AgenticCANN: Automated Ascend C Operator Generation via Knowledge-Augmented Agentic Evolution

2026-07-29

AI Summary

This work addresses the challenge of Ascend C operator optimization, which heavily relies on hardware-specific expertise and faces significant hurdles in low-resource NPU environments due to scarce platform knowledge and discrepancies in programming models. To overcome these limitations, the authors propose a knowledge-enhanced agent evolution framework that enables efficient automated code synthesis through full-cycle, multi-level domain knowledge injection and phase-adaptive interaction strategies with large language models. The core innovations include a pioneering knowledge-orchestrated generation system tailored for Ascend C and a novel agent evolution mechanism that jointly optimizes operator feasibility and performance. Experimental results on the Ascend 910B demonstrate 90–100% feasibility for element-wise and normalization operators, 56% for fused operators, and up to a 6.65× speedup for Pangu 1B inference kernels.

Abstract

Ascend C operator optimization is critical for NPU (Neural Processing Unit) inference performance but requires deep hardware expertise.While large language models (LLMs) have shown promise in automated CUDA kernel generation, the fundamentally different programming model of Ascend C introduces unique challenges that remain unexplored. In this paper, we propose AgenticCANN, a knowledge-augmented agentic evolution framework specifically tailored for automated Ascend C operator synthesis in low-corpus NPU environments.To overcome the severe platform knowledge deficit on unfamiliar hardware, AgenticCANN incorporates a knowledge-orchestrated generation system that delivers structured, multi-level domain insights across the development lifecycle to resolve the upstream feasibility bottleneck.Building on this foundation, it features a stage-adaptive agentic evolution strategy that dynamically aligns LLM interaction modes with specific generation and evolution phases, balancing high-exploration candidate discovery with high-convergence performance tuning.Extensive experiments on Huawei Ascend 910B across six operators spanning five pattern categories demonstrate that our method achieves 90 to 100 percent feasibility on elementwise and normalization operators, 56% on fusion operators, and up to 6.65$\times$ speedup on 1B Pangu model inference kernels. Further analysis reveals that knowledge injection monotonically improves feasibility from 57% to 86% on elementwise operators, demonstrating its general rather than operator-specific benefit.

Problem

• Ascend C
• NPU
• operator generation
• hardware knowledge deficit
• low-corpus environment

Innovation

• Ascend C
• knowledge-augmented generation
• agentic evolution
• automated operator synthesis
• NPU optimization