OpenCL backend for LLVM

[Adrian Chadd]

A lot of the linux work is pushed not by hobbyists, but by large
companies with customers that request support.

Don't mislead yourself by thinking all this Linux work gets done by a
large number of unpaid volunteers. Go look at the contribution
statistics sometime.

If people would like to see CUDA support on FreeBSD then the best
thing they can do is to write up the exact state of things, say what
works and what doesn't, then make it REALLY easy to Nvidia to throw
some resources at it. Having them start at "hm, someone wants BSD
support, no customers, erk" is a no brainer - there's no
justification, there's noone internally championing BSD, so the
chances of it happening are slim. On the other hand, if there's a very
specific set of things that need to occur, if you as an organisation
will state you'll publicly shift to using Nvidia hardware + CUDA on
FreeBSD if/when Nvidia moves, AND you engage an org like the
Foundation to work with NVidia to make this happen.. yes, you may find
it occurs.

Right now if there's a way to run 32 bit CUDA workloads on BSD then
please wrap it up in a port and make it an absolute no brainer to do.

If someone can identify how to make 64 bit CUDA stuff work - eg, if it
only worked on 8.2 but not 9.0, then also please post instructions,
wrap it up in a port and make it work on 8.2, help the rest of the
community try to identify what broke in 9.0, etc.

Thanks,


Adrian