as suggest by Dirk Eddelbuettel in this talk and this answer I tried to profile compiled R code using gperftools. Here is what I did.
I used Dirks profilingSmall.R as script that I want to profile. I repeat it here:
## R Extensions manual, section 3.2 'Profiling R for speed'
## 'N' reduced to 99 here
suppressMessages(library(MASS))
suppressMessages(library(boot))
storm.fm <- nls(Time ~ b*Viscosity/(Wt - c), stormer, start = c(b=29.401, c=2.2183))
st <- cbind(stormer, fit=fitted(storm.fm))
storm.bf <- function(rs, i) {
st$Time <- st$fit + rs[i]
tmp <- nls(Time ~ (b * Viscosity)/(Wt - c), st, start = coef(storm.fm))
tmp$m$getAllPars()
}
rs <- scale(resid(storm.fm), scale = FALSE) # remove the mean
Rprof("boot.out")
storm.boot <- boot(rs, storm.bf, R = 99) # pretty slow
Rprof(NULL)
To profile it I run the following script
LD_PRELOAD="/usr/lib/libprofiler.so.0"
\CPUPROFILE=sample.log \
Rscript profilingSmall.R
Then I tried to parse the log file using
pprof /usr/bin/R sample.log
This returned the following error
Using local file /usr/bin/R.
Using local file sample.log.
substr outside of string at /usr/local/bin/pprof line 3618.
Use of uninitialized value in string eq at /usr/local/bin/pprof line 3618.
substr outside of string at /usr/local/bin/pprof line 3620.
Use of uninitialized value in string eq at /usr/local/bin/pprof line 3620.
sample.log: header size >= 2**16
sample.log is empty. However, a bunch of sample.log_digit were created that contain information that looks reasonable.