(renamed) don't mangle shmem on shape change
At some point, maybe LPD will start sending different numbers of trains all the time. We should think about how to persist buffers rather than reallocating in this scenario.
-
Allow changing shape in the shmem buffer without immediately mangling previous train data -
Allow kernel runner to use partial buffers -
(Maybe out of scope) refactor kernel runner reconfiguration interaction - Have a huge diff stashed somewhere (this is a mess for AGIPD), will try to stick to LPD
-
Persist GPU buffers as long as new data is less than or equal to allocated number of frames
Edited by David Hammer