High performance computing challenges for research and industrial simulation codes

International audienceComputing hardware, from mobile devices to supercomputer clusters, is undergoing a paradigm shift : with the advent of multicore CPUs and accelerators (GPUs, Intel Xeon Phi), parallelism is becoming hierarchical, ubiquitous, heterogeneous, and more and more complex. Performance critical applications need new programming models to exploit efficiently such architectures. Guided by Herb Sutter analysis ([1], [2]) we draw practical conclusions for the design of future simulation codes, and the optimization of legacy codes