First create the document folder and configure repository

sudo apt-get install git
mkdir mydocument
cd mydocument/
git init
vi .git/config

Add the following lines

[core]
	repositoryformatversion = 0
	filemode = true
	bare = false
	logallrefupdates = true
	autocrlf = true

[alias]
    wdiff = diff -w -b --ignore-space-at-eol --word-diff=porcelain --color

Now create .gitignore to skip files which are processing artefacts

vi .gitignore

And add the following entries

# Ignore LaTeX processing files
*.aux
*.log
*~
*.out
*.bbl
*.blg
*.dvi
*.pdf

Now add files to repository and commit

git add *
git commit -m "First commit"

Each time You change something You can use the following commands
git wdiff To see word-by-word changes
git log To see the history of changes
git commit To commit the differences
git help For other commands

Some of my experience with released in April R 2.13 including (still not default because in development phase) compiler.

library(microbenchmark)
library(compiler)
# uglyPCAPred_Function is function which has some nested loops and does svd to predict scores for each level of factor from remaing ones
luglyPCAPred_Function < - cmpfun(uglyPCAPred_Function)
microbenchmark(uglyPCAPred_Function ,times=100000)
#Unit: nanoseconds
#                     expr min  lq median  uq    max
#1 uglyPCAPred_Function   70 559    628 629 255898
microbenchmark(luglyPCAPred_Function ,times=100000)
#Unit: nanoseconds
#                      expr min  lq median  uq    max
#1 luglyPCAPred_Function   69 559    628 629 116355
#> 255898/116355
#[1] 2.199287 !!!!

To test the performance for different implementations of libraries used for matrix operations the following R code was used

library(microbenchmark);
n = 100;
prodAxA = function(A) {
A%*%t(A)
}
mtime = rep(0,n);
for(i in 1:n) {
A = matrix(runif(i^2),nrow=i,ncol=i);
mtime[i] = median(microbenchmark(prodAxA(A))$time)
}

All libraries were O3 optimized for Core2 processor family, by rebuilding packages according to standard rules provided by package maintainer on GNU/Linux Debian operating system.

The performance of following matrix operation
was measured in nano-seconds 100 times and median time was recorded for from 1 to 100 matrix size.

According to results presented in the following figure, OpenBlas preformed best.

OpenBlas was 76% faster then original implementation and 36% faster then Atlas implementation.

The measurements were performed on Intel(R) Core(TM)2 Duo CPU T8300 @ 2.40GHz processor with performance governor.