Building R 4+ for Windows with OpenBLAS

[This article was first published on R – Strange Attractors, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

R-project LogoThis post outlines the steps needed to build R 4+ for Windows with OpenBLAS. The release of R 4.0 includes significant changes to the Windows build system from prior versions—for the better! Before anything, we all owe Jeroen Ooms significant gratitude for the many hours he spent working on the build system. Thank you, Jeroen!!

The build process below creates an installation file for both 32-bit and 64-bit R but makes some compromises. Extra time spent building the installer is traded for reduced time spent adjusting files and minimizing needed file changes. The OpenBLAS library itself is built using the “dynamic” keyword so it will work for all architectures, and it is built to use up to 64 threads. It isn’t completely optimized to the actual building CPU, which is not only reasonable but necessary. A centralized library must be generic enough to serve all comers. Eventually, I hope to refine the process to build just the 64 bit version and to link to a static, locally-compiled, and locally-optimized OpenBLAS library. Lastly, I intend to address the steps needed to add specific packages from source like XML or nloptr in a future post. For now, installation as binaries should work.

Step 1: Install Rtools40

Installing Rtools40 is straightforward. Make sure to download the 64 bit version and add the PATH variable as described on the website.

Step 2: Install the necessary support software

Install MiKTeX and Inno Setup to their default locations if they are not already. Install qpdf anywhere if you wish to use it.

Step 3: Obtaining the R-windows build scripts

Clone or download the r-windows/r-base repository from Github. Personally, I download the ZIP file. I tend to monkey around a lot with the code while figuring out how to customize the build, so I do not want to clone/fork the original repository. When I am ready to suggest changes, I clone the repository and send a pull request. Regardless, there should be a directory with the files listed at the r-base repository, likely called r-base-master.

Step 4: Create blas patch

The build uses the same process which works in R3+; it hooks OpenBLAS into the build via the ATLAS variables in the Makefiles. However, the new system downloads the R source code as part of the build process, so changes to src/extra/blas/Makefile.win are made using patch file instead of manual edits. Therefore, save the following text as a file named blas.diff in the same subdirectory as the extracted/cloned files.

--- a/src/extra/blas/Makefile.win
+++ b/src/extra/blas/Makefile.win
@@ -12,7 +12,7 @@
 ../../../$(BINDIR)/Rblas.dll: blas00.o ../../gnuwin32/dllversion.o
 	@$(ECHO) -------- Building $@ --------
 	$(DLL) -shared $(DLLFLAGS) -o $@ $^ Rblas.def \
-	   -L../../../$(IMPDIR) -lR  -L"$(ATLAS_PATH)" -lf77blas -latlas
+	   -L../../../$(IMPDIR) -lR  -L"$(ATLAS_PATH)" -lopenblas
 else
 ../../../$(BINDIR)/Rblas.dll: blas.o cmplxblas.o ../../gnuwin32/dllversion.o
 	@$(ECHO) -------- Building $@ --------

Step 5: Adjust existing files

Make the following changes to the build files to ensure that OpenBLAS is pulled from pacman (the package manager, not the Namco character) and that the proper libraries are accessed at the right times.

full-build.sh

Add the highlighted line to full-build.sh after the call to cairo, tk, and curl. This is where other packages like nlopt or xml will be added eventually too.

pacman -S --needed --noconfirm mingw-w64-{i686,x86_64}-{cairo,tk,curl}
pacman -S --needed --noconfirm mingw-w64-{i686,x86_64}-openblas

MkRules.local.in

Edit EOPTS (extra optimization flags) in MkRules.local.in as in prior versions of R. Personally, I use -march=native -pipe as I build bespoke R installers for each machine I use. To use the same installer on different computers I recommend -mtune=generic -pipe. See the GCC 8.3 documentation for more on optimization flags.
Next, add and edit the following lines under #Enable features:

USE_ATLAS = YES
ATLAS_PATH="/mingw$(WIN)/lib/"

This tells the build to use the OpenBLAS libraries and where to find both the 32 and 64 bit versions.
To use qpdf, add the following under # For building docs/installer, substituting the proper path.

