Sergei Sergeevich Shumilin

Sergei Sergeevich Shumilin

Auto-generated profile, based on publicly available information

Sergei Sergeevich Shumilin

Межведомственный суперкомпьютерный центр РАН

Research Focus & Metrics

Academic Career Length

9Years (2018-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 May 2019 in Lobachevskii Journal of Mathematics
Verified
Journal Article
Closed access
PDF file
BSB. M. Shabanov
B. M. Shabanov
ARA. A. Rybakov
A. A. Rybakov
SSS. S. Shumilin
S. S. Shumilin
Modern calculation codes used in supercomputing are very demanding of computing resources. For their effective appliance requires the use of parallelization at all levels, starting with the use of...

30 September 2019
Verified
Journal Article
Open access
ARAlexey Anatiljevich Rybakov
Alexey Anatiljevich Rybakov
SSSergey Sergeevich Shumilin
Sergey Sergeevich Shumilin
Numerical methods based on solving the Riemann problem of the decay of an arbitrary discontinuity are extremely demanding of computational resources. Applying the data of numerical methods to modern computational grids requires the use of a supercomputer. Among the various tools for improving the performance of supercomputer applications, we can emphasize the vectorization of program code. The AVX-512 instruction set has a number of unique features allowing to apply vectorization to the Riemann solver software context, which results in a significant acceleration of the solver.

1 December 2020 in Lobachevskii Journal of Mathematics
Verified
Journal Article
Closed access
PDF file
GSG. I. Savin
G. I. Savin
BSB. M. Shabanov
B. M. Shabanov
ARA. A. Rybakov
A. A. Rybakov
SSS. S. Shumilin
S. S. Shumilin
Widespread application of supercomputer technologies in various spheres of life, as well as the need of high-performance calculations allows us to speak about the relevance of the problem of increasing the performance of computer codes on supercomputers of modern architectures. Vectorization of program code is a low-level optimization that can, with a relatively local and compact application, increase the productivity of computational codes by several times. Modern Intel microprocessors have support for a unique set of instructions AVX-512, which, due to its features, allows you to vectorize almost any kind of code written in a predicate form. A set of simple restrictions when developing programs along with vectorization tools to enable the use of the AVX-512 instruction set can significantly speed up the resulting program. The article discusses approaches to vectorization of flat loops—a special-purpose program context, the successful vectorization of which allows to increase the productivity of supercomputer applications even for such program code for which optimizing compilers are powerless.

27 November 2019 in Lobachevskii Journal of Mathematics
Verified
Journal Article
Closed access
PDF file
ARA. A. Rybakov
A. A. Rybakov
SSS. S. Shumilin
S. S. Shumilin
Numerical simulation of the surface ice accretion includes the work of various solvers that are performed iteratively and exchange data with each other. The calculation execution chain consists of...

30 September 2018 in Международный научный журнал "Современные информационные технологии и ИТ-образование"
Verified
Journal Article
Open access
PDF file
АЛА. Бендерский Л
А. Бендерский Л
ААА. Рыбаков А
А. Рыбаков А
ССС. Шумилин С
С. Шумилин С
Современные расчетные пакеты для проведения суперкомпьютерных вычислений крайне требовательны к ресурсам, поэтому остро стоит вопрос повышения их производительности. В то же время появление новых аппаратных архитектур открывает новые возможности по оптимизации программного кода. Набор инструкций AVX-512, появившийся впервые в 2016-м году в микропроцессорах Intel Xeon Phi второго поколения, является мощным инструментом для создании высокопроизводительного параллельного кода. Набор инструкций AVX-512 обладает рядом особенностей, среди которых специальные регистры масок, позволяющие отбирать отдельные элементы векторов для обработки, комбинированные арифметические операции, операции чтения из памяти элементов данных с произвольными смещениями, векторные трансцендентные операции, многообразие операций перемешивания элементов векторов. Эти и многие другие уникальные возможности позволяют достигать кратного ускорения на наиболее важных участках расчетных кодов. Отдельные численные методы оперируют специальными объектами, эффективность обработки которых критически влияет на эффективность всего пакета. Для расчетных кодов RANS/ILES, предназначенных для расчета нестационарных турбулентных течений, такими специфическими объектами являются матрицы размера 5x5, расположенные внутри матриц размера 8x8 в качестве подматриц. Основной операцией для работы с такими матрицами является их перемножение. В статье рассматривается эффективный подход к векторизации перемножения матриц размера 8x8, а также проводится исследование, как понижение размера матриц влияет на эффективность векторизации. Предложенный подход реализован с помощью специальных функций-инстринсиков, и его эффективность проверена экспериментами, проведенными на суперкомпьютере, находящимся в МСЦ РАН. Modern software packages for supercomputer calculations require a large amount of computing resources. At the same time there are new hardware architectures that open up new opportunities for program code optimizing. The AVX-512 instruction set is a unique tool with many useful features that allow creating high-performance parallel code for supercomputer calculations. The most striking features of AVX-512 instruction set are special mask registers that allow selecting the elements of vectors for processing, combined arithmetic instructions, vector transcendental instructions, operations of multiple memory access with different arbitrary offsets, and many others. Some numerical methods use special objects, the processing speed of which critically affects the speed of the entire calculation package. Matrices of size 5 by 5, represented as submatrices of 8-by-8 matrices, are such critical objects for the numerical RANS/ILES method, which is used for nonstationary turbulent flows calculating. The main operation for working with such matrices is multiplication. In this paper we consider an effective approach to vectorizing the multiplication of 8-by-8 matrices. Further, the effect of decreasing the dimension of the matrices on the efficiency of vectorization is estimated. The considered approach is realized with the help of special intrinsic functions for the AVX-512 instruction set and it is checked on a supercomputer located in JSCC RAS.

6 January 2026 in Machine Learning: Science and Technology
Verified
Journal Article
Open access
ARAlexander Ryabov
Alexander Ryabov
VNViacheslav Naumov
Viacheslav Naumov
SSSergei Shumilin
Sergei Shumilin
NYNikolay Yavich
Nikolay Yavich
SRSayan Ranu
Sayan Ranu
VVVladimir Vanovskiy
Vladimir Vanovskiy
Efficient simulations of complex physical systems described by partial differential equations (PDE) require computational methods that can reduce the resource demands without sacrificing the accuracy. Here, we introduce a framework based on graph neural networks (GNN) for learnable self-supervised differentiable coarsening of unstructured computational grids. We leverage graph-based representation of the physical system and offer a computational grid coarsening method which preserves the underlying physical properties together with the stability of the chosen numerical scheme with the help of the designed loss terms. The coarsening model is trained in a self-supervised way by minimizing the error between the output of the simulations on the coarsened and original grids.We evaluate the approach on four PDE systems of different types, covering both linear and nonlinear regimes, including subsurface flow and wave propagation. We demonstrate that the proposed framework exhibits the ability to maintain high fidelity in simulation outputs even after 95% reduction in the number of nodes, significantly reducing computational overhead. We show that the model exhibits generalizability to unseen scenarios, thereby outperforming the baselines. The generality of the developed framework is also proven by its possibility to adapt to implicit numerical schemes used to model stiff systems of PDEs. Thus, the developed approach demonstrates the ability to accelerate physical simulations without compromising accuracy.