GPGPUクラスタ上でのFMO計算の性能評価

Computer program suite for the calculation, storage and manipulation of molecular property and activity descriptors

Chemistry is the field of discussing the behavior of electrons in a molecule. Sixty years ago, the mathematical foundation for predicting the behavior of electrons was given by the quantum mechanics. Now a days, the rapid development of high speed digital computer makes it possible to perform a large scale quantum chemical calculation. Two powerful tools of the quantum chemistry reviewed here are the non-empirical molecular orbital calculation program series: GAUSSIAN and the semi-empirical one: MOPAC. Especially, the treatment of the reaction process in the MOPAC is shown in detail.

Calculation of auger electron spectra of H 2O, HF and C 2H 4 with the semi-empirical HAM/3 molecular orbital method

EMPIRE is a massively parallel semiempirical (NDDO) molecular-orbital program designed to scale well both on single multi-core nodes (using open MP) and on large clusters (using a hybrid open MP/MPI model). The program design and performance are discussed for single self-consistent-field calculations on up to 55,000 (the adamantane crystal shown in the graphic) atoms and on both single- and multi-node machines using either Windows 7 or Linux. EMPIRE currently carries out the full SCF calculation with no local approximations or other linear-scaling techniques. The single-node version is available free of charge to bona fide academic groups.

Modern GPUs have shown promising results in accelerating compute-intensive and numerical workloads with limited data sharing. However, emerging GPU applications manifest ample amount of data sharing among concurrently executing threads. Often data sharing requires mutual exclusion mechanism to ensure data integrity in multithreaded environment. Although modern GPUs provide atomic primitives that can be leveraged to construct fine-grained locks, the existing GPU lock implementations either incur frequent concurrency bugs, or lead to extremely low hardware utilization due to the Single Instruction Multiple Threads (SIMT) execution paradigm of GPUs. To make more applications with data sharing benefit from GPU acceleration, we propose a new locking scheme for GPU architectures. The proposed locking scheme allows lock stealing within individual warps to avoid the concurrency bugs due to the SMIT execution of GPUs. Moreover, it adopts lock virtualization to reduce the memory cost of fine-grain GPU locks. To illustrate the usage and the benefit of GPU locks, we apply the proposed GPU locking scheme to Delaunay mesh refinement (DMR), an application involving massive data sharing among threads. Our lock-based implementation can achieve 1.22x speedup over an algorithmic optimization based implementation (which uses a synchronization mechanism tailored for DMR) with 94% less memory cost.

The increasing availability of GPU accelerated architectures for high‐performance computing presents new opportunities for scientific software but also challenges due to the complexity of porting legacy codes to accelerator platforms. Directive‐based programming models such as OpenACC offer a minimally intrusive pathway to exploit GPU acceleration without requiring extensive rewriting of existing codes. The current work presents a comprehensive performance and portability study of a LatticeBoltzmann Method solver (PyLB) originally written in Python, Mpi4Py, and Fortran for CPU architectures, which is ported to GPUs using OpenACC directives applied to the Fortran routines. The performance of the solver is evaluated on NVIDIA V100, A100, and H100 GPUs available on the Jean Zay supercomputer from Institute for Development and Resources in Intensive Scientific Computing (IDRIS) in France. Roofline analysis and extensive strong and weak scalability tests are conducted, showing that the GPU‐enabled version of PyLB scales efficiently across multiple GPUs. The solver achieves performance on the H100 GPU equivalent to thousands of CPU cores and shows strong energy and carbon efficiency advantages over traditional CPU‐based simulations. The implementation is validated using classical benchmarks, including the decaying Taylor‐Green vortex and the flow over a 3‐D sphere. The results confirm the physical accuracy of the GPU port while highlighting its computational and environmental advantages.

Modern computer systems feature multiple homogeneous or heterogeneous computing units with deep memory hierarchies, and expect a high degree of thread-level parallelism from the software. Exploitation of data locality is critical to achieving scalable parallelism, but adds a significant dimension of complexity to performance optimization of parallel programs. This is especially true for programming models where locality is implicit and opaque to programmers. In this paper, we introduce the hierarchical place tree (HPT) model as a portable abstraction for task parallelism and data movement. The HPT model supports co-allocation of data and computation at multiple levels of a memory hierarchy. It can be viewed as a generalization of concepts from the Sequoia and X10 programming models, resulting in capabilities that are not supported by either. Compared to Sequoia, HPT supports three kinds of data movement in a memory hierarchy rather than just explicit data transfer between adjacent levels, as well as dynamic task scheduling rather than static task assignment. Compared to X10, HPT provides a hierarchical notion of places for both computation and data mapping. We describe our work-in-progress on implementing the HPT model in the Habanero-Java (HJ) compiler and runtime system. Preliminary results on general-purpose multicore processors and GPU accelerators indicate that the HPT model can be a promising portable abstraction for future multicore processors.

