This build of MMC (mmc-trinity) adds the NVIDIA CUDA (-c cuda) and OptiX
(-c optix) backends to the CPU (-c sse) and OpenCL (-c opencl) backends of
the standard mmc package.  The CUDA code is compiled for every GPU from
Maxwell (sm_52) to Ampere (sm_86), and as PTX for newer ones.

It needs an NVIDIA GPU with driver R515 (Linux) or R516 (Windows) or
newer.  OptiX ray-traces with the RT cores of RTX GPUs; it runs the
dual-grid method (-M g); the other backends run every method.

  mmc --bench dmmc-cube60 -c optix       # the OptiX backend
  mmc --bench dmmc-cube60 -c cuda        # the CUDA backend
  mmc -L                                 # list the GPUs

mmclab-trinity (MATLAB/Octave: cfg.compute='cuda' or 'optix') and
pmmc-trinity (Python: cfg['compute']='cuda' or 'optix') are separate
packages.  See README.txt for everything else.
