CUDA Function Decorators Give Warnings

[cliffburdick]

Subject: When compiling the markup files for CUDA code, I see the warnings.

Problem

The sphinx compiler shows:

Invalid definition: Expected identifier in nested name, got keyword: void [error at 15]
global void TL::est(float * a)

I'm guessing this is because the parser doesn't recognize the device and global decorators since it's cuda-specific. Is there a workaround for this? There should be two underscores on global and device, but github's markdown seems to make it bold.

[jakobandersen]

I assume this is with the C++ domain?
In that case, it is possible to tell Sphinx that global should be treated as a valid attribute: http://www.sphinx-doc.org/en/master/usage/configuration.html#confval-cpp_id_attributes.
That is, in conf.py insert something like: cpp_id_attributes = ["__global__", "__device__"]

(Perhaps we should improve the documentation to make this much clearer.)

[cliffburdick]

Thanks @jakobandersen! That seemed to work for that case. Do you know if there's a way to fix it within the function definition itself on C++ functions that use restrict? Here's an example:

If declarator-id:
Invalid definition: Expecting "," or ")" in parameters_and_qualifiers, got "a". [error at 72]
global void TL::est2(float*restrict a)

It looks similar to this issue:

#2682

But cpp_paren_attributes still gives the errors

[dabe5]

Namaste

[dabe5]

Kolkata

[jakobandersen]

The cpp_paren_attributes is for custom attributes that takes an argument, so that is not the right one to use here. Attributes are only parsed in a few places, and as part of declarators was not one of them. On master, which will become version 1.8, it should work now. You just have to add __restrict__ to cpp_id_attributes as well.