The fragment molecular orbital (FMO) scheme is one of the popular fragmentation-based methods and has the potential advantage of making the circuit shallow for quantum chemical calculations on quantum computers. In this study, we used a GPU-accelerated quantum simulator (cuQuantum) to perform the electron correlation part of the FMO calculation as unitary coupled-cluster singles and doubles (UCCSD) with the variational quantum eigensolver (VQE) for hydrogen-bonded (FH) and (FH) -H O systems with the STO-3G basis set. VQE-UCCSD calculations were performed using both canonical and localized MO sets, and the results were examined from the point of view of size-consistency and orbital-invariance affected by the Trotter error. It was found that the use of localized MO leads to better results, especially for (FH) -H O. The GPU acceleration was substantial for the simulations with larger numbers of qubits, and was about a factor of 6.7-7.7 for 18 qubit systems.

Weight pruning methods of deep neural networks (DNNs) have been demonstrated to achieve a good model pruning rate without loss of accuracy, thereby alleviating the significant computation/storage requirements of large-scale DNNs. Structured weight pruning methods have been proposed to overcome the limitation of irregular network structure and demonstrated actual GPU acceleration. However, in prior work, the pruning rate (degree of sparsity) and GPU acceleration are limited (to less than 50%) when accuracy needs to be maintained. In this work, we overcome these limitations by proposing a unified, systematic framework of structured weight pruning for DNNs. It is a framework that can be used to induce different types of structured sparsity, such as filterwise, channelwise, and shapewise sparsity, as well as nonstructured sparsity. The proposed framework incorporates stochastic gradient descent (SGD; or ADAM) with alternating direction method of multipliers (ADMM) and can be understood as a dynamic regularization method in which the regularization target is analytically updated in each iteration. Leveraging special characteristics of ADMM, we further propose a progressive, multistep weight pruning framework and a network purification and unused path removal procedure, in order to achieve higher pruning rate without accuracy loss. Without loss of accuracy on the AlexNet model, we achieve 2.58× and 3.65× average measured speedup on two GPUs, clearly outperforming the prior work. The average speedups reach 3.15× and 8.52× when allowing a moderate accuracy loss of 2%. In this case, the model compression for convolutional layers is 15.0× , corresponding to 11.93× measured CPU speedup. As another example, for the ResNet-18 model on the CIFAR-10 data set, we achieve an unprecedented 54.2× structured pruning rate on CONV layers. This is 32× higher pruning rate compared with recent work and can further translate into 7.6× inference time speedup on the Adreno 640 mobile GPU compared with the original, unpruned DNN model. We share our codes and models at the link http://bit.ly/2M0V7DO.

The goal of Matrices Over Runtime Systems at Exascale (MORSE) project is to design dense and sparse linear algebra methods that achieve the fastest possible time to an accurate solution on large-scale multicore systems with GPU accelerators, using all the processing power that future high end systems can make available. In this poster, we propose a framework for describing linear algebra algorithms at a high level of abstraction and delegating the actual execution to a runtime system in order to design software whose performance is portable accross architectures. We illustrate our methodology on three classes of problems: dense linear algebra, sparse direct methods and fast multipole methods. The resulting codes have been incorporated into Magma, Pastix and ScalFMM solvers, respectively.

A study has been made on the polarization dependence and the temperature dependence of absorption and luminescence spectra of a 2,6-dihydroxyanthraquinone single crystal. The single crystal structure has been investigated by X-ray diffraction and is assigned as monoclinic and the space group P2 <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> /c (#14). Electronic energy levels were calculated by molecular orbital calculation program of WinMOPAC. The absorption band at 3.2eV and the luminescence band at 2.2eV are assigned mainly to the transition between HOMO and LUMO states. The anisotropic effect on the absorption and luminescence was analyzed in view of the crystal structure. The polarization dependence is consistently explained by the direction of transition dipole moment of 2,6-dihydroxyanthraquinone molecule.

