Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Multiple processors”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 343 records · Page 19

Fast multipurpose Monte Carlo simulation for proton therapy using multi- and many-core CPU architectures

Purpose: Accuracy in proton therapy treatment planning can be improved using Monte Carlo (MC) simulations. However the long computation time of such methods hinders their use in clinical routine. This work aims to develop a fast multipurpose Monte Carlo simulation tool for proton therapy using massively parallel central processing unit (CPU) architectures. Methods: A new Monte Carlo, called MCsquare (many-core Monte Carlo), has been designed and optimized for the last generation of Intel Xeon processors and Intel Xeon Phi coprocessors. These massively parallel architectures offer the flexibility and the computational power suitable to MC methods. The class-II condensed history algorithm of MCsquare provides a fast and yet accurate method of simulating heavy charged particles such as protons, deuterons, and alphas inside voxelized geometries. Hard ionizations, with energy losses above a user-specified threshold, are simulated individually while soft events are regrouped in a multiple scattering theory. Elastic and inelastic nuclear interactions are sampled from ICRU 63 differential cross sections, thereby allowing for the computation of prompt gamma emission profiles. MCsquare has been benchmarked with the GATE/GEANT4 Monte Carlo application for homogeneous and heterogeneous geometries. Results: Comparisons with GATE/GEANT4 for various geometries show deviations within 2%–1 mm. In spite of the limited memory bandwidth of the coprocessor simulation time is below 25 s for 10{sup 7} primary 200 MeV protons in average soft tissues using all Xeon Phi and CPU resources embedded in a single desktop unit. Conclusions: MCsquare exploits the flexibility of CPU architectures to provide a multipurpose MC simulation tool. Optimized code enables the use of accurate MC calculation within a reasonable computation time, adequate for clinical practice. MCsquare also simulates prompt gamma emission and can thus be used also for in vivo range verification.

60 APPLIED LIFE SCIENCES↗

Modern approaches for the theoretical description of multiparticle scattering and nuclear reactions

A review of novel approaches to solution of multiparticle scattering problems in the area above three-body breakup together with the review of new computational technologies which provide very effective and ultrafast realization of the novel approaches with ordinary PC are given. The novel direction presented here is based on two key points: a new formulation of the quantum scattering theory in a discrete Hilbert space of stationary wave packets and the massive-parallel solution of the resulted matrix equations with usage of ultrafast graphic processors (the so called GPU-computations). For the reader's convenience, a short review of the modern GPU calculations for the medicine, physics, military applications etc. is presented.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Precise micromotion compensation of a tilted ion chain

Excess micromotion can be a substantial source of errors in trapped-ion based quantum processors and clocks due to the sensitivity of the internal states of the ion to external fields and motion. This problem can be fixed by compensating background electric fields in order to position ions at the RF node and minimize their driven micromotion. Here we describe techniques for compensating ion chains in scalable surface ion traps. These traps are capable of cancelling stray electric fields with fine spatial resolution in order to compensate multiple closely spaced ions due to their large number of relatively small control electrodes. We demonstrate a technique that compensates an ion chain to better than 5 V/m and within 0.1 degrees of chain rotation.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Status Of The National Ignition Campaign And National Ignition Facility Integrated Computer Control System

The National Ignition Facility (NIF) at the Lawrence Livermore National Laboratory is a stadium-sized facility that will contains a 192-beam, 1.8-Megajoule, 500-Terawatt, ultraviolet laser system together with a 10-meter diameter target chamber with room for multiple experimental diagnostics. NIF is the world's largest and most energetic laser experimental system, providing a scientific center to study inertial confinement fusion (ICF) and matter at extreme energy densities and pressures. NIF's laser beams are designed to compress fusion targets to conditions required for thermonuclear burn. NIF is operated by the Integrated Computer Control System (ICCS) in an object-oriented, CORBA-based system distributed among over 1800 frontend processors, embedded controllers and supervisory servers. In the fall of 2010, a set of experiments began with deuterium and tritium filled targets as part of the National Ignition Campaign (NIC). At present, all 192 laser beams routinely fire to target chamber center to conduct fusion and high energy density experiments. During the past year, the control system was expanded to include automation of cryogenic target system and over 20 diagnostic systems to support fusion experiments were deployed and utilized in experiments in the past year. This talk discusses the current status of the NIC and the plan for controls and information systems to support these experiments on the path to ignition.

