The rapid development of RISC-V instruction set architecture presents new opportunities and challenges for software developers. Is it sufficient to simply recompile high-performance software optimized for x86-64 onto RISC-V CPUs? Are current compilers capable of effectively optimizing C and C++ codes or is it necessary to use intrinsics or assembler? Can we analyze and improve performance without well-developed profiling tools? Do standard optimization techniques work? Are there specific RISC-V features that need to be considered? These and other questions require careful consideration. In this paper, we present our experience optimizing four BLAS algorithms for band matrix operations on RISC-V processors. We demonstrate how RISC-V-optimized implementations of OpenBLAS algorithms can be significantly accelerated through improved vectorization of computationally intensive loops. Experiments on Lichee Pi 4A and Banana Pi BPI-F3 devices using RVV 0.7.1 and RVV 1.0 vector instruction sets respectively, show speedups of 1.5x to 10x depending on the operation compared to the OpenBLAS baseline. In particular, the successful use of vector register grouping with RVV can lead to significant performance improvements.

The interaction of multipetawatt lasers with plasma is a complex multiparameter problem, providing a wide field for fundamental research and opening up great opportunities for creating unique sources of high-energy electrons and positrons, dense pair plasma, and gamma-photons. However, to achieve the high efficiency of such a source, it is necessary to use targets with optimized parameters, primarily density and size, for the given laser parameters. With the use of 3D QED-PIC modeling, it is shown that, when targets whose size is comparable with the laser wavelength are irradiated by laser beams with a total power of several tens of PW, the total initial number of target electrons may be regarded as the similarity parameter. In practice, this can significantly simplify the selection of the targets needed for controlling the interaction and, accordingly, for achieving the specified parameters of the developed electron-positron plasma and gamma-photon sources. Based on the similarity parameter, various laser-plasma interaction modes are identified, the necessary conditions for their launch are determined, and the properties of the pair particle and gamma-photon source are revealed. Moreover, qualitative estimates of the quantitative and energy characteristics of such a source are obtained, allowing it to be optimized for various laser beam configurations. (c) 2025 Author(s). All article content, except where otherwise noted, is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International (CC BY-NC) license (https://creativecommons.org/licenses/by-nc/4.0/).

The emergence and rapid development of the open RISC-V instruction set architecture opens up new horizons on the way to efficient devices, ranging from existing low-power IoT boards to future high-performance servers. The effective use of RISC-V CPUs requires software optimization for the target platform. In this paper, we focus on the RISC-V-specific optimization of the CatBoost library, one of the widely used implementations of gradient boosting for decision trees. The CatBoost library is deeply optimized for commodity CPUs and GPUs. However, vectorization is required to effectively utilize the resources of RISC-V CPUs with the RVV 0.7.1 vector extension, which cannot be done automatically with a C++ compiler yet. The paper reports on our experience in benchmarking CatBoost on the Lichee Pi 4a, RISC-V-based board, and shows how manual vectorization of computationally intensive loops with intrinsics can speed up the use of decision trees several times, depending on the specific workload. The developed codes are publicly available on GitHub.

Deep learning methods are widely used to solve problems in different areas, ranging from daily tasks on smartphones to complex scientific and engineering simulations. Most of the time spent on preparing deep models goes into labeling datasets and training neural networks, but the inference performance has a significant impact on the potential use of these models in various scenarios. Therefore, it is important to study this aspect. In this paper, we examine the performance of deep model inference on RISC-V processors. Specifically, we compare the capabilities of three frameworks: OpenVINO, TensorFlow Lite, and Apache TVM, in terms of optimizing deep model inference on these processors. Our study shows that TensorFlow Lite outperformed TVM in terms of inference speed on DenseNet-121 and GoogleNet-v4, with an average improvement of 2.69 and 2.47 times, respectively. The OpenVINO toolkit demonstrates excellent performance on x86 devices, but it is significantly inferior on RISC-V because of the lack of RISC-V-specific optimizations for the main computational kernels. The results show that deep learning frameworks can be already successfully used on RISC-V devices.

