MPI-CUDA sparse matrix–vector multiplication for the conjugate gradient method with an approximate inverse preconditioner

Oyarzun, Guillermo; Borrell, R.; Gorobets, A.; Oliva, A.

doi:10.1016/j.compfluid.2013.10.035

Cited by 36 publications

(23 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The application of the technology and its component approaches to simulating turbulent flows is exemplified in [24], where compressible flows, a combined finite volume/finite difference method, unstructured grids, and a vertex centered scheme were consid ered; in [10,6], where compressible flows, the finite volume method, unstructured hybrid grids, polyno mial reconstruction, and a cell centered scheme were used; in [14], where incompressible flows, the finite volume method, and structured staggered grids were employed; and in [25], where incompressible flows, the finite volume method, and unstructured grids were applied. These algorithms are intended for large scale computations of turbulent flows employing tens of thousands of processor cores and multiple massively parallel accelerators.…”

Section: Resultsmentioning

confidence: 99%

Parallel technology for numerical modeling of fluid dynamics problems by high-accuracy algorithms

Gorobets

2015

Comput. Math. and Math. Phys.

View full text Add to dashboard Cite

Section: Resultsmentioning

confidence: 99%

Parallel technology for numerical modeling of fluid dynamics problems by high-accuracy algorithms

Gorobets

2015

Comput. Math. and Math. Phys.

View full text Add to dashboard Cite

“…Both preconditioners can be represented as a SpMV at the solution stage . Further details can be found in [22]. In conclusion, we express the full time integration step in terms of three algebraic kernels: the SpMV, the linear combination of two vectors (AXPY) and the dot product.…”

Section: Governing Equations and Numerical Methodsmentioning

confidence: 99%

Efficient CFD code implementation for the ARM-based Mont-Blanc architecture

Oyarzun

Borrell

Gorobets

et al. 2018

Future Generation Computer Systems

Self Cite

View full text Add to dashboard Cite

h i g h l i g h t s• Termo Fluids CFD code has been run on up to 128 ARM-based Mont-Blanc nodes.• An heterogeneous implementation has been developed to occupy the overall system.• A dynamic Tabu search load balance algorithm distributes the workload among devices.• The hybrid approach is up to times faster than the CPU-only version of the code.• Mont-Blanc nodes are 41% more energy efficient than Minotauro hybrid nodes. t r a c tSince 2011, the European project Mont-Blanc has been focused on enabling ARM-based technology for HPC, developing both hardware platforms and system software. The latest Mont-Blanc prototypes use system-on-chip (SoC) devices that combine a CPU and a GPU sharing a common main memory. Specific developments of parallel computing software and well-suited implementation approaches are of crucial importance for such a heterogeneous architecture in order to efficiently exploit its potential. This paper is devoted to the optimizations carried out in the TermoFluids CFD code to efficiently run it on the Mont-Blanc system. The underlying numerical method is based on an unstructured finite-volume discretization of the Navier-Stokes equations for the numerical simulation of incompressible turbulent flows. It is implemented using a portable and modular operational approach based on a minimal set of linear algebra operations. An architecture-specific heterogeneous multilevel MPI+OpenMP+OpenCL implementation of such kernels is proposed. It includes optimizations of the storage formats, dynamic load balancing between the CPU and GPU devices and hiding of communication overheads by overlapping computations and data transfers. A detailed performance study shows time reductions of up to 2.1× on the kernels' execution with the new heterogeneous implementation, its scalability on up to 128 MontBlanc nodes and the energy savings (around 40%) achieved with the Mont-Blanc system versus the highend hybrid supercomputer MinoTauro.

show abstract

“…Several works [11]- [13] have presented approaches for optimising matrix multiplications through the usage of GPUs. Both [11], [12] focused on the overlapping between computation and data communication in order to minimise the communication and transfer times. Dang et.…”

Section: Related Workmentioning

confidence: 99%

“…The matrix multiplication implementation considered a bi-directional data transfer. Oyarzun et al [12] proposed an approach for matrix multiplication in the context of the conjugate gradient method in multi-GPU environments. In particular, the proposed solver was tested for the Poisson equation and reported improvements of up to a 200% regarding the CPU-only solver.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Sparse-matrix arithmetic operations in computer clusters: A text feature selection application

Tommasel¹,

Mateos²,

Godoy³

et al. 2014

2014 IEEE Biennial Congress of Argentina (ARGENCON)

View full text Add to dashboard Cite

Arithmetic operations on matrices are frequently used in scientific computing areas. They usually become a performance bottleneck due to their high complexity. In this context, the parallel processing of matrix operations in distributed environments arises as an important field of study. This work presents several strategies for distributing sparse matrix arithmetic operations on computer clusters, focusing on the intrinsic characteristics of the operations and the matrices involved. The performance of the proposed strategies for determining the number of parallel tasks to be executed on the computer cluster was evaluated considering a high-dimensional feature selection approach. Additionally, the performance of two alternatives for efficiently representing bigscale sparse matrices was tested. Experimental results showed that the proposed strategies significantly reduce the computing time of matrix operations, outperforming computations based on serial and multi-thread implementations.Resumen-Las operaciones aritméticas matriciales son frecuentemente utilizadas en muchas áreas de la computación científica. Usualmente, son un cuello de botella debido a su alta complejidad computacional. En este contexto, el procesamiento paralelo de operaciones matriciales en ambientes distribuidos surge como un importante tópico de estudio. Este trabajo presenta diversas estrategias para la distribución de operaciones aritméticas entre matrices ralas en clusters de computadoras. Dichas estrategias se basan en las características intrínsecas de las operaciones y las matrices involucradas. El desempeño de las estrategias propuestas para determinar la cantidad de tareas paralelas a ejecutar, fue evaluado considerando un enfoque de selección de atributos multidimensional. Además, se evaluó el desempeño de dos alternativas para la representación eficiente de matrices ralas a gran escala. Los resultados experimentales mostraron que las estrategias propuestas redujeron los tiempos de cómputo de las operaciones, mejorando así significativamente los tiempos de las implementaciones seriales y multi-thread.

show abstract

MPI-CUDA sparse matrix–vector multiplication for the conjugate gradient method with an approximate inverse preconditioner

Cited by 36 publications

References 18 publications

Parallel technology for numerical modeling of fluid dynamics problems by high-accuracy algorithms

Parallel technology for numerical modeling of fluid dynamics problems by high-accuracy algorithms

Efficient CFD code implementation for the ARM-based Mont-Blanc architecture

Sparse-matrix arithmetic operations in computer clusters: A text feature selection application

Contact Info

Product

Resources

About