Anisotropic mesh adaptation is a powerful way to directly minimise the computational cost of mesh based simulation. It is particularly important for multi-scale problems where the required number of floating-point operations can be reduced by orders of magnitude relative to more traditional static mesh approaches. Increasingly, finite element/volume codes are being optimised for modern multicore architectures. Inter-node parallelism for mesh adaptivity has been successfully implemented by a number of groups using domain decomposition methods. However, thread-level parallelism using programming models such as OpenMP is significantly more challenging because the underlying data structures are extensively modified during mesh adaptation and a greater degree of parallelism must be realised while keeping the code race-free. In this thesis we describe a new thread-parallel implementation of four anisotropic mesh adaptation algorithms, namely edge coarsening, element refinement, edge swapping and vertex smoothing. For each of the mesh optimisation phases we describe how safe parallel execution is guaranteed by processing workitems in batches of independent sets and using a deferredoperations strategy to update the mesh data structures in parallel without data contention. Scalable execution is further assisted by creating worklists using atomic operations, which provides a synchronisation-free alternative to reduction-based worklist algorithms. Additionally, we compare graph colouring methods for the creation of independent sets and present an improved version which can run up to 50% faster than existing techniques. Finally, we describe some early work on an interrupt-driven work-sharing for-loop scheduler which is shown to perform better than existing workstealing schedulers. Combining all aforementioned novel techniques, which are generally applicable to other unordered irregular problems, we show that despite the complex nature of mesh adaptation and inherent load imbalances, we achieve

Inspired by natural evolutionary processes, Evolutionary Computation (EC) has established itself as a cornerstone of Artificial Intelligence. Recently, with the surge in data-intensive applications and large-scale complex systems, the demand for scalable EC solutions has grown significantly. However, most existing EC infrastructures fall short of catering to the heightened demands of large-scale problem solving. While the advent of some pioneering GPU-accelerated EC libraries is a step forward, they also grapple with some limitations, particularly in terms of flexibility and architectural robustness. In response, we introduce EvoX: a computing framework tailored for automated, distributed, and heterogeneous execution of EC algorithms. At the core of EvoX lies a unique programming model to streamline the development of parallelizable EC algorithms, complemented by a computation model specifically optimized for distributed GPU acceleration. Building upon this foundation, we have crafted an extensive library comprising a wide spectrum of 50+ EC algorithms for both single-and multi-objective optimization. Furthermore, the library offers comprehensive support for a diverse set of benchmark problems, ranging from dozens of numerical test functions to hundreds of reinforcement learning tasks. Through extensive experiments across a range of problem scenarios and hardware configurations, EvoX demonstrates robust system and model performances. EvoX is open-source and accessible at: https://github.com/EMI-Group/EvoX.

GPU accelerators are ubiquitous in modern HPC systems. To program them, users have the choice between vendor-specific, native programming models, such as CUDA, which provide simple parallelism semantics with minimal runtime support, or portable alternatives, such as OpenMP, which offer rich parallel semantics and feature an extensive runtime library to support execution. While the operations of such a runtime can easily limit performance and drain resources, it was to some degree regarded an unavoidable overhead. In this work we present a co-design methodology for optimizing applications using a specifically crafted OpenMP GPU runtime such that most use cases induce near-zero overhead. Specifically, our approach exposes runtime semantics and state to the compiler such that optimization effectively eliminating abstractions and runtime state from the final binary. With the help of user provided assumptions we can further optimize common patterns that otherwise increase resource consumption. We evaluated our prototype build on top of the LLVM/OpenMP GPU offloading infrastructure with multiple HPC proxy applications and benchmarks. Comparison of CUDA, the original OpenMP runtime, and our co-designed alternative show that, by our approach, performance is significantly improved and resource consumption is significantly lowered. Oftentimes we can closely match the CUDA implementation without sacrificing the versatility and portability of OpenMP.

In this work, we present a first step towards the efficient implementation of polarizable molecular mechanics force fields with GPU acceleration. The computational bottleneck of such applications is found in the treatment of electrostatics, where higher-order multipoles and a self-consistent treatment of polarization effects are needed. We have coded these sections, for the case of a non-periodic simulation, with the CUDA programming model. Results show a speedup factor of 21 for a single precision GPU implementation, when comparing to the serial CPU version. A discussion of the optimization and parameterization steps is included. Comparison between different graphic cards and a shared memory parallel CPU implementation is also given. The current work demonstrates the potential usefulness of GPU programming in accelerating this field of applications.