Numerical simulation of strong-field quantum electrodynamics (SFQED) processes is an essential step towards current and future high-intensity laser experiments. The complexity of SFQED phenomena and their stochastic nature make them extremely computationally challenging, requiring the use of supercomputers for realistic simulations. Recently, we have presented a novel approach to numerical simulation of SFQED processes based on an accurate approximation of precomputed rates, which minimizes the number of rate calculations per QED event. The current paper is focused on the high-performance implementation of this method, including vectorization of resource-intensive kernels and improvement of parallel computing efficiency. Using two codes, PICADOR and Hi-. (the latter being free and publicly available), we demonstrate significant reduction in computation time due to these improvements. We hope that the proposed approach can be applied in other codes for the numerical simulation of SFQED processes.

Significant challenges in further increasing processors performance are caused by physical and technological limitations at the current level of hardware development. In this context, software-hardware co-design technologies gain particular importance. These technologies represent one of the promising research and development directions for the coming decades. This paper explores the prospects of implementing a new cross-product instruction in processors based on the free and open RISC-V architecture, as well as possible performance gains when solving computational tasks. We demonstrate the complete flow of performed actions, starting from the idea of implementing a new command, its prototyping, evaluating performance using the Gem5 simulator, and culminating with embedding the instruction into the GCC compiler and providing a specific usage example.

The study of the relationship between brain neuron activity and behavioral responses of humans and other animals is an area of interest, although it has received relatively little attention from scientific biology and medical research centers. In this paper, we consider the problem of determining a mouse position in a circular track based on its neural activity data, and investigate the use of machine learning for solving this problem. The study is conducted in two parts: a classification task, where the model predicts which sector of the track the mouse is in at a particular time, and a regression task, where it predicts exact coordinates for each time step. We propose a neural network-based solution for both tasks, based on a graph of brain neuron activity. Accuracy results were obtained: 89% for classification and 93% for regression.

Special quantum circuits make it possible to collect experimental data to study the dynamics of quantum processor qubits [1]. The harmonic inversion method restores a set of eigenvalues that form a diagram qualitatively similar to the full spectrum of the open quantum system Liouvillian [2]. The Lindblad tomography method [3] evaluates initial state preparation and measurement error (SPAM), Kraus operators, non-Markovian measure, Hamiltonian and Lindblad operators describing the evolution of an open quantum system. We estimated SPAM errors, reconstructed the evolution using Kraus operators for discrete times, and estimated the non-Markovianity of the first qubits of the OriginQ Wuyuan 1 (Origin Quantum Cloud) and ibmq_belem (IBM Quantum Computing) quantum computers. The obtained results demonstrate the comparability of the platforms parameters and a low degree of non-Markovian behavior. The nearest future challenges are related to describing the observed processes in the form of the time-independent Lindblad equation along with the experiments involving two-qubit interaction.

The development of an open and free RISC-V architecture is of great interest for a wide range of areas, including high-performance computing and numerical simulation in mathematics, physics, chemistry and other problem domains. In this paper, we discuss the possibilities of accelerating computations on available RISC-V processors by improving the vectorization of several computer vision and machine learning algorithms in the widely used OpenCV library. It is shown that improved vectorization speeds up computations on existing prototypes of RISC-V devices by tens of percent.

The emergence of a new, open, and free instruction set architecture, RISC-V, has heralded a new era in microprocessor architectures. Starting with low-power, low-performance prototypes, the RISC-V community has a good chance of moving towards fully functional high-end microprocessors suitable for high-performance computing. Achieving progress in this direction requires comprehensive development of the software environment, namely operating systems, compilers, mathematical libraries, and approaches to performance analysis and optimization. In this paper, we analyze the performance of two available RISC-V devices when executing three memory-bound applications: a widely used STREAM benchmark, an in-place dense matrix transposition algorithm, and a Gaussian Blur algorithm. We show that, compared to x86 and ARM CPUs, RISC-V devices are still expected to be inferior in terms of computation time but are very good in resource utilization. We also demonstrate that well-developed memory optimization techniques for x86 CPUs improve the performance on RISC-V CPUs. Overall, the paper shows the potential of RISC-V as an alternative architecture for high-performance computing.

