Construction and Performance Evaluation of an Arbitrary-Precision Floating-Point Arithmetic Environment on CUDA

DOI10.48550/arXiv.2608.00085MaRDI QIDQ7284582arXivFDO

Authors Tomonori Kouya

Publication date 30 July 2026

Published in arXiv

Tldr We develop mpc_cuda, a CUDA-based arbitrary-precision floating-point arithmetic environment for scientific computing, using generative AI, achieving 40--105$\times$ speedup over CPU-based MPFR/MPC while preserving bit-exactness; we also implement cu_freal/cu_fcomplex types and outperform existing GPU libraries at low-to-medium precision.

Full work available at URL https://doi.org/10.48550/arXiv.2608.00085

arXiv classification

cs.MS

Cites work

• A high-order iterative formula for simultaneous determination of zeros of a polynomial
• CAMPARY: CUDA multiple precision arithmetic library and applications
• GMRES: A Generalized Minimal Residual Algorithm for Solving Nonsymmetric Linear Systems
• Iterative Krylov Methods for Large Linear Systems
• Iterative Methods for Sparse Linear Systems
• MPFR

This page was built for publication: Construction and Performance Evaluation of an Arbitrary-Precision Floating-Point Arithmetic Environment on CUDA

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7284582)