Pith. sign in

REVIEW 2 major objections 3 minor 46 references

Beyond Fast Contractions: Attenuation and Recovery of Matrix-Engine Speedups in High-Order Finite Elements

T0 review · 2 major / 3 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Matrix-engine speedups in finite-element codes are governed by the whole operator path, not the contraction kernel.

desk verdict Central attenuation numbers are solid; the 1.6x recovery headline is a best-to-best comparison that should not be sold as a controlled effect. read the letter →

arxiv 2608.09175 v1 pith:CPBP5UK4 submitted 2026-08-10 cs.DC

Yinuo Wang , Lin Gan , Tianqi Mao , Zeyu Song , Wubing Wan , Jiayu Fu , Zekun Yin , Yuyang Jin
show 3 more authors
Xiaohui Duan Wei Xue Guangwen Yang
This is my paper · ORCID
classification cs.DC
keywords ArmSMESVEhigh-orderfiniteelementsspectralelementmethodmatrix-freeoperatorstensorcontractionsdatalayoutperformanceanalysis
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to show why a processor's matrix engine can look overwhelmingly fast on paper yet deliver little inside a real scientific operator. On the LX2 CPU, the matrix engine's 4x single-precision peak advantage over conventional SIMD falls to about 2.2x for isolated tensor contractions and to about 1.1x inside the conventional shared-degree-of-freedom operator. The loss is attributed to pointwise physics work, indirect field movement, shared-degree-of-freedom reconciliation, and irregular coefficient delivery. Coordinated fixes—explicit SIMD pointwise code, field layouts that trade indirection for duplication, and vector-blocked coefficient streaming—recover about 1.6x at high polynomial order, while a contraction-free control bounds any further contraction-only gain at 1.11–1.32x. If correct, this means matrix hardware pays off only when the entire operator path is co-designed.

What carries the argument

The carrying mechanism is a factorized diagnostic methodology that decomposes the stiffness operator into independently controllable layers—contraction backend, pointwise physics and geometry, field layout, coefficient layout, and reconciliation—and compares matched null, SIMD, and matrix-engine backends using measured time, throughput, and effective-bandwidth deltas rather than hardware counters. The null backend is the key control: it removes contraction instructions while preserving the surrounding operator path, so its 1.11–1.32x margin over the recovered matrix-engine operator is both a diagnosis of non-contraction cost and a counterfactual ceiling on making contractions free. The recovery uses three concrete layout mechanisms: S-vector and E-vector field storage that trade in-kernel indirection and conflicts for duplication and explicit reconciliation, and array-of-structures-of-vectors coefficient storage that aligns streams with SIMD consumption.

What would settle it

An independent reimplementation of the SIMD contraction and operator path that raises single-core utilization well above 65% of SIMD FMA peak would test the claim: if the full-operator matrix-engine/SIMD speedup then approaches the ~2.2x contraction proxy at p=15, the attenuation cascade would be an artifact of baseline quality rather than operator structure.

Watch

Extended reading notes

Core claim

The central claim is that useful matrix-engine speedup is a property of the whole operator, not of the contraction kernel. On LX2, the matrix engine's 4x FP32 peak over SIMD becomes about 2.2x for matched standalone tensor contractions, because small spectral-element contractions only partially occupy the matrix tile, and about 1.1x inside the conventional L-vector stiffness operator, because pointwise computation, indirect gather/scatter, shared-degree-of-freedom reconciliation, and irregular coefficient traffic dominate elapsed time. Explicit SIMD for pointwise physics raises the full-operator ratio to about 1.3x; moving fields to S/E-vector layouts and coefficients to vector-blocked streaming raises it to about 1.6x at p=15. A contraction-free null backend runs only 1.11–1.32x faster than the recovered matrix-engine operator, giving an empirical ceiling on further contraction-only optimization.

Load-bearing premise

The central ratios are measured against the authors' hand-written SIMD baseline, which reaches about 65% of SIMD peak; if a substantially better SIMD implementation exists, the reported attenuation, recovery, and ceiling numbers would all change.

