At the first Sberbank SmartDev technology conference on 20 May, Iosif Meerov and Valentin Volokitin made their report “Porting Computational Codes to DPC++: First Results and Examples of Use”. A team of researchers from the oneAPI Competence Center of the UNN Institute of Information Technologies, Mathematics and Mechanics, the Institute of Applied Physics of the Russian Academy of Sciences and the Gothenburg University presented their first results on adaptation of scientific software to the new heterogeneous computing model oneAPI introduced in 2020 by Intel.
The report describes the ways to port parallel software to the new Data Parallel C++ (DPC++) programming language.
According to the developers, one of the main advantages of DPC++ is its versatility – the ability to compile and build programs for working on different computing architectures built on the basis of traditional CPUs, graphics accelerators and FPGAs. Using the example of a computational physics algorithm, the authors describe the mechanisms that affect the performance of the DPC++ code and show that no significant performance loss occurred when switching to DPC++. The first results of calculations on Intel graphics accelerators are presented.
More information about the current stage of research can be found here.
The oneAPI Competence Center is a joint project of Lobachevsky University and Intel Corporation. The Center was opened at the UNN in 2021. One of the key tasks of the Center is to conduct research in modern physics using systems based on various architectures and hardware accelerators using the capabilities of the oneAPI cross-platform programming model.
Source: Lobachevsky University