I am wrapping a number of R functions in a library (purely for syntax consistency with other systems, not performance or anything else). Something like:

ABS_1 <- function (x) { abs(x) }
ABS_2 <- function (x) { .Primitive("abs")(x) }

And I noticed this:

x <- 1e5

microbenchmark::microbenchmark(abs(x), ABS_1(x), ABS_2(x), times = x)

Unit: nanoseconds
     expr min  lq                   mean median  uq     max neval
   abs(x)   0   0  51.100760000065896804      0   0 4464203 1e+05
 ABS_1(x)  41 123 142.329860000001815479    123 123  326196 1e+05
 ABS_2(x) 287 369 398.285890000000563305    369 369  532426 1e+05

I understand there is overhead when wrapping abs(), but what explains the performance difference between the two wrapping functions ?

This difference doesn’t seem to scale when computing the function on a vector instead:

n <- 1e5
x <- 1:n

microbenchmark::microbenchmark(abs(x), ABS_1(x), ABS_2(x), times = n)

Unit: microseconds
     expr   min    lq  mean median    uq   max neval
   abs(x) 7.626 24.03 51.25  24.68 27.18 59664 1e+05
 ABS_1(x) 7.790 24.19 52.38  24.89 27.31 59591 1e+05
 ABS_2(x) 8.036 24.48 50.11  25.21 27.55 58694 1e+05

I am simply curious what causes the difference.