Editorial extensions

If this is right

  • On LX2, swapping in matrix-engine contractions alone is not enough; the conventional irregular operator gains only about 1.1x, so pointwise SIMD and layout changes are required to approach 1.6x.
  • Further tuning of the contraction kernels has at most 1.11–1.32x left to give at high order, since even a contraction-free operator is only that much faster than the recovered matrix-engine configuration.
  • The shared-degree-of-freedom cost is relocated, not eliminated: L-, S-, and E-vector layouts trade indirection and conflicts for duplication and reconciliation, and reconciliation time must be counted in any claimed speedup.
  • In an iso-accuracy wave-propagation test, the recovered throughput favors higher polynomial order: p=7 is clearly beneficial, while p=15 helps only when the accuracy-selected timestep stays above the CFL limit.
  • The methodology ports to another matrix-engine CPU with the same experimental controls, so the diagnostic can identify where acceleration is masked before large optimization efforts are invested.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the same attenuation cascade appears on GPU tensor cores, isolated microkernel speedups over conventional GPU cores likely overstate end-to-end gains for high-order finite elements; the LX2 evidence suggests the operator path, not the kernel, is the bottleneck.
  • The null-backend ceiling implies that future work should target pointwise and data-movement stages, whose combined cost is several times larger than the contraction-only savings, rather than further matrix-engine microkernel tuning.
  • A testable extension would run the same factorized diagnostic with an FP64 or different matrix-engine configuration; the ratios would shift, but the claim that layout and reconciliation dominate at low-to-mid order could be checked directly.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 3 minor

Summary. The paper studies why the Arm SME matrix unit's large FP32 peak advantage (4x over SVE on the LX2 processor) is not realized end to end in SPECFEM3D's stiffness operator. Using matched SVE, SME, and null contraction backends embedded in a factorized full-operator benchmark, the authors measure attenuation from peak to isolated tensor contractions (about 2.2x at p=15), then to the conventional irregular L-vector operator (about 1.1x), and finally report partial recovery through explicit-SIMD pointwise work (about 1.3x) and through field-layout and coefficient-streaming changes (about 1.6x at high order). A contraction-free null backend is used to bound further contraction-only gains at 1.11-1.32x. The paper also includes a smooth-wave-problem case study suggesting that recovered high-order operator throughput can make higher polynomial order more favorable in an iso-accuracy sense.

Significance. If the central findings hold, the paper makes a valuable contribution to the growing literature on matrix engines in scientific computing: it demonstrates, on a production-relevant CPU and a widely used application operator, that the realizable speedup from a matrix unit is determined by the whole operator path—pointwise physics, field movement, coefficient delivery, and shared-DoF reconciliation—rather than by the contraction kernel alone. The methodological contribution is strong: the authors provide matched build contracts for SVE and SME, canonical-output validation for every optimized path, binary hashes, a clearly described filtering and remeasurement procedure for outlier samples, and an artifact repository. They also explicitly flag limitations of their own measurements (the best-to-best nature of the envelope and the non-strictness of the null-backend ceiling). These transparency practices raise confidence in the reported numbers. If the concerns below are addressed, the paper would be a useful reference for both hardware-software co-design and high-order FEM framework design.

