Split out of #1131.
Placement and MemoryDomain exist and AllocationSpec carries them, but nothing chooses — a tensor's home is decided by whoever constructs it, not by a policy that can see the whole model.
Research question
Where should placement decisions come from: a per-model policy, the PlannerProfile, or something derived from the graph? Does this need a graph to reason about at all, or can it work over the eager construction order?
Depends on
Nothing blocking — the storage model (#932) landed the types this needs. Related to #1131's open question of whether P7 is wanted for the eager path at all, or only for the recorded path where IREE already does allocation planning.
Split out of #1131.
PlacementandMemoryDomainexist andAllocationSpeccarries them, but nothing chooses — a tensor's home is decided by whoever constructs it, not by a policy that can see the whole model.Research question
Where should placement decisions come from: a per-model policy, the
PlannerProfile, or something derived from the graph? Does this need a graph to reason about at all, or can it work over the eager construction order?Depends on
Nothing blocking — the storage model (#932) landed the types this needs. Related to #1131's open question of whether P7 is wanted for the eager path at all, or only for the recorded path where IREE already does allocation planning.