Computing:

R Packages: C/C++ Code, Lapack, MKL, ATLAS, ACML, OpenMP, Profiling and more...

Document to appear here soon.

Installing R with Intel Compilers (icc, icpc, ifort) and MKL

I've always had impressive results with MKL and intel compilers on the Intel-based clusters that I have used in recent years. Given that I was just given access to a new Intel-based cluster that did not have a supported R installation, I decided to have some fun and build R with only intel compilers and MKL linked for all matrix operations (rather than the standard BLAS/Lapack implementations included with R). For anyone interested in doing the same, here are some pointers:

• Remove any remnants of other R versions from the environment variables (PATH, LD_LIBRARY_PATH etc)
• Edit the file 'config.site' included with R to customize your R installation as follows (you can tweak this as desired but it took me a few tries to find a configuration that worked):

  R_PAPERSIZE=letter

  CC="icc -std=c99"

  CFLAGS="-g -O2 -fPIC -DPIC"

  F77=ifort

  FFLAGS="-g -O2 -fPIC"

  MAIN_LD="icc -std=c99"

  MAIN_LDFLAGS="-Wl,--export-dynamic"

  CPICFLAGS="-fPIC -DPIC"

  FPICFLAGS="-fPIC"

  SHLIB_LD="icc -std=c99"

  SHLIB_LDFLAGS="-shared"

  DYLIB_LD="icc -std=c99"

  DYLIB_LDFLAGS="-shared"

  LDFLAGS="-L/usr/local/lib64 -L/usr/local/intel/mkl/10.0.3.020/lib/em64t" # Change to your MKL lib location

  CXX=icpc

  SHLIB_CXXLD=icpc

  SHLIB_CXXLDFLAGS="-shared"

  FC=ifort

• Set up your MKL specifications for R as follows:

setenv MKL_LIB_PATH "${MKL_HOME}/lib/em64t" # Again, modify depending on whether 32/64-bit

setenv LD_LIBRARY_PATH ${LD_LIBRARY_PATH}:${MKL_LIB_PATH}

setenv MKL "-L${MKL_LIB_PATH} -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_lapack -liomp5 -lpthread -lguide -lmkl_def -lmkl_core -lmkl"

Depending on your favourite shell, you can tweak as desired. I linked to multi-threaded MKL here, you could also link to serial MKL by modifying the linking options given above.

• Run: ./configure --with-blas="${MKL}" --with-lapack
• Run: make
• Cross your fingers. :)

Software:

RxCEcolInf

D. James Greiner, Paul D. Baines and and Kevin M. Quinn (2009). RxCEcolInf: R x C Ecological Inference With Optional Incorporation of Survey Information. R package version 0.1-1.

Recommended download from CRAN. Source code is available here.

cmd

Paul D. Baines (2009). cmd: Color-Magnitude Diagrams. R package version 0.1-1.

Source code is available here.