List view
Currently user has to employ their own memory fences/barriers to ensure the memory the thread receiving the ref is valid (or just ignore that and not care about undefined behaviour); loony only provides the basis for a memory safe MPMC queue to pass references around and makes no assurances about the memories validity. Being worked on (specifically with CPS in mind) is having push/pop have a moderate assurance that the memory being accessed by other cores is valid. Realistically, the current thinking is this is going to be done by either: - Copying/moving the data atomically on push/pop - Invalidating the cache lines the data is stored on in push/pop Implementations will be prototyped and benchmarked to decide upon the best approach.
No due date•3/3 issues closed