major comments (2)
  1. [Section 6.3, Figure 11, and abstract] The paper's most prominent quantitative recovery claim—that coordinated pointwise, field-layout, and coefficient-layout changes 'together raise speedup to 1.6x'—is not supported by a controlled cumulative experiment. Section 6.3 explicitly states that Figure 11 is 'a best-to-best rather than fixed-layout comparison,' because each backend independently selects its fastest configuration at every order. The controlled comparisons in Figures 7 and 9A-D isolate individual changes with other factors fixed, but no experiment applies the three recovery stages cumulatively while holding all other factors fixed for both backends. As written, the 1.6x ratio could reflect differential configuration selection (e.g., SME choosing E-vector with mixed-AoSoV while SVE chooses S-vector with SoA) as much as the named operator-path changes. This is load-bearing because the abstract and conclusion present the 1.6x as the result of the specific recovery sequence. I recommend either adding a controlled cumulative comparison (e.g., start from the matched L-vector baseline and apply explicit SIMD, then S/E-vector, then AoSoV with all else fixed) or explicitly reframing the claim as 'the best validated SME configuration reaches 1.6x over the best validated SVE configuration,' without attributing the gain to the sequence.
  2. [Section 4, SVE baseline quality] The load-bearing premise of all SME/SVE ratios is that the hand-written SVE contraction backend is a genuinely high-performance SIMD implementation. The paper reports that the SVE backend reaches about 65 GFLOP/s, roughly 65% of the single-core SVE FMA peak, and states that the matched SVE path ensures later comparisons measure hardware datapaths rather than a strong SME implementation against a weak SIMD baseline. However, no independent evidence is provided that 65% of peak is near the attainable limit for these small, direction-dependent tensor contractions. A stronger SVE kernel—for example, with better register tiling, software pipelining, or a different direction-specific strategy—would directly reduce the reported 2.2x contraction speedup and the 1.6x recovery ratio. This is a correctness-risk concern rather than a demonstrated flaw. I would like to see a sensitivity analysis: vary one or two SVE kernel parameters (e.g., unroll factors or pipeline depth) and report the resulting SME/SVE ratios, or cite an independent high-quality SVE implementation for the same kernels.
minor comments (3)
  1. [Section 6.3, null-backend ceiling] The abstract says the contraction-free control 'bounds further contraction-only gains at 1.11–1.32x,' but Section 6.3 correctly notes that this 'is not a strict hardware bound because removing contraction can perturb caching and overlap.' Please use more precise wording in the abstract, such as 'suggests an empirical ceiling' or 'indicates an upper range,' to avoid overstating the strength of the null-backend result.
  2. [Section 7, iso-accuracy projection] The iso-accuracy cost projection relies on a least-squares fit of station-wise error to c1*x + c2*x^2 and on the assumption that the scalar-wave calibration transfers to elastic propagation. The paper states this assumption explicitly, but the sensitivity of the projected speedups to the error-fit choice is not reported. A brief sentence on how the frontier changes if the error model is changed (e.g., using a single power law) would strengthen the case-study conclusions.
  3. [Throughout] There are several typographical and formatting issues from the LaTeX source, including missing spaces around numbers and symbols (e.g., 'SME’s4×' in the abstract, '1.1×' and 'SVE' in Section 1) and inconsistent use of en-dashes in ranges. These are cosmetic but should be cleaned up in the final version.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: all headline SME/SVE ratios are direct matched measurements; the only fitted parameters are diagnostic or calibration tools, not sources of the speedup claims.

full rationale

The paper's central quantitative claims are direct measurements between matched backends on the same hardware: the 4x peak-to-2.2x contraction-to-1.1x operator attenuation and the recovery to 1.3x/1.6x are timing results from Figures 6, 7, 9, and 11. The only free parameters are alpha_b in the partial-overlap model (Eq. 6) and c1/c2 in the iso-accuracy error fits (Section 7), and neither generates the headline speedups: alpha_b is explicitly labeled 'diagnostic rather than cycle accurate,' and the c1/c2 fits are a calibration for the discretization projection, not a first-principles prediction. The CPI-style bound (Eqs. 4-5) uses stated issue-cost constants (CPI_OPA=2, CPI_LSU=1) rather than values extracted from the measured SME curves, so the bound is not fitted to the data. The null-backend ceiling (1.11-1.32x) is a measured counterfactual, not a derived equivalence. The author self-citations ([14] for vector-blocked AoSoA layout and [37] for stencil-related work) are contextual and non-load-bearing; the recovery numbers are measured with the paper's own implementations. The paper also transparently discloses that the final envelope (Figure 11) is a 'best-to-best rather than fixed-layout comparison,' which is a validity caveat rather than a circular reduction and does not affect the matched L-vector comparisons of Figure 7. No derivation step reduces to its inputs by construction.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claims rest on measured hardware behavior, so the ledger is light. The only genuine free parameters are the fitted overlap coefficient alpha_b and the error-distance polynomial coefficients c1/c2 used in the secondary discretization projection. The stated hardware issue costs and the partial-overlap model structure are explicit modeling assumptions, and the scalar-to-elastic transfer is an acknowledged assumption for the discretization implication only. No new physical or computational entities are introduced.

