Search

Graph engines have been used in many applications to handle big graphs efficiently. The majority of the research to improve their performance has focused primarily on the design of efficient graph processing. This paper claims, however, the focus should be given also to graph storage design. This is because good storage design can improve both CPU performance and I/O performance of graph engines. In this paper, we propose an efficient data layout for single-machine based graph engines. We identify the common node access pattern of the graph algorithms running on single-machine based graph engines. Based on this finding, we propose the breadth-first (BF) data layout which places the nodes processed together in the same or adjacent storage space so that they can be accessed together as much as possible. The experimental results show that the BF data layout improves both CPU and I/O performances significantly in all single-machine based graph engines. [ABSTRACT FROM AUTHOR]

In the controller development process, it is required to verify whether control functions operate normally, without fault or anomaly. Many control functions with complex system structures require various actual tests and test data analysis for verification. This paper presents an anomaly detection algorithm to verify control functions and to more efficiently analyze test data of a hybrid control unit (HCU) of a hybrid electric vehicle (HEV). The anomaly detection algorithm automatically detects anomaly of control functions from the test data, instead of manual labor by engineers. The examined target HCU control functions in this paper are the engine clutch engagement/disengagement control function and the engine start cooperative control function, which are crucial functions of transmission mounted electric drive (TMED) HEVs. A data-driven approach using a one-class support vector machine (SVM) is used to make it easy to apply various control functions. Actual vehicle test data is examined by the algorithm to verify the control functions at an actual vehicle level. The developed anomaly detection algorithm demonstrates feasibility and effectiveness of the proposed algorithm in detecting not only prior-known anomalies but also prior-unknown anomalies. Because other vehicle control functions have similar characteristics with the target functions of this paper, this algorithm is expected to be successfully applied to other HCU control functions. [ABSTRACT FROM AUTHOR]

This paper presents a novel active power control scheme for SRMs that are used as S/G units at MEAs; the proposed method uses a vector-controlled SAF, facilitating the acceleration of the SRM during starter mode and the efficient power control of the SRG at various operational points during generator mode. Moreover, an energy storage unit is incorporated in the SAF, aiming at enhancing the SRM's starting torque production and taking full advantage of the SRG's energy generation capability, through the appropriate phases’ magnetization regulation that is imposed by the developed adaptive SAF vector control. In addition, the proposed scheme provides enhanced FRTC during both modes of operation. The mathematical analysis is verified through MATLAB/Simulink simulations, while the performance of the S/G concept under study is evaluated via real-time CHIL tests, with the use of the dSPACE 1202 (MicroLabBox) platform and the 16-bit dsPIC30F4011 microcontroller. [ABSTRACT FROM AUTHOR]

In the modern era of artificial intelligence, increasingly sophisticated artificial neural networks (ANNs) are implemented, which pose challenges in terms of execution speed and power consumption. To tackle this problem, recent research on reduced-precision ANNs opened the possibility to exploit analog hardware for neuromorphic acceleration. In this scenario, photonic-electronic engines are emerging as a short-medium term solution to exploit the high speed and inherent parallelism of optics for linear computations needed in ANN, while resorting to electronic circuitry for signal conditioning and memory storage. In this paper we introduce a precision-scalable integrated Photonic-Electronic Multiply-Accumulate Neuron (PEMAN). The proposed device relies on (i) an analog photonic engine to perform reduced-precision multiplications at high speed and low power, and (ii) an electronic front-end for accumulation and application of the nonlinear activation function by means of a nonlinear encoding in the analog-to-digital converter (ADC). The device has been numerically validated through cosimulations to perform multiply-accumulate operations (MAC). Simulations are based on the iSiPP50G SOI process for the photonic engine and a commercial 28 nm CMOS process for the electronic front-end. The PEMAN exhibits a multiplication accuracy of 6.1 ENOB up to 10 GMAC/s, while it can perform computations up to 56 GMAC/s with a reduced accuracy down to 2.1 ENOB. The device can trade off speed and power consumption with resolution, significantly outperforming its analog electronics counterparts both in terms of speed and energy consumption. With respect to other photonic ANNs, the PEMAN has comparable speed and energy consumption with a higher resolution, while outperforming them by a hundredfold in the fan-in, which opens the possibility to accelerate more complex networks. [ABSTRACT FROM AUTHOR]

