Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
mlir/include/mlir/Dialect/Shape/IR/ShapeOps.td | ||
---|---|---|
330 | Why this rename? I think "body" sounds better. | |
mlir/include/mlir/Dialect/Shape/Transforms/Passes.td | ||
15 | Can we find a better name for this pass? The goal is to reduce shape ops to other more primitive shape ops. That this can simplify converting to standard is a byproduct but not essential. Maybe "ShapeToShapeLowering? | |
mlir/lib/Dialect/Shape/Transforms/LegalizeShapeOpsForExport.cpp | ||
2 ↗ | (On Diff #268864) | nit: this looks like it got line wrapped. |
mlir/lib/Dialect/Shape/Transforms/PassDetail.h | ||
17 | Is AtomicRMWOp needed? |
mlir/include/mlir/Dialect/Shape/Transforms/Passes.h | ||
---|---|---|
24 | Could we expand here on what the pass does too? | |
mlir/lib/Dialect/Shape/Transforms/LegalizeShapeOpsForExport.cpp | ||
1 ↗ | (On Diff #268864) | Could we fit the next along with this line? |
28 ↗ | (On Diff #268864) | Prefer to keep anonymous namespaces small (https://llvm.org/docs/CodingStandards.html#anonymous-namespaces) |
mlir/include/mlir/Dialect/Shape/IR/ShapeOps.td | ||
---|---|---|
330 | This is to avoid confusion, since SingleBlockImplicitTerminator generates a getBody() method that return Block*, i.e. body().front() if using the old naming. scf.for and scf.parallel also call it just region. | |
mlir/include/mlir/Dialect/Shape/Transforms/Passes.td | ||
15 | Naming is hard. Thanks, I like ShapeToShapeLowering! | |
mlir/lib/Dialect/Shape/Transforms/PassDetail.h | ||
17 | well, yes, but not here :) |
Why this rename? I think "body" sounds better.