jointprof

The goal of jointprof is to assist profiling R packages that include native code (C++, C, Fortran, …). It collects profiling output simultaneously using Rprof and gperftools and provides a unified view of profiling data.

See the guide for a more detailed overview, or take a look at the internals document if you’re curious.

Installation

Ubuntu

Other Linux distributions may work if you install the right system dependencies (let me know which!).

1. Install system dependencies:

   sudo apt install \
     libgoogle-perftools-dev \
     libprotoc-dev libprotobuf-dev protobuf-compiler \
     golang-go \
     graphviz

2. Install pprof:

   go get github.com/google/pprof

3. Install the package:

   # install.packages("remotes")
   remotes::install_github("r-prof/jointprof")

OS X

1. Install system dependencies:

   brew install graphviz

2. Install gperftools (currently from a branch, pull request pending):

   git clone https://github.com/krlmlr/gperftools.git -b f-export-stack
   cd gperftools
   ./autogen.sh
   ./configure
   make
   sudo make install
   cd ..

3. Install pprof:

   go get github.com/google/pprof

4. Install the package:

   # install.packages("remotes")
   remotes::install_github("r-prof/jointprof")

Other OS

Windows and Solaris are not supported.

Usage

library(jointprof)

target_file <- "Rprof.out"

# Collect profile data
start_profiler(target_file)
## code to be profiled
stop_profiler()

# Analyze profile data
summaryRprof(target_file)
profvis::profvis(prof_input = target_file)
proftools::readProfileData(target_file)
prof.tree::prof.tree(target_file)

# Convert to pprof format and analyze
pprof_target_file <- "Rprof.pb.gz"
profile_data <- profile::read_rprof(target_file)
profile::write_rprof(profile_data, pprof_target_file)
system2(
  find_pprof(),
  c(
    "-http",
    "localhost:8080",
    shQuote(pprof_target_file)
  )
)

Acknowledgment

This project is being realized with financial support from the