free parameters (2)
  • alpha_b non-overlap parameter = not reported explicitly
    Fitted per backend in the partial-overlap model (Eq. 6) to separate contraction time from memory/control time in the L-vector diagnosis (Section 5.1).
  • c1, c2 error-distance fit coefficients = not reported
    Least-squares fit of waveform L2 error vs. propagation distance for each polynomial order in the iso-accuracy calibration (Section 7 Group 2).
assumptions (5)
  • domain assumption LX2 hardware characteristics: SME provides 4x FP32 peak over SVE, ZA tile is 64 rows of 64 bytes, and the measured FP32 configuration uses VL=16.
    Vendor architecture parameters used for the modeled bounds (Eqs. 4-5) and for the 4x peak ratio (Sections 2.1 and 4).
  • domain assumption Single-core issue costs CPI_OPA=2 and CPI_LSU=1 for the SME outer-product and load/store units.
    Assumed issue costs used in the attainable-rate model in Section 4, Eq. 4-5, to derive the 2/3 and 2/5 attainable-rate factors.
  • ad hoc to paper Partial-overlap model structure: T_b approx max(T_mem, T_contr) + alpha_b min(T_mem, T_contr) in Eq. 6.
    Introduced as a diagnostic model in Section 5.1 to attribute observed full-operator time between memory and contraction; the authors state it is diagnostic rather than cycle accurate.
  • domain assumption Order-to-order spatial-resolution trend from the homogeneous scalar wave calibration transfers to smooth elastic propagation.
    Explicitly stated in Section 7: the scalar calibration is used as a 'relative points-per-wavelength proxy for the elastic projection, assuming its order-to-order spatial-resolution trend transfers to smooth elastic propagation.'
  • standard math Tensor-product sum factorization expresses SEM directional derivatives as batches of small tensor contractions.
    Background mathematics of spectral element methods used throughout Section 2.2 to define the contraction targets.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond Fast Contractions: Attenuation and Recovery of Matrix-Engine Speedups in High-Order Finite Elements." pith.science (2026). https://pith.science/paper/CPBP5UK4

@misc{pith2026260809175,
  author       = {Pith},
  title        = {Pith review of: Beyond Fast Contractions: Attenuation and Recovery of Matrix-Engine Speedups in High-Order Finite Elements},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CPBP5UK4}},
  note         = {Machine review of arXiv:2608.09175}
}
abstract

Modern processors increasingly provide matrix engines whose peak arithmetic throughput greatly exceeds conventional SIMD, but scientific applications rarely realize this advantage end to end. We examine this gap in SPECFEM3D's dominant stiffness operator on the Arm LX2 CPUs that power the flagship Lineshine supercomputer. Against a matched, high-performance SVE baseline on the same cores, SME's $4\times$ single-precision peak advantage falls to $2.2\times$ for isolated tensor contractions and $1.1\times$ for the complete operator. Our factorized diagnostic attributes the loss to pointwise computation, indirect field movement and synchronization, and irregular coefficient delivery. Explicit SIMD mitigates pointwise work, raising the full-operator speedup to $1.3\times$. Field-layout changes mitigate indirect movement and synchronization, while vector-blocked coefficient streaming reduces irregular-access costs; together they raise speedup to $1.6\times$ at high order. A contraction-free control bounds further contraction-only gains at $1.11$--$1.32\times$. Realizing matrix-engine performance therefore requires co-designing the entire operator path, not merely replacing its contraction kernel.

Figures

Figures reproduced from arXiv: 2608.09175 by the authors.