Pixel-level 2D object semantic understanding is an important topic in computer vision and could help machine deeply understand objects (e.g., functionality and affordance) in our daily life. However, most previous methods directly train on correspondences in 2D images, which is end-to-end but loses plenty of information in 3D spaces. In this paper, we propose a new method on predicting image corresponding semantics in 3D domain and then projecting them back onto 2D images to achieve pixel-level understanding. In order to obtain reliable 3D semantic labels that are absent in current image datasets, we build a large scale keypoint knowledge engine called KeypointNet, which contains 103,450 keypoints and 8,234 3D models from 16 object categories. Our method leverages the advantages in 3D vision and can explicitly reason about objects self-occlusion and visibility. We show that our method gives comparative and even superior results on standard semantic benchmarks. [ABSTRACT FROM AUTHOR]

Domain-independent planning decouples a planning task specification from planning engines. As the specification is usually describing only the physics of the environment, actions and a goal, the planning engines being generic solvers designed to solve any planning task tend to struggle with tasks that can be easily solved by domain-specific algorithms. Additional control knowledge can, to large extent, bridge such a performance gap. Instead of providing a specific planner supporting a given form of control knowledge, control knowledge can be directly encoded within the planning task specification and thus can be exploited by generic planners. In this paper, we propose Attributed Transition-Based Domain Control Knowledge (ATB-DCK) that is represented by a finite state automaton with attributed states, referring to specific states of objects, connected by transitions imposing constraints on action applicability. ATB-DCK, roughly speaking, represents the “grammar” of solution plans that guides the search. We show that ATB-DCK can be compiled into a classical planning task and thus it complements domain-independent planning techniques. Using several domains from the International Planning Competitions as benchmarks, we demonstrate that this approach often considerably improves efficiency of existing state-of-the-art planning engines. [ABSTRACT FROM AUTHOR]

In the past decade, high performance compute capabilities exhibited by heterogeneous GPGPU platforms have led to the popularity of data parallel programming languages such as CUDA and OpenCL. Developing high performance parallel programming solutions using such languages involve a steep learning curve due to the complexity of the underlying heterogeneous compute devices and their impact on performance. This has led to the emergence of several High Performance Computing frameworks which provide high-level abstractions for easing the development of data-parallel applications on heterogeneous platforms. However, the scheduling decisions undertaken by such frameworks only exploit coarse-grained concurrency in data parallel applications. In this paper, we propose PySchedCL, a framework which explores fine-grained concurrency aware scheduling decisions that harness the power of heterogeneous CPU/GPU architectures efficiently. We showcase the efficacy of such scheduling mechanisms over existing coarse-grained dynamic scheduling schemes by conducting extensive experimental evaluations for a diverse set of popular Deep Learning benchmarks. [ABSTRACT FROM AUTHOR]

Ferry contributing a significant amount of greenhouse gas is one of the critical vessels to be electrified. Designing a power system for a ferry with hybrid-shaft generators is different from designing a power system for other vessels because of its fixed route. More clearly, ferries repeatedly travel between their port of origin and port of destination, and before the next voyage, the battery must be recharged to the initial state so that the optimal energy management scheme can be repeatedly applied. Furthermore, the flexibility of hybrid-shaft generators, which allow more fuel saving, increases design complexity. In this paper, a mixed-integer non-linear programming problem is first formulated, and a power management algorithm with an initialization step for fulfilling the battery recharging requirement and a refinement step for minimizing the fuel consumption is proposed. The simulation results obtained from data of an actual ferry show that the proposed power management algorithm can fully recharge the battery and consumes less fuel than a rule-based power management scheme. Simulations also reveal that fuel consumption depends on available shore power, highlighting the necessity to develop charging infrastructure for practical electrification. Because of its speed, the algorithm can support hardware sizing, e.g., battery sizing. [ABSTRACT FROM AUTHOR]

The design of fusion engines is a subject of great importance in a variety of fields. In this paper, we focus on the problem of linear fusion at the feature level for multiple signal matrices with noises, with the features being extremal eigenvectors. When given multiple similarity matrices, the objective is to find an estimate of the latent signal eigenspace. The concentration result for the inner product of features from different matrix samples is developed, utilizing the random matrix theory. Based on of the theoretical results, we proposed an efficient algorithm, EigFuse, to solve the constrained data-driven optimization problem with different level of noises. Our method is of high efficiency by comparing it with state-of-the-art baseline approaches with multiple noise levels. Comprehensive experiments on several synthetic as well as real-life networks demonstrate our method’s superior performance. [ABSTRACT FROM AUTHOR]

