Automatic program parallelization with block data distribution
Lev Romanovich Gervich, Evgeniy Nikolaevich Kravchenko, Boris Yakovlevich Steinberg, Mikhail Viktorovich Yurushkin
Southern Federal University, 105/42 Bolshaya Sadovaya Str., Rostov-on-Don, 344006
Keywords: automatic parallelization, tiling, memory, distributed memory, block distribution of arrays, optimization of memory, distribution with overlapping
Abstract
This paper discusses several automated methods of acceleration programs. The acceleration is achieved by parallelization and optimization of memory access. Optimization of accesses to RAM is achieved by switching to a block code and block placements arrays. When using a distributed memory, the automated distribution of arrays and array distribution with overlapping are employed. Automation is implemented using the C language with pragmas in Open Parallelizing System. This paper presents the numerical results for linear algebra and mathematical physics. Some features of this demonstration converter have a remote access to the Internet.
|
