Performance Modeling of Vectorized SNAP Inter-Atomic Potentials on CPU Architectures.
- Carnegie Mellon Univ., Pittsburgh, PA (United States); Sandia National Lab. (SNL-NM), Albuquerque, NM (United States)
- Sandia National Lab. (SNL-NM), Albuquerque, NM (United States)
SNAP potentials are inter-atomic potentials for molecular dynamics that enable simulations at accuracy levels comparable to density functional theory(DFT) at a fraction of the cost. As such, SNAP scales to on the order of 104 — 106 atoms. In this work, we explore CPU optimization of potentials computation using SIMD. We note that efficient use of SIMD is non-obvious as the application features an irregular iteration space for various potential terms, necessitating use of SIMD across atoms in a cross matrix, batched fashion. We present a preliminary analytic model to determine the correct batch size for several CPU architectures across several vendors, and show end-to-end speedups between 1.66x and 3.22x compared to the original.
- Research Organization:
- Sandia National Lab. (SNL-NM), Albuquerque, NM (United States)
- Sponsoring Organization:
- USDOE National Nuclear Security Administration (NNSA)
- DOE Contract Number:
- AC04-94AL85000; NA0003525
- OSTI ID:
- 1564037
- Report Number(s):
- SAND-2019-10915R; 679419
- Country of Publication:
- United States
- Language:
- English
Similar Records
A Step towards Energy Efficient Computing: Redesigning a Hydrodynamic Application on CPU-GPU, In: 2014 IEEE 28th International Parallel and Distributed Processing Symposium
Parallel architectures and parallel algorithms for integrated vision systems