Rethinking Agentic Kernel Generation for Emerging Accelerators

2026-08-01

AI Summary

This work addresses the inefficiency in kernel generation for emerging accelerators, which stems from the absence of mature compiler backends and necessitates repeated construction of workload-agnostic machine semantics. To overcome this limitation, the authors propose Zomboss, a novel framework that, for the first time, encapsulates compiler-defined symbolic machine semantics and validity constraints into a reusable mapping interface. By integrating neural agents to efficiently explore scheduling strategies within a verified search space, Zomboss enables correct, comprehensive, and high-performance kernel synthesis. Evaluated on 56 workloads across Gemmini and PLENA accelerators, the approach consistently generates valid code, achieving average speedups of 3.34× on Gemmini and 1.10× on PLENA compared to the default compiler, while reducing inference overhead by over 50%.

Abstract

Emerging accelerators often lack mature compiler backends, motivating neural agents that generate and repair kernels from architectural documentation and simulator feedback. This approach repeatedly reconstructs workload-invariant machine semantics--including instruction behavior, legality constraints, synchronization rules, and memory protocols--for every workload. We argue that these semantics should be compiled once into a persistent symbolic artifact, while neural reasoning should focus on workload-dependent mapping decisions. We present Zomboss, a compiler-mediated agentic framework for kernel generation that places neural search within a verified compiler boundary. Zomboss compiles machine semantics and legality constraints into a reusable mapping interface, then uses a neural agent to optimize workload-dependent decisions within the validated mapping space. Across 20 Gemmini and 36 PLENA workload instances, Zomboss returns a correct verified kernel on all 56 instances. Relative to the compiler default, Zomboss achieves geometric-mean speedups of $3.34\times$ on Gemmini and $1.10\times$ on PLENA. Relative to direct agentic generation, it reduces inference tokens by 71.2% on Gemmini and 54.2% on PLENA. These results show that a compiler-defined symbolic interface turns native kernel synthesis into verified design-space exploration: compiler infrastructure preserves legality and correctness, while neural guidance improves workload-specific performance with lower search cost and complete coverage.

Problem

• emerging accelerators
• kernel generation
• machine semantics
• compiler backend
• neural agents

Innovation

• agentic kernel generation
• compiler-mediated framework
• symbolic interface
• verified design-space exploration
• workload-dependent mapping