You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We start to realize that our heuristic ad-hoc dependency tracking is an improvement over the inconsistent explicit loops where one has no dependency abstraction but fails shorts of delivering a really good way of organizing the work. At the current stage the memory requirement scales with the number of threads p as O(p) which severely limits us on most machines. Additionally we don't know how much memory we really need for what.
By introducing a dependency graph we would have full knowledge of what to build, would know how much memory every node needs and could do optimal subgraph traversal for maximal re-use.
This issue shall contain the discussion and the accompanying Dependency Graph wiki page shall contain the design of that feature such that it can serve as documentation later on.
The text was updated successfully, but these errors were encountered:
We start to realize that our heuristic ad-hoc dependency tracking is an improvement over the inconsistent explicit loops where one has no dependency abstraction but fails shorts of delivering a really good way of organizing the work. At the current stage the memory requirement scales with the number of threads p as O(p) which severely limits us on most machines. Additionally we don't know how much memory we really need for what.
By introducing a dependency graph we would have full knowledge of what to build, would know how much memory every node needs and could do optimal subgraph traversal for maximal re-use.
This issue shall contain the discussion and the accompanying Dependency Graph wiki page shall contain the design of that feature such that it can serve as documentation later on.
The text was updated successfully, but these errors were encountered: