Our tool
With the latest innovative research from reference labs such as IETR and Inria, Kaktus proposes its first tool to help embedded systems engineers to reduce their time to market while optimizing their applications. So as to be able to embed millions of code lines in embedded systems, optimization becomes a must-have. So as to get an interesting trade-off between application cost and performance, we propose optimizing the data word-lengths of your application.
Tool flow
The automated optimization software tool is composed of 4 main blocks.
The first step is composed of block 4 (source code transformation and source to source transformation). This step allows the proposed tool to take as input any C/C++ code and to modify it so as to benefit as much as possible of the proposed optimization. At the end of this step, a similar and readable C++ code is generated with generic finite precision arithmetic data-types.
The second step is composed of block 1 (Integer Word-Lengths (IWL) determination). This step runs the obtained source code on all the possible inputs so as to determine the range of each variable to convert. At the end of this step, histograms of the range of each variable are obtained as well as the deduced IWL.
The last step is composed of blocks 2 and 3 (Word-Lengths (WL) optimization, accuracy and cost evaluation). During this step, different optimization algorithms can be run so as to optimize a user-defined criterion (energy, area) subject to a maximum quality loss. At the end of this step, each variable in the application is optimized, and the source code is generated with the optimized data-types and can be injected as input of a High-Level Synthesis tool.
Money savings
Choose the most suited hardware for your application leading to conception costs gains.
Sustainability
Optimize the energy consumption of your application.
Speedup
Reduce your time-to-market with our automated optimization tool leading to production costs gains.
On demand
Kaktus also proposes services to optimize your application subject to quality, energy, area or timing constraints.
Training
Learn the best practices for software developing with the aim of hardware implementation
Integration in your environment
Optimize the energy consumption of your application.
Suited to the features of your application
Reduce your time-to-market with our automated optimization tool leading to production costs gains.
Our team
Justine Bonnot
CEO and Approximate computing expert
After an electrical engineering degree and a signal processing PhD in INSA Rennes (France), Justine decided to take her research on approximate computing out of the lab with the creation of Kaktus. She leads this project since almost 2 years. She is convinced that this project can make embedded software efficiency take a step forward.
Olivier Maffeis
Compiling Expert
After a PhD in IRISA/Inria and an ERCIM postdoc research, Olivier worked as an IT Architect and as a Technical Manager in IT Services companies for large projects in CRM and OSS/BSS. In 2020, Olivier joined Justine's exiting Kaktus project to provide a software solution to help customers improving efficiency and power-consumption of their embedded applications.
They talk about us
News
Optimiser la consommation des applications embarquées
Pour économiser la batterie, les applications embarquées doivent consommer le minimum d'énergie et fonctionner avec une faible empreinte mémoire. D'où la nécessité d'optimiser le code afin d'alléger le travail du processeur. Dans ce domaine, des résultats de recherche d'Inria et de l'IETR vont aboutir à la création d'une startup autour d'un compilateur facilitant l'identification du meilleur compromis qualité/performance.
Emergences, 2019
Read more
05/02/2020
Contact us
Inria, 263 Avenue Général Leclerc, 35042 Rennes
Mon - Fri, 8:30-18:30
+ 33 2 99 84 75 20
Supported by
