Code Transformation and Optimization Project (A.Y. 2013/2014)

This directory contains the project developed for the CTO course at Politecnico di Milano, A.Y. 2013/2014.

Aim of the project is to develop a set of LLVM passes that, starting from a specification of the ranges of the local - floating point - variables, convert the code to fixed point with a predictable loss of precision.

Some notes about the structure and functionalities of the project are in the doc folder.

Quick Start

To create IR with the metadata used within the project, it is necessary to compile a modified version of clang as per the following instructions.

The project was developed and tested with LLVM 3.4 under Linux.

In the following, we assume that

• LLVM_SRC is the directory where the llvm source code is located
• LLVM_BUILD is the llvm build directory (separated from the source tree)

Download the LLVM 3.4 and clang sources:

$ wget http://llvm.org/releases/3.4/llvm-3.4.src.tar.gz
$ tar xvf llvm-3.4.src.tar.gz
$ mv llvm-3.4 $LLVM_SRC
$ rm llvm-3.4.src.tar.gz
$ cd $LLVM_SRC
$ wget http://llvm.org/releases/3.4/clang-3.4.src.tar.gz
$ tar xvf clang-3.4.src.tar.gz
$ mv clang-3.4 tools/clang
$ rm clang-3.4.src.tar.gz

Clone this repository in the projects/ directory:

$ git clone https://github.com/pogliamarci/llvm-float-range $LLVM_SRC/projects/float-range

Apply the provided patch (adding support for the float_range attribute):

$ cd $LLVM_SRC
$ patch -p1 < projects/float-range/llvm-float-range.patch
$ cd tools/clang
$ patch -p1 < $LLVM_SRC/projects/float-range/clang-float-range.patch

Configure LLVM, clang and the float-range project in LLVM_BUILD:

$ mkdir -p $LLVM_BUILD
$ cd $LLVM_BUILD
$ $LLVM_SRC/configure
$ mkdir -p $LLVM_BUILD/projects/float-range
$ cd $LLVM_BUILD/projects/float-range
$ $LLVM_SRC/projects/float-range/configure

Finally, configure and compile it all:

$cd $LLVM_BUILD
$ make

When running configure, you can as usual specify an installation prefix using --prefix, and install the binaries running make install. If you have a multi-core machine, you can speed up compilation passing the usual -j $N parameter to make. Beware that the LLVM and clang compilation and linking process require quite an high amount of memory.

The LLVM build system should automatically compile llvm and clang in release mode ($LLVM_BUILD/Release+Asserts/{bin,lib}), and the float-range project in debug mode ($LLVM_BUILD/projects/float-range/Debug+Asserts/lib). The build mode is chosen based on the directory being a repository checkout or not (it checks for .git or .svn directories).