PyMicroTracer

This is a trace-based micro-architecture simulator. This framework is able to approximate both static, dynamic ILPs at this phase.

Installation

---

Clone Capstone dissambler tools and and branch it to next and install its python bindings:

git clone https://github.com/aquynh/capstone.git
cd capstone
git branch next
./make.sh
sudo ./make.sh install
cd bindings/python/
sudo python3 setup.py install

Install following Ubuntu packages:

sudo apt-get install python3-tk unixodbc unixodbc-dev unixodbc-bin unixodbc libsqlite3-dev

Next install following python modules:

sudo pip3 install networkx numpy matplotlib pyodbc pydot colour xdot

Download odbc driver sqliteodbc-0.9995.tar.gz from http://www.ch-werner.de/sqliteodbc/ and install it:

tar xvf sqliteodbc-0.9995.tar.gz
./configure && make && sudo make install

Cerate odbc configuration file: sudo nano /etc/odbcinst.ini And copy following content to odbcinst.ini

[SQLite3]
Description=SQLite ODBC Driver
Driver=/usr/local/lib/libsqlite3odbc.so
Setup=/usr/local/lib/libsqlite3odbc.so
Threading=2

[SQLite]
Description=SQLite ODBC Driver
Driver=/usr/local/lib/libsqliteodbc.so
Setup=/usr/local/lib/libsqliteodbc.so
Threading=2

Usage

---

Please refer to example file and execute the runme.sh file

./runme.sh

All the generated graph and result will be saved in ./res folder

Switches

PyMicroTracer can simulate three different scheduling methods: Static, Out-of-order (O3), and Hybrid ((IC)²IS). As an example, for following command:

./main.py hello_trace.db 100 5,6 ./res/ hello -b 5,6 -s h,o,s -p -d

'hello_trace.db' is the run time trace, 100 is the coverage. You can use this value if the db file is huge and you want to read a portion of it. The 5,6 is the range of instruction window size which is going to change from 2^5 to 2^6, ./res/ is the output folder used for saveing results, -b 5,6 defines the basic block winddow, -s h,o,s defines the scheduling method: h for hybrid, s for static scheduling and o for O3 scheduling. The -p will plot the final results, and -d will generate the the dependency graph.

Extracting 'db' file out of your benchmark

---

You can clone 'Tracer' tool from https://github.com/SideChannelMarvels/Tracer to extract instruction db file from your application. If your application is going to leave a long trace of instruction, it is a good idea to modify Tracer.cpp and create an index for bbl_id in ins table.