This paper presents a constrained robust output feedback $\mathrm{{H}}\infty /\mathrm{{H}}_{2}$ control-based coordinated delay controller (ROHD) to deal with the controller area network (CAN) induced asynchronous delays which exist in the mode transition process (MTP) of a dual motor plug-in hybrid electric vehicle (DM-PHEV). Due to the multiple actuators of DM-PHEV and the limitation of network transmission rate, the control of MTP from electric driving mode (EM) to hybrid driving mode (HM) based on CAN bus inevitably induces random asynchronous delays. Firstly, the problem of MTP with CAN-induced delay is formulated and a mathematical model of MTP with asynchronous delays is built. In order to deal with the model uncertainty and the interference introduced by the measurement signals, ROHD controller is designed to achieve a fast, smooth and stable MTP in the presence of CAN-induced delays. Simulation and hardware-in-the-loop (HiL) test demonstrate that the torque demand and ride comfort of MTP with network-induced delays controlled by proposed controllers are better than those controlled by traditional robust controller. [ABSTRACT FROM AUTHOR]

Although real-time analytics on the up-to-date dataset has become an emerging demand, many big data systems are still designed for offline analytics. Particularly, for critical applications like Fintech, transactional data ingestion ensures a timely, always-correct, and scalable dataset. To carry out append-only ingestion, existing OLTP/HTAP systems are based on strict transactions with imperfect scalability, while NoSQL-like systems support scalable but relaxed transactions. How to ensure essential transactional guarantees without harming scalability seems to be a non-trivial issue. This paper proposes Karst to bring transactional data ingestion for existing offline analytics. We notice that blocking two-phase commit (2PC) to resolve transactional data ingestion is a performance killer for the partitioned analytical systems. Karst introduces a scalable protocol called metadata-oriented commit (MOC) that converts each distributed transaction into multiple partial transactions to avoid 2PC. Moreover, to ingest massive data into plenty of partitions, Karst also employs lazy persistence, lightweight logging, and optimized data traffic. In experiments, Karst could achieve up to about 2x $\sim$ ∼ 10x performance over relevant systems and also shows remarkable scalability. [ABSTRACT FROM AUTHOR]

According to a recent Deloitte study, the COVID-19 pandemic continues to place a huge strain on the global health care sector. Covid-19 has also catalysed digital transformation across the sector for improving operational efficiencies. As a result, the amount of digitally stored patient data such as discharge letters, scan images, test results or free text entries by doctors has grown significantly. In 2020, 2314 exabytes of medical data was generated globally. This medical data does not conform to a generic structure and is mostly in the form of unstructured digitally generated or scanned paper documents stored as part of a patient’s medical reports. This unstructured data is digitised using Optical Character Recognition (OCR) process. A key challenge here is that the accuracy of the OCR process varies due to the inability of current OCR engines to correctly transcribe scanned or handwritten documents in which text may be skewed, obscured or illegible. This is compounded by the fact that processed text is comprised of specific medical terminologies that do not necessarily form part of general language lexicons. The proposed work uses a deep neural network based self-supervised pre-training technique: Robustly Optimized Bidirectional Encoder Representations from Transformers (RoBERTa) that can learn to predict hidden (masked) sections of text to fill in the gaps of non-transcribable parts of the documents being processed. Evaluating the proposed method on domain-specific datasets which include real medical documents, shows a significantly reduced word error rate demonstrating the effectiveness of the approach. [ABSTRACT FROM AUTHOR]

High-performance multicast-enabled packet forwarding engines (PFEs), as an essential component of high-end switches, use a polynomial-time membership query algorithm to determine which port(s) the data packet should be forwarded. The currently widely used query algorithm is Bloom Filter (BF), which has been proven to have many fatal flaws. Another error-free membership query algorithm includes Scalar-pair Vectors Routing Forwarding (SVRF), Fractional-N Scalar-pair Vectors Routing Forwarding (Frac-N SVRF), and the Per-Port Prime Filter Array (P3FA) also have some shortcomings in space and time efficiencies. In this paper, we proposed a hybrid strategy: Divaricate Heterogeneous SVRF (DH-SVRF) scheme, which based on the P3FA and Frac-N SVRF, which randomly divides all member ships into N groups, and each group has the same structure and is independent of each other to obtain higher time efficiency and space utilization. Finally, we also discussed the selection of the optimal egress-diversity threshold. Through mathematical modeling and simulation, we validate that the proposed DH-SVRF scheme is superior to the SVRF/Frac-N SVRF and traditional BF in terms of scalability, space utilization, and time efficiency in specific conditions such as appropriate egress-diversity thresholds. [ABSTRACT FROM AUTHOR]

