FR: faster range

[m-muecke]

Not sure if this fits the scope of the project, but it would be lovely to get a faster range function i.e. frange(), since base R just calls min() and max() instead of doing it in a single loop. R implementation for reference: https://github.com/r-devel/r-svn/blob/d56b2b2cbfc1ca3699dcab9d0c3d571ed71e4853/src/library/base/R/range.R#L22 and a faster C implementation from collapse: https://github.com/SebKrantz/collapse/blob/f319f91853721b18fee45ac52d56359d8db7427e/src/programming.c#L1033

[MichaelChirico]

Could you provide some motivating benchmarks showing how inefficient it is to call min()/max() separately vs. calculate both in a single loop?

[m-muecke]

# integer
bench::press(
  n = c(10, 1e3, 1e6),
  {
    x <- sample(n)
    bench::mark(
      range = range(x),
      frange = collapse::frange(x)
    )
  }
)
#> Running with:
#>         n
#> 1      10
#> 2    1000
#> 3 1000000
#> # A tibble: 6 × 7
#>   expression       n      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>   <dbl> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 range           10   1.11µs   1.19µs   781631.    14.2KB      0  
#> 2 frange          10    410ns 492.01ns  1888247.    6.25MB    189. 
#> 3 range         1000   3.12µs    3.4µs   269203.    3.95KB     26.9
#> 4 frange        1000   1.02µs   1.11µs   871791.        0B      0  
#> 5 range      1000000   2.02ms   2.21ms      447.    3.81MB     45.2
#> 6 frange     1000000 600.12µs 612.38µs     1622.        0B      0

# real
bench::press(
  n = c(10, 1e3, 1e6),
  {
    x <- rnorm(n)
    bench::mark(
      range = range(x),
      frange = collapse::frange(x)
    )
  }
)
#> Running with:
#>         n
#> 1      10
#> 2    1000
#> 3 1000000
#> # A tibble: 6 × 7
#>   expression       n      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>   <dbl> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 range           10   1.11µs   1.23µs   735998.        0B     73.6
#> 2 frange          10    410ns 492.01ns  1889826.        0B      0  
#> 3 range         1000   4.22µs    4.8µs   200672.    7.86KB     20.1
#> 4 frange        1000   1.52µs    1.6µs   600102.        0B      0  
#> 5 range      1000000   3.64ms   3.73ms      267.    7.63MB     89.1
#> 6 frange     1000000   1.12ms   1.14ms      874.        0B      0

^{Created on 2025-01-03 with reprex v2.1.1}

[ben-schwen]

But base::range can also handle other types like character? We could ofc implement grange to support range for grouped calculations but without a strong need this just gets at the end of our backlog.

[m-muecke]

But base::range can also handle other types like character? We could ofc implement grange to support range for grouped calculations but without a strong need this just gets at the end of our backlog.

Yes ofc, thought it was a long shot. Perhaps trying to create an issue for R-devel makes more sense.