Figure 1
Figure 1. SPECFEM3D weak scaling and solver runtime com [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Peak-to-operator attenuation of SME over SVE. Across polynomial order, horizontal lines mark the LX2 FP32 matrix-FMA peak ratio (4×) and parity (1×); the measured curves expose the successive loss from matched contractions to the conventional compiler-vectorized irregular-elastic L￾vector operator. gains even when the contractions themselves map efficiently, creat￾ing a gap between matrix-unit peak, contraction perf… view at source ↗
Figure 3
Figure 3. LX2 organization and per-core SME execution. The dual-die package contains four 38-core NUMA domains per die, connected to HBM stacks, DDR controllers, and I/O. Every CPU core contains an SME matrix unit. The inset illustrates the outer product of vectors A and B accumulated into ZA using 𝑉𝐿 = 8 for visual clarity (our measured FP32 configura￾tion uses 𝑉𝐿 = 16). the complete operator pays for pointwise work, field m… view at source ↗
Figures from the paper (8 more)
Figure 5
Figure 5. Figure 5: Direction-specific data movement in the SME contraction backend. The 𝜂 path packs short lines, the 𝜉 path reorganizes discontinuous memory operands through ZA-tile transposition, and the 𝜁 path exploits naturally contiguous slices. 4 5 6 7 9 11 13 15 Polynomial order p…
Figure 6
Figure 6. Figure 6: Single-core directional tensor contractions and full-contraction proxy. The [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Factorized diagnosis of L-vector attenuation for elastic stiffness. The sidebar traces contraction backend, pointwise [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: Field storage and accumulation. L-vector uses shared DoFs with colored execution; S-vector and E-vector duplicate [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Field- and coefficient-layout recovery for elastic stiffness. (A) Regular-geometry L-vector, S-vector, and E-vector results [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: AoSoV coefficient semantics and mixed-precision conversion. Each block stores one SIMD width of points con￾tiguously by component. Mixed AoSoV packs selected com￾ponents in FP16, then shuffles and converts each load into two FP32 vectors for pointwise computation. 4 5…
Figure 11
Figure 11. Figure 11: Best validated elastic-irregular speedup envelope. [PITH_FULL_IMAGE:figures/full_fig_p009_11.png]
Figure 13
Figure 13. Figure 13: From measured operator throughput to iso-accuracy stiffness cost. Group 1 reports reconciliation-inclusive elastic [PITH_FULL_IMAGE:figures/full_fig_p010_13.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 21 canonical work pages

  1. [1]

    Ahmad Abdelfattah, Valeria Barra, Natalie Beams, Ryan Bleile, Jed Brown, Jean- Sylvain Camier, Robert Carson, Noel Chalmers, Veselin Dobrev, Yohann Dudouit, Paul Fischer, Ali Karakus, Stefan Kerkemeier, Tzanio Kolev, Yu-Hsiang Lan, Elia Merzari, Misun Min, Malachi Phillips, Thilina Rathnayake, Robert Rieben, Thomas Stitt, Ananias Tomboulides, Stanimire To...

    arXiv 2021

  2. [2]

    Mark Ainsworth. 2004. Discrete Dispersion Relation forℎ𝑝-Version Finite Ele- ment Approximation at High Wave Number.SIAM J. Numer. Anal.42, 2 (2004), 553–575. doi:10.1137/S0036142903423460

    work page doi:10.1137/s0036142903423460 2004

  3. [3]

    Mark Ainsworth and Hafiz Abdul Wajid. 2009. Dispersive and Dissipative Behavior of the Spectral Element Method.SIAM J. Numer. Anal.47, 5 (2009), 3910–3937. doi:10.1137/080724976

    work page doi:10.1137/080724976 2009

  4. [4]

    Robert Anderson, Julian Andrej, Andrew Barker, Jamie Bramwell, Jean-Sylvain Camier, Jakub Cerveny, Veselin Dobrev, Yohann Dudouit, Aaron Fisher, Tzanio Kolev, Will Pazner, Mark Stowell, Vassileios Tomov, Ido Akkerman, Johann Dahm, David Medina, and Stefano Zampini. 2021. MFEM: A Modular Finite Element Methods Library.Computers & Mathematics with Applicati...

    doi:10.1016/j.camwa.2020.06.009 2021

  5. [5]

    Julian Andrej, Nabil Atallah, Jan-Phillip Bäcker, Jean-Sylvain Camier, Dylan Copeland, Veselin Dobrev, Yohann Dudouit, Tobias Duswald, Brendan Keith, Dohyun Kim, Tzanio Kolev, Boyan Lazarov, Ketan Mittal, Will Pazner, Socratis Petrides, Syun’ichi Shiraiwa, Mark Stowell, and Vladimir Tomov. 2024. High- Performance Finite Elements with MFEM.The Internationa...

    doi:10.1177/10943420241261981 2024

  6. [6]

    2022.Arm Architecture Reference Manual Supplement: The Scalable Matrix Extension

    Arm Ltd. 2022.Arm Architecture Reference Manual Supplement: The Scalable Matrix Extension. Arm Ltd. https://developer.arm.com/documentation/ddi0616/ latest

    work page 2022

  7. [7]