42 ENGINEERING↗

Compiling Quantum Circuits for Dynamically Field-Programmable Neutral Atoms Array Processors

Dynamically field-programmable qubit arrays (DPQA) have recently emerged as a promising platform for quantum information processing. In DPQA, atomic qubits are selectively loaded into arrays of optical traps that can be reconfigured during the computation itself. Leveraging qubit transport and parallel, entangling quantum operations, different pairs of qubits, even those initially far away, can be entangled at different stages of the quantum program execution. Such reconfigurability and non-local connectivity present new challenges for compilation, especially in the layout synthesis step which places and routes the qubits and schedules the gates. In this paper, we consider a DPQA architecture that contains multiple arrays and supports 2D array movements, representing cutting-edge experimental platforms. Within this architecture, we discretize the state space and formulate layout synthesis as a satisfiability modulo theories problem, which can be solved by existing solvers optimally in terms of circuit depth. For a set of benchmark circuits generated by random graphs with complex connectivities, our compiler OLSQ-DPQA reduces the number of two-qubit entangling gates on small problem instances by 1.7x compared to optimal compilation results on a fixed planar architecture. To further improve scalability and practicality of the method, we introduce a greedy heuristic inspired by the iterative peeling approach in classical integrated circuit routing. Using a hybrid approach that combined the greedy and optimal methods, we demonstrate that our DPQA-based compiled circuits feature reduced scaling overhead compared to a grid fixed architecture, resulting in 5.1X less two-qubit gates for 90 qubit quantum circuits. These methods enable programmable, complex quantum circuits with neutral atom quantum computers, as well as informing both future compilers and future hardware choices.

Physics↗

Partitioning Models for Scaling Parallel Sparse Matrix-Matrix Multiplication

We investigate outer-product--parallel, inner-product--parallel, and row-by-row-product--parallel formulations of sparse matrix-matrix multiplication (SpGEMM) on distributed memory architectures. For each of these three formulations, we propose a hypergraph model and a bipartite graph model for distributing SpGEMM computations based on one-dimensional (1D) partitioning of input matrices. Here, we also propose a communication hypergraph model for each formulation for distributing communication operations. The computational graph and hypergraph models adopted in the first phase aim at minimizing the total message volume and balancing the computational loads of processors, whereas the communication hypergraph models adopted in the second phase aim at minimizing the total message count and balancing the message volume loads of processors. That is, the computational partitioning models reduce the bandwidth cost and the communication hypergraph models reduce the latency cost. Our extensive parallel experiments on up to 2048 processors for a wide range of realistic SpGEMM instances show that although the outer-product--parallel formulation scales better, the row-by-row-product--parallel formulation is more viable due to its significantly lower partitioning overhead and competitive scalability. For computational partitioning models, our experimental findings indicate that the proposed bipartite graph models are attractive alternatives to their hypergraph counterparts because of their lower partitioning overhead. Finally, we show that by reducing the latency cost besides the bandwidth cost through using the communication hypergraph models, the parallel SpGEMM time can be further improved up to 32%.

97 MATHEMATICS AND COMPUTING↗

Improving energy efficiency of Embedded DRAM Caches for High-end Computing Systems

With increasing system core-count, the size of last level cache (LLC) has increased and since SRAM consumes high leakage power, power consumption of LLCs is becoming a significant fraction of processor power consumption. To address this, researchers have used embedded DRAM (eDRAM) LLCs which consume low-leakage power. However, eDRAM caches consume a significant amount of energy in the form of refresh energy. In this paper, we propose ESTEEM, an energy saving technique for embedded DRAM caches. ESTEEM uses dynamic cache reconfiguration to turn-off a portion of the cache to save both leakage and refresh energy. It logically divides the cache sets into multiple modules and turns-off possibly different number of ways in each module. Microarchitectural simulations confirm that ESTEEM is effective in improving performance and energy efficiency and provides better results compared to a recently-proposed eDRAM cache energy saving technique, namely Refrint. For single and dual-core simulations, the average saving in memory subsystem (LLC+main memory) on using ESTEEM is 25.8% and 32.6%, respectively and average weighted speedup are 1.09X and 1.22X, respectively. Additional experiments confirm that ESTEEM works well for a wide-range of system parameters.

