How to link together generated CPP and CUDA code about slang HOT 6 CLOSED

I think the cleanest way would be to build separate libraries and have the libraries interact with each other through a clearly defined C interface.

from slang.

But basically I need at least to build two separate dynamic libraries (one for CUDA and one for CPU) and load them dynamically during runtime based on some switch for example. Because due to the symbol clashes I won't be able to link statically those two separate libraries generated from Slang source.

What about some slang compiler option to prefix all symbols with some namespace (or at least some prefix), so we can statically link together multiple implementations generated from the same slang source?

It could be useful to link together CUDA and CPU implementation or just multiple implementations with some -D flags passed to the Slang compiler.

from slang.

Also it is not currently possible to include both slang-cpp-prelude.h and slang-cuda-prelude.h in a single file, to use them as headers for internal types like StructuredBuffer.

from slang.

Cpp and Cuda are treated like distinct targets, and we have not architected the preludes and the compiler to support using them in a single program. It may be possible to rename the preludes and use some name mangling logic to avoid the name clashes, but it is going to be a pervasive and potentially breaking change that I am not sure we are ready to make right now.

Is it possible for you to wrap the entire generated source inside a namespace in your application?

from slang.

There were some issues with wrapping the code in a namespace. I would probably have to add some code wrapper with a custom interface and make all other Slang symbols hidden using --visibility=hidden.

It would be nice in a future, if we can have possibility to link multiple targets together and create single app that can switch backend based on the available hardware.

from slang.

It should be possible to have both cpu and CUDA code in your application if you use CUDA driver api to load the recompiled PTX modules, right? That way you won't have any c++ linker problem.

from slang.