The all-observation arm of the space-group search merges on a copy of the engine (CloneIngest), and
its IngestDevice staged every observation from the ~92-byte source reflections a second time -
concurrently with the main engine's own staging of the same reflections. Now the main engine
stages, and the copy takes its card arrays with cudaMemcpyPeer (a device copy when both engines
share the card, as on a one-card machine). CloneIngest no longer copies the predicted partiality,
which only that staging read.
Output identical (cytc p.hkl md5, and with --dump-observations the dump md5).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EizBKhTqrAYago9KJG3dA3
leonarski_fandClaude Opus 5.5
b46b4a4657