In this work we investigate vacuum breakdown in multipetawatt laser fields of the multibeam configuration corresponding to a converging magnetic dipole wave. Three-dimensional numerical simulation via the particle-in-cell method was used to determine the threshold total laser radiation power required for vacuum breakdown depending on the number of laser beams. It is shown that the minimal vacuum breakdown threshold power of 14 PW is attained when there are 12 beams while for a lower number of beams (from two to six) the threshold increases to 18 PW. The dependence of the growth rate of the quantum-electrodynamic cascade on the number of laser beams is determined in a wide range of total radiation power. Fluctuations in the power of each of the beams are shown to have little effect on both the breakdown threshold and the cascade growth rate.

Probabilistic generation of photons and electron-positron pairs due to the processes of strong-field quantum electrodynamics (SFQED) is often the most resource-intensive part of the kinetic simulations required in order to model current and future experimental studies at high-intensity laser facilities. To reduce its computational demands one can exploit tabulation of the precomputed rates, time-step sub-cycling, dynamic down-sampling of particle/photon ensembles and other approaches. As the culmination of previous improvements, the method described here provides the opportunity to make the minimal possible number of rate computations per QED event and, therefore, this method can increase performance by more than an order of magnitude. The computational routine is publicly available as a part of the open-source framework hi-$\chi$ designed as a Python-controlled toolbox for collaborative development.

The possibility of developing an ultra-high-brightness photon source with energies up to several GeV based on promising multipetawatt multibeam laser systems have been studied. For a maximum photon energy to be reached and for a directed photon beam to be generated, it is proposed to use an electric dipole field structure as the main configuration, which maximizes the amplitude of the electric field in the focus region. Dependences of the characteristics of the generated gamma radiation on the number of laser beams irradiating the plasma target and their focusing are investigated. It is shown that for the laser system parameters close to the stated XCELS facility parameters and a total power of 36 PW, the maximum energy of gamma photons can reach 2 GeV, the efficiency of conversion of laser radiation into photons with an energy of more than 1 MeV is 40

Currently, machine learning (ML) methods are widely used to process the results of physical experiments. In some cases, due to the limited amount of real experimental data, ML-models can be pre-trained on synthetic data simulated based on the analytical theory and then fine-tuned using real experimental data. A limitation of this approach is the presence of the latent parameters of the analytical model, which values are difficult or impossible to estimate. Setting these parameters incorrectly may induce a dataset shift even when applied to synthetic data. To overcome this problem, we train the ML-model on a dataset with randomly varied latent parameters of the analythical model to force the ML-model to concentrate on more general patterns that depend weakly on the latent parameters. We applied this approach to the problem of tight focusing of a laser pulse with the complex structure of the wavefront. We observed good accuracy of reconstructing of the tilt parameters when training and testing the ML-model on datasets generated for different values of the latent parameters. This confirms that the ML-model was able to select relevant information without over-fitting for specific features inherent in certain values of the latent parameters. We believe that this approach will enrich possible applications of ML-methods to an experimental diagnostics of laser pulses.

The article deals with the problem of taking into account radiation losses in the description of the dynamics of a charged particle in electromagnetic fields. The solution of this fundamental problem is becoming more and more in demand due to the development of ultrahigh-power lasers. Since the generation of radiation by a charged particle depends both on the magnitude of the fields in which it moves and on their structure, the developed multipetawatt multibeam laser systems, such as XCELS, can open up unique ways to solve this problem. This paper proposes one of such ways based on using a magnetic dipole field configuration formed by several laser pulses. We show by means of numerical simulations that when a solid target in the form of a nanowire is irradiated with several petawatt laser pulses that maximize the magnetic field in focus, there may appear an anomalous radiative trapping regime. The formed distributions of accelerated electrons and generated gamma photons in this regime have distinctive properties on the basis of which it is possible to experimentally determine the applicability of various theoretical approaches to the description of radiation losses.

