Conveners
Software development and machines
- Xiangyu Jiang (Indiana University)
Software development and machines
- Gunnar Bali (Universität Regensburg)
Software development and machines
- Anthony Grebe (University of Maryland, College Park)
-
Dr Issaku Kanamori (R-CCS, RIKEN)27/07/2026, 14:004Software development and machinesContributed talk
We present a stabilized FP16 mixed-precision solver. Because the dynamic range of FP16 is much smaller than that of FP32, straightforward use of FP16 arithmetic may suffer from underflow and overflow. To address this issue, we introduce a rescaling procedure for the working vectors. The performance of the solver is evaluated on the Supercomputer Fugaku, which provides native FP16 SIMD support,...
Go to contribution page -
Anthony Grebe (University of Maryland, College Park)27/07/2026, 14:203Software development and machinesContributed talk
Historically, the cost of lattice-QCD calculations was dominated by applications of the Dslash matrix on a single right-hand side, which is constrained by memory-bandwidth limits to run at a small fraction (10-20%) of the available processing power. The advent of multigrid methods and multi-right-hand-side solvers mean that these limits are no longer the binding constraints they once were. ...
Go to contribution page -
Frank Winter (Jefferson Lab)27/07/2026, 14:401Software development and machinesContributed talk
Maintaining software portability across diverse accelerated hardware architectures remains a critical challenge for the lattice QCD community. The QDP-JIT library addresses this by providing a transparent, drop-in replacement for QDP++, enabling Chroma applications to run with high efficiency on modern GPUs without requiring any modifications to high-level code.
In this work, we present the...
Go to contribution page -
Leon Hostetler (Indiana University)27/07/2026, 15:002Software development and machinesContributed talk
We introduce PerfAdvisor, a lightweight Python package that uses an LLM-driven agent loop to diagnose GPU performance bottlenecks from supercomputer profiles. Modern HPC profiles contain millions of events across kernels, MPI calls, memory transfers, and idle gaps, making manual diagnosis slow and expert-intensive. PerfAdvisor automates this process by producing ranked, evidence-grounded...
Go to contribution page -
Gunnar Bali (Universität Regensburg), Giovanni Pederiva (FZ Jülich)27/07/2026, 15:202Software development and machinesContributed talk
The International Lattice Data Grid (ILDG) is a global infrastructure for data sharing. It currently mainly aims for sharing ensembles of gauge field configurations. We present the status of the redesigned version of this infrastructure (ILDG 2.0), together with first user-experiences. A short live demonstration of basic functionalities will be carried out.
Go to contribution page -
Xiangyu Jiang (Indiana University)27/07/2026, 16:104Software development and machinesContributed talk
We developed PyQUDA, a Python wrapper for QUDA written in Cython, designed to facilitate lattice QCD calculations using the Python programming language. PyQUDA leverages the optimized linear algebra capabilities of NumPy/CuPy/PyTorch, along with the highly optimized lattice QCD operations provided by QUDA to accelerate research. This integration simplifies the process of writing calculation...
Go to contribution page -
Justus Kuhlmann (ETH Zürich)27/07/2026, 16:304Software development and machinesContributed talk
We present recent developments in the openQxD code, developed by the RC* collaboration. We focus on first steps towards offloading parts of the HMC algorithm using the recently completed interface of our code to the solvers implemented in the QUDA library. As a first high-impact step, the inversions necessary for the pseudo fermion force are offloaded. We provide a breakdown of the time spent...
Go to contribution page -
Constantinos Costa (Rinnoco Ltd)27/07/2026, 16:505Software development and machinesContributed talk
We present a data-storage and compression tool for managing the large expressions generated in lattice perturbative calculations. A similar strategy may also be applied to numerical nonperturbative data, and this extension is planned for a future version of the tool. In perturbative calculations involving stout-smeared links and improved lattice actions, intermediate expressions may contain...
Go to contribution page -
Nikolai Husung (CERN)27/07/2026, 17:105Software development and machinesContributed talk
Constructing minimal operator bases with prescribed transformation properties under discrete symmetries is a recurring task in lattice QCD. Examples include lattice artifacts of local operator insertions, such as the axial vector current, or the construction of interpolating fields with given quantum numbers. The Python package opbasis automates this derivation, with common lattice...
Go to contribution page -
Alexis Harilaos Verney-Provatas (Swansea University/The University of Edinburgh)28/07/2026, 16:101Software development and machinesContributed talk
We present hmcdj, a new tool developed to ease the process of generating large numbers of gauge ensembles for a given theory. Building on top of the Grid library, hmcdj allows quick specification of theories in C++ with minimal boilerplate, and precise specification of per-ensemble parameters to be controlled at run time from a YAML input file. Additionally, it supports many quality-of-life...
Go to contribution page -
Nobuyuki Matsumoto (Boston University)28/07/2026, 16:302Software development and machinesPoster presentation
Riemannian manifold HMC (RMHMC), following the basic idea of Fourier acceleration, has been studied as a viable algorithm toward mitigating critical slowing down. In fact, by using multilevel integration, the algorithmic overhead becomes additive to the fermionic inversion cost that dominates in production runs, especially with domain-wall fermions. Given the rapid development of AI/ML, we...
Go to contribution page -
Max Hansen28/07/2026, 16:504Software development and machinesContributed talk
AmPyL is an open-source Python package for relating finite-volume lattice data to infinite-volume scattering amplitudes, with particular emphasis on three-particle systems. The library provides an object-oriented framework for constructing all aspects of the relativistic-field-theory three-particle formalism, including flavor and spectator channel spaces, finite-volume kinematics,...
Go to contribution page