    Daniel Arndt, Wolfgang Bangerth, Denis Davydov, Timo Heister, Luca Heltai, Martin Kronbichler, Matthias Maier, Jean-Paul Pelteret, Bruno Turcksin, and David Wells. 2021. The deal.II Finite Element Library: Design, Features, and Insights.Computers & Mathematics with Applications81 (2021), 407–422. doi:10. 1016/j.camwa.2020.02.022

    work page 2021

  8. [8]

    Jed Brown, Ahmad Abdelfattah, Valeria Barra, Natalie Beams, Jean-Sylvain Camier, Veselin Dobrev, Yohann Dudouit, Leila Ghaffari, Tzanio Kolev, David Medina, Will Pazner, Thilina Ratnayaka, Jeremy Thompson, and Stan Tomov

    work page

Show all 46 references
  1. [9]

    Canuto, M

    Claudio G. Canuto, M. Yousuff Hussaini, Alfio M. Quarteroni, and Thomas A. Zang. 2007.Spectral Methods: Evolution to Complex Geometries and Applications to Fluid Dynamics. Springer-Verlag, Berlin, Heidelberg

    2007

  2. [10]

    Yuetao Chen, Kun Li, Yuhao Wang, Donglin Bai, Lei Wang, Lingxiao Ma, Liang Yuan, Yunquan Zhang, Ting Cao, and Mao Yang. 2024. ConvStencil: Transform Stencil Computation to Matrix Multiplication on Tensor Cores. InProceedings of the 29th ACM SIGPLAN Annual Symposium on Principl...

    2024

  3. [11]

    Cu Cui. 2024. Acceleration of Tensor-Product Operations with Tensor Cores. ACM Transactions on Parallel Computing11, 4, Article 34 (2024). doi:10.1145/ 3695466

    2024

  4. [12]

    Deville, Paul F

    Michel O. Deville, Paul F. Fischer, and Ernest H. Mund. 2002.High-Order Meth- ods for Incompressible Fluid Flow. Cambridge University Press. doi:10.1017/ CBO9780511546792

    2002

  5. [13]

    Jens Domke, Emil Vatai, Aleksandr Drozd, Peng Chen, Yosuke Oyama, Lingqi Zhang, Shweta Salaria, Daichi Mukunoki, Artur Podobas, Mohamed Wahib, and Satoshi Matsuoka. 2021. Matrix Engines for High Performance Computing: A Paragon of Performance or Grasping at Straws?. InProceedi...

    2021

  6. [14]

    Xiaohui Duan, Kai Xu, Yuandong Chan, Christian Hundt, Bertil Schmidt, Pavan Balaji, and Weiguo Liu. 2017. S-Aligner: Ultrascalable Read Mapping on Sunway Taihu Light. In2017 IEEE International Conference on Cluster Computing. IEEE, 36–46. doi:10.1109/CLUSTER.2017.16

    2017 doi

  7. [15]

    Ruibo Fan, Wei Wang, and Xiaowen Chu. 2024. DTC-SpMM: Bridging the Gap in Accelerating General Sparse Matrix Multiplication with Tensor Cores. In Proceedings of the 29th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volu...

    2024

  8. [16]

    Paul Fischer, Stefan Kerkemeier, Misun Min, Yu-Hsiang Lan, Malachi Phillips, Thilina Rathnayake, Elia Merzari, Ananias Tomboulides, Ali Karakus, Noel Chalmers, and Tim Warburton. 2022. NekRS, a GPU-Accelerated Spectral Ele- ment Navier–Stokes Solver.Parallel Comput.114 (2022),...

    2022

  9. [17]

    Qiqi Gu, Chenpeng Wu, Heng Shi, Jianguo Yao, and Haibing Guan. 2026. Do We Need Tensor Cores for Stencil Computations? arXiv:2603.00477 [cs.DC]

    2026

  10. [18]

    Alexander Heinecke, Greg Henry, Maxwell Hutchinson, and Hans Pabst. 2016. LIBXSMM: Accelerating Small Matrix Multiplications by Runtime Code Genera- tion. InProceedings of the International Conference for High Performance Comput- ing, Networking, Storage and Analysis. IEEE Pre...

    2016

  11. [19]

    Stefan Henneking, Sreeram Venkat, Veselin Dobrev, John Camier, Tzanio Kolev, Milinda Fernando, Alice-Agnes Gabriel, and Omar Ghattas. 2025. Real-Time Bayesian Inference at Extreme Scale: A Digital Twin for Tsunami Early Warning Applied to the Cascadia Subduction Zone. InProcee...