It is proposed to use multipetawatt multibeam systems that reproduce dipole waves of electric configuration to maximize the electric field in the interaction region and the energy of accelerated particles. The breakdown of vacuum in fields of this configuration leads to the formation of beams of electrons and positrons accelerated to several GeV. It is shown that when use is made of 12 laser pulses with a duration of 30 fs and a total power of 36 PW, the total charge of accelerated electrons/positrons can reach 250 to 270 nC and the charge of particles with energies above 1 GeV can exceed 1 nC. It is demonstrated that the electric dipole structure of the fields makes it possible to obtain an extremely narrow (several mrad) distribution of particles over the polar angle measured from the field symmetry axis. It is also shown that the distribution of particles over the azimuthal angle can be used to diagnose interaction modes during vacuum breakdown. The obtained results can be useful in preparing experiments on the promising XCELS laser system.

In studies of interaction of matter with laser fields of extreme intensity there are two limiting cases of a multibeam setup maximizing either the electric field or the magnetic field. In this work attention is paid to the optimal configuration of laser beams in the form of an m-dipole wave, which maximizes the magnetic field. We consider in such highly inhomogeneous fields the advantages and specific features of laser-matter interaction, which stem from individual particle trajectories that are strongly affected by gamma photon emission. It is shown that in this field mode qualitatively different scenarios of particle dynamics take place in comparison with the mode that maximizes the electric field. A detailed map of possible regimes of particle motion (ponderomotive trapping, normal radiative trapping, radial, and axial anomalous radiative trapping), as well as angular and energy distributions of particles and gamma photons, is obtained in a wide range of laser powers up to 300 PW, and it reveals signatures of radiation losses experimentally detectable even with subpetawatt lasers.

Modern high-performance computing systems are mostly heterogeneous. The development of parallel programs that can use the full potential of such systems is fraught with significant difficulties. It is required not only to use the appropriate programming languages and technologies, but also to take into account the features of central and graphic processors that affect, among other things, the implementation of parallel schemes and memory management. The oneAPI heterogeneous programming model presented by Intel is aimed at simplifying the process of developing such programs, and its key component is the Data Parallel C ++ language, which allows developing portable high-performance programs for CPU, GPU, FPGA and other devices. The article presents a training course on oneAPI, developed at the Lobachevsky University. The course is aimed at studying a wide range of issues related to high-performance computing using models, methods and tools for parallel programming on Intel platforms. The article presents the concept of the course, describes its structure, categories of listeners who may be interested in it, and options for building a course depending on the level of preparation of the audience.

The Black-Scholes option pricing problem is one of the widely used financial benchmarks. We explore the possibility of developing a high-performance portable code using the SYCL (Data Parallel C++) programming language. We start from a C++ code parallelized with OpenMP and show optimization techniques that are beneficial on modern Intel Xeon CPUs. Then, we port the code to SYCL and consider important optimization aspects on CPUs and GPUs (device-friendly memory access patterns, relevant data management, employing vector data types). We show that the developed SYCL code is only 10% inferior to the optimized C++ code when running on CPUs while achieving reasonable performance on Intel GPUs. We hope that our experience of developing and optimizing the code on SYCL can be useful to other researchers who plan to port their high-performance C++ codes to SYCL to get all the benefits of single-source programming.

Similar to its classical version, quantum Markovian evolution can be either time-discrete or time-continuous. Discrete quantum Markovian evolution is usually modeled with completely positive trace-preserving maps, while time-continuous evolution is often specified with superoperators referred to as "Lindbladians." Here, we address the following question: Being given a quantum map, can we find a Lindbladian that generates an evolution identical-when monitored at discrete instances of time-to the one induced by the map? It was demonstrated that the problem of getting the answer to this question can be reduced to an NP-complete (in the dimension N of the Hilbert space, the evolution takes place in) problem. We approach this question from a different perspective by considering a variety of machine learning (ML) methods and trying to estimate their potential ability to give the correct answer. Complimentarily, we use the performance of different ML methods as a tool to validate a hypothesis that the answer to the question is encoded in spectral properties of the so-called Choi matrix, which can be constructed from the given quantum map. As a test bed, we use two single-qubit models for which the answer can be obtained using the reduction procedure. The outcome of our experiment is that, for a given map, the property of being generated by a time-independent Lindbladian is encoded both in the eigenvalues and the eigenstates of the corresponding Choi matrix.