cache reconfiguration↗

Open Radiation Monitoring: Histogram Builder Module Design

The Open Radiation Monitoring Project seeks to develop and demonstrate a modular radiation detection architecture designed specifically for use in arms control treaty verification (ACTV) applications that will facilitate rapid development of trusted systems to meet the needs of potential future treaties. A modular architecture can be used to reduce more complex systems to a series of single purpose building blocks, thereby facilitating equipment inspection and in turn building trust in the equipment by all treaty parties. Furthermore, a modular architecture can be used to control data flow within the measurement system, reducing the risk of "hidden switches" and constraining the amount of sensitive information that could potentially be inadvertently leaked. This report details the first revision of a prototype circuit that will convert analog pulses directly into a histogrammed data set for further processing. The circuit was designed with both spectroscopy and multiplicity analysis in mind but can, in principle, be used to reduce any raw data stream into a histogram. The number of output channels is limited, and the histogram bin ranges are user configurable to allow for non-uniform and discontinuous bins, which makes it possible to restrict the information being passed down stream if desired. Pulse processing relies entirely on analog circuitry and non- programmable logic, which enables operation without the need for a central processor or other programmable control unit. The circuit remains untested under the Open Radiation Monitoring project due to the closure of the sponsoring program. However, further development and testing is scheduled to take place in support of a purpose-built trusted verification system development effort known as COGNIZANT, which demonstrates the potential benefit of developing a suite of modular trusted system components.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Evict on write, a management strategy for a prefetch unit and/or first level cache in a multiprocessor system with speculative execution

In a multiprocessor system with at least two levels of cache, a speculative thread may run on a core processor in parallel with other threads. When the thread seeks to do a write to main memory, this access is to be written through the first level cache to the second level cache. After the write though, the corresponding line is deleted from the first level cache and/or prefetch unit, so that any further accesses to the same location in main memory have to be retrieved from the second level cache. The second level cache keeps track of multiple versions of data, where more than one speculative thread is running in parallel, while the first level cache does not have any of the versions during speculation. A switch allows choosing between modes of operation of a speculation blind first level cache.

97 MATHEMATICS AND COMPUTING↗

Data Locality Enhancement of Dynamic Simulations for Exascale Computing (Final Report)