    2025

  12. [20]

    Han Huang, Jiabin Xie, Guangnan Feng, Xianwei Zhang, Dan Huang, Zhiguang Chen, and Yutong Lu. 2025. HStencil: Matrix–Vector Stencil Computation with Interleaved Outer Product and MLA. InProceedings of the International Conference for High Performance Computing, Networking, Sto...

    2025

  13. [21]

    Jones and Paul E

    Mark T. Jones and Paul E. Plassmann. 1993. A Parallel Graph Coloring Heuristic. SIAM Journal on Scientific Computing14, 3 (1993), 654–669. doi:10.1137/0914041

    1993 doi

  14. [22]

    Hesthaven

    Andreas Klöckner, Tim Warburton, and Jan S. Hesthaven. 2011. Viscous Shock Capturing in a Time-Explicit Discontinuous Galerkin Method.Mathematical Modelling of Natural Phenomena6, 3 (2011), 57–83. doi:10.1051/mmnp/20116303 11 Yinuo Wang, Lin Gan, Tianqi Mao, Zeyu Song, Wubing ...

    2011

  15. [23]

    Tzanio Kolev, Paul Fischer, Misun Min, Jack Dongarra, Jed Brown, Veselin Dobrev, Tim Warburton, Stanimire Tomov, Mark S. Shephard, Ahmad Abdelfattah, Valeria Barra, Natalie Beams, Jean-Sylvain Camier, Noel Chalmers, Yohann Dudouit, Ali Karakus, Ian Karlin, Stefan Kerkemeier, Y...

    2021

  16. [24]

    Dimitri Komatitsch, Gordon Erlebacher, Dominik Göddeke, and David Michéa

  17. [25]

    Dimitri Komatitsch and Jeroen Tromp. 1999. Introduction to the Spectral Element Method for Three-Dimensional Seismic Wave Propagation.Geophysical Journal International139, 3 (1999), 806–822. doi:10.1046/j.1365-246x.1999.00967.x

    1999

  18. [26]

    Xiaoyan Liu, Yi Liu, Hailong Yang, Jianjin Liao, Mingzhen Li, Zhongzhi Luan, and Depei Qian. 2022. Toward Accelerated Stencil Computation by Adapting Tensor Core Unit on GPU. InProceedings of the 36th ACM International Conference on Supercomputing. Association for Computing Ma...

    2022

  19. [27]

    Yuechen Lu, Hongwei Zeng, Marc Casas, and Weifeng Liu. 2026. Characterizing Matrix Multiplication Units across General Parallel Patterns in Scientific Com- puting. InProceedings of the 31st ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming. Associ...

    2026

  20. [28]

    McCalpin

    John D. McCalpin. 1995. Memory Bandwidth and Machine Balance in Cur- rent High Performance Computers.IEEE Technical Committee on Computer Architecture Newsletter(1995), 19–25

    1995

  21. [29]

    Anthony T. Patera. 1984. A Spectral Element Method for Fluid Dynamics: Laminar Flow in a Channel Expansion.J. Comput. Phys.54, 3 (1984), 468–488. doi:10. 1016/0021-9991(84)90128-1

    1984

  22. [30]

    Daniel Peter, Dimitri Komatitsch, Yang Luo, Roland Martin, Nicolas Le Goff, Emanuele Casarotti, Pieyre Le Loher, Federica Magnoni, Qinya Liu, Céline Blitz, Tarje Nissen-Meyer, Piero Basini, and Jeroen Tromp. 2011. Forward and Adjoint Simulations of Seismic Wave Propagation on ...

    2011 arXiv

  23. [31]

    Jinliang Shi, Shigang Li, Youxuan Xu, Rongtian Fu, Xueying Wang, and Tong Wu

  24. [32]

    Kasia Swirydowicz, Noel Chalmers, Ali Karakus, and Tim Warburton. 2019. Ac- celeration of Tensor-Product Operations for High-Order Finite Element Methods. The International Journal of High Performance Computing Applications33, 4 (2019), 735–757. doi:10.1177/1094342018816368

    2019 doi

  25. [33]

    TOP500. 2026. TOP500 List, June 2026. https://www.top500.org/lists/top500/ 2026/06/. Accessed July 2026

    2026

  26. [34]

    Sara Touhami, Filippo Gatti, Fernando Lopez-Caballero, David Alejandro Cas- tro Cruz, and Didier Clouteau. 2022. SEM3D: A 3D High-Fidelity Numerical Earthquake Simulator for Broadband (0–10 Hz) Seismic Response Prediction at a Regional Scale.Geosciences12, 3 (2022), 112. doi:1...

