The aim of this thesis is acceleration of the process of calculating generalized eigenvalues of a paired matrix using the QZ algorithm and using parallel processing capabilities in this algorithm.
We used SuperGlue, a new framework with parallelism and very little overhead. The operations of the algorithm are defined as SuperGlue tasks which resolve data dependencies with data versioning.
By specifying dependencies, we could perform tasks more efficiently.
We chose the reduction process to Hessenberg form for parallesion as a part of the algorithm. This is a two part process: the first part is a reduction to block Hessenburg form, and the second part is reduction from block hessenburg, to hessenburg form. The first part is just parallelised by the new framework, and the second part is replaced by a new algorithm that is more efficient and parallelised by SuperGlue.