Skip to content
A dev version of MAESTRO
C++ MATLAB Python
Branch: master
Clone or download
Hyoukjun Kwon Hyoukjun Kwon
Latest commit 70d13b4 Oct 14, 2019
Type Name Latest commit message Commit time
Failed to load latest commit information.
lib Improved stability. L1/L2 size constraints re-activated Oct 15, 2019
.gitignore Dev version update Jul 19, 2019
SConstruct Initial commit Jun 1, 2019
maestro-top.cpp Improved stability. L1/L2 size constraints re-activated Oct 15, 2019 Improved stability. L1/L2 size constraints re-activated Oct 15, 2019


A dev version of MAESTRO. We do not prorvide technical support for this version.

Package Dependences

C++ compiler (g++)

SCONS build system (scons)

Boost libarary (libboost-all-dev)

Python 2.7 or later

How to compile the code


How to run the program


How to change the parameters

Change the contents of "" For parameters other than listed below, please ignore it; active development is going on them so correct functionailty is not guaranteed.

--print_res=true/false : If set true, MAESTRO prints out detailed cost information to the screen

--print_res_csv_file=true/false : If set true, MAESTRO prints out a csv file that contains various statistics

--print_log_file=true/false : If set true, MAESTRO prints out a log file that contains various information of detailed computation patterns to "log.txt"

--DFSL_file='data/DFSL_description/Resnet50_dla.m' : Specify the target dataflow and layer description file

--noc_bw=64 : NoC bandwidth

--noc_hop_latency=1 : NoC latency per hops

--noc_mc_support=true : NoC multicast support (In current dev version it's always on)

--num_pes=256 : Number of PEs

--num_pe_alus=1 : PE ALU vector width

--l1_size=32 : l1 buffer size

--l2_size=512 : l2 buffer size

How to change the DNN model and dataflow

Create a DFSL file under "data/DFSL_description" and point the file using --DFSL_file parameter in ""

For syntax of the DFSL file, please refer to other DFSL files in data/DFSL_description.

How to convert Pytorch model to MAESTRO dataflow

cd data/pytorch_example

Use torchvision.models

Run with default setting


Run MAESTRO with the converted dataflow file

Change the contents of ""




Change the input arguement

python --input_size 3,224,224 --model mobilenet_v2 --dataflow os --outfile out.m

--input_size: the input image size of the first layer

--model: the model name from torchvision.models

--dataflow: the dataflow for each layer, choose from "os, ws, rs, dla"

--outfile: the MAESTRO dataflow output file name


Hyoukjun Kwon ( Main developer, developed core framework and functionalities

Prasanth Chatarasi ( Improved APIs

Felix (Sheng-Chun) Kao ( Implemented Pytorch front-end

You can’t perform that action at this time.