[cmake-developers] FindCUDA: Clang/CUDA Support

[Huebl, Axel]

Hi everyone,


I was wondering if someone is already working on Clang/CUDA support for
the FindCUDA module.

With clang 3.8 the --cuda-arch and --cuda-path directives allow to
compile .cu files as an alternative to nvcc. Maybe we could just specify
a steering variable such as

  set(CUDA_COMPILE_CHAIN "clang")
  set(CUDA_COMPILE_CHAIN "nvcc")

before find_package(CUDA) to steer it (and of course adding the
implementation).

Alternatively,
  CUDA_HOST_COMPILER
and adding
  CUDA_DEVICE_COMPILER
  (or reusing the advanced CUDA_NVCC_EXECUTABLE)

might also be a way to go.

Currently, you still need to have the CUDA toolkit from Nvidia installed
to compile .cu files to ptx with clang, so I guess my question is really
just about replacing nvcc and it's options with clang (using different
options).

What do you think?


Best regards,
Axel Huebl
-- 

Axel Huebl
Phone +49 351 260 3582
https://www.hzdr.de/crp
Computational Radiation Physics
Laser Particle Acceleration Division
Helmholtz-Zentrum Dresden - Rossendorf e.V.

Bautzner Landstrasse 400, 01328 Dresden
POB 510119, D-01314 Dresden
Vorstand: Prof. Dr.Dr.h.c. R. Sauerbrey
          Prof. Dr.Dr.h.c. P. Joehnk
VR 1693 beim Amtsgericht Dresden

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 5119 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://public.kitware.com/pipermail/cmake-developers/attachments/20160316/23e3f149/attachment.bin>