Skip to content

Add hrr-replay-findings skill for read-only HRR replay triage#91

Draft
akshayv wants to merge 6 commits into
amd:mainfrom
akshayv:hrr-replay-findings
Draft

Add hrr-replay-findings skill for read-only HRR replay triage#91
akshayv wants to merge 6 commits into
amd:mainfrom
akshayv:hrr-replay-findings

Conversation

@akshayv

@akshayv akshayv commented Jul 15, 2026

Copy link
Copy Markdown
Contributor

Summary

Adds hrr-replay-findings — a read-only Agent Skill for triaging HIP Runtime Replay (HRR) captures and replay logs across diverse GPU workloads (vLLM, PyTorch finetune, hipBLASLt StreamK).

  • Classifies replay outcomes: read-only/OOB fault, illegal access, NaN/Inf D2H divergence, hang, OOM, pass
  • Extracts structured fields: fault address, failing event/call index, kernel name, kernarg address, D2H stats
  • Includes capture layout explainer (reference.md) and worked examples from MAF repro sessions
  • Ships with scripts/analyze_replay_finding.py (log + optional archive --info + multi-replay sweep TSV → JSON/Markdown)

Designed as the first, low-risk triage step before deeper crash analysis / patching workflows.

Skill layout

skills/hrr-replay-findings/
├── SKILL.md
├── reference.md
├── examples.md
├── skill-card.md
└── scripts/analyze_replay_finding.py

Test plan

  • Parser tested on real multi-GPU MAF replay logs (StreamK Cijk_*_SK3_*, d2h_fail=0)
  • Parser tested on multi-replay sweep TSV (6/6 MAF aggregation)
  • Install via npx skills add amd/skills --skill hrr-replay-findings after merge
  • Run against PASS replay log to verify replay_pass classification

Usage

python3 skills/hrr-replay-findings/scripts/analyze_replay_finding.py \
  --log replay.log \
  --archive capture.hrr/pid-NNN \
  --format markdown

Made with Cursor

Structured parser classifies replay faults (read-only/OOB, illegal access,
NaN/Inf divergence, hang, OOM) and extracts fault address, event index,
kernel, and kernarg from logs and archives across diverse GPU workloads.
@akshayv
akshayv force-pushed the hrr-replay-findings branch from 54aedd4 to 11435ff Compare July 15, 2026 15:34
AMD HRR Team added 5 commits July 15, 2026 15:38
Add run_hrr_replay.sh (native or Docker) with optional --analyze step;
update SKILL workflow to archive-first replay then structured finding.
Remove HRR_REPO_ROOT and internal lab script deps; auto-pick GPU by free
VRAM; default docker replay uses stock container ROCm without HIP_SO paths.
Replay runs hrr-playback on the host with auto GPU selection; no container
image or docker dependency in the skill or run_hrr_replay.sh.
Auto-find hrr-playback on PATH and /opt/rocm/bin; auto LD_LIBRARY_PATH;
ask user once only for archive or playback if discovery fails. No Docker,
no hardcoded lab paths, no run-this-script user instructions.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant