Visual Studio doesn't like TRISYCL_WEAK_ATTRIB_PREFIX a.k.a. __declspec(selectany) in front of function declarations

[Lotte1990]

Visual Studio doesn't like TRISYCL_WEAK_ATTRIB_PREFIX a.k.a. __declspec(selectany) in front of function declarations. Apparently this is only for initialized global data. @keryell Since C++17 has been around for a while now it is maybe possible to switch to C++17 inline as mentioned by the documentation for TRISYCL_WEAK_ATTRIB_PREFIX?

See also #269

[j-stephan]

Reference link

I hope I'll have some free time this evening so I can play around with godbolt and various inlining possibilities...

[j-stephan]

After doing some research I've come to the conclusion that this will be harder than I thought.

Visual Studio officially doesn't support weak linking of functions. __declspec(selectany) only works with variables.

We could use an undocumented linker pragma (alternatename) as shown here (both files are relevant) but we'd need to come up with a Macro-way of obtaining the mangled names. And since all of this is undocumented it is probably not too future-proof.

@keryell @MathiasMagnus What do you think about this? Currently I'm in favour of ditching the Windows weak symbols for now and just annotate the functions with [[gnu::weak]]. We don't have a device compiler on Win32 anyway, right?

[keryell]

Thank you for the research!
There is a simpler solution: do not use this code with VisualStudio because anyway ComputeCpp and Intel DPC++ are already product-quality solutions while this is just an untested triSYCL device compiler path on Windows. :-)
So the solution would be to just #if out the device-compiler related part of the run time that would require massive work to have working on Windows anyway and just keep triSYCL as is: as a pure CPU implementation on Windows.
Any disagreement?

[j-stephan]

No disagreement from my side.