The robot’s mobility and intelligence have expanded its application in recent years. Specifically, indoor tracking is a fundamental function of public service robots in nursing homes, hospitals, and warehouses. Existing vision-based tracking requires visual information, which may be unavailable and introduce privacy issues in practical deployment. To this end, in this paper, we propose $\mathsf {LTrack}$ , a long-range tracking system based on LoRa, an emerging low-power wide-area networking (LPWAN) technology, with a single transceiver pair. Note that commodity LoRa devices cannot estimate the angle of arrival (AoA) of signals due to hardware limitations. We design a virtual circular antenna array in the mobile rotating anchor via a lightweight hardware modification to multiplex the only RF channel in the low-cost LoRa device. The difference of time of flight (TDoF) measured in the circular antenna array is fused with the rotating orientation to estimate the target AoA. We also redesign and optimize the primitive LoRa ranging engine based on systematic analysis. Further, we present a real-time mobile target tracking algorithm based on the Doppler frequency shift to combat the uncertainty introduced by the target movement. We have developed the prototype of $\mathsf {LTrack}$ , which consists of a mobile rotating anchor, a LoRa tag, and a commercial robot. The system is evaluated in both LOS and NOLS indoor scenarios. Experiments show that $\mathsf {LTrack}$ supports robust tracking with a median error of 0.12 m and 0.45 m in a $\text{137}\,\text{m}^2$ lab space and a $\text{600}\,\text{m}^2$ corridor, respectively. [ABSTRACT FROM AUTHOR]

Many role-playing games feature character creation systems where players are allowed to edit the facial appearance of their in-game characters. This paper proposes a novel method to automatically create game characters based on a single face photo. We frame this “artistic creation” process under a self-supervised learning paradigm by leveraging the differentiable neural rendering. Considering the rendering process of a typical game engine is not differentiable, an “imitator” network is introduced to imitate the behavior of the engine so that the in-game characters can be smoothly optimized by gradient descent in an end-to-end fashion. Different from previous monocular 3D face reconstruction which focuses on generating 3D mesh-grid and ignores user interaction, our method produces fine-grained facial parameters with a clear physical significance where users can optionally fine-tune their auto-created characters by manually adjusting those parameters. Experiments on multiple large-scale face datasets show that our method can generate highly robust and vivid game characters. Our method has been applied to two games and has now provided over 10 million times of online services. [ABSTRACT FROM AUTHOR]

This paper presents a design of driveline torque observer with dual clutch transmission (DCT). A goal of this study is to improve estimation performance of the observer. To increase the observer’s performance, it is important to make an accurate driveline model and a robust observer. First, to increase the accuracy of the model, we propose lumped driveline efficiency which is a parameter that expresses driveline transmission loss. In addition, a recursive least square estimation (RLSE) algorithm is proposed to estimate the parameter using driving data. Then, the model is updated using the estimated parameter value. Second, a reduced order observer is proposed to estimate transmitted torques of driveline in both driving and gear shifting process. Moreover, the observer, which is robust against measurement noise and disturbance, is designed through frequency domain analysis. For frequency domain analysis, an eigenstructure assignment method is applied to tune observer gains. Performance of the proposed RLSE algorithm and the observer are verified through simulation and testbench experiments. [ABSTRACT FROM AUTHOR]

With the advent of the Internet and Internet-connected devices, modern business applications can experience rapid increases as well as variability in transactional workloads. Database replication has been employed to scale performance and improve availability of relational databases but past approaches have suffered from various issues including limited scalability, performance versus consistency tradeoffs, and requirements for database or application modifications. This paper presents Hihooi, a replication-based middleware system that is able to achieve workload scalability, strong consistency guarantees, and elasticity for existing transactional databases at a low cost. A novel replication algorithm enables Hihooi to propagate database modifications asynchronously to all replicas at high speeds, while ensuring that all replicas are consistent. At the same time, a fine-grained routing algorithm is used to load balance incoming transactions to available replicas in a consistent way. Our thorough experimental evaluation with several well-established benchmarks shows how Hihooi is able to achieve almost linear workload scalability for transactional databases. [ABSTRACT FROM AUTHOR]

