The first compiler of its kind in the world with machine learning features, intended for free use by developers
Researchers at the IBM research laboratory in Haifa have developed a compiler, the first in the world, for machine learning (Machine Learning) written in open source code, which makes it possible to shorten development times and improve the work of new application developers.
The new software - Milepost GCC, revealed by IBM researchers performs intelligent analysis and optimization of applications, and thus the pace of development of software systems can be significantly accelerated. Thus, for example, the researchers were able to show an average improvement of 18% in embedded applications running on Power6 processors beyond the performance of code that used the standard optimizations.
The new tools enable software developers to streamline and shorten the development of software systems for various products ranging from mobile phones to supercomputers.
Dr. Belha Mendelson, manager of code optimization technologies at the IBM research laboratory in Haifa, stated that the new system knows how to automatically learn how to extract maximum performance from a given hardware - whether it is mobile phones, desktop computers or complete systems. As a result of the optimization process, the code will run faster - and will require less energy. The decision to open the compiler environment allows access to artificial intelligence and the instructions of machine learning systems - to the point of automatically determining the nature of the specific optimization that will yield the maximum improvement in performance.
The new compiler which was developed as an extension of the standard compiler in the open source community, GCC, and in collaboration with a pan-European consortium for research and development, as part of a research program of the European community with partners from the UK and France. The availability of the new tool will allow, for example, to shorten the development times of new software, while adjusting the applications and adapting them to different target architectures. Thus, for example, today the development of a mobile phone requires many months of adjusting the software until a reasonable level of performance is guaranteed. The new tool can reduce this time by up to a tenth.
As a companion product to the new system, the European consortium also introduced a code performance tuning website, available for use by the developer community. Developers can now upload their software code to this site - and automatically receive notifications and suggestions about ways to improve application speed.
One response
What is the URL?