R integration

[eddelbuettel]

R integration

[rstub]

The release has happened some time ago, so I am closing this now.

[eddelbuettel]

It's been a while since I tried this. I think it used to work, and I am wondering if the function registration now gets in the way...

[eddelbuettel]

And indeed it still works with version 0.1.2 (random pick I made).

And I see user_norm_rand in the shared library when I call nm on it. R calls this directly, so I am not sure how we get in the way, but we do.

Let me ponder this.

[eddelbuettel]

It is missing in the registration setup. I think I can fix that -- good catch!

[rstub]

I have only now been able to test your patch edfbfc2. The R integration is working fine, which is why I am closing this issue. Unfortunately the speed increase is not as impressive as with zrnormR or zrnorm:

library(RcppZiggurat)
is.loaded("user_norm_rand")
#> [1] TRUE
system.time(rnorm(10^8))
#>        User      System verstrichen 
#>       6.040       0.124       6.165
system.time(zrnormR(10^8))
#>        User      System verstrichen 
#>       1.104       0.120       1.223
system.time(zrnorm(10^8))
#>        User      System verstrichen 
#>       0.656       0.128       0.782
RNGkind(normal.kind = "user")
system.time(rnorm(10^8))
#>        User      System verstrichen 
#>       3.132       0.132       3.263

[eddelbuettel]

Thank you for that too! I had been meaning to circle back to you.

I think I did the same / similar benchmark test when I first cooked this up and was similarly disappointed. Thinking about a little I realized that ... this is likely due to the overhead from R's state keeping which is much more involved than what we do here with Ziggurat. Still a gain though...

I re-opened just to remind myself to release this. I need one more pass over the vignette, not sure when I get to that. I could drop the package into a drat repo if that helped you, but you seem fine with GitHub installation too :)

Gruesse aus Chicago.