At the risk of derailing this thread into the quicksand of programming
style ....
Except when required by the specific application, my preference would be
to stay with the Rcpp idiom. Mixing in the older C API seems to risk
more programming error. Of course, each application is different and we
have to make decisions about efficiency (machine) vs possible
inefficiency (human).
The "Rf_" part of the API in particular is ugly and somewhat of an
add-on forced in a few examples by the use of some common names in the
macro files.
This year in our Stanford grad course, I'm planning to push Rcpp, so in
a few months I may have a different view. :-)
Cheers,
John
On 1/6/12 10:15 AM, Dirk Eddelbuettel wrote:
On 6 January 2012 at 13:00, Steve Lianoglou wrote:
| Hi,
|
| 2012/1/6 Douglas Bates<ba...@stat.wisc.edu>:
| [snip]
|> As I mentioned in another thread, I prefer the idiom
|>
|> int n2 = ::Rf_asInteger(n);
|>
|> because asInteger is part of the R API (in Rcpp it must be called as
|> Rf_asInteger - the :: is a hint to the compiler that it will be found
|> in the global namespace). Functions like asInteger and asReal are
|> used in thousands of places in the base R code and have been optimized
|> to death as well as being as general as they can possibly be.
|
| Cool ... I actually missed that tip, thanks for pointing it out again.
And if one overcomes the "ick" factor of mixing APIs<grin>, it saves a
little:
R> library(inline)
R> library(microbenchmark)
R>
R> fr<- cxxfunction(signature(xs="integer"), plugin="Rcpp", body='
+ int x = ::Rf_asInteger(xs);
+ return Rcpp::wrap(x);
+ ')
R>
R> frcpp<- cxxfunction(signature(xs="integer"), plugin="Rcpp", body='
+ int x = Rcpp::as<int>(xs);
+ return Rcpp::wrap(x);
+ ')
R>
R> microbenchmark(fr(123), frcpp(123))
Unit: nanoseconds
expr min lq median uq max
1 fr(123) 841 878.5 908.5 976 15684
2 frcpp(123) 976 999.5 1017.0 1085 6589
R>
A good chunk here is "fixed" cost of calling a function, returning a
value, ... so that the "variable" gain of ::Rf_asInteger() looks pretty good.
Dirk, somewhat wondering why we're debating 100ns gains in the context of R
_______________________________________________
Rcpp-devel mailing list
Rcpp-devel@lists.r-forge.r-project.org
https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel
