Add 'gprofiler/' from commit '9d31be4531101a56ac8732a687016229a72029e6'

git-subtree-dir: gprofiler
git-subtree-mainline: 6ffd7bf
git-subtree-split: 9d31be4

gprofiler/.Rbuildignore

@@ -0,0 +1,17 @@
+^.*\.Rproj$
+^\.Rproj\.user$
+^CMakeLists\.txt$
+^cmake-build-debug$
+^clion-test\.R$
+^.*\.prof$
+^.*\.prof\.out$
+^README\.Rmd$
+^README-.*\.png$
+^\.travis\.yml$
+^appveyor\.yml$
+^tic\.R$
+^src/pprof$
+^src/go/src/github.com/google/pprof/internal/driver/testdata$
+^src/go/src/github.com/google/pprof/\.
+^src/go/src/\.idea$
+^src/main\..*$

gprofiler/.gitignore

@@ -0,0 +1,12 @@
+.Rproj.user
+.Rhistory
+.RData
+/.idea
+clion-test.R
+cmake-build-debug
+*.prof
+*.prof.out
+inst/doc
+src/pprof
+src/go/src/.idea/
+src/main.*

gprofiler/.travis.yml

@@ -0,0 +1,63 @@
+# R for travis: see documentation at https://docs.travis-ci.com/user/languages/r
+# Default configuration for use with tic package
+# Usually you shouldn't need to change the first part of the file
+
+# DO NOT CHANGE THE CODE BELOW
+before_install: R -q -e 'install.packages("remotes"); remotes::install_github("ropenscilabs/tic"); tic::prepare_all_stages(); tic::before_install()'
+install: R -q -e 'tic::install()'
+after_install: R -q -e 'tic::after_install()'
+before_script: R -q -e 'tic::before_script()'
+script: R -q -e 'tic::script()'
+after_success: R -q -e 'tic::after_success()'
+after_failure: R -q -e 'tic::after_failure()'
+before_deploy: R -q -e 'tic::before_deploy()'
+deploy:
+  provider: script
+  script: R -q -e 'tic::deploy()'
+  on:
+    all_branches: true
+after_deploy: R -q -e 'tic::after_deploy()'
+after_script: R -q -e 'tic::after_script()'
+# DO NOT CHANGE THE CODE ABOVE
+
+# Custom parts:
+
+# Header
+language: r
+sudo: false
+dist: trusty
+cache: packages
+latex: true
+
+#env
+env:
+  global:
+  - _R_CHECK_FORCE_SUGGESTS_=false
+  - MAKEFLAGS="-j 2"
+
+jobs:
+  include:
+  - addons:
+      apt:
+        packages:
+        - libgoogle-perftools-dev
+    env:
+    - KIND=apt
+    - BUILD_PKGDOWN=true
+  - addons:
+      apt:
+        packages:
+        - libunwind8-dev
+    env:
+    - KIND=github
+    before_install:
+    - ( cd /tmp && git clone --depth 1 https://github.com/gperftools/gperftools.git )
+    - ( cd /tmp/gperftools && ./autogen.sh && ./configure --prefix=${HOME} && make && make install )
+    - mkdir -p ~/.R
+    - echo "PKG_CPPFLAGS = -I${HOME}/include" >> ~/.R/Makevars
+    - echo "LDFLAGS := ${LDFLAGS} -shared -L${HOME}/lib" >> ~/.R/Makevars
+    - echo "LD_RUN_PATH = ${HOME}/lib" >> ~/.R/Makevars
+    - echo "export LD_RUN_PATH" >> ~/.R/Makevars
+
+#services
+services:

gprofiler/CMakeLists.txt

@@ -0,0 +1,23 @@
+cmake_minimum_required(VERSION 3.6)
+project(gprofiler)
+
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
+
+set(R_LIBRARY "~/R-library")
+set(R_INCLUDE "/usr/share/R/include")
+
+include_directories(${R_LIBRARY}/Rcpp/include)
+include_directories(${R_INCLUDE})
+include_directories(src)
+
+add_definitions(-D__CLION__)
+
+
+file(GLOB SOURCE_FILES src/*.cpp src/*.c)
+add_library(gprofiler ${SOURCE_FILES})
+
+add_custom_target(
+    Rinstall
+    # Using MAKEFLAGS=-j$(nproc) didn't work
+    COMMAND MAKEFLAGS=-j8 nice R CMD INSTALL --libs-only --no-test-load ${gprofiler_SOURCE_DIR}
+)

gprofiler/DESCRIPTION

@@ -0,0 +1,31 @@
+Package: gprofiler
+Title: Joint Profiling of Native and R Code
+Version: 0.0-1
+Authors@R: c(
+      person("Kirill", "Müller", role = c("aut", "cre"), email = "[email protected]"),
+      person("Google, Inc.", role = "cph", comment = "Bundled pprof sources")
+    )
+Description: Simultaneously collects profiling data at both R and native levels
+    (using 'Rprof' and 'gprofiler') and provides a unified view of the collected data. Work in progress.
+License: GPL-3
+Encoding: UTF-8
+LazyData: true
+Date: 2017-01-18
+BugReports: https://github.com/r-prof/gprofiler/issues
+URL: https://github.com/r-prof/gprofiler, http://r-prof.github.io/gprofiler
+LinkingTo: Rcpp
+Imports:
+    dplyr,
+    profvis,
+    tidyr,
+    Rcpp
+SystemRequirements: C++11
+Suggests: 
+    DBI,
+    knitr,
+    rmarkdown,
+    RSQLite,
+    testthat
+VignetteBuilder: knitr
+Roxygen: list(markdown = TRUE)
+RoxygenNote: 6.0.1

gprofiler/NAMESPACE

@@ -0,0 +1,12 @@
+# Generated by roxygen2: do not edit by hand
+
+export(callback1_r)
+export(callback2_r)
+export(callback3_r)
+export(get_profiler_traces)
+export(show_profiler_pdf)
+export(start_profiler)
+export(stop_profiler)
+importFrom(Rcpp,sourceCpp)
+importFrom(tidyr,"%>%")
+useDynLib(gprofiler,.registration = TRUE)

gprofiler/NEWS.md

@@ -0,0 +1,11 @@
+## gprofiler 0.0-1 (2017-01-18)
+
+Initial release
+
+- Proof of concept.
+- Functions `start_profiler()` and `stop_profiler()` to initialize and terminate joint profiling.
+    - Daisy-chain `SIGPROF` handlers, the `gperftools` profiler offers a filter which can be used to sneak in the call to the original handler installed by `Rprof()`.
+- Function `get_profiler_traces()` returns joint profiling data.
+    - Parsing `gperftools` data with a bleeding-edge version of `pprof` (needs `-traces` switch), Ubuntu binary included.
+    - Parsing `Rprof` data with a call to an unexported `profvis` function.
+- Convenience function `show_profiler_pdf()` shows a PDF of the `gperftools` data.

gprofiler/R/RcppExports.R

@@ -0,0 +1,27 @@
+# Generated by using Rcpp::compileAttributes() -> do not edit by hand
+# Generator token: 10BE3573-1514-4C36-9D1C-5A225CD40393
+
+callback1_cpp <- function() {
+    invisible(.Call(`_gprofiler_callback1_cpp`))
+}
+
+callback2_cpp <- function() {
+    invisible(.Call(`_gprofiler_callback2_cpp`))
+}
+
+callback3_cpp <- function() {
+    invisible(.Call(`_gprofiler_callback3_cpp`))
+}
+
+init_profiler_impl <- function() {
+    .Call(`_gprofiler_init_profiler_impl`)
+}
+
+start_profiler_impl <- function(ldc, path) {
+    invisible(.Call(`_gprofiler_start_profiler_impl`, ldc, path))
+}
+
+stop_profiler_impl <- function(ldc) {
+    invisible(.Call(`_gprofiler_stop_profiler_impl`, ldc))
+}
+

gprofiler/R/callback.R

@@ -0,0 +1,23 @@
+sleep <- function() {
+  stats::runif(2e7)
+  NULL
+}
+
+#' @export
+callback1_r <- function() {
+  sleep()
+}
+
+#' @export
+callback2_r <- function() {
+  sleep()
+  callback1_cpp()
+  sleep()
+}
+
+#' @export
+callback3_r <- function() {
+  sleep()
+  callback2_cpp()
+  sleep()
+}

gprofiler/R/gprofiler-package.R

@@ -0,0 +1,3 @@
+#' @useDynLib gprofiler,.registration = TRUE
+#' @importFrom Rcpp sourceCpp
+"_PACKAGE"

