-
Notifications
You must be signed in to change notification settings - Fork 27.5k
[RFC] Back symm_mem.empty with an implicit MemPool #172050
Copy link
Copy link
Closed
Labels
module: cudaRelated to torch.cuda, and CUDA support in generalRelated to torch.cuda, and CUDA support in generalmodule: symm_memIssues and PRs of Symmetric MemoryIssues and PRs of Symmetric MemorytriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module
Milestone
Metadata
Metadata
Assignees
Labels
module: cudaRelated to torch.cuda, and CUDA support in generalRelated to torch.cuda, and CUDA support in generalmodule: symm_memIssues and PRs of Symmetric MemoryIssues and PRs of Symmetric MemorytriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module
🚀 The feature, motivation and pitch
So that it has a similar UX as
torch.empty.Could simplify auto-generated codes, like those by Inductor.
For eager codes, explicit MemPool would work better for
y = foo(x)where y is desired to be in symm mem.Alternatives
No response
Additional context
No response
cc @ptrblck @msaroufim @eqy @jerryzh168 @tinglvv @nWEIdia