.png){kind=icon}
Return
Automatic Generation of Cross-Platform Vectorization Kernels for Cloud Microphysics Parameterization
.png){kind=icon}
PRE
AI
DOI：10.1016/j.jpdc.2026.105273.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• An auto-generation framework (Auto-CLOUDSC) is proposed for optimizing the vectorization of the CLOUDSC cloud microphysics scheme. • The method introduces a physics-combine (loop fusion) algorithm to improve temporal data locality and reduce computational redundancy. • A cache-aware tiling algorithm is designed to enhance spatial locality and cache utilization for column-major data structures. • The framework improves performance, achieving 1.3-2.1 × speedup on ARM and 1.9-3.4 × on Intel CPUs. • Auto-CLOUDSC automates Fortran-to-vector code transformation, supporting modern architectures.
Keywords:
Auto-CLOUDSC
vectorization
cloud microphysics
loop fusion
cache-aware tiling
Journal
IF:
4
Papers:
3.8K
Citations:
4.8K
Organization
Cited Papers
.png){kind=icon}
ShareSaveShareSaveShareSaveShareSave
ShareSaveShareSave
View more