gprofiler/R/profiler.R

@@ -0,0 +1,97 @@
+.my_env <- new.env(parent = emptyenv())
+
+#' Starts and stops profiling
+#'
+#' `start_profiler()` initiates profiling.
+#'
+#' @param path Path to the output file. Other files based on this path
+#'   may be created.
+#'
+#' @export
+start_profiler <- function(path = "1.prof") {
+  prof_data <- init_profiler_impl()
+  utils::Rprof(filename = paste0(path, ".out"), line.profiling = TRUE)
+  start_profiler_impl(prof_data, path)
+  .my_env$prof_data <- prof_data
+}
+
+#' `stop_profiler()` terminates profiling. The results are available with
+#' [get_profiler_traces()].
+#'
+#' @export
+#' @rdname start_profiler
+stop_profiler <- function() {
+  on.exit(.my_env$prof_data <- NULL, add = TRUE)
+
+  utils::Rprof(NULL)
+  stop_profiler_impl(.my_env$prof_data)
+}
+
+#' Parse profiler output
+#'
+#' Combines the profiler output obtained from [start_profiler()] into
+#' a nested tibble.
+#'
+#' @param path The path to the profiler output.
+#' @importFrom tidyr %>%
+#' @export
+get_profiler_traces <- function(path = "1.prof") {
+  traces <- system2(
+    get_pprof_path(),
+    c("-unit", "us", "-lines", "-traces", shQuote(path)),
+    stdout = TRUE)
+
+  pprof_nested <-
+    strsplit(paste(traces, collapse = "\n"), "\n-+[+]-+\n?")[[1L]][-1L] %>%
+    tibble::enframe(name = "index", value = "gprofiler") %>%
+    dplyr::mutate(count = as.numeric(sub("^ +([0-9]+).*$", "\\1", gprofiler)) / 10000) %>%
+    dplyr::slice(., rep(seq_len(nrow(.)), count)) %>%
+    dplyr::transmute(time = seq_along(gprofiler), gprofiler = parse_pprof(gprofiler))
+
+  rprof_nested <-
+    profvis:::parse_rprof(paste0(path, ".out"))$prof %>%
+    tibble::as_tibble() %>%
+    tidyr::nest(-time, .key = rprof)
+
+  stopifnot(pprof_nested$time == rprof_nested$time)
+  tibble::as_tibble(cbind(pprof_nested, rprof_nested[-1]))
+}
+
+parse_pprof <- function(output) {
+  lapply(output, parse_pprof_one)
+}
+
+parse_pprof_one <- function(output_item) {
+  output_lines <- strsplit(output_item, "\n", fixed = TRUE)[[1]]
+  rx <- "^(?:| *[0-9][^ ]+) +(?:[?][?]|[<]unknown[>]|(.+) +(?:[?][?]|([^ ]+):([0-9]+)))$"
+  invalid <- grep(rx, output_lines, invert = TRUE)
+  if (length(invalid) > 0) {
+    stop(
+      "Unexpected profiler output:\n",
+      paste0(format(invalid), ": ", output_lines[invalid]),
+      call. = FALSE
+    )
+  }
+
+  tibble::tibble(
+    label = gsub(rx, "\\1", output_lines),
+    filename = gsub(rx, "\\2", output_lines),
+    line = gsub(rx, "\\3", output_lines)
+  )
+}
+
+#' @export
+show_profiler_pdf <- function(path = "1.prof", focus = NULL) {
+  pprof_exit_code <- system2(
+    get_pprof_path(),
+    c("-lines", "-evince", if (!is.null(focus)) paste0("-focus=", focus),
+      shQuote(path)),
+    wait = FALSE)
+  if (pprof_exit_code != 0) {
+    warning("pprof exited with ", pprof_exit_code)
+  }
+}
+
+get_pprof_path <- function() {
+  system.file("bin", "pprof", package = utils::packageName())
+}

gprofiler/README.Rmd

@@ -0,0 +1,61 @@
+---
+output:
+  github_document
+---
+
+<!-- README.md is generated from README.Rmd. Please edit that file -->
+
+```{r, echo = FALSE}
+knitr::opts_chunk$set(
+  collapse = TRUE,
+  comment = "#>",
+  fig.path = "README-"
+)
+```
+
+# gprofiler
+
+The goal of gprofiler is to assist profiling R packages that include native code (C++, C, Fortran, ...).
+It collects profiling output simultaneously using [Rprof](https://www.rdocumentation.org/packages/utils/versions/3.3.2/topics/Rprof) and [gperftools](https://github.com/gperftools/gperftools) and provides a unified view of profiling data.
+At present this is but a feasibility study: it requires Ubuntu Linux 64-bit and the `libgoogle-perftools-dev` package.
+
+
+## Example
+
+The following example writes the `iris` data 100 times to an in-memory SQLite database and collects profiling data.
+By default, the profiling data are collected in two files, `1.prof` (native) and `1.prof.out` (R).
+
+```{r}
+library(DBI)
+
+gprofiler::start_profiler()
+con <- dbConnect(RSQLite::SQLite(), ":memory:")
+invisible(lapply(1:100, function(x)
+  dbWriteTable(con, paste0("iris", x), iris)))
+dbDisconnect(con)
+gprofiler::stop_profiler()
+```
+
+A unified view is created with `get_profiler_traces()`. Currently this returns a nested data frame with two list columns, one for the native trace and one for the R trace. Each row represents one sample:
+
+```{r}
+gprofiler::get_profiler_traces()
+```
+
+
+Below is another example where an R function calls a C++ function that calls back into R.
+
+```{r}
+gprofiler::start_profiler()
+gprofiler::callback2_r()
+gprofiler::stop_profiler()
+gprofiler::get_profiler_traces()
+```
+
+Eventually, the result will be an `Rprof`-compatible data format which can be consumed by `profvis` and other existing packages.
+
+### Acknowledgment
+
+This project is being realized with financial support from the
+
+<img src="https://www.r-consortium.org/wp-content/uploads/sites/13/2016/09/RConsortium_Horizontal_Pantone.png" width="400">