl instead.

[Gnimuc]

Wait. Is __half really a C++ thing?

https://stackoverflow.com/questions/73873870/is-the-half-a-native-c-type