dot -- dot product
  -1 -- float
  -2 -- complex<float>
  -3 -- float          -- generic backend
  -4 -- complex<float> -- generic backend
  -5 -- float          -- BLAS backend
  -6 -- complex<float> -- BLAS backend
  -7 -- float          -- CUDA backend
  -8 -- complex<float> -- CUDA backend

 Parameters:
  -start N      -- starting problem size 2^N (default 4 or 16 points)
  -loop_start N -- initial number of calibration loops (default 5000)