noc_mesh

---

Network on Chip implemented on a Mesh Infrastructure

This is a implementation of 3x3 mesh type Network on Chip, with total 7 processing nodes implemented in Bluespec System Verilog. Each processing element is a primitive implementation of SMIPS, (from MIT 6.884) running a simple Matrix Vector Product. Each packet routing is done through a sophisticated arc routing generated from running multi commodity flow solver over the network specification to minimize congestion. The idea is to benchmark this routing process over a normal X-Y routing, and analyse the total cycle time and average packet hops taken.

Dependencies

• Bluespec
• Python
• Map and Route flow maintained at https://github.com/ssk1328/mcmf/
• src/Lookup.bsv from /bsv folder in git repo https://github.com/ssk1328/mcmf

Using Bluespec

• Run this on machine with bluespec installed (hpc28)
• Setup bluespec on your machine
• Copy the /opt/Bluespec-/ directory from hpc28 to your machine
• Add bluespec library paths as specified in .bashrc of hpc28 to .bashrc of your machine
• To verify $ bluespec this will open GUI window of bluespec

Setting up map and route git repo

cd ..

git clone https://github.com/ssk1328/mcmf.git

mcmf and noc_mesh should be initiated in the same common directory. More information on generating these results in mcmf repository

While in the noc_mesh git directory, for the route choosen: mcf or xy

Run

cp ../mcmf/bsv/Lookupmcf.bsv ../noc_mesh/src/

mv bsv/Lookupmcf.bsv bsv/Lookup.bsv

$ cd sim

$ make

This will clean compile link and simulate the source files and give the cycle count as one of the results of system. Data and processor dump files in /dump copied to appropiate folder in scripts

Scripts

$ cd ..

$ cp -r dump/proc scripts/proc_mcf

$ cd scripts

$ python analyseProcDump.py mcf

• Generated data report files reportSim_mcf.txt
• This gives average packet inflight time in the network

Multi commodity flow done to generate src/Lookup.bsv, preproceesing for that done here https://github.com/ssk1328/mcmf/

This is done as a part of DDP Stage 2 project work.

Shashank Gangrade
High Performance Computing Lab
Department of Electrical Engineering
IIT Bombay