The development of modern processors exhibits two trends that complicate the optimizations of modern software. The first is the increasing sensitivity of processors' throughput to irregularities in computation. With more processors produced through a massive integration of simple cores, future systems will increasingly favor regular data-level parallel computations, but deviate from the needs of applications with complex patterns. Some evidences are already shown on Graphic Processing Units (GPU): Irregular data accesses (e.g., indirect references A[D[i]]) and conditional branches are limiting many GPU applications' performance at a level an order of magnitude lower than the peak of GPU. The second hardware trend is the growing gap between memory bandwidth and the aggregate speed—that is, the sum of all cores' computing power—of a Chip Multiprocessor (CMP). Despite the capped growth of the peak CPU speed, the aggregate speed of a CMP keeps increasing as more cores get into a single chip. It is expected that by 2018, node concurrency in an exascale system will increase by hundreds of times, whereas, memory bandwidth will expand by only 10 to 20 times. Consequently, data movement and storage is expected to consume more than 70% of the total system power. Bridging this gap is difficult; the complexities of modern CMP memory hierarchy make it even harder: Data cache becomes shared among computing units, and the sharing is often non-uniform—whether two computing units share a cache depends on their proximity and the level of the cache. On the recent IBM Power7 architecture, for instance, four hardware contexts (or SMT threads) in a core share the entire memory hierarchy, all cores in one chip share an on-chip L3 cache, and cores across chips share L3 and main memory through off-chip connections. These two trends complicate the translation of computing power into performance, especially for a program with either intensive data accesses or complex patterns in data accesses or control flow paths. Unfortunately, both attributes present and will persist in a class of important applications. For instance, many scientific simulations deal with a large volume of data. And meanwhile, as most real-world processes are non-uniform and evolving (e.g., the evolution of a galaxy or the process of a drug injection), both the computations and data accesses of these programs tend to be irregular and dynamically changing. Currently, the lack of support to these applications on modern CMP severely limits their performance. On GPU, as our recent study shows and other studies echo, performance enhancement of a factor of integers is possible when memory accesses or control flows are streamlined for a set of GPU applications. On multicore CPU, our studies show that traditional locality enhancement, for being oblivious to the new features of multicore memory hierarchy, may even cause large slowdown to data-intensive dynamic applications. The severity of the issues is expected to worsen as the two hardware trends continue. Some recent studies try to match software with the trends, but in a limited scope or manner. For irregularities on GPU, most studies focus on irregularities analyzable through static analysis (e.g., data accesses in regular loops). Dynamic irregularities are harder to address because the needed analysis and transformations typically have to happen at run time. Some other research resorts to hardware extensions, an actual adoption of which is unclear for the entailed space cost and complexity. For data locality, recent years have seen some exploitations of the new memory hierarchy on multicore for performance, but most of them are on process or thread scheduling, rather than program transformations. Our recent study reveals that program-level transformations may magnify the scheduling benefits by a factor of seven, concluding that program-level transformation should play a central role for data locality enhancement on modern CMP. But research in this direction has been sparse, and most have focused on data layout or cache performance modeling, rather than program transformations to match with the new memory hierarchy features. Overall, it is still an open question how to bridge the gap between dynamic computations and the two prominent properties of modern processors. The goal of this project is to develop a set of techniques and software tools to enhance the matching between memory accesses in dynamic simulations and the prominent features of modern and future CMP, alleviating the memory performance issues for petascale and exascale computing. This report summarizes the discoveries and products produced throughout this project. It includes free launch, a new software approach to overcoming the shortcomings of both methods; coherence-free multiview, an approach that allows multiple views of a single data object to co-exist on GPU memory during a GPU kernel execution; algorithmic optimizations to data analytics problems, especially those that involve lots of distance calculations. Supporting dynamic parallelism is important for GPU to benefit a broad range of applications. There are currently two fundamental ways for programs to exploit dynamic parallelism on GPU: a software-based approach with software-managed worklists, and a hardware-based approach through dynamic subkernel launches. Neither is satisfactory. The former is complicated to program and is often subject to some load imbalance; the latter suffers large runtime overhead. In this work, we propose free launch, a new software approach to overcoming the shortcomings of both methods. It allows programmers to use subkernel launches to express dynamic parallelism. It employs a novel compiler-based code transformation named subkernel launch removal to replace the subkernel launches with the reuse of parent threads. Coupled with an adaptive task assignment mechanism, the transformation reassigns the tasks in the subkernels to the parent threads with a good load balance. The technique requires no hardware extensions, immediately deployable on existing GPUs. It keeps the programming convenience of the subkernel launch-based approach while avoiding its large runtime overhead. Meanwhile, its superior load balancing makes it outperform manual worklist-based techniques by 3X on average. The work was published at the 48th Annual IEEE/ACM International Symposium on Microarchitecture (Micro'2015). A Graphic Processing Unit (GPU) system is typically equipped with many types of memory (e.g., global, constant, texture, shared, cache). Data placement determines what data are placed on which type of memory, essential for GPU memory performance. Prior optimizations of data placement always require a single view of a data object on memory, which limits the optimization effectiveness. In this work, we propose coherence-free multiview, an approach that allows multiple views of a single data object to co-exist on GPU memory during a GPU kernel execution. We demonstrate that under certain conditions, the multiple views can remain incoherent while facilitating enhanced data placement. We present a theorem and some compiler support to ensure the soundness of the usage of coherence-free multiview. We further develop reference-discerning data placement, a new way to enhance data placements on GPU. It enables more flexible data placements by using coherence-free multiview to leverage the slack in coherence requirement of some GPU programs. Experiments on three types of GPU systems show that, with less than 200KB space cost, the new data placement technique can provide a 1.6X average (up to 4.27X) speedup. In addition, we have examined algorithmic optimizations to data analytics problems, especially those that involve lots of distance calculations. Computing distances among data points is an essential part of many important algorithms in data analytics, graph analysis, and other domains. In each of these domains, developers have spent significant manual effort optimizing algorithms, often through novel applications of the triangle equality, in order to minimize the number of distance computations in the algorithms. In this work, we observe that many algorithms across these domains can be generalized as an instance of a generic distance-related abstraction. Based on this abstraction, we derive seven principles for correctly applying the triangular inequality to optimize distance-related algorithms. Guided by the findings, we develop Triangular OPtimizer (TOP), the first software framework that is able to automatically produce optimized algorithms that either matches or outperforms manually designed algorithms for solving distance-related problems. TOP achieves up to 237x speedups and 2.5X on average. The work has been published at the 32nd International Conference on Machine Learning (ICML'15) and the 41st International Conference on Very Large Data Bases (VLDB'15).

