Re: [boost] [build] using nvcc as toolset target?

Subject: Re: [boost] [build] using nvcc as toolset target?
From: Oliver Kowalke (oliver.kowalke_at_[hidden])
Date: 2017-09-09 15:30:58

2017-09-09 16:33 GMT+02:00 Jason Roehm <jasonr_at_[hidden]>:

>
> > On Sep 9, 2017, at 3:46 AM, Oliver Kowalke via Boost <
> boost_at_[hidden]> wrote:
> >
> > Hello,
> > a grep over the boost.config showed some code related to nvcc - but I
> could
> > not figure out how to use nvcc together with b2.
> > What is the toolset target for nvcc ('b2 toolset=nvcc' doesn't work)?
> > Oliver
>
> What’s your goal in doing so? You only need to use nvcc when compiling
> code that uses CUDA language extensions (e.g. kernels that will run on the
> GPU). I’ve used Boost a lot with CUDA-based projects, and I just built
> Boost with gcc and linked it to my nvcc-compiled code. In some more exotic
> cases, CUDA linking can be a bit tricky to get working, but for simple
> cases, it’s pretty simple.
>

I'd like to extend one of my boost libraries to support CUDA - so one part
of the library has to be compiled for the host and the other part has to be
compiled for the device.
That means the library code will contain  __host__, __device__ etc.
keywords ...