The wording was wrong and suggested that operands to memref.clone may not be mutated.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td | ||
---|---|---|
428–431 | This is just completely twisting and directly contradicting the statement above "Clones the data in the input view into an implicitly defined output view.". Is this op misnamed? clone_read_only_view? Even with that it doesn't explain disallowing mutations to the source. |
mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td | ||
---|---|---|
428–431 | Yes the naming is very misleading, I'm very supportive of a renaming (if you want to send a patch @bondhugula ?) |
mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td | ||
---|---|---|
428–431 | I just posted on the discord: https://llvm.discourse.group/t/reasoning-about-memref-mutability/3830/24?u=bondhugula The best I could think of based on my undertsanding is freeze_and_clone_read_only --- as in it freezes the source memref and makes a read-only clone. Pretty awkward but the semantics are as well! |
This is just completely twisting and directly contradicting the statement above "Clones the data in the input view into an implicitly defined output view.". Is this op misnamed? clone_read_only_view? Even with that it doesn't explain disallowing mutations to the source.