LLM spots useless 95GB-per-frame memcpy in ffmpeg dblur, cutting runtime from 2452ms to 49ms
IgorCarron · x · 2026-10-16
The ffmpeg "dblur" filter saga continues: Romain Hedouin used an LLM to spot a glaring inefficiency — the filter copied the same memory chunk once per pixel-width of the frame for no reason, moving 95GB of data per frame. After the fix, the human-written code took 2452ms while the human+LLM version runs in just 49ms, a 50x speedup. A strong showcase of LLMs hunting down low-level performance bugs rather than writing new features.
More from coding & agent
- Jev tutorial: 106x cheaper, 180ms yes/no gates replace PASS/FAIL string matching — max_paperclips · 2026-10-16
- Six-month prediction: hiring for generic agent harnesses built from scratch will cool — yi111 · 2026-10-16
- Omarchy ISO wouldn't boot on NVIDIA, so he had Claude patch it — indragie · 2026-10-16
- Dev uses Grok to clean up 3D printer firmware, catches it sneaking in nonsense code — yacineMTB · 2026-10-16
- A rude AI agent took 89% of a shared resource in 102-run multi-agent personality experiment — Brave-History-6502 · 2026-10-16
- Guardrails can make risky delegation work — but it takes a full-time team, says iandanforth — iandanforth · 2026-10-16