FusionRCG: Orchestrating Recursive Computation Graphs across GPU Memory Hierarchies
摘要
Evaluating high-dimensional integrals via deep hierarchical recurrences is a dominant cost in quantum chemistry. While CPUs manage these efficiently, GPUs suffer a critical mismatch: limited per-thread memory is quickly overwhelmed by an explosion of simultaneously live intermediate variables. As recurrence scales, this forces massive data spilling to global memory, collapsing performance into a severe memory-bound regime. We present FusionRCG, a framework that jointly optimizes computation graph structure and GPU memory mapping. Exploiting the inherent topological flexibility of recurrence graphs, using electron repulsion integrals as an example, we contribute: (1) liveness-aware graph orchestration to minimize peak live intermediates; (2) algebraic dimensionality reduction via stepwise Cartesian-to-spherical fusion, shrinking intermediate footprints by up to ; and (3) an adaptive multi-tier kernel architecture routing graphs across the memory hierarchy. Evaluated on NVIDIA A100 GPUs, FusionRCG achieves up to end-to-end SCF speedup over GPU4PySCF and maintains parallel efficiency at 64~GPUs, successfully rescuing these workloads from memory-bound limits.
引用
@article{arxiv.2605.10312,
title = {FusionRCG: Orchestrating Recursive Computation Graphs across GPU Memory Hierarchies},
author = {Yihong Zhang and Xinran Wei and Junshi Chen and Fusong Ju and Wei Hu and Jinlong Yang and Huanhuan Xia},
journal= {arXiv preprint arXiv:2605.10312},
year = {2026}
}