42 ENGINEERING↗

Multi-sensor fusion for robust autonomous flight in indoor and outdoor environments with a rotorcraft micro-aerial vehicle (MAV)

The subject matter described herein includes a modular and extensible approach to integrate noisy measurements from multiple heterogeneous sensors that yield either absolute or relative observations at different and varying time intervals, and to provide smooth and globally consistent estimates of position in real time for autonomous flight. We describe the development of the algorithms and software architecture for a new 1.9 kg MAV platform equipped with an IMU, laser scanner, stereo cameras, pressure altimeter, magnetometer, and a GPS receiver, in which the state estimation and control are performed onboard on an Intel NUC 3rd generation i3 processor. We illustrate the robustness of our framework in large-scale, indoor-outdoor autonomous aerial navigation experiments involving traversals of over 440 meters at average speeds of 1.5 m/s with winds around 10 mph while entering and exiting buildings.

42 ENGINEERING↗

OpenMP in VASP: Threading and SIMD

Abstract The Vienna Ab initio Simulation Package (VASP) is a widely used electronic structure code that originally exploits process‐level parallelism through the Message Passing Interface (MPI) for work distribution within and across nodes. Architectural changes of modern parallel processors urge programmers to address thread‐ and data‐level parallelism as well to benefit most from the available compute resources within a node. We describe for VASP how to approach for an MPI + OpenMP parallelization including data‐level parallelism through OpenMP SIMD constructs together with a generic high‐level vector coding scheme. We can demonstrate an improved scalability of VASP and more than 20% gain over the MPI‐only version as well as a 2× increased performance of collective operations using the multiple‐endpoint MPI feature. The high‐level vector coding scheme applied to VASP's general gradient approximation routine gives up 9× performance gain on AVX512 platforms with the Intel compiler.

Wende, Florian↗

Extending Moore's Law via Computationally Error Tolerant Computing.

Dennard scaling has ended. Lowering the voltage supply (V dd ) to sub-volt levels causes intermittent losses in signal integrity, rendering further scaling (down) no longer acceptable as a means to lower the power required by a processor core. However, it is possible to correct the occasional errors caused due to lower V dd in an efficient manner and effectively lower power. By deploying the right amount and kind of redundancy, we can strike a balance between overhead incurred in achieving reliability and energy savings realized by permitting lower V dd . One promising approach is the Redundant Residue Number System (RRNS) representation. Unlike other error correcting codes, RRNS has the important property of being closed under addition, subtraction and multiplication, thus enabling computational error correction at a fraction of an overhead compared to conventional approaches. We use the RRNS scheme to design a Computationally-Redundant, Energy-Efficient core, including the microarchitecture, Instruction Set Architecture (ISA) and RRNS centered algorithms. Finally, from the simulation results, this RRNS system can reduce the energy-delay-product by about 3× for multiplication intensive workloads and by about 2× in general, when compared to a non-error-correcting binary core.

97 MATHEMATICS AND COMPUTING↗

Control systems and methods to enable autonomous drilling

A system or method for drilling includes autonomously controlling a rotary or percussive drilling process as it transitions through multiple materials with very different dynamics. The method determines a drilling medium based on real-time measurements and comparison to prior drilling data, and identifies the material type, drilling region, and approximately optimal setpoint based on data from at least one operating condition. The controller uses these setpoints initially to execute an optimal search to maximize performance by minimizing mechanical specific energy. Near-bit depth-of-cut estimations are performed using a machine learning prediction deployed in an embedded processor to provide high-speed ROP estimates. The sensing capability is coupled with a near-bit clutching mechanism to support drilling dysfunction mitigation.

