AscendCraft: Automatic Ascend NPU Kernel Generation via DSL-Guided Transcompilation

TL;DR

AscendCraft introduces a DSL-guided method for automatic generation of AscendNPU kernels, significantly improving correctness and performance, and bridging the gap in kernel generation for specialized AI accelerators.

Contribution

It presents a novel DSL-guided transcompilation approach that enables LLMs to generate high-quality AscendC kernels for NPUs, addressing domain-specific challenges.

Findings

Achieves 98.1% compilation success rate.

Attains 90.4% functional correctness.

Generated kernels match or outperform PyTorch eager execution.

Abstract

The performance of deep learning models critically depends on efficient kernel implementations, yet developing high-performance kernels for specialized accelerators remains time-consuming and expertise-intensive. While recent work demonstrates that large language models (LLMs) can generate correct and performant GPU kernels, kernel generation for neural processing units (NPUs) remains largely underexplored due to domain-specific programming models, limited public examples, and sparse documentation. Consequently, directly generating AscendC kernels with LLMs yields extremely low correctness, highlighting a substantial gap between GPU and NPU kernel generation. We present AscendCraft, a DSL-guided approach for automatic AscendC kernel generation. AscendCraft introduces a lightweight DSL that abstracts non-essential complexity while explicitly modeling Ascend-specific execution semantics. Kernels are first generated in the DSL using category-specific expert examples and then transcompiled into AscendC through structured, constraint-driven LLM lowering passes. Evaluated on MultiKernelBench across seven operator categories, AscendCraft achieves 98.1% compilation success and 90.4% functional correctness. Moreover, 46.2% of generated kernels match or exceed PyTorch eager execution performance, demonstrating that DSL-guided transcompilation can enable LLMs to generate both correct and competitive NPU kernels. Beyond benchmarks, AscendCraft further demonstrates its generality by successfully generating two correct kernels for newly proposed mHC architecture, achieving performance that substantially surpasses PyTorch eager execution.