AgentKernelArena

AgentKernelArena is a standardized evaluation arena, built by AMD, that measures how well AI coding agents perform on real GPU kernel optimization tasks. It runs LLM-powered agents side-by-side on the same tasks in isolated workspaces and scores them with objective, reproducible metrics for compilation, correctness, and GPU performance.

• Source: AMD-AGI/AgentKernelArena

• License: MIT

Role in Hyperloom

AgentKernelArena is not part of Hyperloom’s default install or optimization loop. It is a companion evaluation harness for comparing Hyperloom’s kernel-optimization agents against other AI coding agents on standardized GPU kernel tasks. Its task categories (hip2hip, triton2triton, flydsl2flydsl, and others) overlap with the kernel types that Hyperloom’s kernel agent handles through GEAK, but Hyperloom does not clone or invoke AgentKernelArena in the normal runtime path.

AgentKernelArena documentation

For detailed documentation on AgentKernelArena, see AgentKernelArena on ROCm Docs.