Jan Zapletal presented a poster describing the implementation of boundary element quadrature schemes for modern architectures at the PASC17 conference in Lugano, Switzerland.
Results of optimization for Knights Corner and Knighs Landing architectures were presented at Intel Xeon Phi User Group meeting in Chicago.
A paper describing our experience with vectorization for Knights Landing architecture has been submitted. Preprint is available.
A new approach to ACA system matrix distribution among computational nodes for the Laplace and Lame equations has been implemented in the library.
Recently, our team has been focusing on the optimization of the code for the latest Intel's many-core Xeon Phi architecture. Currently, there are two generations available - Knights Corner (KNC) and Knights Landing (KNL). The first of them is available in the form of PCIe coprocessor cards, while the second can be used as a stand-alone processors. They both share several common traits, including large number of cores equipped with wide SIMD vector registers.