    2022 doi

  27. [35]

    Jiqun Tu, Ian Karlin, John Camier, Veselin Dobrev, Tzanio Kolev, Stefan Hennek- ing, and Omar Ghattas. 2026. Accelerating High-Order Finite Element Simula- tions at Extreme Scale with FP64 Tensor Cores. arXiv:2603.09038 [cs.DC]

    2026 arXiv

  28. [36]

    Peter E. J. Vos, Spencer J. Sherwin, and Robert M. Kirby. 2010. From ℎ to𝑝 Efficiently: Implementing Finite and Spectral/ℎ𝑝 Element Methods to Achieve Optimal Performance for Low- and High-Order Discretisations.J. Comput. Phys. 229, 13 (2010), 5161–5181. doi:10.1016/j.jcp.2010.03.031

    2010 doi

  29. [37]

    Yinuo Wang, Tianqi Mao, Lin Gan, Wubing Wan, Zeyu Song, Jiayu Fu, Lanke He, Wenqiang Wang, Zekun Yin, Wei Xue, and Guangwen Yang. 2026. SMES- tencil: Optimizing High-Order Stencils on Arm Multicore Using SME Unit. IEEE Transactions on Parallel and Distributed Systems37, 3 (202...

    2026

  30. [38]

    Z. J. Wang, Krzysztof Fidkowski, Rémi Abgrall, Francesco Bassi, Doru Caraeni, Andrew Cary, Herman Deconinck, Ralf Hartmann, Koen Hillewaert, H. T. Huynh, Norbert Kroll, Georg May, Per-Olof Persson, Bram van Leer, and Miguel Visbal

  31. [39]

    Samuel Williams, Andrew Waterman, and David Patterson. 2009. Roofline: An Insightful Visual Performance Model for Multicore Architectures.Commun. ACM52, 4 (2009), 65–76. doi:10.1145/1498765.1498785

    2009

  32. [40]

    Orestis Zachariadis, Nitin Satpute, Juan Gómez-Luna, and Joaquín Olivares

  33. [41]

    Yiwei Zhang, Kun Li, Liang Yuan, Jiawen Cheng, Yunquan Zhang, Ting Cao, and Mao Yang. 2024. LoRAStencil: Low-Rank Adaptation of Stencil Computation on Tensor Cores. InProceedings of the International Conference for High Performance Computing, Networking, Storage, and Analysis....

    2024 arXiv

  34. [2010]

    High-Order Finite-Element Seismic Wave Propagation Modeling with MPI on a Large GPU Cluster.J. Comput. Phys.229 (2010), 7692–7714. doi:10.1016/j. jcp.2010.06.024

    2010 doi

  35. [2013]

    doi:10.1002/fld.3767

    High-Order CFD Methods: Current Status and Perspective.International Journal for Numerical Methods in Fluids72, 8 (2013), 811–845. doi:10.1002/fld.3767

    2013 doi

  36. [2020]

    Computers & Electrical Engineering88 (2020), 106848

    Accelerating Sparse Matrix–Matrix Multiplication with GPU Tensor Cores. Computers & Electrical Engineering88 (2020), 106848. doi:10.1016/j.compeleceng. 2020.106848

    2020

  37. [2021]

    Journal of Open Source Software6, 63 (2021), 2945

    libCEED: Fast Algebra for High-Order Element-Based Discretizations. Journal of Open Source Software6, 63 (2021), 2945. doi:10.21105/joss.02945

    2021 doi

  38. [2025]

    InProceedings of the 30th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming

    FlashSparse: Minimizing Computation Redundancy for Fast Sparse Matrix Multiplications on Tensor Cores. InProceedings of the 30th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming. Association for Computing Machinery, 312–325. doi:10.1145/3710848.3710858

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.