[R2]

R2 is lightweight languages extensions for Java 
to reduce manual annotations and automate approximate programming. 
R2 extensions enable programmers to implicitly relax the accuracy 
of program data and operations.

[Compiling R2]

1. Setting Environmental Variables

	To use the R2 framework, first you need to set all the necessary environment variables. 

	JAVA_HOME and R2_HOME should be set according to your own machine configurations. 

		export JAVA_HOME=your/java/installation/path
		export R2_HOME=your/r2.code/path

	Once your JAVA_HOME and R2_HOME are set properly, 
	the following five variables can be just copied and pasted.

		export JSR308=$R2_HOME/checker-framework/
		export R2_INST=$R2_HOME/r2.optimization
		export CHORD_MAIN=$R2_HOME/chord/main
		export R2_ANALYSIS=$R2_HOME/r2.analysis
		export R2_BENCH=$R2_HOME/r2.apps

	You can manage your memory configuration of analysis using the following variable. 
	You can see the chord's reference for detailed information. (https://code.google.com/p/jchord/)

		export CHORD_MAX_HEAP=8192m

2. Building R2

	We provide a script to build everything called "build.sh".
	What you have to do is only typing the following:

		$ ./build.sh 

	This command will make the framework to be ready for use. 

[Using R2]

  1. Approximation Safety Analysis

    You can find all R2 benchmarks at "r2.apps" directory.
    There are nine benchmarks in r2.apps: mc, fft, lu, sor, smm, simpleRaytracer, jmeint, zxing.

    Let me explain with an example, mc.
    If you see inside of mc (MonteCarlo) directory, you would see a source directory, src.
    You can add or remove annotations from source codes in src.

    Now, you can run analysis using a tool called "analysis.py" in r2.apps as described in the following command.

		$ ./analyze.py mc

    Generally, you can put any benchmark name and the name should be matched with the corresponding directory name.

		$ ./analyze.py [bench]

    There are three phases:
      (1) Static Analysis
      (2) Back Annotation
      (3) Compilation

    The tool automatically generates another copy of src named "src-marked".
    The only disparity betwteen src-marked and original src is comments describing
    relaxed data and operations found by the static analysis .
    The comments are placed at the end of source code line corresponding the data and operations.

  2. Stochastic Optimization

    You can also run the stochastic optimization using "optimize.py" in r2.apps as following:

		$ ./optimize.py mc 1

    [0|1|2|3] represent error and energy models: low, medium, high, and aggressive.
    Therefore, the above command will run the optimization for mc (MonteCarlo) on medium specification.

    The following is the general command to run optimization:

		$ ./optimize.py [bench] [0|1|2|3]