cheuerde

#!/bin/bash
# We are going to compile the C/C++ parts with intel's icc/icpc
export CC="icc"
export CXX="icpc"
export AR="xiar"
export LD="xild"

export F77='gfortran'
export FC='gfortran'

cheuerde

#### (Generalized) linear mixed models in R for continous and categorical traits for animal models ###
#
# General TasK: Fit (G)LMM in R for an animal model with hundreds 
# of thousands of observations.
# This is generally a feasable task, as the pedigree-based covariance structure of
# the main random term is very sparse (A-Inverse).
# Nevertheless, this is far from trivial directly in R and there are a lot of 
# caveats. 
#
# Problems:

cheuerde

#####################################################
### How to compile R and link it against OpenBLAS ###
#####################################################

# Assuming no root priviledges on the destination machine,
# so we have to compile our software locally

# set default installation directory:
INSTDIR=$HOME/opt

cheuerde

### as of september 2014 Eigen offers in its devel branch avx vectorization
### see: http://comments.gmane.org/gmane.comp.lib.eigen/4545

### get the devel-headers

wget -O - http://bitbucket.org/eigen/eigen/get/default.tar.gz > Eigen-Devel.tar.gz

### or

hg clone https://bitbucket.org/eigen/eigen/

cheuerde

####################################################
### Claas Heuer, Sep. 2014                       ###
### Example of performing Gauss-Seidel with      ###
### residual updating using NetCDF stored data.  ###
### Algorithm adopted from: de los Campos (2012) ###
####################################################


# this library allows us to read/write NetCDF files in R
# Note: for compiling (on linux) the netcdf headers are required:

cheuerde

// Claas Heuer, August 2013
// Computing a dense crossproduct 
// with distributed data using MPI and OpenMP.
// Depends Eigen (http://eigen.tuxfamily.org/index.php?title=Main_Page)

// compile:
// mpicxx -fopenmp -O2 -I/path/to/Eigen crossproduct_mpi_mp.cpp -o cross.o
//
// run the program (in shell):
// # MP = number of cores for openMP

cheuerde

# it is straightforward to change the BLAS-library that is used by an R-instance.
# For this we can use the environmental variable LD_PRELOAD.
# This variable will load any library specified prior to all others, so we
# can define a priority for certain libraries
#
# So in terms of R we would do something like this:

# imagine my BLAS-library is located in /home/user/myBLAS/BLAS.so

# invoke R with 'myBLAS'

cheuerde

# cuBLAS/cuBLAS-xt is a full CUDA implementation of BLAS 1-3
# ACML is AMD's BLAS 1-3 optimized for AMD GPUs and fma4-CPUs

### cuBLAS
#
# get cuBLAS (note: non-free, closed source) 
https://developer.nvidia.com/cuda-downloads

# example of how to use: http://devblogs.nvidia.com/parallelforall/drop-in-acceleration-gnu-octave/

cheuerde

### using ddot and daxpy
# system("curl -O -s https://gist.githubusercontent.com/cheuerde/2f59dd5087e201daee78/raw/1222222c5b4b38c04c55571138da7a962a74f4d0/ddot_daxpy_R.c ")
# system("R CMD SHLIB ddot_daxpy_R.c")
# dyn.load("ddot_daxpy_R.so")
#
# in R:
#
# ddot:
# .Call("ddot",x,y)
#

cheuerde

###############################################
### How to link R against OpenBLAS on a Mac ###
###############################################


# this will install brew to /usr/local
ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"

brew doctor