id	author	title	date	pages	extension	mime	words	sentence	flesch	summary	cache	txt
ap-1795	Hladík, Jakub; Lórencz, Róbert; Šimeček, Ivan	Clock Math — a System for Solving SLEs Exactly	2013	5	.pdf	application/pdf	2978	172	60	As we now have a working system, we would like to proceed to: • add support for both single and double precision for matrix elimination, as the β vector module count and GPU performance may differ; • add support for larger matrices than 4096× 4096 – optimize SMP shared memory usage; • add support for automatic kernel group size tuning for larger matrices, as the group size lowers the mem- ory access time on different GPUs/architectures; • examine the modulus operation performance across different GPU architectures and further opti- mize the SAXPY, respectively DAXPY functions (modm); • utilize OpenMPI library to add cluster support; • adjust and run the solver on our university STAR cluster to test AMD’s OpenCL CPU implementa- tion. The goal of matrix scaling is to adjust all the floating-point numbers of the matrix to their corresponding integer versions.	cache/ap-1795.pdf	txt/ap-1795.txt