Repository navigation
Tags
Tags
v0.2.0: Optimize ffmpeg normalization pipeline and improve GPU offload Normalize engine: - Consolidate probe/duration/metadata into single ffprobe call (ProbeFile) - Split encode step so copy-mode fallback reuses measured loudness data instead of restarting the entire analysis pipeline - Coalesce progress callbacks (1% delta threshold) to reduce downstream churn - Always drain stdout pipe to prevent ffmpeg hangs when progress is unused Handler orchestration: - Throttle progress DB writes to 500ms intervals during encode phase - Batch job counter updates (every 5 items instead of every item) - SSE delta events: send only changed items between periodic full snapshots - Add active-file lock preventing concurrent normalization of same file - Deduplicate items by canonical path in StartNormalize - Carry per-item target_lufs forward on retry GPU acceleration: - Remove pointless -hwaccel flags from copy mode (video is stream-copied) - Fix VAAPI re-encode: use -hwaccel_output_format vaapi without conflicting format=nv12,hwupload so frames stay on GPU surface - Fix NVENC re-encode: add -hwaccel_output_format cuda for full GPU pipeline - Preserve all streams (-map 0) instead of dropping subs/attachments/extras Config and deployment: - Raise default parallel from 1 to 2 (HW caps still apply) - Smart entrypoint chown: skip recursive walk when /data ownership matches - Clearer docker-compose GPU profiles with documentation - Sonarr integration, handler improvements, settings enhancements