How do I profile the underlying C-code in an R package?

Question

I have a pretty good understanding of C but very little of R. I have installed an R-package (coxme) into R-studio on Windows7 from source which included compiling several C files into a .dll. This package is not standalone and depends on at least 2 other packages (maybe more). It runs too slow so I would like to see if I can speed it up by profiling it first. I used the R-package "profvis" to profile the R-code but it only tells me that 98% of the time get used by the "coxme" package without giving any further details. The call to the R package looks something like:

frail.xg=coxme(Surv(y.vec,delta.vec)~x.mat+g.vec+(1|fiid.vec),varlist=list(K.mat))

The entire package has no function named coxme(...), so I have no idea what goes on after I call this package as shown above. I tagged every function inside the "src" directory of the package to print (append) to a file, but only the init() function fires upon installation when it registers the package. Non of the other functions print to that file once I run coxme from within R-Studio. Besides the init() function it seems that non of the C-functions in the package even get executed when I call coxme(...). My approach is like:

remove.packages("coxme")

manipulate C-code in coxme "src" directory to print to some external file "I am function XYZ, etc.". create coxme.tar.gz

install.packages('E:/Workdir/AOO/R/coxme/coxme.tar.gz', repos = NULL, type="source")
library("coxme")
frail.xg=coxme(Surv(y.vec,delta.vec)~x.mat+g.vec(1|fiid.vec),varlist=list(K.mat))

this produces some expected result. But non of the C-functions seem to get hit, nor does my profiling code. So my questions are:

What exactly gets executed?
How do I go about and find out what's eating up most of the time within the coxme package?

Maybe try to answer first a more immediate concern: https://stackoverflow.com/help/formatting — s_baldur, May 06 '19 at 14:46
`coxme` calls `coxme.fit` which calls a C function: https://github.com/cran/coxme/blob/master/R/coxme.fit.R#L152 — Stéphane Laurent, May 06 '19 at 14:46
Maybe this will help [how to find the source of a function](https://stackoverflow.com/questions/19226816/how-can-i-view-the-source-code-for-a-function) — MrFlick, May 06 '19 at 15:14
Also maybe this would help: https://www.jimhester.com/2018/08/22/debugging-rstudio/ — MrFlick, May 06 '19 at 15:23
I have the source code. It just won't print to an external file. fopen() does return a valid filehandle but still no printing takes place. — mschmidt, May 06 '19 at 15:45

score 0 · Answer 1 · answered May 06 '19 at 18:00

0

The main problem is that Rstudio (at least on Windows7) does not update any .dll once it was loaded. So my edits of the C-code never took effect since the old .dll was never replaced, even when I unloaded and recompiled via install.packages(). I had to completely exit Rstudio and manually delete the package directory (coxme). Then restart Rstudio in install from my edited source. Now my edits deliver the desired results.

answered May 06 '19 at 18:00

mschmidt

81
1
2

You might try the `More > Clean & Rebuild` option in the `Build` pane to force erasing the `.o`/`.so` objects before reinstalling – MichaelChirico Sep 29 '19 at 15:41
Did you success to speed up the `coxme` package ? I'm using it and it takes 3 seconds to get an estimation for 2500 individuals which is too slow for me. I have to bootstrap it. – Flora Grappelli Jan 07 '20 at 10:31

How do I profile the underlying C-code in an R package?

1 Answers1