This is a short package designed to help profile Rcpp packages based on the excellent advice from this blog post by Minimally Sufficient.

Rgperftools provides two functions start_profiler and stop_profiler, to be used as follows:

library(Rgperftools)

# Force devtools to compile without debug flags for optimal performance profiling
library(pkgbuild)
flags <- pkgbuild::compiler_flags(debug=FALSE)
new_compiler_flags <- function(debug=FALSE) {flags}
assignInNamespace('compiler_flags', new_compiler_flags, 'pkgbuild')

# Load your library
devtools::load_all()

# Collect profiling data
start_profiler("/tmp/profile.out")
run_your_cpp_stuff()
stop_profiler()

It can then be used with Google's pprof to visualize the profiler results in a convenient web interface.

pprof --http=localhost:[port_num] src/MyPackage.so /tmp/profile.out

I recommend using the current Go version of pprof for results visualization. This will require installing Go if you haven't already (instructions here), then running the following command to install pprof to $GOPATH/bin/pprof:

go install github.com/google/pprof@latest

Note that if $GOPATH is not set, this will install to ~/go/bin/pprof

Next, you'll need to install the gperftools system library, then the Rgperftools R package:

conda install -c conda-forge gperftools

# Set compiler variables, so we'll find the conda-based gperftools install
Sys.setenv(
    CPATH=sprintf("%s:%s/include", Sys.getenv("CPATH"), Sys.getenv("CONDA_PREFIX")),
    LIBRARY_PATH=sprintf("%s:%s/lib", Sys.getenv("CPATH"), Sys.getenv("CONDA_PREFIX"))
)
remotes::install_github("bnprks/Rgperftools")

sudo apt install libgoogle-perftools-dev

remotes::install_github("bnprks/Rgperftools")