Scientific Computing and Algorithms in Industrial Simulations by Michael Griebel Anton Schüller & Marc Alexander Schweitzer

Author:Michael Griebel, Anton Schüller & Marc Alexander Schweitzer
Language: eng
Format: epub
Publisher: Springer International Publishing, Cham

On the bottom, all external libraries are listed that are (optionally) used by LAMA for the support of the different architectures and inter-node communication. Primarily, these are the runtime libraries for supported devices (OpenMP®; thread, CUDA®; 5 driver and runtime, OpenMP6 for Accelerator)7 as well as vendor specific BLAS,8 sparse BLAS and LAPACK9 implementations. Currently we provide interfaces to Intel MKL10/BLAS, cuBLAS and cuSPARSE. For inter-node parallelization LAMA makes use of the communication standards MPI11 (message passing programming mode) or GPI12 (partitioned global address space programming model). Efficient implementations are provided on the respective sides.

The next level indicates our main concept for the design of portable parallel applications. A strict separation between memory management, kernel routines, tasking and communication creates a common layer for all supported devices, asynchronous execution and distributing data. As a consequence, all routines implemented on the next higher levels are independent of the underlying device and communication model.

More details about the memory and the kernel management will follow in Sects. 2.1 and 2.2. Section 2.3 explains the principles of tasking and Sect. 2.4 details how data is distributed across node boundaries. The top two levels with the most abstraction from the hardware model exhibit the functionality for dealing with matrices and vectors and finally linear equation solvers. The main functionalities are described in Sects. 2.5 and 2.6.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.