QPDF = /path/to/qpdf

PKGBUILD

Adjusting this file is more complicated. Add the highlighted lines below to PKGBUILD in their appropriate locations:

# Maintainer: Jeroen Ooms <[email protected]>
_realname=r-installer
pkgbase=${_realname}
pkgname="${_realname}"
pkgver=4.0.9000
pkgrel=1
pkgdesc="The R Programming Language"
arch=('any')
makedepends=("${MINGW_PACKAGE_PREFIX}-bzip2"
             "${MINGW_PACKAGE_PREFIX}-gcc"
             "${MINGW_PACKAGE_PREFIX}-gcc-fortran"
             "${MINGW_PACKAGE_PREFIX}-cairo"
             "${MINGW_PACKAGE_PREFIX}-curl"
             "${MINGW_PACKAGE_PREFIX}-icu"
             "${MINGW_PACKAGE_PREFIX}-libtiff"
             "${MINGW_PACKAGE_PREFIX}-libjpeg"
             "${MINGW_PACKAGE_PREFIX}-libpng"
             "${MINGW_PACKAGE_PREFIX}-pcre2"
             "${MINGW_PACKAGE_PREFIX}-tcl"
             "${MINGW_PACKAGE_PREFIX}-tk"
             "${MINGW_PACKAGE_PREFIX}-xz"
             "${MINGW_PACKAGE_PREFIX}-zlib"
             "${MINGW_PACKAGE_PREFIX}-openblas"
             "texinfo"
             "texinfo-tex"
             "sed")
options=('staticlibs')
license=("GPL")
url="https://www.r-project.org/"

# Default source is R-devel (override via $rsource_url)
source=(R-source.tar.gz::"${rsource_url:-https://cran.r-project.org/src/base-prerelease/R-devel.tar.gz}"
    https://curl.haxx.se/ca/cacert.pem
    MkRules.local.in
    shortcut.diff
    create-tcltk-bundle.sh
    blas.diff)

# Automatic untar fails due to embedded symlinks
noextract=(R-source.tar.gz)

sha256sums=('SKIP'
            'SKIP'
            'SKIP'
            'SKIP'
            'SKIP'
            'SKIP')

prepare() {
  # Verify that InnoSetup is installed
  INNOSETUP="C:/Program Files (x86)/Inno Setup 6/ISCC.exe"
  msg2 "Testing for $INNOSETUP"
  test -f "$INNOSETUP"
  "$INNOSETUP" 2>/dev/null || true

  # Put pdflatex on the path (assume Miktex 2.9)
  msg2 "Checking if pdflatex and texindex can be found..."
  export PATH="$PATH:/c/progra~1/MiKTeX 2.9/miktex/bin/x64"
  pdflatex --version
  texindex --version

  # Extract tarball with symlink workarounds
  msg2 "Extracting R source tarball..."
  rm -rf ${srcdir}/R-source
  mkdir -p ${srcdir}/R-source
  MSYS="winsymlinks:lnk" tar -xf ${srcdir}/R-source.tar.gz -C ${srcdir}/R-source --strip-components=1
  cd "${srcdir}/R-source"

  # Ship the CA bundle
  cp "${srcdir}/cacert.pem" etc/curl-ca-bundle.crt

  # Ship the TclTk runtime bundle
  msg2 "Creating the TclTk runtime bundle"
  mkdir -p Tcl/{bin,bin64,lib,lib64}
  ${srcdir}/create-tcltk-bundle.sh  

  # Add your patches here
  patch -Np1 -i "${srcdir}/shortcut.diff"
  patch -Np1 -i "${srcdir}/blas.diff"
}

build() {
  msg2 "Copying source files for 32-bit build..."
  rm -Rf ${srcdir}/build32
  MSYS="winsymlinks:lnk" cp -Rf "${srcdir}/R-source" ${srcdir}/build32

