Remove usage WITH_CYCLES_CUDA_BINARIES in code, use check for
authorSergey Sharybin <sergey.vfx@gmail.com>
Mon, 14 Jan 2013 17:30:33 +0000 (17:30 +0000)
committerSergey Sharybin <sergey.vfx@gmail.com>
Mon, 14 Jan 2013 17:30:33 +0000 (17:30 +0000)
commite5179bfefc649cca07077d34d13c574e712d82d0
tree255aa6003028cd403eab00d015c3a768e92f1cc4
parent36f44c6e02a5e35bb4572f232b867bb055dbb2d8
Remove usage WITH_CYCLES_CUDA_BINARIES in code, use check for
precompiled cubins instead,

Logic here is following now:
- If there're precompiled cubins, assume CUDA compute is available,
  otherwise
- If cuda toolkit found, assume CUDA compute is available
- In all other cases CUDA compute is not available

For windows there're still check for only precompiled binaries,
no runtime compilation is allowed.

Ended up with such decision after discussion with Brecht. The thing
is, if we'll support runtime compilation on windows we'll end up
having lots of reports about different aspects of something doesn't
work (you need particular toolkit version, msvc installed, environment
variables set properly and so) and giving feedback on such reports
will waste time.
intern/cycles/CMakeLists.txt
intern/cycles/SConscript
intern/cycles/blender/addon/__init__.py
intern/cycles/device/device_cuda.cpp
intern/cycles/util/util_cuda.cpp
intern/cycles/util/util_cuda.h