Date: Fri, 22 Nov 2019 12:42:27 -0800 From: Conrad Meyer <cem@freebsd.org> To: Theron <theron.tarigo@gmail.com> Cc: Gleb Popov <arrowd@freebsd.org>, freebsd-hackers <freebsd-hackers@freebsd.org> Subject: Re: Linking Linux library to FreeBSD program. Message-ID: <CAG6CVpWkPJMtk7SXXNMzEQL6TJ4GLjR6z1eSP78EiGdK7V5WEA@mail.gmail.com> In-Reply-To: <7679128e-d565-0132-2fe2-cc1083abfa8d@gmail.com> References: <CALH631m8OD95VFWV7=znTebBrLNBQYWwXnSDMaEcWZriBQEiQA@mail.gmail.com> <7679128e-d565-0132-2fe2-cc1083abfa8d@gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Nov 22, 2019 at 5:35 AM Theron <theron.tarigo@gmail.com> wrote: > > On 2019-11-22 02:36, Gleb Popov wrote: > > What course of action could be taken in this situation? My plan is to > > generate a native wrapper library for libcudart.so in which every exported > > function would call Linux binary linked to real libcudart.so and route > > function arguments via RPC or some other mechanism. This looks awful for > > sure, but I don't see any other solutions. > > https://github.com/shkhln/nvshim appears to allow using Nvidia Linux > binary by FreeBSD executable, but for Vulkan. Does that look like it > could be extended to support CUDA? That looks really promising. I was going to suggest looking at https://github.com/taviso/loadlibrary for rough inspiration, but nvshim looks a lot closer (and is specifically aimed at Nvidia userspace Linux libraries already). Best, Conrad
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAG6CVpWkPJMtk7SXXNMzEQL6TJ4GLjR6z1eSP78EiGdK7V5WEA>