  # Build 32 bit version
  msg2 "Building 32-bit version of base R..."
  cd "${srcdir}/build32/src/gnuwin32"
  sed -e "s|@win@|32|" -e "s|@texindex@||" -e "s|@home32@||" "${srcdir}/MkRules.local.in" > MkRules.local
  #make 32-bit SHELL='sh -x'
  make 32-bit
  
  # Build 64 bit + docs and installers
  msg2 "Building 64-bit distribution"
  cd "${srcdir}/R-source/src/gnuwin32"
  TEXINDEX=$(cygpath -m $(which texindex))  
  sed -e "s|@win@|64|" -e "s|@texindex@|${TEXINDEX}|" -e "s|@home32@|${srcdir}/build32|" "${srcdir}/MkRules.local.in" > MkRules.local
  make distribution
}

check(){
  # Use cloud mirror for CRAN unit test
  #export R_CRAN_WEB="https://cran.rstudio.com"

  # Run 64 bit checks in foreground
  cd "${srcdir}/R-source/src/gnuwin32"
  echo "===== 64 bit checks ====="
  make check-all
}

package() {
  # Derive output locations
  REVISION=$((read x; echo ${x:10}) < "${srcdir}/R-source/SVN-REVISION")
  CRANDIR="${srcdir}/R-source/src/gnuwin32/cran"

  # This sets TARGET variable
  $(sed -e 's|set|export|' "${CRANDIR}/target.cmd")

  # Copy CRAN release files
  cp "${srcdir}/R-source/SVN-REVISION" "${pkgdir}/SVN-REVISION.${target}"
  cp "${CRANDIR}/NEWS.${target}.html" ${pkgdir}/
  cp "${CRANDIR}/README.${target}" ${pkgdir}/

  # Determine which webpage variant to ship from target (for example "R-3.4.1beta")
  case "$target" in
  *devel|*testing)
    cp "${CRANDIR}/rdevel.html" "${pkgdir}/"
    ;;
  *patched|*alpha|*beta|*rc)
    cp "${CRANDIR}/rpatched.html" "${pkgdir}/"
    cp "${CRANDIR}/rtest.html" "${pkgdir}/"
    ;;
  R-4*)
    cp "${CRANDIR}/index.html" "${pkgdir}/"
    cp "${CRANDIR}/md5sum.txt" "${pkgdir}/"
    cp "${CRANDIR}/rw-FAQ.html" "${pkgdir}/"
    cp "${CRANDIR}/release.html" "${pkgdir}/"
    REVISION="$target"
    ;;
  *)
    echo "Unknown release type: $target"
    exit 1
    ;;
  esac

  # Helper for appveyor script
  echo "set revision=${REVISION}" >> "${CRANDIR}/target.cmd"
  cp "${CRANDIR}/target.cmd" ${pkgdir}/
}

Step 6: Build R

Launch Rtools40 via msys.exe which creates a shell window. Navigate to the subdirectory where the build files are using Unix style. For example, if the files are in “c:\R\R40”, navigate to “/c/R/R40”. Invoke the build via “./full-build.sh” which will probably run for hours. The process updates all the necessary components, builds both the 32 bit and 64 bit versions of R using OpenBLAS, checks the results for each version, and packages the build into an executable installer file. When done, the desired R-devel-win.exe will be either in the above subdirectory or in /src/R-source/src/gnuwin32/installer, if not both. This the executable which installs R 4+ for Windows with OpenBLAS!

Final notes, for now

Installing and using this modified version of R should pose no problems, nor should installing most packages from source. There are some packages which rely on compiled libraries—such as XML or nlopt—for which extra steps are needed to build from source. However, these are few and may be installed as binaries for now. I’d appreciate thoughts and corrections in the comments; good luck!

The post Building R 4+ for Windows with OpenBLAS appeared first on Strange Attractors.

To leave a comment for the author, please follow the link and comment on their blog: R – Strange Attractors.

R-bloggers.com offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Never miss an update!
Subscribe to R-bloggers to receive
e-mails with the latest R posts.
(You will not see this message again.)

Click here to close (This popup will not appear again)