Note: Default behavior is to compile operators operators that use
float 32bits precision in computation. Hence, if you do not modify rkeops
options, you have to call the function compile4float64 to
compile operators using float 64bits precision.
Using float 64bits (or double) precision is likely to result in a loss of
performance regarding computing time on GPU. If you want to get the best
performance but worry about computation precision, you can use float 32bits
precision and compensated sums that are implemented in KeOps.