Large-scale Marchenko imaging with distance-aware matrix reordering, tile low-rank compression, and mixed-precision computations

Abstract

A variety of wave-equation-based seismic processing algorithms rely on the repeated application of the Multi-Dimensional Convolution (MDC) operator. For large-scale 3D seismic surveys, this comes with severe computational challenges due to the sheer size of high-density, full-azimuth seismic datasets required by such algorithms. We present a three-fold solution that greatly alleviates the memory footprint and computational cost of 3D MDC by leveraging a combination of i) distance-aware matrix reordering, ii) Tile Low-Rank (TLR) matrix compression, and iii) computations in mixed floating-point precision. By applying our strategy to a 3D synthetic dataset, we show that the size of kernel matrices used in the Marchenko redatuming and Multi-Dimensional Deconvolution equations can be reduced by a factor of 34 and 6, respectively. We also introduce a TLR Matrix-Vector Multiplication (TLR-MVM) algorithm that, as a direct consequence of such compression capabilities, is consistently faster than its dense counterpart by a factor of 4.8 to 36.1 (depending on the selected hardware). As a result, the associated inverse problems can be solved at a fraction of cost in comparison to state-ofthe-art implementations that require a pass through the entire data at each MDC operation. This is achieved with minimal impact on the quality of the processing outcome.

Can tile low-rank compression live up to expectations? An application to 3D multidimensional deconvolution

“…This ultimately leads to a faster and less memory demanding MDD process. This corroborates our previous findings in the context of Marchenko-based redatuming for a synthetic dataset modelled in a much simpler geological setting (Hong et al, 2021;Ravasi et al, 2022a).…”

High performance computing seismic redatuming by inversion with algebraic compression and multiple precisions

“…Our MP TLR-MVM implementation renders SRI competitive with industry standard methods from a computational point of view, and therefore attractive in practice. Related work (Ravasi et al, 2022) has recently shown the versatility of our approach and its applicability to other geophysical problems. However, whilst Ravasi et al (2022) are mostly concerned with the geophysical aspects of the problem, our article focuses on detailing the implementation of our synergistic MP TLR-MVM approach as well as providing an extensive suite of performance benchmarks.…”