Buerger, Stephen↗

Control systems and methods to enable autonomous drilling

A system or method for drilling includes autonomously controlling a rotary or percussive drilling process as it transitions through multiple materials with very different dynamics. The method determines a drilling medium based on real-time measurements and comparison to prior drilling data, and identifies the material type, drilling region, and approximately optimal setpoint based on data from at least one operating condition. The controller uses these setpoints initially to execute an optimal search to maximize performance by minimizing mechanical specific energy. Near-bit depth-of-cut estimations are performed using a machine learning prediction deployed in an embedded processor to provide high-speed ROP estimates. The sensing capability is coupled with a near-bit clutching mechanism to support drilling dysfunction mitigation.

Buerger, Stephen↗

Enabling Parallel Execution of System-level Simulations in SAM

This report summarizes the recent code updates related to “element ghosting” in SAM to enable the parallel execution of system-level simulations using multiple processors/cores. Unlike typical MOOSE-based applications, for system-level simulations, SAM mostly deals with a collection of discrete small pieces of meshes, and the connection of physics on these meshes are realized by using “connector” types of components/code structures, such as conjugate heat transfer and flow junctions. The required code implementation is to correctly mark the necessary ghost elements for each type of such components/code structures; thus, the lower-level libraries can correctly perform the necessary data transfer between processors (CPUs) when executed in parallel mode. After the code updates, SAM can now run system-level simulations in the parallel mode. The parallel execution capability was then tested with an ABTR input model with 23k DOFs. Significant speedup was demonstrated when the optimal number of CPUs were used in parallel mode. Future systematic studies on parallelization performance using additional test cases covering different physics/scenarios will be needed to provide additional insights into the scalability of SAM parallelization.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

A Metascalable Computing Framework for Large Spatiotemporal-Scale Atomistic Simulations

A metascalable (or 'design once, scale on new architectures') parallel computing framework has been developed for large spatiotemporal-scale atomistic simulations of materials based on spatiotemporal data locality principles, which is expected to scale on emerging multipetaflops architectures. The framework consists of: (1) an embedded divide-and-conquer (EDC) algorithmic framework based on spatial locality to design linear-scaling algorithms for high complexity problems; (2) a space-time-ensemble parallel (STEP) approach based on temporal locality to predict long-time dynamics, while introducing multiple parallelization axes; and (3) a tunable hierarchical cellular decomposition (HCD) parallelization framework to map these O(N) algorithms onto a multicore cluster based on hybrid implementation combining message passing and critical section-free multithreading. The EDC-STEP-HCD framework exposes maximal concurrency and data locality, thereby achieving: (1) inter-node parallel efficiency well over 0.95 for 218 billion-atom molecular-dynamics and 1.68 trillion electronic-degrees-of-freedom quantum-mechanical simulations on 212,992 IBM BlueGene/L processors (superscalability); (2) high intra-node, multithreading parallel efficiency (nanoscalability); and (3) nearly perfect time/ensemble parallel efficiency (eon-scalability). The spatiotemporal scale covered by MD simulation on a sustained petaflops computer per day (i.e. petaflops {center_dot} day of computing) is estimated as NT = 2.14 (e.g. N = 2.14 million atoms for T = 1 microseconds).

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Communication Lower Bounds and Optimal Algorithms for Multiple Tensor-Times-Matrix Computation

Multiple tensor-times-matrix (Multi-TTM) is a key computation in algorithms for computing and operating with the Tucker tensor decomposition, which is frequently used in multidimensional data analysis. Here, we establish communication lower bounds that determine how much data movement is required (under mild conditions) to perform the Multi-TTM computation in parallel. The crux of the proof relies on analytically solving a constrained, nonlinear optimization problem. We also present a parallel algorithm to perform this computation that organizes the processors into a logical grid with twice as many modes as the input tensor. We show that, with correct choices of grid dimensions, the communication cost of the algorithm attains the lower bounds and is therefore communication optimal. Finally, we show that our algorithm can significantly reduce communication compared to the straightforward approach of expressing the computation as a sequence of tensor-times-matrix operations when the input and output tensors vary greatly in size.

HBL-inequalities↗