This paper aims to explore the performance potential of a novel dual-mode power-split hybrid powertrain. First, the steady-state power split characteristics for the proposed hybrid powertrain in different modes are analyzed. The mode switching strategy is developed to maximize powertrain efficiency, which is verified by using dynamic programming based on direct transmit points (DTPs). Moreover, a novel multi-objective evolutionary algorithm based on the decomposition (MOEA/D) method used in a nested way with dynamic programming is proposed to solve the multi-objective optimization of the PS-HEVs for the first time, and the computational efficiency and superiority of the proposed algorithm is compared with the commonly used NSGA-II algorithm. The results show that the MOEA/D based multi-objective optimization framework has similar performance in the search for the Pareto frontier but significantly higher computational efficiency than that of the NSGA-II algorithm. The obtained Pareto frontier provides optimal design candidates for hybrid powertrain systems. [ABSTRACT FROM AUTHOR]

In the power control process of the range extender, both the engine and the generator are at risk of reaching the torque output limit. The limitation not only comes from the restriction of their own torque output capacity, but also comes from the additional constraints that are established to avoid power reverse undershoot. In order to improve the system performance under time-varying torque constraints, a real-time model predictive controller (MPC) that manipulates the torque commands of the engine and the generator to track the reference power is designed. Different from the traditional penalty function method, the quadratic function is taken as the penalty function in this paper, which greatly improves the efficiency of solving the optimal solution. Meanwhile, the steady-state torque command in the feasible region is taken as the extreme point of the quadratic penalty function, which avoids the penalty coefficient tending to infinity. By adjusting the penalty coefficients of different constraints, the overall optimal solution approaches the boundary of the active inequality constraints. And according to the Karush-Kuhn-Tucker (KKT) condition, it is proved that the optimal solution of the original problem can be obtained through iteration. The effectiveness and practicability of the proposed strategy are evaluated through numerical simulations on Simulink and experiments on a range extender. [ABSTRACT FROM AUTHOR]

Recent improvements in vehicle-to-everything (V2X) communication and onboard computing power have enabled the development of control algorithms that jointly optimize the vehicle velocity and powertrain control in Connected and Automated Vehicles (CAVs), commonly referred to as the Eco-Driving problem. This paper presents a novel and computationally efficient algorithm to optimize the velocity planning and energy management in a CAV with a hybrid electric powertrain. The Eco-Driving problem is formulated as a dynamic, constrained optimization problem in the spatial domain, where information about the upcoming speed limits and road topography is assumed known. This problem is solved by embedding an Equivalent Consumption Minimization Strategy (ECMS) into a Dynamic Programming (DP) optimization to obtain a sub-optimal solution that provides results close to the global optimum at a fraction of the computational cost. Further, a multi-layer hierarchical control architecture is proposed as a path to a causal, real-time implementation. The DP-ECMS algorithm is converted into a Model Predictive Control (MPC) framework by using principles of Approximate Dynamic Programming (ADP). This causal implementation is finally benchmarked to a global optimal solution obtained with DP for different scenarios. [ABSTRACT FROM AUTHOR]

Advanced algorithms can promote the development of energy management strategies (EMSs) as a key technology in hybrid electric vehicles (HEVs). Reinforcement learning (RL) with distributed structure can significantly improve training efficiency in complex environments, and multi-threaded parallel computing provides a reliable algorithm basis for promoting adaptability. Dedicated to trying more efficient deep reinforcement learning (DRL) algorithms, this paper proposed a deep q-network (DQN)-based energy and emission management strategy (E&EMS) at first. Then, two distributed DRL algorithms, namely asynchronous advantage actor-critic (A3C) and distributed proximal policy optimization (DPPO), were adopted to propose EMSs, respectively. Finally, emission optimization was taken into account and then distributed DRL-based E&EMSs were proposed. Regarding dynamic programming (DP) as the optimal benchmark, simulation results show that three DRL-based control strategies can achieve near-optimal fuel economy and outstanding computational efficiency, and compared with DQN, two distributed DRL algorithms have improved the learning efficiency by four times. [ABSTRACT FROM AUTHOR]

