memory-debug

Diagnose data-corruption and memory-related failures in the tt-emule software emulator.

4|1|Updated Apr 1, 2026
One-click install
npx skills add https://github.com/tenstorrent/tt-emule --skill memory-debug-tenstorrent
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: memory-debug
Source: https://github.com/tenstorrent/tt-emule/tree/main/.claude/skills/memory-debug
Command: npx skills add https://github.com/tenstorrent/tt-emule --skill memory-debug-tenstorrent

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This skill provides a systematic diagnostic framework for identifying and fixing data-corruption failures in the tt-emule software emulator, specifically when kernel operations produce incorrect output values or memory-related regressions.

Core Features & Use Cases

  • Corruption Pattern Analysis: Tools to characterize output discrepancies, such as exact-zero regions versus random-wrong values, to isolate the fault site.
  • Layered Trace Methodology: A structured 5-step diagnostic process to trace data flow from host I/O through L1 buffers, compute kernels, and writer stages.
  • Use Case: When a regression test reports an ATOL failure or output corruption, use this skill to instrument specific pipeline hops and identify whether the bug originates in bank topology, address arithmetic, or synchronization logic.

Quick Start

Use the memory-debug skill to trace the dataflow pipeline and identify the root cause of the observed output corruption in the current kernel execution.

Frequently Asked Questions about memory-debug

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I debug data-corruption and memory failures in the tt-emule software emulator?

To debug data-corruption in the tt-emule emulator, use a structured diagnostic process to trace data flow from host I/O through L1 buffers to writer stages, isolating whether bugs originate in bank topology or address arithmetic.

How do I trace L1 memory access and NOC async operations during kernel execution?

Trace L1 memory access and NOC async operations by instrumenting specific pipeline hops using a layered trace methodology, targeting the precise execution traces of JIT-compiled kernels to identify output discrepancies.

Why does my kernel regression test report an ATOL failure with incorrect output values?

An ATOL failure indicates kernel operations produce incorrect output values. Characterize the output discrepancies, such as exact-zero regions versus random-wrong values, to isolate the fault site in the data flow pipeline.

What is the best way to analyze exact-zero regions versus random-wrong values in kernel output?

Analyzing exact-zero regions versus random-wrong values involves characterizing corruption patterns to isolate the fault site. This approach helps distinguish between address resolution errors and synchronization logic bugs.

Can I diagnose bank topology mapping and address arithmetic regressions in tt-emule?

Yes, you can diagnose bank topology mapping and address arithmetic regressions in tt-emule by targeting regression analysis for kernel operations and instrumenting L1 memory access to identify the root cause of data corruption.

When do I need to instrument JIT-compiled kernel execution traces for memory debugging?

You need to instrument JIT-compiled kernel execution traces when a regression test reports output corruption, requiring precise instrumentation of L1 memory access and NOC async operations to identify the fault site.