PENE

PENE (Pin Enabled Numerical Exploration) is a dynamic instrumentation tool based on Intel Pin that instruments floating-point operations. It is compatible for Linux and Windows systems.

PENE allows to change the rounding mode for floating-point operations by choosing one of the IEEE rounding mode or random rounding mode.

Changing the rounding mode was possible thanks to the implementation of the library that allows to change rounding mode by Verrou developers.

Requirements

• cmake version >= 3.20
• make
• g++ compiler for Linux and MSVC compiler for Windows
• python3
• jinja2
• pytest

Installation

First clone the repository.

Then, go inside the repository then configure and build the project and finally run the tests.

Installation for Linux

Run via your command line on Linux :

cd [PENE repository]
mkdir build
cd build 
cmake -DCMAKE_BUILD_TYPE=Debug ..
make 
ctest -C Debug --output-on-failure

Installation for Windows

Run via your command line on your powershell on Windows :

cd [PENE repository] 
New-Item build -ItemType Directory
cd build

##runing cmake 
cmake -DCMAKE_BUILD_TYPE=Debug ..

##build solution
cmake --build .

##Testing
ctest -C Debug --output-on-failure

You can also install the tool via a code editor.

How to execute your code with PENE?

To use PENE to execute your application, run:

path/to/pin -t path/to/pene.so/or/pene.dll [OPTIONS] -- path/to/executable

• The path/to/pin should be in your PENE repository under /Pin/Linux/pin if you are on Linux machine and /Pin/Windows/pin if you are on Windows machine.
• The path to the PENE tool is the path to the shared object file that corresponds to PENE tool, which was created during the build. Without specifying any options, this command will execute your code under PENE without any instrumentation.

How to instrument your code with PENE?

To instrument your code, you can use the fp-replace option:

path/to/pin -t path/to/pene.so/or/pene.dll -fp-replace [0 or 1] -- path/to/executable

-fp-replace 0 is the default mode when no instrumentation is done -fp-replace 1 corresponds to instrumenting the application and replacing floating-point operations by operations that use also IEEE default rounding mode.

To be able to change the rounding mode for floating-point operations while instrumenting, you should use:

path/to/pin -t path/to/pene.so/or/pene.dll -fp-replace 3 -rounding-mode [0 to 5] -- path/to/executable

where:

• if rounding-mode is not specified or set to 1 : nearest rounding mode
• rounding-mode = 2 : upward rounding
• rounding-mode = 3 : downward rounding
• rounding-mode = 4 : rounding to zero
• rounding-mode = 5 : random rounding

How to exclude parts of the code from instrumentation?

You can filter which parts of the code your want to instrument using the option -exclude along with an exclusion file containing a list of the functions you want to exclude.

The exclusion file should follow the following format:

• one exclusion rule per line
• each line should be of the form pathsymbol fctname , where pathsymbol is the path for the symbol: the shared library or the executable, and fctname is the function name.

To generate the list of symbols with floating-point instructions present in your code, you can use the option -gen-sym-list:

path/to/pin -t path/to/pene.so/or/pene.dll -gen-sym-list list.lst -- path/to/executable

The list of symbols will be saved in list.lst in that case. You can rename this file.

There is also an option -gen-sym-list that sets the mode for symbols generation:

path/to/pin -t path/to/pene.so/or/pene.dll -gen-sym-list list.txt -gen-sym-mode [0 or 1] -- path/to/executable

When the mode is set to 0, it will list all the symbols with floating-point instructions that have been loaded during execution. If set to 1, it will list all the symbols with floating-point instructions that have been executed. Default mode is 0.