For flrunif()
and flrnorm()
, the data is produced without a
double precision copy. That is, it is not (computationally) equivalent to
fl(matrix(runif(...)))
, though the operations are conceptually the
same. For these, To produce a vector instead of a matrix, leave argument
n
blank. Setting n=1
will produce an mx1 matrix.
For flrand()
, the data is generated in double precision in 4KiB
batches and copied over to a pre-allocated vector. This will be slower than
generating all of the data up front and copying it, although it uses far less
memory most of the time. So you can think of flrunif()
and
flrnorm()
as highly optimized versions of flrand()
for uniform
and normal generators specifically.