Belief propagation (BP)-based stereo matching has popular owing to its regularity and ability to yield promising results. Some commonly observed hardware-implementation challenges pertaining to the use of this algorithm are large memory requirements and trade-offs between speed and chip area, along with an increasing disparity range. The paper presents a hardware- and memory-efficient architecture for building a BP-based disparity estimation system capable of overcoming issues associated with large disparity ranges. The proposed architecture is memory-efficient owing to the regularity of its underlying algorithm. In addition, the improved hardware efficiency can be attributed to processing element modifications to demonstrate shareable characteristics. Results obtained in this study reveal a 67.8% reduction in required memory corresponding to a time–area term complexity of $O(L(logL)^{2})$ , where L denotes the disparity range. This result is in stark contrast to the $O(L^{2}logL)$ and $O(L^{2})$ complexities observed in extant studies. Compared to state-of-the-art implementations, the proposed architecture offers an 86.2% gate count reduction for message update units at a disparity range of 512. These results confirm the proposed architecture’s suitability for use in large disparity scenarios. [ABSTRACT FROM AUTHOR]

A proper channel modeling methodology that characterizes the statistics of extreme events is key in the design of a system at an ultra-reliable regime of operation. The strict constraint of ultra-reliability corresponds to the packet error rate (PER) in the range of 10−9−10−5 within the acceptable latency on the order of milliseconds. Extreme value theory (EVT) is a robust framework for modeling the statistical behavior of extreme events in the channel data. In this paper, we propose a methodology based on EVT to model the extreme events of a non-stationary wireless channel for the ultra-reliable regime of operation. This methodology includes techniques for splitting the channel data sequence into multiple groups concerning the environmental factors causing non-stationarity, and fitting the lower tail distribution of the received power in each group to the generalized Pareto distribution (GPD). The proposed approach also consists of optimally determining the time-varying threshold over which the tail statistics are derived as a function of time, and assessing the validity of the derived Pareto model. Finally, the proposed approach chooses the best model with minimum complexity that represents the time variation behavior of the non-stationary channel data sequence. Based on the data collected within the engine compartment of Fiat Linea under various engine vibrations and driving scenarios, we demonstrate the capability of the proposed methodology in providing the best fit to the extremes of the non-stationary data. The proposed approach significantly outperforms the channel modeling approach using the stationary channel assumption in characterizing the extreme events. [ABSTRACT FROM AUTHOR]

This paper presents predictive energy management of hybrid electric vehicles (HEVs) via computationally efficient multi-layer control. First, we formulate an optimization problem by considering driveability and a penalty for using service brakes in the objective function to optimize gear, engine on/off, engine clutch state, and power-split decisions subject to constraints on the battery state of charge (SOC) and charge sustenance. Then, we split it into two control layers, including a supervisory control in a higher layer and a local power-split control in a lower layer. In the supervisory layer, a gear and powertrain mode manager (PM) is designed, and optimal gear, engine on/off and clutch states are obtained by using a combination of dynamic programming (DP) and Pontryagin's minimum principle (PMP). Moreover, a real-time iteration Secant method is proposed to calculate optimal battery costate such that the constraint on charge sustenance is satisfied. In the local controller layer, a linear quadratic tracking method (LQT) is used to optimally split power between the engine and the electric machine and keep battery SOC within its bounds. [ABSTRACT FROM AUTHOR]

The matrix-multiply assist (MMA) facility is the latest addition to IBM's power instruction set architecture and first shipped in the recently introduced POWER10 processor. MMA is designed to accelerate matrix–matrix operations, such as matrix multiplication and convolution, using instructions that compute the outer product of vector-register operands. Outer product computations have been used for decades in linear algebra libraries to deliver high-performance implementations of matrix operations. Such libraries use conventional single-instruction–multiple-data (SIMD) instructions to emulate outer product operations. MMA in POWER10 is the first hardware with direct support for outer product operations released in the market. MMA operates with the widest diversity of data types compared to any accelerator design currently announced. Unleashing the high-performance enabled by MMA requires careful code generation. Two key considerations for optimal MMA code performance are 1) the choice of accumulation layout when maximizing the using the accumulators and 2) the selection of matrix access order. This article shows that over 92% of peak performance in POWER10 with MMA can be achieved when the code generation makes the right choices. [ABSTRACT FROM AUTHOR]