Few fixes for CUDA

[krzysg]

Hello,

Here are few fixes:

1. Removed 'extra_smooth' parameter as agreed with Joel some time ago (+/- 1 year ;-) ).
2. Added missing explicit instantiations for types needed by APRTest.cpp
3. Switching off two big pipeline test - I will work on them in the near future and enable when whole pipeline is working again...
4. Because pipeline is not working I turned it off in APRConverter.cpp (so CPU is always used regardless off APR_USE_CUDA). It allows to have working tests and progress on CUDA impl.

Those changes allows to compile LibAPR and run tests for CUDA which is minimum for later development and would be nice to have it on develop.

[joeljonsson]

Thanks @krzysg !

The windows tests should pass after updating vcpkg (e.g. after #163 or #168 are merged in)

[joeljonsson]

I'm thinking more along the lines of

public:
    void get_apr(...)         // CPU version

#ifdef APR_USE_CUDA
    void get_apr_cuda(...)    // CUDA version
#endif

such that both versions are callable by the user. This just requires the initial checks from get_apr to be moved into the cpu/gpu methods. Alternatively, if you want to keep the behavior of get_apr the way it is, you can make both get_apr_cpu and get_apr_gpu public in the same way, and then just alias get_apr to one of them based on cuda availability.

[krzysg]

I think that get_apr should stay as it was in terms of behavior (so used pipeline should depend on APR_USE_CUDA value).
I have change get_apr_cpu(cuda) to public so they can be used explicitly to force using specific pipeline.

[joeljonsson]

I think that get_apr should stay as it was in terms of behavior (so used pipeline should depend on APR_USE_CUDA value). I have change get_apr_cpu(cuda) to public so they can be used explicitly to force using specific pipeline.

Excellent