High-Performance Seismic Modeling with Finite-Difference Using Spatial and Temporal Cache Blocking

Etienne, V.; Tonellot, Thierry; Malas, Tareq B.; Ltaief, Hatem; Kortas, Samuel; Thierry, Philippe; Keyes, David E.

doi:10.3997/2214-4609.201702324

Cited by 9 publications

(7 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To increase data locality, TB through tiling techniques (Bandishti et al, 2012;Grosser et al, 2014b;Malas et al, 2015;Orozco and Gao, 2009;Strzodka et al, 2011;Wellein et al, 2009;Wonnacott, 2000;Yuan et al, 2017;Zhou, 2013) has been widely considered using various advanced programming models to favor asynchronous execution. Performance tuning using roofline models (Datta, 2009;Etienne et al, 2017;Nguyen et al, 2010;Titarenko and Hildyard, 2017) remains an important assessment step for stencil computations to ensure a good utilization of the underlying hardware resources. Some of these efforts have translated into software releases (e.g.…”

Section: Prior Work and Current Contributionsmentioning

confidence: 99%

Asynchronous computations for solving the acoustic wave propagation equation

Akbudak

Ltaief

Etienne

et al. 2020

The International Journal of High Performance Computing Applica

View full text Add to dashboard Cite

The aim of this study is to design and implement an asynchronous computational scheme for solving the acoustic wave propagation equation with absorbing boundary conditions (ABCs) in the context of seismic imaging applications. While the convolutional perfectly matched layer (CPML) is typically used for ABCs in the oil and gas industry, its formulation further stresses memory accesses and decreases the arithmetic intensity at the physical domain boundaries. The challenges with CPML are twofold: (1) the strong, inherent data dependencies imposed on the explicit time-stepping scheme render asynchronous time integration cumbersome and (2) the idle time is further exacerbated by the load imbalance introduced among processing units. In fact, the CPML formulation of the ABCs requires expensive synchronization points, which may hinder the parallel performance of the overall asynchronous time integration. In particular, when deployed in conjunction with the multicore-optimized wavefront diamond temporal blocking (MWD-TB) approach for the inner domain points, it results in a major performance slow down. To relax CPML’s synchrony and mitigate the resulting load imbalance, we embed CPML’s calculation into MWD-TB’s inner loop and carry on the time integration with fine-grained computations in an asynchronous, holistic way. This comes at the price of storing transient results to alleviate dependencies from critical data hazards while maintaining the numerical accuracy of the original scheme. Performance and scalability results on various x86 architectures demonstrate the superiority of MWD-TB with CPML support against the standard spatial blocking on various grid sizes. To our knowledge, this is the first practical study that highlights the consolidation of CPML ABCs with asynchronous temporal blocking stencil computations.

show abstract

Section: Prior Work and Current Contributionsmentioning

confidence: 99%

Asynchronous computations for solving the acoustic wave propagation equation

Akbudak

Ltaief

Etienne

et al. 2020

The International Journal of High Performance Computing Applica

View full text Add to dashboard Cite

show abstract

“…Finding the optimal block size is not straightforward since it depends on the FD scheme stencil, the model size and the underlying hardware (e.g., number of threads and cache size). We follow a simple approach proposed by Etienne et al (2017), to find the optimal setting. We perform a series of computations where the block size is parameterized and different for each computation.…”

Section: Spatial Blockingmentioning

confidence: 99%

“…We assume it is better not to divide the domain along z, which corresponds to the inner index in order to keep an efficient hardware vectorization through Intel AVX instructions. We perform the test using a 512x512x512 grid points without CPML and determine that the optimal block size is 16 points along x, and 5 points along y (Etienne et al 2017). These results are obtained using 32 OpenMP threads on one computing node of the supercomputer Shaheen II at KAUST, based on a dual-socket 16-core Intel Haswell CPU.…”

Section: Spatial Blockingmentioning

confidence: 99%

“…Below, we investigate how spatial (Etienne et al 2017) and temporal (Malas et al 2015 cache blocking techniques can speed-up the FDTD computations on homogeneous multicores CPU x86 architectures. Both cache blocking techniques aim at increasing locally cached data, which is one of the most critical aspects for performance efficiency, especially when it comes to memory-bound algorithms like FDTD.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Application of High Performance Asynchronous Acoustic Wave Equation Stencil Solver into a Land Survey

Abdelkhalak

Akbudak

Etienne

et al. 2019

Day 3 Wed, March 20, 2019

View full text Add to dashboard Cite

This template is provided to give authors a basic shell for preparing your manuscript for submittal to an SPE meeting or event. Styles have been included (Head1, Head2, Para, FigCaption, etc) to give you an idea of how your finalized paper will look before it is published by SPE. All manuscripts submitted to SPE will be extracted from this template and tagged into an XML format; SPE's standardized styles and fonts will be used when laying out the final manuscript. Links will be added to your manuscript for references, tables, and equations. Figures and tables should be placed directly after the first paragraph they are mentioned in.

show abstract

“…We investigate the extension of Multicore-optimized Wavefront Diamond Temporal Blocking (MWD-TB) approach (Malas et al, 2015(Malas et al, , 2017 -already introduced for the second-order differential formulation (Etienne et al, 2017) -to the first order formulation. This is achieved by doubling the total number of time iterations and performing an update of the pressure wavefield and the velocity wavefield alternately ( Figure 3).…”

Section: Temporal Cache Blockingmentioning

confidence: 99%