White-box debug a custom CUDA/CUTLASS kernel producing a WRONG numeric result (over-amplification, NaN, coherence) after black-box bisection is EXHAUSTED: every external lever varied (grid, pipeline depth, layout, dtype), the result deterministic + lever-independent, yet the structure reads correct. Two tracks: (A) an in-kernel operand + accumulator trace (if constexpr-gated to survive megakernel pipelining), (B) a standalone reproducer .cu instantiating the EXACT kernel template + TMA descriptors with controlled inputs (all-ones then real-dump) vs a host GEMM. Localizes the defect to operand-load vs MMA vs epilogue vs descriptor. Escalation tier above inference-kernel-ncu-profile when H/P are fine but the output is numerically wrong. Triggers on "white-box kernel", "kernel over-amplifies / NaN", "standalone reproducer", "in-kernel trace", "operand dump", "cutlass UMMA bug", or any combination of "white-box / reproducer / in-kernel / operand-trace" with "kernel / cutlass / UMMA / megakernel / vllm".
Install with
Claude CodeCodexCursorGitHub CopilotGemini CLI
Supported by the Atlas install prompt. Skill-specific requirements vary.
Skill stats
Skills1CategoryGeneralStars0
About this skill
White-box debug a custom CUDA/CUTLASS kernel producing a WRONG numeric result (over-amplification, NaN, coherence) after black-box bisection is EXHAUSTED: every external lever varied (grid, pipeline depth, layout, dtype), the result deterministic + lever-independent, yet the structure reads correct. Two tracks: (A) an in-kernel operand + accumulator trace (if constexpr-gated to survive megakernel pipelining), (B) a standalone reproducer .cu instantiating the EXACT kernel template + TMA descriptors with controlled inputs (all-ones then real-dump) vs a host GEMM. Localizes the defect to operand-load vs MMA vs epilogue vs descriptor. Escalation tier above inference-kernel-ncu-profile when H/P are fine but the output is numerically wrong. Triggers on "white-box kernel", "kernel over-amplifies / NaN", "standalone reproducer", "in-kernel trace", "operand dump", "cutlass UMMA bug", or any combination of "white-box / reproducer / in-kernel / operand-trace" with "kernel / cutlass / UMMA / megakernel / vllm".
Quick start
Copy the install prompt. Use the green button above this README.
Send it to your agent. Paste it into your coding agent’s chat and send.
Review the setup. Your agent checks the skill’s instructions, requirements, and installation scope.
White-box debug a custom CUDA/CUTLASS kernel producing a WRONG numeric result (over-amplification, NaN, coherence) after black-box bisection is EXHAUSTED: every external lever varied (grid, pipeline depth, layout, dtype), the result deterministic + lever-independent, yet the structure reads correct. Two tracks: (A) an in-kernel operand + accumulator trace (`if constexpr`-gated to survive megakernel pipelining), (B) a standalone reproducer .cu instantiating the EXACT kernel template + TMA descriptors with controlled inputs (all-ones then real-dump) vs a host GEMM. Localizes the defect to operand-load vs MMA vs epilogue vs descriptor. Escalation tier above `inference-kernel-ncu-profile` when H/P are fine but the output is numerically wrong. Triggers on "white-box kernel", "kernel over-amplifies / NaN", "standalone reproducer", "in-kernel trace", "operand dump", "cutlass UMMA bug", or any combination of "white-box / reproducer / in-kernel / operand-trace" with "kernel / cutlass / UMMA / megakernel / vllm".
About this listing
Category
General
Reported license
MIT
Listing language (estimated)
English
Review status
Not individually reviewed
Agent compatibility
Atlas installation supports Claude Code, Codex, and Cursor. Skill-specific dependencies and features are checked during setup.
Claude Code
Codex
Cursor
GitHub Copilot
Gemini CLI