Authors:
Jānis Kampars
1
;
Guntis Mosāns
1
;
Jānis Zuters
2
;
Aldis Gulbis
3
and
Rasa Gulbe
3
Affiliations:
1
Institute of Information Technology, Riga Technical University, 10 Zunda krastmala, Riga, Latvia
;
2
Faculty of Computing, University of Latvia, 19 Raiņa bulvāris, Riga, Latvia
;
3
Dati Group, 80 A Balasta dambis, Riga, Latvia
Keyword(s):
Software Optimization, Vectorization, SIMD.
Abstract:
To cope with the growing volume of data and complexity of the associated processing logic, modern CPU capabilities such as vector registers and SIMD (Single Instruction Multiple Data) instructions need to be taken advantage of. Although from a technical point of view, usage of SIMD instructions is not complicated, building computing tasks with good SIMD capabilities has always been a challenging task. Modern compilers assist developers to some extent with solutions like Compiler Automatic Vectorization, which is not always sufficient, and several researchers demonstrate that manual code optimization is still necessary. The paper gives an overview of the existing computing task optimization approaches, designs and describes development of a cloud-based software optimization platform and demonstrates its usage by optimizing a software correlator.